TW202207241A - Systems and methods for artificial intelligence-based image analysis for detection and characterization of lesions - Google Patents

Abstract

Presented herein are systems and methods that provide for improved detection and characterization of lesions within a subject via automated analysis of nuclear medicine images, such as positron emission tomography (PET) and single photon emission computed tomography (SPECT) images. In particular, in certain embodiments, the approaches described herein leverage artificial intelligence (AI) to detect regions of 3D nuclear medicine images corresponding to hotspots that represent potential cancerous lesions in the subject. The machine learning modules may be used not only to detect presence and locations of such regions within an image, but also to segment the region corresponding to the lesion and/or classify such hotspots based on the likelihood that they are indicative of a true, underlying cancerous lesion. This AI-based lesion detection, segmentation, and classification can provide a basis for further characterization of lesions, overall tumor burden, and estimation of disease severity and risk.

Description

Artificial intelligence-based image analysis system and method for detection and characterization of lesions

The present invention generally relates to systems and methods for generating, analyzing and/or presenting medical imaging data. More particularly, in certain embodiments, the present invention relates to systems and methods for automated analysis of medical images used to identify and/or characterize cancerous lesions.

Nuclear medicine imaging involves the use of radiolabeled compounds (called radiopharmaceuticals). Radiopharmaceuticals are administered to a patient and accumulate in a manner that depends on, and is therefore indicative of, the biophysical and/or biochemical properties of tissues in various regions of the body, such as those affected by the presence and/or state of a disease such as cancer in various regions of the body. For example, certain radiopharmaceuticals accumulate in abnormal osteogenic areas associated with malignant bone lesions indicative of metastasis after administration to a patient. Other radiopharmaceuticals can bind to specific receptors, enzymes and proteins in the body that are altered during disease evolution. After administration to a patient, these molecules circulate in the blood until they find their intended target. The bound radiopharmaceutical remains at the disease site while the remainder of the agent is cleared from the body.

Nuclear medicine imaging techniques acquire images by detecting radiation emitted from the radioactive portion of a radiopharmaceutical. The use of the accumulated radiopharmaceuticals as beacons allows images depicting disease location and concentration to be obtained using common nuclear medicine modalities. Examples of nuclear medicine imaging modalities include bone scan imaging (also known as scintigraphy), single photon emission computed tomography (SPECT), and positron emission tomography (PET). Bone scan, SPECT and PET imaging systems are found in most hospitals around the world. The choice of a particular imaging modality depends on and/or dictates the particular radiopharmaceutical used. For example, Xun 99m ( ^99m Tc)-labeled compounds are compatible with bone scan imaging and SPECT imaging, while PET imaging typically uses 18F-labeled fluorinated compounds. The compound99mTc ^{methylenediphosphonate (99mTc MDP} ) is a popular radiopharmaceutical for bone scan imaging to detect metastatic cancer. Radiolabeled prostate-specific membrane antigen (PSMA) targeting compounds such ^as99mTc -labeled 1404 and PyL ^™ (also known as [18F]DCFPyL) can be used with SPECT and PET imaging, respectively, and provide high specificity Potential for prostate cancer detection.

Thus, nuclear medicine imaging is a valuable technique for providing physicians with information that can be used to determine the presence and extent of disease in a patient. Physicians can use this information to advise patients on the course of treatment and to track disease progression.

For example, an oncologist may use nuclear medicine images from a patient's study as her assessment of whether a patient has a particular disease (eg, prostate cancer), what stage of the disease is apparent, why the recommended course of treatment (if any), whether it is indicated Input of surgical intervention and possible prognosis. The oncologist can use the radiologist report in this evaluation. A radiologist's report is a technical evaluation of nuclear medicine images prepared by a radiologist for a physician requesting an imaging study and includes, for example, the type of study performed, clinical history, comparisons between images, data used to perform the study. technique, the radiologist's observations and findings, and the overall impression and recommendation that the radiologist may have based on the results of the imaging study. The signed radiologist's report is sent to the physician who arranged the study for review by the physician, followed by a discussion between the physician and the patient regarding treatment results and recommendations.

Thus, the procedure involves having the radiologist perform an imaging study on the patient, analyzing the images obtained, creating a radiologist report, forwarding the report to the requesting physician, having the physician make evaluation and treatment recommendations, and having the physician communicate the results to the patient , Recommendations and Risks. Procedures may also involve repeating imaging studies due to inconclusive results, or arranging further testing based on initial results. If imaging studies show that the patient has a particular disease or condition (eg, cancer), the physician discusses various treatment options, including surgery, and doing nothing or taking watchful waiting or active surveillance approaches rather than risking surgery.

Thus, the process of viewing and analyzing multiple patient images over time plays a critical role in the diagnosis and treatment of cancer. There is a significant need for improved tools to facilitate and improve the accuracy of image viewing and analysis for cancer diagnosis and treatment. In this way, the toolbox utilized by physicians, radiologists, and other health care professionals is improved for significantly improving standards of care and the patient experience.

It is proposed herein to provide improved detection and characterization of lesions within a subject through automated analysis of nuclear medicine images, such as positron emission tomography (PET) and single photon emission computed tomography (SPECT) images system and method. In particular, in certain embodiments, the methods described herein utilize artificial intelligence (AI) techniques to detect regions of 3D nuclear medicine images corresponding to hot spots representing potential cancerous lesions in the subject. In certain embodiments, these regions correspond to localized regions (hot spots) of increased intensity relative to their surroundings due to increased uptake of the radiopharmaceutical within the lesion. The systems and methods described herein may employ one or more machine learning modules that are used not only to detect the presence and location of such hotspots within an image, but also to segment the corresponding The area of hotspots and/or the hotspots are classified based on the likelihood that the hotspots do correspond to real, potentially cancerous lesions. These AI-based lesion detection, segmentation and classification methods can provide a basis for further characterization of lesions, overall tumor burden, and estimation of disease severity and risk.

For example, once image hot spots representing lesions are detected, segmented, and classified, lesion index values can be computed to provide a measure of radiopharmaceutical uptake within the underlying lesion and/or the size (eg, volume) of the underlying lesion. The computed lesion index values can then be aggregated to provide an overall estimate of the subject's tumor burden, disease severity, risk of metastasis, and the like. In some embodiments, the lesion index value is calculated by comparing the intensity within the segmented hotspot volume with a measure of the intensity of a particular reference organ, such as the liver and aortic portions. Using a reference organ in this way allows measuring lesion index values on a normalized scale that can be compared between images of different subjects. In certain embodiments, the methods described herein include methods for inhibiting radiation from regions corresponding to organs and tissues in which radiopharmaceuticals normally accumulate at high levels, such as the kidneys, liver, and bladder (eg, urinary bladder) The technique of intensity bleed of multiple image regions. Intensities in the regions of nuclear medicine images corresponding to these organs are often high (even for normal, healthy subjects) and are not necessarily indicative of cancer. In addition, high levels of radiopharmaceutical accumulation in these organs result in high levels of emitted radiation. The increased emitted radiation can scatter, resulting not only in high intensities within the region corresponding to the nuclear medicine image of the organ itself, but also in nearby external voxels. This intensity exudation into regions of the image corresponding to regions outside and around organs associated with high uptake can hinder detection of nearby lesions and cause inaccuracies in measuring uptake therein. Therefore, correcting for these intensity exudation effects improves the accuracy of lesion detection and quantification.

In certain embodiments, the AI-based lesion detection techniques described herein augment data obtained from nuclear medicine images with anatomical information obtained from anatomical images, such as x-ray computed tomography (CT) images. Feature information. For example, a machine learning module utilized in the methods described herein may receive multiple input channels, including a first channel corresponding to a portion of functional, nuclear medicine, images (eg, PET images; eg, SPECT images), and Additional channels corresponding to portions of co-aligned anatomical (eg, CT) images and/or anatomical information derived therefrom. Adding anatomical context in this way can improve the accuracy of the lesion detection method. Anatomical information can also be incorporated into lesion classification methods applied after detection. For example, in addition to computing lesion index values based on the intensity of the detected hotspots, anatomical markers may also be assigned to hotspots based on their location. For example, a detected hot spot can be automatically assigned a marker (eg, an alphanumeric marker based on whether the location of the detected hot spot corresponds to the prostate, pelvic lymph nodes, non-pelvic lymph nodes, bone, or locations within the soft tissue region outside the prostate and lymph nodes). ).

In certain embodiments, detected hotspots and associated information (such as computed lesion index values and anatomical markers) are displayed with an interactive graphical user interface (GUI) to allow medical professionals (such as physicians, radiologists, physicians, technicians, etc.) to check. Thus, a medical professional can use the GUI to view and confirm the accuracy of detected hot spots and corresponding index values and/or anatomical landmarks. In some embodiments, the GUI may also allow the user to identify and segment (eg, manually) additional hot spots within the medical image, thereby allowing the medical professional to identify additional potential lesions that he/she believes may be missed by the automated detection process. Once identified, lesion index values and/or anatomical landmarks may also be determined for these manually identified and segmented lesions. Once users are satisfied with the collection of detected hotspots and information computed from them, they can confirm their approval and generate a final, signed report that can be viewed, for example, and used for Discuss the results and diagnosis with the patient, and evaluate prognosis and treatment options.

In this way, the methods described herein provide AI-based tools for lesion detection and analysis that can improve the accuracy and simplify the status and progression of a subject's disease (eg, cancer) its evaluation. This facilitates diagnosis, prognosis, and assessment of response to treatment, thereby improving patient outcomes.

In one aspect, the present invention relates to a method for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the method comprising: (a) receiving (eg, and/or accessing) by the processor of the computing device using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography ( SPECT)] of the subject's 3D functional image [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having a channel representing an emission from the specific physical volume detecting an intensity value of radiation (eg, a standard uptake value (SUV)), wherein at least a portion of the plurality of voxels of the 3D functional image represents a physical volume within the target tissue region]; (b) using a machine by the processor A learning module [eg, a pre-trained machine learning module (eg, with predetermined (eg, and fixed) parameters that has been determined by a training program)] automatically detects one or more hot spots within the 3D functional image, each hot spot Corresponds to a local area of increased intensity relative to its surroundings and represents (eg, indicates) a potential cancerous lesion within the subject, thereby establishing one or both of the following (i) and (ii): (i) a list of hot spots [eg, a list of coordinates (eg, image coordinates; eg, physical space coordinates); eg, a mask identifying the voxels of the 3D-enabled image corresponding to the locations (eg, centroids) of detected hotspots], which For each hotspot, the location of the hotspot is identified, and (ii) a 3D heatmap that identifies, for each hotspot, the corresponding 3D hotspot volume within the 3D functional image {eg, wherein the 3D heatmap is a segmentation map (eg, includes one or more a segmentation mask) that identifies, for each hotspot, voxels within the 3D functional image corresponding to the 3D hotspot volume for each hotspot [eg, where the 3D heatmap is created through artificial intelligence-based segmentation of the functional image] obtained (eg, using a machine learning module that receives at least the 3D functional image as input and generates the 3D heatmap as output to segment hotspots)]; eg, wherein the 3D heatmap depicts the 3D boundary of the hotspot for each hotspot (eg, , an irregular boundary) (eg, the 3D boundary encloses the 3D hotspot volume, eg, and distinguishes the voxels of the 3D functional image that make up the 3D hotspot volume from other voxels of the 3D functional image)}; and (c) storing and/or providing the hotspot list and/or the 3D heatmap for display and/or further processing.

In some embodiments, the machine learning module receives at least a portion of the 3D-enabled image as input and automatically detects the one or more hot spots based at least in part on intensities of voxels in the received portion of the 3D-enabled image. In certain embodiments, the machine learning module receives as input a 3D segmentation map that identifies one or more volumes of interest (VOIs) within the 3D functional image, each VOI corresponding to a volume of interest within the subject Specific target tissue regions and/or specific anatomical regions [eg, soft tissue regions (eg, prostate, lymph nodes, lung, breast); eg, one or more specific bones; eg, entire skeletal region].

In certain embodiments, the method includes receiving (eg, and/or accessing), by the processor, using an anatomical imaging modality [eg, x-ray computed tomography (CT); eg, magnetic resonance imaging ( MRI); e.g., ultrasound] obtained 3D anatomical image of the subject, wherein the 3D anatomical image includes a graphical representation of tissue (e.g., soft tissue and/or bone) within the subject, and the machine learning module receiving at least two input channels, the input channels including a first input channel corresponding to at least a portion of the 3D anatomical image and a second input channel corresponding to at least a portion of the 3D functional image [eg, wherein the machine learning module receives PET Image and CT image as separate channels (eg, separate channels representing the same volume) (eg, similar to receiving two color channels (RGB) of photographic color images by a machine learning module)].

In some embodiments, the machine learning module receives as input a 3D segmentation map that identifies one or more volumes of interest (VOIs) within the 3D functional image and/or the 3D anatomical image, each VOI corresponding to in specific target tissue areas and/or specific anatomical areas. In some embodiments, the method includes automatically segmenting, by the processor, the 3D anatomical image, thereby creating the 3D segmentation map.

In certain embodiments, the machine learning module is a region-specific machine learning module that receives as input specific portions of the 3D functional image corresponding to one or more specific tissue regions and/or anatomical regions of the subject .

In some embodiments, the machine learning module generates the hotspot list as an output [eg, wherein the machine learning module implements training to determine one or more locations based on intensities of at least a portion of the voxels of the 3D functional image ( For example, a machine learning algorithm (eg, an artificial neural network (ANN)) for 3D coordinates, each location corresponds to the location of one of the one or more hotspots].

In some embodiments, the machine learning module generates the 3D heat map as output [eg, wherein the machine learning module implements training to segment the 3D functional image (eg, based at least in part on voxels of the 3D functional image) intensity) to identify the 3D heat map (e.g., the 3D heat map delineates the 3D boundaries (e.g., irregular boundaries) of the hotspot for each hotspot), thereby identifying the 3D hotspot volumes (e.g., by the 3D hotspot boundaries) A machine learning algorithm (eg, an artificial neural network (ANN)) of the 3D hotspot volumes that encloses)); for example, wherein the machine learning module implementation is trained for each voxel of at least a portion of the 3D functional image A machine learning algorithm that determines a hotspot likelihood value representing the likelihood that the voxel corresponds to a hotspot (eg, and step (b) includes performing one or more subsequent post-processing steps, such as qualifying, to use the hotspot likelihood The property value identifies the 3D hotspot volumes of the 3D heatmap (eg, the 3D heatmap delineates the 3D boundaries (eg, irregular boundaries) of the hotspot for each hotspot, thereby identifying the 3D hotspot volumes (eg, by the etc. 3D hotspot boundary enclosure )))].

In certain embodiments, the method includes: (d) determining, by the processor for each of at least a portion of the hotspots, a lesion likelihood classification corresponding to the likelihood that the hotspot represents a lesion within the subject [eg, , a binary classification indicating whether the hot spot is a true lesion; eg, a likelihood value on a scale (eg, a floating point value in the range zero to one) representing the likelihood that the hot spot represents a true lesion].

In certain embodiments, step (d) includes using a second machine learning module to determine the lesion likelihood classification for each hotspot of the portion [eg, wherein the machine learning module implements training to detect hotspots (eg, A hotspot list and/or 3D heatmap is generated as output) and a machine learning algorithm for each hotspot to determine the lesion likelihood classification for that hotspot]. In some embodiments, step (d) includes using a second machine learning module (eg, a hotspot classification module) to determine a lesion likelihood classification for each hotspot [eg, based at least in part on an intensity selected from the 3D functional image, one or more members of the group consisting of the hotspot list, the 3D heatmap, the intensity of the 3D anatomical image, and the 3D segmentation map; for example, wherein the second machine learning module receives the intensity, hotspot list, One or more members of the group consisting of the 3D heat map, the intensity of the 3D anatomical image, and the 3D segmentation map, one or more input channels].

In certain embodiments, the method includes determining, by the processor for each hotspot, a set of one or more hotspot features and using the set of the one or more hotspot features as input to the second machine learning module .

In certain embodiments, the method includes: (e) selecting, by the processor, the one or the other corresponding to hotspots having a high probability of corresponding to cancerous lesions based at least in part on the lesion likelihood classifications of the hotspots A subset of multiple hotspots (eg, for inclusion in a report; eg, for computing one or more risk index values for a subject).

In certain embodiments, the method includes: (f) [eg, prior to step (b)] adjusting, by the processor, intensities of voxels of the 3D functional image to correct one or more from the 3D functional image Intensity exudation (e.g., crosstalk) of high intensity volumes, each of the one or more high intensity volumes corresponding to high radiopharmaceutical uptake in the subject associated with normal conditions (e.g., not necessarily indicative of cancer) high uptake tissue areas within. In some embodiments, step (f) includes correcting the intensity bleeds from the plurality of high-intensity volumes in a sequential manner one at a time [eg, first adjusting the intensities of the voxels of the 3D functional image to correct for the intensities from the first high-intensity volume Intensity bleed to generate a first corrected image, then the intensities of voxels of the first corrected image are adjusted to correct for intensity bleed from a second high-intensity volume, etc.]. In certain embodiments, the one or more high-intensity volumes correspond to one or more high-uptake tissue regions selected from the group consisting of kidney, liver, and bladder (eg, urinary bladder).

In certain embodiments, the method includes: (g) determining, by the processor, for each of at least a portion of the one or more hotspots, a level indicative of radiopharmaceutical uptake within the underlying lesion corresponding to the hotspot and/or Or the corresponding lesion index for the size (eg, volume) of the underlying lesion. In certain embodiments, step (g) includes comparing (eg, at and/or near the location of the hotspot; eg, within the volume of the hotspot) one or more voxels (several ) intensity (eg, corresponding to a standard uptake value (SUV)) and one or more reference values, each reference value being associated with a particular reference tissue region (eg, liver; eg, aortic portion) within a subject and based on The intensity (eg, SUV value) of the reference volume corresponding to the reference tissue region is determined [eg, as a mean (eg, a robust mean, such as the mean of values within an interquartile range)]. In certain embodiments, the one or more reference values comprise selected from the group consisting of an aortic reference value associated with an aortic portion of the subject and a liver reference value associated with the subject's liver One or more members of the group.

In certain embodiments, for at least one particular reference value associated with a particular reference tissue region, determining the particular reference value comprises fitting intensities of voxels within a particular reference volume corresponding to the particular reference tissue region [eg, , fitting a distribution of voxel intensities (e.g., fitting a histogram of voxel intensities)] to a multicomponent mixture model (e.g., a two-component Gaussian model) [e.g., and identifying one of the distributions of voxel intensities or A plurality of secondary peaks that correspond to voxels associated with abnormal uptake and that are excluded from reference value determinations (eg, to be considered in a particular portion of a reference tissue region such as a portion of the liver) the effect of abnormally low radiopharmaceutical uptake)].

In certain embodiments, the method includes computing (eg, automatically by the processor) a subject's overall risk index that indicates the subject's cancer status and/or risk using the determined lesion index values.

In certain embodiments, the method includes determining, by the processor, for each hot spot (eg, automatically) corresponding to a potential cancerous lesion represented by the hot spot [eg, by the processor (eg, based on received and 3D segmentation map)] for localization [eg, within the prostate, pelvic lymph nodes, non-pelvic lymph nodes, bone (eg, areas of bone metastases), and soft tissue regions not located in the prostate or lymph nodes] within subjects An anatomical classification of a specific anatomical region and/or group of anatomical regions.

In some embodiments, the method includes: (h) causing, by the processor, at least a portion of a graphical representation of the one or more hot spots to be displayed within a graphical user interface (GUI) for viewing by a user. In certain embodiments, the method includes: (i) receiving, by the processor, via the GUI, children of the one or more hotspots identified by the user for viewing as likely to represent a potential cancerous lesion in the subject User selection of the set.

In certain embodiments, the 3D functional image comprises a PET or SPECT image obtained after administration of an agent (eg, a radiopharmaceutical; eg, an imaging agent) to the subject. In certain embodiments, the agent comprises a PSMA binding agent. In certain embodiments, ^the agent includes18F. In certain embodiments, the agent comprises [18F]DCFPyL. In certain embodiments, the agent includes PSMA-11 (eg, ^68Ga -PSMA-11). In certain embodiments, the reagent comprises one or more members selected from the group consisting ^of99mTc , ^68Ga , ^177Lu ,225Ac, ^111In , ^123I , ^{124I , and131I} .

In certain embodiments, the machine learning module implements a neural network [eg, artificial neural network (ANN); eg, convolutional neural network (CNN)].

In some embodiments, the processor is a processor of a cloud-based system.

In another aspect, the invention relates to a method for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the method comprising: (a) receiving (eg, and/or accessing) by the processor of the computing device using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography ( SPECT)] of the subject's 3D functional image [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having a channel representing an emission from the specific physical volume detecting intensity values of radiation, wherein at least a portion of the plurality of voxels of the 3D functional image represents a physical volume within the target tissue region]; (b) receiving (eg, and/or accessing) by the processor using anatomy A 3D anatomical image of the subject obtained by an imaging modality [eg, x-ray computed tomography (CT); eg, magnetic resonance imaging (MRI); eg, ultrasound], wherein the 3D anatomical image includes the subject A graphical representation of tissue (eg, soft tissue and/or bone) within the image; (c) using a machine learning module by the processor to automatically detect one or more hot spots within the 3D functional image, each hot spot corresponding to a relative A local area of increased intensity around it and representing (eg, indicative of) a potentially cancerous lesion within the subject, thereby establishing one or both of the following (i) and (ii): (i) a list of hot spots for Each hotspot identifies the location of the hotspot, and (ii) a 3D heatmap that, for each hotspot, identifies the corresponding 3D hotspot volume within the 3D functional image {eg, where the 3D heatmap is a segmentation map (eg, includes one or more segmentation mask) that identifies, for each hotspot, the voxels within the 3D functional image corresponding to the 3D hotspot volume for each hotspot [eg, where the 3D heatmap is obtained via artificial intelligence-based segmentation of the functional image] (eg, using a machine learning module that receives at least the 3D functional image as input and generates the 3D heatmap as output to segment hotspots)]; eg, where the 3D heatmap depicts the 3D boundary of the hotspot for each hotspot (eg, irregular boundary) (eg, the 3D boundary encloses the 3D hotspot volume, eg, and distinguishes the voxels of the 3D functional image that make up the 3D hotspot volume from other voxels of the 3D functional image)}, where the The machine learning module receives at least two input channels, the input channels include a first input channel corresponding to at least a portion of the 3D anatomical image and a second input channel corresponding to at least a portion of the 3D functional image [eg, wherein the machine learning The module receives PET images and CT images as separate channels (eg, separate channels representing the same volume) (eg, similar to receiving two color channels (RGB) of photographic color images by a machine learning module)] and/or from Its derived anatomical information [eg, a 3D segmentation map that identifies one or more volumes of interest within the 3D functional image (V OI), each VOI corresponds to a specific target tissue region and/or a specific anatomical region]; and (d) storing and/or providing the hotspot list and/or the 3D heatmap for display and/or further processing.

In another aspect, the invention relates to a method for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the method comprising: (a) receiving (eg, and/or accessing) by the processor of the computing device using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography ( SPECT)] of the subject's 3D functional image [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having a channel representing an emission from the specific physical volume Detecting intensity values of radiation, wherein at least part of the plurality of voxels of the 3D functional image represents a physical volume within the target tissue region]; (b) automatically detecting the 3D by the processor using a first machine learning module One or more hotspots within the functional image, each hotspot corresponding to a local area of increased intensity relative to its surroundings and representing (eg, indicative of) a potential cancerous lesion within the subject, thereby establishing a method for identifying the hotspot for each hotspot. A list of hotspots of locations [eg, where the machine learning module implements a machine learning algorithm (eg, artificial Neural Network (ANN)), each location corresponding to the location of one of the one or more hotspots]; (c) using a second machine learning module and the hotspot list by the processor for the one or more hotspots Each of the hotspots automatically determines the corresponding 3D hotspot volume within the 3D-enabled image, thereby creating a 3D heatmap [eg, wherein the second machine learning module implementation is trained to be based at least in part on the hotspot list and the volume of the 3D-enabled image A machine learning algorithm (eg, an artificial neural network (ANN)) that segments the 3D functional image to identify the 3D hotspot volumes of the 3D heatmap; for example, wherein the machine learning module implementation is trained for Each voxel of at least a portion of the 3D functional image determines a machine learning algorithm for a hotspot likelihood value representing the likelihood that the voxel corresponds to a hotspot (eg, and step (b) includes performing one or more subsequent steps such as defining post-processing step to identify the 3D hotspot volumes of the 3D heatmap using the hotspot likelihood values] [eg, wherein the 3D heatmap is performed using (eg, based on and/or corresponding to from) a second machine learning model A set of (the output of) generated segmentation maps (e.g., including one or more segmentation masks) that identify, for each hotspot, voxels within the 3D functional image of the 3D hotspot volume corresponding to each hotspot; e.g. , where the 3D heat map depicts, for each hotspot, the 3D boundary (eg, an irregular boundary) of the hotspot (eg, the 3D boundary enclosing the 3D hotspot volume, eg, and the 3D functional image of the 3D hotspot volume that constitutes the 3D hotspot volume). voxels are distinguished from other voxels of the 3D functional image)]; and (d) storing and/or providing the hotspot list and/or the 3D heatmap for display and/or or further processing.

In certain embodiments, the method includes: (e) determining, by the processor, for each of at least a portion of the hotspots, a lesion likelihood classification corresponding to the likelihood that the hotspot represents a lesion in the subject. In certain embodiments, step (e) includes using a third machine learning module (eg, a hotspot classification module) to determine the lesion likelihood classification for each hotspot [eg, based at least in part on an intensity selected from a 3D functional image] , a list of hot spots, a 3D heat map, one or more members of the group consisting of the intensity of the 3D anatomical image and the 3D segmentation map; List, 3D heat map, intensity of 3D anatomical image, and one or more members of the group consisting of the 3D segmentation map, one or more input channels].

In certain embodiments, the method includes: (f) selecting, by the processor, the one or the other corresponding to hotspots having a high probability of corresponding to cancerous lesions based at least in part on the lesion likelihood classifications of the hotspots A subset of multiple hotspots (eg, for inclusion in a report; eg, for computing one or more risk index values for a subject).

In another aspect, the invention pertains to a measure of intensity values within a reference volume (eg, a liver volume associated with a subject's liver) corresponding to a reference tissue region in order to avoid errors from low (eg, abnormal) A method for the effect of a tissue region associated with low) radiopharmaceutical uptake (eg, due to a tumor without tracer uptake), the method comprising: (a) receiving by a processor of a computing device (eg, and/or access) a 3D functional image of the subject obtained using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] [ For example, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having an intensity value representing detected radiation emitted from the specific physical volume, wherein the 3D functional image at least a portion of the plurality of voxels represent a solid volume within the target tissue region]; (b) identify, by the processor, the reference volume within the 3D functional image; (c) combine, by the processor, a multicomponent mixture model ( For example, a two-component Gaussian mixture model is fitted to the intensities of voxels within the reference volume [eg, the multi-component mixture model is fitted to a distribution (eg, a histogram) of the intensities of voxels within the reference volume )]; (d) identifying, by the processor, the dominant mode of the multicomponent model; (e) determining, by the processor, the (e.g., mean, maximum, mode) corresponding to the dominant mode , median, etc.) measure of intensity to determine a reference intensity value corresponding to a measure of intensity for voxels that are (i) within the reference tissue volume and (ii) associated with the dominant mode (e.g., and exclude voxels with intensities associated with minor patterns from reference value calculations (eg, to avoid effects from tissue regions associated with low radiopharmaceutical uptake); (f) by the processor in the function Intra-image detection of one or more hot spots corresponding to potentially cancerous lesions; and (g) determining, by the processor, a lesion index value using at least the reference intensity value for each of at least a portion of the detected hot spots [eg, , the lesion index value is based on (i) a measure of the intensity of the voxel corresponding to the detected hot spot and (ii) the reference intensity value].

In another aspect, the invention pertains to a correction for intensities derived from regions of high uptake tissue within a subject associated with high radiopharmaceutical uptake under normal conditions (eg, and not necessarily indicative of cancer) A method of exuding (eg, crosstalk), the method comprising: (a) receiving (eg, and/or accessing), by a processor of a computing device, a 3D functional image of a subject, the 3D functional image using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] obtains [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing the subject within a specific physical volume and having an intensity value representing the detected radiation emitted from the specific physical volume, wherein at least a portion of the plurality of voxels of the 3D functional image represent the physical volume within the target tissue region]; (b) by A high-intensity volume within the 3D functional image is identified by the processor, the high-intensity volume corresponding to a particular high-uptake tissue region (eg, kidney; eg, liver; eg, bladder) in which high radiopharmaceutical uptake normally occurs (c) identifying, by the processor, a suppression volume within the 3D functional image based on the identified high-intensity volume, the suppression volume corresponding to outside the boundaries of the identified high-intensity volume and at a distance from the identified high-intensity volume a volume within a predetermined attenuation distance of the boundary of the high-intensity volume; (d) determining, by the processor, a background image corresponding to the 3D functional image, wherein using a voxel based on the 3D functional image within the suppression volume (e) by the processor by subtracting the intensities of the voxels of the background image from the intensities of the voxels from the 3D functional image ( For example, performing voxel-wise subtraction) to determine an estimated image; (f) determining, by the processor, a suppression map by extrapolating to the suppression the intensities of voxels of the estimated image corresponding to the high-intensity volume the position of the voxels within the volume to determine the intensity of the voxels of the suppression map corresponding to the suppression volume; and setting the intensity of the voxels of the suppression map corresponding to the position outside the suppression volume to zero; and ( g) adjusting, by the processor, based on the suppression map, the intensities of voxels of the 3D functional image (eg, by subtracting the intensities of voxels of the suppression map from the intensities of voxels of the 3D functional images), thereby Correction for intensity bleed from the high intensity volume.

In certain embodiments, the method includes performing steps (b) through (g) for each of the plurality of high-intensity volumes in a sequential manner, thereby correcting for intensity bleed from each of the plurality of high-intensity volumes.

In certain embodiments, the plurality of high-intensity volumes comprise one or more members selected from the group consisting of kidney, liver, and bladder (eg, urinary bladder).

In another aspect, the invention relates to a method for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the method comprising: (a) receiving (eg, and/or accessing) by the processor of the computing device using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography ( SPECT)] of the subject's 3D functional image [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having a channel representing an emission from the specific physical volume Detecting intensity values of radiation, wherein at least part of the plurality of voxels of the 3D functional image represents a physical volume within the target tissue region]; (b) automatically detecting one or more of the 3D functional images by the processor a hotspot, each hotspot corresponding to a local area of increased intensity relative to its surroundings and representing (eg, indicative of) a potential cancerous lesion within the subject; (c) causing, by the processor, to list the one or more a graphical representation of hotspots for display within an interactive graphical user interface (GUI) (eg, a quality control and reporting GUI); (d) received by the processor via the interactive GUI including at least a portion (at most all) ) user selection of the final set of hotspots of the one or more auto-detected hotspots (eg, for inclusion in reports); and (e) storing and/or providing the final set of hotspots for display and/or or further processing.

In certain embodiments, the method includes: (f) receiving, by the processor, via the GUI, user selections of one or more additional, user-identified hotspots for inclusion in the final set of hotspots; and ( g) updating, by the processor, the final set of hotspots to include the one or more additional user-identified hotspots.

In certain embodiments, step (b) includes using one or more machine learning modules.

In another aspect, the invention relates to a method for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the method comprising: (a) receiving (eg, and/or accessing) by the processor of the computing device using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography ( SPECT)] of the subject's 3D functional image [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having a channel representing an emission from the specific physical volume Detecting intensity values of radiation, wherein at least part of the plurality of voxels of the 3D functional image represents a physical volume within the target tissue region]; (b) automatically detecting one or more of the 3D functional images by the processor (c) targeting, by the processor, at least a portion of the one or more Each of the hotspots is automatically determined to correspond to a potential cancerous lesion in which the hotspot represents is determined [eg, by the processor (eg, based on the received and/or determined 3D segmentation map)] as localized [eg, in the prostate, pelvis, Anatomical classification of specific anatomical regions and/or groups of anatomical regions within a subject of lymph nodes, non-pelvic lymph nodes, bone (eg, areas of bone metastases), and soft tissue regions not located in the prostate or lymph nodes]; and (d ) stores and/or provides the identification of the one or more hotspots and, for each hotspot, an anatomical classification corresponding to the hotspot for display and/or further processing.

In certain embodiments, step (b) includes using one or more machine learning modules.

In another aspect, the invention relates to a system for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the system comprising: A processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive (eg, and/or access) use functions A 3D functional image of the subject obtained by an imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having an intensity value (eg, a standard uptake value (SUV)) representing the detected radiation emitted from the specific physical volume, wherein the 3D functional image at least a portion of the plurality of voxels represent a physical volume within the target tissue region]; (b) using a machine learning module [e.g., a pre-trained machine learning module (e.g., with a predetermined value that has been determined by a training program (e.g., and fixed) parameters)] automatically detects one or more hotspots within the 3D functional image, each hotspot corresponding to a local area of increased intensity relative to its surroundings and representing (eg, indicative of) a potential cancerousness within the subject lesions, thereby creating one or both of the following (i) and (ii): (i) a list of hot spots [eg, a list of coordinates (eg, image coordinates; eg, physical space coordinates); eg, identifying the 3D functional image A mask of voxels, each voxel corresponding to the location of a detected hotspot (eg, centroid)], which identifies the location of the hotspot for each hotspot, and (ii) a 3D heat map, which identifies the 3D hotspot for each hotspot The corresponding 3D hotspot volume within the functional image {eg, where the 3D heatmap is a segmentation map (eg, including one or more segmentation masks) that identifies, for each hotspot, the 3D feature corresponding to the 3D hotspot volume for each hotspot Voxels within an image [eg, where the 3D heat map is obtained via artificial intelligence-based segmentation of the functional image (eg, segmenting hot spots using receiving at least the 3D functional image as input and generating the 3D heat map as output) machine learning module)]; for example, wherein the 3D heat map depicts, for each hotspot, a 3D boundary (eg, an irregular boundary) of the hotspot (eg, the 3D boundary encloses the 3D hotspot volume, eg, and the 3D distinguish the voxels of the functional image that make up the 3D hotspot volume from other voxels of the 3D functional image)}; and (c) store and/or provide the hotspot list and/or the 3D heatmap for display and/or further processing.

In some embodiments, the machine learning module receives at least a portion of the 3D-enabled image as input and automatically detects the one or more hot spots based at least in part on intensities of voxels in the received portion of the 3D-enabled image.

In certain embodiments, the machine learning module receives as input a 3D segmentation map that identifies one or more volumes of interest (VOIs) within the 3D functional image, each VOI corresponding to a volume of interest within the subject Specific target tissue regions and/or specific anatomical regions [eg, soft tissue regions (eg, prostate, lymph nodes, lung, breast); eg, one or more specific bones; eg, entire skeletal region].

In certain embodiments, the instructions cause the processor to: receive (eg, and/or access) the use of an anatomical imaging modality [eg, x-ray computed tomography (CT); eg, magnetic resonance imaging ( MRI); e.g., ultrasound] obtained 3D anatomical image of the subject, wherein the 3D anatomical image includes a graphical representation of tissue (e.g., soft tissue and/or bone) within the subject, and the machine learning module receiving at least two input channels, the input channels including a first input channel corresponding to at least a portion of the 3D anatomical image and a second input channel corresponding to at least a portion of the 3D functional image [eg, wherein the machine learning module receives PET Image and CT image as separate channels (eg, separate channels representing the same volume) (eg, similar to receiving two color channels (RGB) of photographic color images by a machine learning module)].

In some embodiments, the machine learning module receives as input a 3D segmentation map that identifies one or more volumes of interest (VOIs) within the 3D functional image and/or the 3D anatomical image, each VOI corresponding to in specific target tissue areas and/or specific anatomical areas.

In some embodiments, the instructions cause the processor to automatically segment the 3D anatomical image, thereby creating the 3D segmentation map.

In certain embodiments, the machine learning module is a region-specific machine learning module that receives as input specific portions of the 3D functional image corresponding to one or more specific tissue regions and/or anatomical regions of the subject .

In some embodiments, the machine learning module generates the hotspot list as an output [eg, wherein the machine learning module implements training to determine one or more locations based on intensities of at least a portion of the voxels of the 3D functional image ( For example, a machine learning algorithm (eg, an artificial neural network (ANN)) for 3D coordinates, each location corresponds to the location of one of the one or more hotspots].

In some embodiments, the machine learning module generates the 3D heat map as output [eg, wherein the machine learning module implements training to segment the 3D functional image (eg, based at least in part on voxels of the 3D functional image) intensity) to identify the 3D heat map (e.g., the 3D heat map delineates the 3D boundaries (e.g., irregular boundaries) of the hotspot for each hotspot), thereby identifying the 3D hotspot volumes (e.g., by the 3D hotspot boundaries) A machine learning algorithm (eg, an artificial neural network (ANN)) of the 3D hotspot volumes that encloses)); for example, wherein the machine learning module implementation is trained for each voxel of at least a portion of the 3D functional image A machine learning algorithm that determines a hotspot likelihood value representing the likelihood that the voxel corresponds to a hotspot (eg, and step (b) includes performing one or more subsequent post-processing steps, such as qualifying, to use the hotspot likelihood The property value identifies the 3D hotspot volumes of the 3D heatmap (eg, the 3D heatmap delineates the 3D boundaries (eg, irregular boundaries) of the hotspot for each hotspot, thereby identifying the 3D hotspot volumes (eg, by the etc. 3D hotspot boundary enclosure )))].

In certain embodiments, the instructions cause the processor to: (d) determine, for each of at least a portion of the hotspots, a lesion likelihood classification corresponding to the likelihood that the hotspot represents a lesion within the subject [eg, , a binary classification indicating whether the hot spot is a true lesion; eg, a likelihood value on a scale (eg, a floating point value in the range zero to one) representing the likelihood that the hot spot represents a true lesion].

In certain embodiments, at step (d), the instructions cause the processor to use a machine learning module to determine the lesion likelihood classification for each hot spot of the portion [eg, wherein the machine learning module implements training to detect hot spots (eg, generating a list of hotspots and/or a 3D heatmap as output) and a machine learning algorithm for each hotspot to determine the lesion likelihood classification for that hotspot].

In certain embodiments, in step (d), the instructions cause the processor to use a second machine learning module (eg, a hotspot classification module) to determine a lesion likelihood classification for each hotspot [eg, based at least in part on a 3D One or more members of the group consisting of the intensity of the functional image, the hotspot list, the 3D heat map, the intensity of the 3D anatomical image, and the 3D segmentation map; One or more members of the group consisting of intensity, hotspot list, 3D heat map, intensity of 3D anatomical image, and 3D segmentation map, one or more input channels].

In some embodiments, the instructions cause the processor to determine, for each hotspot, a set of one or more hotspot features and use the set of the one or more hotspot features as input to the second machine learning module.

In certain embodiments 55-58, wherein the instructions cause the processor to: (e) for each of at least a portion of the hotspots, determine a lesion likelihood classification corresponding to the likelihood that the hotspot represents a lesion within the subject .

In some embodiments, the instructions cause the processor to: (f) [eg, prior to step (b)] adjusting, by the processor, the intensities of voxels of the 3D-enabled image to correct one of the 3D-enabled images or Intensity exudation (eg, crosstalk) of a plurality of high-intensity volumes, each of the one or more high-intensity volumes corresponding to high radiopharmaceutical uptake associated with normal conditions (eg, not necessarily indicative of cancer) in the subject. Areas of high uptake tissue within the subject.

In some embodiments, at step (f), the instructions cause the processor to correct the intensity bleeds from the plurality of high-intensity volumes one at a time in a sequential manner [eg, first adjust the intensities of the voxels of the 3D functional image to correct for the intensities from the The intensity of a high-intensity volume bleeds out to generate a first corrected image, then the intensities of the voxels of the first corrected image are adjusted to correct for the intensity bleed-out from a second high-intensity volume, etc.].

In certain embodiments, the one or more high-intensity volumes correspond to one or more high-uptake tissue regions selected from the group consisting of kidney, liver, and bladder (eg, urinary bladder).

In certain embodiments, the instructions cause the processor to: (g) determine, for each of at least a portion of the one or more hotspots, a level indicative of radiopharmaceutical uptake within the underlying lesion to which the hotspot corresponds and/or the potential The corresponding lesion index for the size (eg, volume) of a sexual lesion.

In certain embodiments, at step (g), the instructions cause the processor to compare one or more of the ones associated with the hotspot (eg, at and/or near the location of the hotspot; eg, within the volume of the hotspot). The intensity(s) of a voxel (eg, corresponding to a standard uptake value (SUV)) and one or more reference values, each reference value being associated with a specific reference tissue region within a subject (eg, liver; eg, aortic portion) ) is associated and determined based on the intensity (eg, SUV value) of the reference volume corresponding to the reference tissue region [eg, as a mean (eg, a robust mean, such as the mean of values within an interquartile range)] .

In certain embodiments, the one or more reference values comprise selected from the group consisting of an aortic reference value associated with an aortic portion of the subject and a liver reference value associated with the subject's liver One or more members of the group.

In some embodiments, for at least one particular reference value associated with a particular reference tissue region, the instructions cause the processor to determine the particular reference value by: placing the values within the particular reference volume corresponding to the particular reference tissue region Intensity fitting of voxels [e.g., by fitting a distribution of voxel intensities (e.g., fitting a histogram of voxel intensities)] to a multicomponent mixture model (e.g., a two-component Gaussian model) [e.g., and Identify one or more secondary peaks in the distribution of voxel intensities that correspond to voxels associated with abnormal uptake, and exclude those voxels (from the reference value determination) from the reference value determination (eg, to account for the effect of abnormally low radiopharmaceutical uptake in specific parts of a reference tissue region such as parts of the liver)].

In certain embodiments, the instructions cause the processor to compute (eg, automatically) a subject's overall risk index that indicates the subject's cancer status and/or risk using the determined lesion index values.

In certain embodiments, the instructions cause the processor to determine for each hot spot (eg, automatically) that a potential cancerous lesion corresponding to the hot spot is represented is determined [eg, by the processor (eg, based on received and/or determined) 3D segmentation map)] is a specific anatomy within a subject located [eg, within the prostate, pelvic lymph nodes, non-pelvic lymph nodes, bone (eg, areas of bone metastases), and soft tissue regions not located in the prostate or lymph nodes] Anatomical classification of regions and/or groups of anatomical regions.

In some embodiments, the instructions cause the processor to: (h) cause at least a portion of the graphical representation of the one or more hot spots to be listed for display within a graphical user interface (GUI) for viewing by a user.

In certain embodiments, the instructions cause the processor to: (i) receive, via the GUI, use of a subset of the one or more hotspots identified by the user for viewing that are likely to represent potential cancerous lesions in the subject user's choice.

In certain embodiments, the 3D functional image comprises a PET or SPECT image obtained after administration of an agent (eg, a radiopharmaceutical; eg, an imaging agent) to the subject. In certain embodiments, the agent comprises a PSMA binding agent. In certain embodiments, ^the agent includes18F. In certain embodiments, the agent comprises [18F]DCFPyL. In certain embodiments, the agent includes PSMA-11 (eg, ^68Ga -PSMA-11). In certain embodiments, the reagent comprises one or more members selected from the group consisting ^of99mTc , ^68Ga , ^177Lu ,225Ac, ^111In , ^123I , ^{124I , and131I} .

In certain embodiments, the machine learning module implements a neural network [eg, artificial neural network (ANN); eg, convolutional neural network (CNN)].

In some embodiments, the processor is a processor of a cloud-based system.

In another aspect, the invention relates to a system for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the system comprising: A processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive (eg, and/or access) use functions A 3D functional image of the subject obtained by an imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having an intensity value representing detected radiation emitted from the specific physical volume, wherein at least a portion of the plurality of voxels of the 3D functional image represents a target a solid volume within a tissue region]; (b) receive (eg, and/or access) using anatomical imaging modalities [eg, x-ray computed tomography (CT); eg, magnetic resonance imaging (MRI); eg , ultrasound] a 3D anatomical image of the subject obtained, wherein the 3D anatomical image includes a graphical representation of tissue (eg, soft tissue and/or bone) within the subject; (c) using a machine learning module to automatically detect Detecting one or more hot spots within the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing (eg, indicative of) a potential cancerous lesion within the subject, establishing (i) as follows and (ii) one or both of: (i) a list of hotspots, which identifies the location of the hotspot for each hotspot, and (ii) a 3D heatmap, which identifies, for each hotspot, the corresponding 3D hotspot volume within the 3D functional image { For example, where the 3D heat map is a segmentation map (eg, including one or more segmentation masks) that identifies, for each hotspot, voxels within the 3D functional image corresponding to the 3D hotspot volume for each hotspot [eg, where The 3D heat map is obtained through artificial intelligence-based segmentation of the functional image (eg, using a machine learning module that receives at least the 3D functional image as input and generates the 3D heat map as output to segment hot spots); for example , where the 3D heat map depicts, for each hotspot, the 3D boundary (eg, an irregular boundary) of the hotspot (eg, the 3D boundary enclosing the 3D hotspot volume, eg, and the 3D functional image of the 3D hotspot volume that constitutes the 3D hotspot volume). voxels are distinguished from other voxels in the 3D functional image)}, wherein the machine learning module receives at least two input channels, the input channels include a first input channel corresponding to at least a portion of the 3D anatomical image and a first input channel corresponding to at least a portion of the 3D anatomical image A second input channel of at least a portion of the 3D functional image [eg, wherein the machine learning module receives PET images and CT images as separate channels (eg, separate channels representing the same volume) (eg, similar to that obtained by a machine learning module) receive two color channels (RGB) of photographic color images] and/or anatomical information derived therefrom [e.g., 3D a cut map that identifies one or more volumes of interest (VOIs) within the 3D functional image, each VOI corresponding to a specific target tissue region and/or a specific anatomical region]; and (d) storing and/or providing the hotspot list and /or the 3D heat map for display and/or further processing.

In another aspect, the invention relates to a system for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the system comprising: A processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive (eg, and/or access) use functions A 3D functional image of the subject obtained by an imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having an intensity value representing detected radiation emitted from the specific physical volume, wherein at least a portion of the plurality of voxels of the 3D functional image represents a target (b) use a first machine learning module to automatically detect one or more hot spots within the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing (e.g. , indicating) potential cancerous lesions within the subject, thereby establishing a list of hotspots identifying the location of the hotspot for each hotspot [eg, wherein the machine learning module implements a voxel trained to base at least a portion of the 3D functional image The strength of a machine learning algorithm (eg, an artificial neural network (ANN)) that determines one or more locations (eg, 3D coordinates), each location corresponding to the location of one of the one or more hotspots]; (c ) use a second machine learning module and the hotspot list to automatically determine the corresponding 3D hotspot volume within the 3D functional image for each of the one or more hotspots, thereby creating a 3D heatmap [eg, wherein the second machine learning module A group implements a machine learning algorithm (eg, an artificial neural network) trained to segment the 3D functional image based at least in part on the hotspot list and the intensities of voxels of the 3D functional image to identify the 3D hotspot volumes of the 3D heat map (ANN)); for example, wherein the machine learning module implements a machine learning algorithm trained to determine, for each voxel of at least a portion of the 3D functional image, a hotspot likelihood value representing the likelihood that the voxel corresponds to a hotspot ( For example, and step (b) includes performing one or more subsequent post-processing steps, such as defining, to identify the 3D hotspot volumes of the 3D heatmap using the hotspot likelihood values] [eg, wherein the 3D heatmap Using (eg, based on and/or corresponding to an output from) a second machine learning module generated segmentation maps (eg, including one or more segmentation masks), the 3D heat map for each hot spot identified corresponds to voxels within the 3D functional image of the 3D hotspot volume for each hotspot; eg, where the 3D heatmap depicts the 3D boundary (eg, an irregular boundary) of the hotspot for each hotspot (eg, the 3D boundary encloses the 3D hotspot volume, for example, and distinguish the voxels of the 3D functional image that make up the 3D hotspot volume from other voxels of the 3D functional image)]; and (d) storage The hotspot list and/or the 3D heatmap are stored and/or provided for display and/or further processing.

In certain embodiments, the instructions cause the processor to: (e) determine, for each of at least a portion of the hotspots, a lesion likelihood classification corresponding to the likelihood that the hotspot represents a lesion within the subject.

In certain embodiments, at step (e), the instructions cause the processor to use a third machine learning module (eg, a hotspot classification module) to determine the lesion likelihood classification for each hotspot [eg, based at least in part on selected from one or more members of the group consisting of the intensity of the 3D functional image, the hotspot list, the 3D heat map, the intensity of the 3D anatomical image, and the 3D segmentation map; Image intensity, hotspot list, 3D heat map, 3D anatomical image intensity, and one or more members of the group consisting of 3D segmentation maps, one or more input channels].

In certain embodiments, the instructions cause the processor to: (f) select the one or more hotspots corresponding to hotspots having a high likelihood corresponding to cancerous lesions based at least in part on the lesion likelihood classifications of the hotspots A subset of (eg, for inclusion in a report; eg, for computing one or more risk index values for a subject).

In another aspect, the invention relates to a method for measuring intensity values within a reference volume corresponding to a reference tissue region (eg, a liver volume associated with a subject's liver) in order to avoid , abnormally low) radiopharmaceutical uptake (eg, due to tumors without tracer uptake) associated with a tissue region of a system comprising: a processor of an arithmetic device; and a processor having instructions stored thereon memory, wherein the instructions, when executed by the processor, cause the processor to: (a) receive (eg, and/or access) a 3D functional image of the subject, the 3D functional image using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] obtains [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing the subject within a specific physical volume and having intensity values representing detected radiation emitted from the specific physical volume, wherein at least a portion of the plurality of voxels of the 3D functional image represent the physical volume within the target tissue region]; (b) identifying the reference volume within the 3D functional image; (c) fitting a multi-component mixture model (eg, a two-component Gaussian mixture model) to the intensities of voxels within the reference volume [eg, the multi-component mixture fit the model to the distribution (eg, histogram) of the intensities of voxels within the reference volume]; (d) identify the dominant mode of the multicomponent model; (e) determine the number, maximum, mode, median, etc.) intensity to determine a reference intensity value corresponding to a measure of intensity for voxels that are (i) within the reference tissue volume and (ii) associated with the Major mode association (eg, and exclusion of voxels with intensities associated with minor modes from reference value calculations) (eg, to avoid effects from tissue regions associated with low radiopharmaceutical uptake); (f) in and (g) using at least the reference intensity value to determine a lesion index value for each of at least a portion of the detected hotspots [eg, the The lesion index value is based on (i) a measure of the intensity of the voxel corresponding to the detected hot spot and (ii) the reference intensity value].

In another aspect, the invention pertains to a method for correcting for tissue regions of high uptake within a subject attributable to high uptake of radiopharmaceuticals associated with normal conditions (eg, and not necessarily indicative of cancer) A system for exuding (eg, crosstalk) of the intensity of Emission tomography (PET); e.g., single photon emission computed tomography (SPECT)] obtained [e.g., wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and has an intensity value representing the detected radiation emitted from the specific physical volume, wherein at least a portion of the plurality of voxels of the 3D functional image represents the physical volume within the target tissue region]; (b) identifying the 3D functional image within the (c) based on the identified high intensity An intensity volume identifies a suppression volume within the 3D functional image, the suppression volume corresponding to a volume located outside the boundary of the identified high-intensity volume and within a predetermined attenuation distance from the boundary of the identified high-intensity volume; (d) determining a background image corresponding to the 3D functional image, wherein the intensities of the voxels within the high intensity volume are replaced by interpolation based on the determination of the intensities of the voxels of the 3D functional image within the suppression volume; (e) ) determine the estimated image by subtracting the intensities of the voxels of the background image from the intensities of the voxels from the 3D functional image (eg, performing a voxel-by-voxel subtraction); (f) determine the suppression map by: Extrapolate the intensities of the voxels of the estimated image corresponding to the high-intensity volume to the positions of the voxels within the suppression volume to determine the intensities of the voxels of the suppression map corresponding to the suppression volume; and will correspond to the suppression The intensity of the voxels of the suppression map at locations outside the volume are set to zero; and (g) the intensity of the voxels of the 3D functional image is adjusted based on the suppression map (eg, by voxels from the 3D functional image) minus the intensities of the voxels of the suppression map) to correct for intensity bleed from the high-intensity volume.

In certain embodiments, the instructions cause the processor to perform steps (b) through (g) for each of the plurality of high-intensity volumes in a sequential manner, thereby correcting for intensity bleed from each of the plurality of high-intensity volumes.

In certain embodiments, the plurality of high-intensity volumes comprise one or more members selected from the group consisting of kidney, liver, and bladder (eg, urinary bladder).

In another aspect, the invention relates to a system for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the system comprising: A processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive (eg, and/or access) use functions A 3D functional image of the subject obtained by an imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having an intensity value representing detected radiation emitted from the specific physical volume, wherein at least a portion of the plurality of voxels of the 3D functional image represents a target Physical volume within tissue region]; (b) automatically detect one or more hot spots within the 3D functional image, each hot spot corresponding to a localized region of increased intensity relative to its surroundings and representing (eg, indicative of) the subject potentially cancerous lesions within the receiving, via the interactive GUI, a user selection (eg, for inclusion in a report) of a final set of hotspots including at least a portion, and at most all, of the one or more auto-detected hotspots; and (e) storing and/or provide the final set of hotspots for display and/or further processing.

In certain embodiments, the instructions cause the processor to: (f) receive, via the GUI, user selections of one or more additional, user-identified hotspots for inclusion in the final set of hotspots; and (g) update The final set of hotspots includes the one or more additional user-identified hotspots.

In certain embodiments, in step (b), the instructions cause the processor to use one or more machine learning modules.

In another aspect, the invention relates to a method for automatically processing 3D images of a subject to identify and/or characterize (eg, grade) cancerous lesions in the subject, the method comprising: (a) receiving (eg, and/or accessing), by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality [eg, positron emission tomography (PET); eg , Single Photon Emission Computed Tomography (SPECT)]; (b) receiving (eg, and/or accessing), by the processor, a 3D anatomical image of the subject obtained using an anatomical imaging modality [eg, computed tomography (CT) images; magnetic resonance (MR) images]; (c) receiving (eg, and/or accessing), by the processor, a 3D segmentation map that identifies within the 3D functional image and/or one or more specific tissue regions or groups of tissue regions within the 3D anatomical image (e.g., a group of tissue regions corresponding to a specific anatomical region; e.g., including one of the organs in which high or low radiopharmaceutical uptake occurs (d) using one or more machine learning modules by the processor to automatically detect and/or segment a set of one or more hotspots within the 3D functional image, each hotspot corresponding to a relative A local area of increased intensity surrounding it and representing a potential cancerous lesion within the subject, creating one or both of the following (i) and (ii): (i) a list of hotspots that identifies the location of the hotspot for each hotspot [eg, as detected by one or more machine learning modules], and (ii) a 3D heat map that identifies, for each hot spot, the corresponding 3D hot spot volume within the 3D functional image [eg, as detected by one or more Segmentation performed by multiple machine learning modules to determine] [eg, where the 3D heat map is a segmentation map that, for each hotspot, delineates the 3D hotspot boundary (eg, an irregular boundary) of the 3D Hotspot Volume)], wherein at least one (eg, up to all) of the one or more machine learning modules receives as input (i) 3D functional images, (ii) 3D anatomical images, and (iii) 3D segmentation maps; and (e) storing and/or providing a list of hot spots and/or a 3D heat map for display and/or further processing.

In certain embodiments, the method includes receiving, by the processor, an initial 3D segmentation map that identifies one or more (eg, a plurality) within the 3D anatomical image and/or the 3D functional image a specific tissue region (eg, an organ and/or a specific bone); and identifying, by the processor, at least a portion of the one or more specific tissue regions as belonging to one or more tissue groups (eg, a predefined group) specifying and updating the 3D segmentation map by the processor to indicate the identified specific regions as belonging to the specific tissue group; and using the updated 3D segmentation map by the processor as sending the one or the Input to at least one of a plurality of machine learning modules.

In certain embodiments, the one or more tissue groups include a soft tissue group such that a particular tissue region representing soft tissue is identified as belonging to the soft tissue group. In certain embodiments, the one or more tissue groups include a bone tissue group such that a particular tissue region representing bone is identified as belonging to the bone tissue group. In certain embodiments, the one or more tissue populations comprise high uptake organ populations such that one or more associated with high radiopharmaceutical uptake (eg, under normal circumstances, and not necessarily due to the presence of lesions) Multiple organs were identified as belonging to this high uptake group.

In certain embodiments, the method includes, for each detected and/or segmented hotspot, determining, by the processor, a classification of the hotspot [eg, according to anatomical location, eg, classifying the hotspot as bone, lymph, or prostate For example, an alphanumeric code is assigned based on the determined (eg, by a processor) location of hotspots in a subject, such as the labeling scheme in Table 1].

In certain embodiments, the method includes using at least one of one or more machine learning modules to determine the classification of the hot spot for each detected and/or segmented lesion (eg, wherein a single machine learning module performs detection, segmentation and classification).

In certain embodiments, the one or more machine learning modules include: (A) a systemic lesion detection module that detects and/or segments hot spots throughout the body; and (B) detects and/or segments Prostate lesions model of hot spots in the prostate. In certain embodiments, the method includes generating a hotspot list and/or map using each of (A) and (B) and combining the results.

In certain embodiments, step (d) comprises: segmenting and classifying the set of one or more hotspots to create a labeled 3D heatmap that identifies for each hotspot the Corresponding 3D hotspot volumes within the 3D functional image and wherein each hotspot volume is marked as belonging to a specific hotspot category of the plurality of hotspot categories [eg, each hotspot category identifies the specific anatomy and/or tissue to which the lesion represented by the hotspot is determined to be located Region (eg, lymph, bone, prostate)]: segment a first initial set of one or more hotspots within the 3D functional image using a first machine learning module to create a first initial set of volumes identifying the first initial hotspot volume 3D heat map, wherein the first machine learning module segments the hot spots of the 3D functional image according to a single hotspot category [eg, identifying all hotspots as belonging to a single hotspot category in order to distinguish background regions from hotspot volumes (eg, but not different type of hotspots) (eg, such that each hotspot volume identified by the first 3D heatmap is marked as belonging to a single hotspot class, as opposed to background)]; segment the 3D functional image using a second machine learning module a second initial set of one or more hotspots, thereby establishing a second initial 3D heat map identifying the second initial set of hotspot volumes, wherein the second machine learning module segments the 3D functional image according to a plurality of different hotspot categories such that The second initial 3D heat map is a multi-class 3D heat map in which each hotspot volume is marked as belonging to a particular one of a plurality of different hotspot classes (eg, to distinguish between hotspot volumes corresponding to different hotspot classes, and to distinguish between hotspot volumes and backgrounds) area); and combining, by the processor, the first initial 3D heat map and the second initial 3D heat map by performing the following operations on at least a portion of the hot spot volumes identified by the first initial 3D heat map: Identifying the matching hotspot volume of the second initial 3D heat map (eg, by identifying the substantially overlapping hotspot volumes of the first and second initial 3D heatmap), the matching hotspot volume of the second 3D heatmap has been marked is a specific hotspot category belonging to a plurality of different hotspot categories; and marking the specific hotspot volume of the first initial 3D heat map as belonging to (eg, the matching hotspot volume is marked as belonging to) the specific hotspot category, thereby establishing a merged 3D heat map that includes segmented hotspot volumes of the first 3D heatmap that have been labeled according to the identified class to which the matching hotspot volumes of the second 3D heatmap belong; and step (e) includes The merged 3D heat map is stored and/or provided for display and/or further processing.

In some embodiments, the plurality of different hotspot categories include one or more members selected from the group consisting of: (i) bone hotspots, the others of which are determined (eg, by a second machine learning model) group) to represent lesions localized in bone; (ii) lymphatic hotspots, which were determined (eg, by a second machine learning module) to represent lesions localized in lymph nodes; and (iii) prostate hotspots, which Equals are determined (eg, by a second machine learning module) to represent lesions localized in the prostate.

In certain embodiments, the method further comprises: (f) receiving and/or accessing the hotspot list; and (g) for each hotspot in the hotspot list, segmenting the hotspot using an analytical model [eg, to create an analyzed A 3D map of the segmented hotspots (eg, the 3D map identifies, for each hotspot, a hotspot volume that includes voxels of the 3D anatomical and/or functional images enclosed by the segmented hotspot region)].

In certain embodiments, the method further comprises: (h) receiving and/or accessing the heat map; and (i) for each hot spot in the heat map, segmenting the hot spot using an analytical model [eg, to create an analyzed A 3D map of the segmented hotspots (eg, the 3D map identifies, for each hotspot, a hotspot volume that includes voxels of the 3D anatomical and/or functional images enclosed by the segmented hotspot region)].

In certain embodiments, the analytical model is an adaptive bounding method, and step (i) includes determining one or more reference values, each reference value based on a reference value located within a specific reference volume corresponding to a specific reference tissue region A measure of the intensity of the voxels of the 3D functional image (eg, blood pool reference values are determined based on intensities within the aortic volume corresponding to portions of the subject's aorta; eg, liver reference values are based on and for each specific hotspot volume of the 3D heat map: by the processor to determine the corresponding hotspot intensity based on the intensities of voxels within the specific hotspot volume [eg, where the the hotspot intensity is the maximum value of the intensities of the voxels (eg, representing SUVs) within the particular hotspot volume]; and by the processor for the particular hotspot based on (i) the corresponding hotspot intensity and (ii) at least one of the or multiple reference values to determine hotspot specific thresholds.

In some embodiments, the hotspot specific threshold is determined using a specific threshold function selected from a plurality of threshold functions, the specific threshold function being selected based on a comparison of the corresponding hotspot intensity with the at least one reference value [For example, where each of the plurality of bounding functions is associated with a particular range of intensity (eg, SUV) values, and the particular bounding function falls within the hotspot intensity and/or its (eg, predetermined) percentage (eg, and wherein each particular range of intensity values is at least partially bounded by a multiple of the at least one reference value)].

In some embodiments, the hotspot specific threshold is determined (eg, by the specific threshold function) as a variable percentage of the corresponding hotspot intensity, wherein the variable percentage decreases as the hotspot intensity increases [ For example, where the variable percentage itself corresponds to a function of hot spot intensity (eg, a decreasing function)].

In another aspect, the invention relates to a method for automatically processing a 3D image of a subject to identify and/or characterize (eg, grade; eg, classify; eg, as represent a particular lesion type) the subject A method of a cancerous lesion within, comprising: (a) receiving (eg, and/or accessing), by a processor of a computing device, a 3D functional image of the subject obtained using a functional imaging modality [eg, Positron Emission Tomography (PET); eg, Single Photon Emission Computed Tomography (SPECT)]; (b) automatically segmenting one of the 3D functional images by the processor using a first machine learning module or a first initial set of hotspots, thereby creating a first initial 3D heatmap identifying a first initial set of hotspot volumes, wherein the first machine learning module segments the hotspots of the 3D functional image according to a single hotspot class [eg, dividing the All hotspots are identified as belonging to a single hotspot category in order to distinguish background regions from hotspot volumes (eg, but not different types of hotspots) (eg, so that each hotspot volume identified by the first 3D heat map is marked as belonging to a single hotspot category, such as the opposite of the background)]; (c) by the processor using a second machine learning module to automatically segment a second initial set of one or more hot spots in the 3D functional image, thereby establishing a second initial set of identification A second initial 3D heat map of the set of hotspot volumes, wherein the second machine learning module identifies a specific anatomical and/or tissue region where the lesion represented by the hotspot is determined to be located according to a plurality of different hotspot categories [eg, each hotspot category (eg, lymph, bone, prostate)] segment the 3D functional image such that the second initial 3D heat map is a multi-class 3D heat map, where each hotspot volume is marked as belonging to a particular one of the plurality of different hotspot classes (eg (d) by the processor by targeting at least a portion of the first initial hot spot identified by the first initial 3D heat map Each particular hotspot volume of the volume set merges the first initial 3D heatmap and the second initial 3D heatmap by identifying matching hotspot volumes of the second initial 3D heatmap (eg, by identifying the first and a substantially overlapping hotspot volume of the second initial 3D heatmap), the matching hotspot volume of the second 3D heatmap has been marked as belonging to a particular hotspot class of a plurality of different hotspot classes; and the first initial 3D heatmap The specific hotspot volume is marked as belonging to the specific hotspot category (to which the matching hotspot volume is marked as belonging), thereby creating a merged 3D heatmap that includes matching hotspots that have been based on the second 3D heatmap a segmented hotspot volume of the first 3D heatmap labeled with the identified class; and (e) storing and/or providing the merged 3D heatmap for display and/or further processing.

In some embodiments, the plurality of different hotspot categories include one or more members selected from the group consisting of: (i) bone hotspots, the others of which are determined (eg, by a second machine learning model) group) to represent lesions localized in bone; (ii) lymphatic hotspots, which were determined (eg, by a second machine learning module) to represent lesions localized in lymph nodes; and (iii) prostate hotspots, which Equals are determined (eg, by a second machine learning module) to represent lesions localized in the prostate.

In another aspect, the invention relates to a method for automatically processing 3D images of a subject to identify and/or characterize (eg, grade; eg, classify; eg, such as representing a particular lesion) via an adaptive bounding method type) a method of a cancerous lesion in the subject, the method comprising: (a) receiving (eg, and/or accessing), by a processor of a computing device, an image of the subject obtained using a functional imaging modality 3D functional images [eg, Positron Emission Tomography (PET); eg, Single Photon Emission Computed Tomography (SPECT)]; (b) received (eg, and/or accessed) by the processor identifying, within the 3D functional image, a preliminary 3D heat map of one or more preliminary hot spot volumes; (c) determining, by the processor, one or more reference values, each reference value based on a specific location corresponding to a specific reference tissue region A measure of the intensity of the voxels of the 3D functional image within the reference volume (eg, blood pool reference values are determined based on the intensity within the aortic volume corresponding to the portion of the subject's aorta; eg, liver reference values are determined based on the corresponding determined by the intensity within the liver volume of the subject's liver); (d) by the processor by each particular preliminary hotspot volume for at least a portion of the one or more preliminary hotspot volumes identified by the preliminary 3D heat map The hotspot volume creates a refined 3D heatmap based on the preliminary hotspot volumes and using adaptive bound-based segmentation by determining the corresponding hotspot intensities based on the intensities of voxels within that particular preliminary hotspot volume [eg, where the hotspot intensity is the maximum value of the intensities of voxels (eg, representing SUVs) within the particular preliminary hotspot volume]; for the particular preliminary hotspot volume based on (i) the corresponding hotspot intensity and (ii) at least one of the one or multiple reference values to determine hotspot specific thresholds; segment at least a portion of the 3D functional image (e.g., approximately a sub-volume within a specific preliminary hotspot volume) using a threshold-based segmentation algorithm using a threshold-based segmentation algorithm for Perform image segmentation [eg, and identify clusters of voxels that have intensities above the hotspot-specific threshold and include the highest intensity voxels of the preliminary hotspot (eg, connect components with n). 3D clusters of voxels connected to each other in a manner (e.g., where n =6, n =18, etc.)], to determine the refinement corresponding to a particular preliminary hotspot volume, the hotspot volume of the analysis segmentation; and the hotspot for that refinement The volume is included in the refined 3D heat map; and (e) storing and/or providing the refined 3D heat map for display and/or further processing.

In some embodiments, the hotspot specific threshold is determined using a specific threshold function selected from a plurality of threshold functions, the specific threshold function being selected based on a comparison of the corresponding hotspot intensity with the at least one reference value [For example, where each of the plurality of bounding functions is associated with a particular range of intensity (eg, SUV) values, and the particular bounding function falls within the hotspot intensity and/or its (eg, predetermined) percentage (eg, and wherein each particular range of intensity values is at least partially bounded by a multiple of the at least one reference value)].

In some embodiments, the hotspot specific threshold is determined (eg, by the specific threshold function) as a variable percentage of the corresponding hotspot intensity, wherein the variable percentage decreases as the hotspot intensity increases [ For example, where the variable percentage itself corresponds to a function of hot spot intensity (eg, a decreasing function)].

In another aspect, the invention relates to a method for automatically processing a 3D image of a subject to identify and/or characterize (eg, grade; eg, classify; eg, as represent a particular lesion type) the subject A method of cancerous lesions within, the method comprising: (a) receiving (eg, and/or accessing) by a processor of a computing device using an anatomical imaging modality [eg, x-ray computed tomography (CT) 3D anatomical image of the subject obtained by magnetic resonance imaging (MRI); eg, ultrasound], wherein the 3D anatomical image includes a graphical representation of tissue (eg, soft tissue and/or bone) within the subject (b) automatically segmenting the 3D anatomical image by the processor to create a 3D segmentation map that identifies a plurality of volumes of interest (VOIs) in the 3D anatomical image, including those corresponding to the subject's liver; Liver volume and aortic volume corresponding to aortic portions (eg, thoracic and/or abdominal portions); (c) received (eg, and/or accessed) by the processor using functional imaging modalities [eg, positive A 3D functional image of the subject obtained by electron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] [eg, wherein the 3D functional image includes a plurality of voxels, each voxel Represents a specific physical volume within the subject and has an intensity value representing the detected radiation emitted from the specific physical volume, wherein at least a portion of the plurality of voxels of the 3D functional image represent the physical volume within the target tissue region] (d) automatically segmenting, by the processor, one or more hotspots within the 3D functional image, each segmented hotspot corresponding to a localized region of increased intensity relative to its surroundings and representing (eg, indicative) within the subject (e) causing, by the processor, to transfer a graphical representation of the one or more automatically segmented hot spot volumes for use in interactive displayed within a graphical user interface (GUI) (eg, quality control and reporting GUI); (f) receiving, by the processor, via the interactive GUI, including at least a portion of the one or more automatically segmented hotspot volumes (eg, user selection of a final set of hotspots up to all); (g) by the processor for each hotspot volume of the final set based on (i) the intensities of the voxels of the functional image and (ii) determining a lesion index value using one or more reference values determined by the intensities of the voxels of the functional image corresponding to the liver volume and the aortic volume; and (e) storing and /or provide a final set of hot spots and/or lesion index values for display and/or further processing.

In certain embodiments, step (b) includes segmenting the anatomical image such that the 3D segmentation map identifies one or more bone volumes corresponding to one or more bones of the subject, and step (d) includes Use the one or more bone volumes within the functional image to identify the bone volume and segment one or more bone hot spot volumes located within the bone volume (eg, by applying one or more difference values of a Gaussian filter and comparing the bone volume limit).

In certain embodiments, step (b) includes segmenting the anatomical image such that the 3D segmentation map identifies soft tissue organs corresponding to the subject [eg, left/right lungs, left/right gluteus maximus, urinary bladder, liver , left/right kidney, gallbladder, spleen, thoracic and abdominal aorta, and optionally (eg, for patients who have not undergone radical prostatectomy) prostate] one or more organ volumes, and step (d) is included in Use the one or more segmented organ volumes within the functional image to identify one or more soft tissue (eg, lymph and optionally prostate) volumes and segment one or more lymphatic and/or prostate hotspots located within the soft tissue volume Volume (eg, by applying one or more Laplacians of Gaussian filters and qualifying the soft tissue volume).

In certain embodiments, step (d) further comprises, prior to segmenting the one or more lymphoid and/or prostate hot spot volumes, adjusting the intensity of the functional image to suppress the intensity from one or more high uptake tissue regions ( For example, using one or more of the inhibition methods described herein).

In certain embodiments, step (g) includes determining a liver reference value using the intensities of the voxels of the functional image corresponding to the liver volume.

In certain embodiments, the method includes fitting a two-component Gaussian mixture model to a histogram of intensities of functional image voxels corresponding to the liver volume, using the two-component Gaussian mixture model fitting to derive the liver volume from the Voxels with intensities associated with regions of abnormally low uptake are identified and excluded, and the intensities of the remaining (eg, not excluded) voxels are used to determine liver reference values.

In another aspect, the invention relates to a method for automatically processing a 3D image of a subject to identify and/or characterize (eg, grade; eg, classify; eg, as represent a particular lesion type) the subject A system of cancerous lesions within, the system comprising: a processor of a computing device; and a memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) Receive (eg, and/or access) a 3D functional image of the subject obtained using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography ( SPECT)]; (b) receiving (eg, and/or accessing) 3D anatomical images of the subject obtained using anatomical imaging modalities [eg, computed tomography (CT) images; magnetic resonance (MR) image]; (c) receiving (eg, and/or accessing) a 3D segmentation map identifying one or more specific tissue regions or groups of tissue regions within the 3D functional image and/or within the 3D anatomical image group (e.g., a group of tissue regions corresponding to a particular anatomical region; e.g., including a group of tissue regions of organs in which high or low radiopharmaceutical uptake occurs); (d) automated detection using one or more machine learning modules Detecting and/or segmenting a set of one or more hotspots within the 3D functional image, each hotspot corresponding to a local area of increased intensity relative to its surroundings and representing a potential cancerous lesion within the subject, thereby establishing (i) as follows and (ii) one or both of: (i) a list of hotspots that, for each hotspot, identifies the location of that hotspot [eg, as detected by one or more machine learning modules], and (ii) a 3D heat map , which for each hotspot identifies the corresponding 3D hotspot volume within the 3D functional image [eg, as determined by segmentation performed by one or more machine learning modules] [eg, where the 3D heatmap is a segmentation map, which is for Each hotspot delineates a 3D hotspot boundary (eg, an irregular boundary) for the hotspot (eg, the 3D boundary encloses the 3D hotspot volume)], wherein at least one of the one or more machine learning modules (eg, up to all ) receive as input (i) 3D functional images, (ii) 3D anatomical images and (iii) 3D segmentation maps; and (e) store and/or provide a list of hot spots and/or 3D heat maps for display and/or further processing .

In certain embodiments, the instructions cause the processor to: receive an initial 3D segmentation map that identifies one or more (eg, a plurality) of specific tissue regions within the 3D anatomical image and/or the 3D functional image (eg, organs and/or specific bones); identify at least a portion of the one or more specific tissue regions as specific ones belonging to one or more tissue groups (eg, predefined groups) and update the 3D segmentation map to Indicating the identified specific regions as belonging to the specific tissue group; and using the updated 3D segmentation map as an input to at least one of the one or more machine learning modules.

In certain embodiments, the one or more tissue groups include a soft tissue group such that a particular tissue region representing soft tissue is identified as belonging to the soft tissue group. In certain embodiments, the one or more tissue groups include a bone tissue group such that a particular tissue region representing bone is identified as belonging to the bone tissue group. In certain embodiments, the one or more tissue populations comprise high uptake organ populations such that one or more associated with high radiopharmaceutical uptake (eg, under normal circumstances, and not necessarily due to the presence of lesions) Multiple organs were identified as belonging to this high uptake group.

In certain embodiments, the instructions cause the processor, for each detected and/or segmented hotspot, to determine a classification of the hotspot [eg, according to anatomical location, eg, classifying the lesion as bone, lymphoid, or prostate, eg, An alphanumeric code is assigned based on the determined (eg, by a processor) location of the hot spot relative to the subject, such as the labeling scheme in Table 1].

In certain embodiments, the instructions cause the processor to use at least one of one or more machine learning modules to determine the classification of the hotspot for each detected and/or segmented hotspot (eg, wherein a single machine learning module group to perform detection, segmentation and classification).

In certain embodiments, the one or more machine learning modules include: (A) a systemic lesion detection module that detects and/or segments hot spots throughout the body; and (B) detects and/or segments Prostate lesions model of hot spots in the prostate. In some embodiments, the instructions cause the processor to use each of (A) and (B) to generate a hotspot list and/or map and combine the results.

In certain embodiments, in step (d), the instructions cause the processor to segment and classify the set of one or more hot spots to create a labeled 3D heat map for Each hotspot identifies a corresponding 3D hotspot volume within the 3D functional image and wherein each hotspot is marked as belonging to a specific hotspot category of a plurality of hotspot categories [eg, each hotspot category identifies the specific anatomy where the lesion represented by the hotspot is determined to be located and /or tissue region (eg, lymph, bone, prostate)]: segment a first initial set of one or more hotspots within the 3D functional image using a first machine learning module, thereby creating a set of volumes identifying the first initial hotspot volume A first initial 3D heat map in which the first machine learning module segments the hotspots of the 3D functional image according to a single hotspot category [eg, identifying all hotspots as belonging to a single hotspot category in order to distinguish background regions from hotspot volumes (eg, but do not distinguish between different types of hotspots) (eg, so that each hotspot volume identified by the first 3D heatmap is marked as belonging to a single hotspot class, as opposed to background)]; segment the 3D using a second machine learning module a second initial set of one or more hotspots within the functional image, thereby creating a second initial 3D heat map identifying the second initial set of hotspot volumes, wherein the second machine learning module segments the 3D features according to a plurality of different hotspot categories image such that the second initial 3D heat map is a multi-class 3D heat map, where each hotspot volume is marked as belonging to a particular one of a plurality of different hotspot classes (eg, to distinguish between hotspot volumes corresponding to different hotspot classes, and to distinguish between hotspots volume and background area); and merging the first initial 3D heat map and the second initial 3D heat map by performing the following operations on at least a portion of the hot spot volumes identified by the first initial 3D heat map: identifying the The matching hotspot volume of the second initial 3D heat map (eg, by identifying the substantially overlapping hotspot volumes of the first and second initial 3D heatmaps) that has been marked as belonging to the second 3D heatmap a specific hotspot category of a plurality of different hotspot categories; and marking the specific hotspot volume of the first initial 3D heat map as belonging to (to which the matching hotspot volume is marked) the specific hotspot category, thereby creating a merged 3D hotspot Figure, the merged 3D heat map includes segmented hotspot volumes of the first 3D heatmap that have been labeled according to the identified class to which matching hotspots of the second 3D heatmap belong; and in step (e), the instructions cause the processor The merged 3D heat map is stored and/or provided for display and/or further processing.

In some embodiments, the plurality of different hotspot categories include one or more members selected from the group consisting of: (i) bone hotspots, the others of which are determined (eg, by a second machine learning model) group) to represent lesions localized in bone; (ii) lymphatic hotspots, which were determined (eg, by a second machine learning module) to represent lesions localized in lymph nodes; and (iii) prostate hotspots, which Equals are determined (eg, by a second machine learning module) to represent lesions localized in the prostate.

In certain embodiments, the instructions further cause the processor to: (f) receive and/or access the hotspot list; and (g) for each hotspot in the hotspot list, segment the hotspot using an analytical model [eg, to establish A 3D map of the segmented hotspots is analyzed (eg, the 3D map identifies, for each hotspot, a hotspot volume that includes voxels of the 3D anatomical and/or functional images enclosed by the segmented hotspot region)].

In certain embodiments, the instructions further cause the processor to: (h) receive and/or access the heat map; and (i) for each hot spot in the heat map, segment the hot spot using an analytical model [eg, to establish A 3D map of the segmented hotspots is analyzed (eg, the 3D map identifies, for each hotspot, a hotspot volume that includes voxels of the 3D anatomical and/or functional images enclosed by the segmented hotspot region)].

In certain embodiments, the analytical model is an adaptive bounding method, and in step (i), the instructions cause the processor to: determine one or more reference values, each reference value based on a location corresponding to a particular reference tissue region A measure of the intensity of the voxels of the 3D functional image within a specific reference volume (eg, a blood pool reference value, which is determined based on the intensity within the aortic volume corresponding to the portion of the subject's aorta; eg, the liver reference value, which is determined based on the intensity within the liver volume corresponding to the subject's liver); and for each specific hotspot volume of the 3D heat map: the corresponding hotspot is determined based on the intensity of the voxels within the specific hotspot volume intensity [eg, where the hot spot intensity is the maximum value of the intensities of the voxels (eg, representing an SUV) within the particular hot spot volume]; and for the particular hot spot, based on (i) the corresponding hot spot intensity and (ii) at least one The one or more reference values determine the hotspot specific threshold value.

In some embodiments, the hotspot specific threshold is determined using a specific threshold function selected from a plurality of threshold functions, the specific threshold function being selected based on a comparison of the corresponding hotspot intensity with the at least one reference value [For example, where each of the plurality of bounding functions is associated with a particular range of intensity (eg, SUV) values, and the particular bounding function falls within the hotspot intensity and/or its (eg, predetermined) percentage (eg, and wherein each particular range of intensity values is at least partially bounded by a multiple of the at least one reference value)].

In some embodiments, the hotspot specific threshold is determined (eg, by the specific threshold function) as a variable percentage of the corresponding hotspot intensity, wherein the variable percentage decreases as the hotspot intensity increases [ For example, where the variable percentage itself corresponds to a function of hot spot intensity (eg, a decreasing function)].

In another aspect, the invention relates to a method for automatically processing a 3D image of a subject to identify and/or characterize (eg, grade; eg, classify; eg, as represent a particular lesion type) the subject A system of cancerous lesions within, the system comprising: a processor of a computing device; and a memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) Receive (eg, and/or access) a 3D functional image of the subject obtained using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography ( SPECT)]; (b) using a first machine learning module to automatically segment a first initial set of one or more hot spots in the 3D functional image, thereby establishing a first initial set of hot spot volumes that identify the corresponding correspondence in the 3D functional image A first initial 3D heat map of a 3D hotspot volume, wherein the first machine learning module segments the hotspots of the 3D functional image according to a single hotspot category [eg, all hotspots are identified as belonging to a single hotspot category in order to distinguish background regions from hotspot volumes (eg, but not distinguishing between different types of hotspots) (eg, such that each hotspot volume identified by the first 3D heatmap is marked as belonging to a single hotspot category, as opposed to the background)]; (c) using the second The machine learning module automatically segments a second initial set of one or more hot spots in the 3D functional image, thereby creating a second initial 3D heat map identifying the second initial hot spot volume set, wherein the second machine learning module is based on the complex number different hotspot categories (eg, each hotspot category identifies a specific anatomical and/or tissue region (eg, lymph, bone, prostate) where the lesion represented by the hotspot is determined to be located) to segment the 3D functional image such that the second initial A 3D heat map is a multi-class 3D heat map, where each hotspot volume is marked as belonging to a particular one of the plurality of different hotspot classes (e.g., to distinguish between hotspot volumes corresponding to different hotspot classes, and to distinguish between hotspot volumes and background regions); (d) merging the first initial 3D heat map and the second initial 3D hotspot by performing the following operations for each specific hotspot volume of at least a portion of the first initial hotspot volume set identified by the first initial 3D heatmap Figure: Identify the matching hotspot volume of the second initial 3D heat map (eg, by identifying substantially overlapping hotspot volumes of the first and second initial 3D heatmaps), the matching hotspot volume of the second 3D heatmap a particular hotspot category that has been marked as belonging to a plurality of different hotspot categories; and marking the particular hotspot volume of the first initial 3D heat map as belonging to (to which the matching hotspot volume is marked) the particular hotspot category, thereby establishing a merged 3D heat map comprising segmented hotspot volumes of the first 3D heatmap that have been labeled according to the class to which matching hotspot identifications of the second 3D heatmap belong; and (e) storing and/or The merged 3D heat map is provided for display and/or further processing.

In some embodiments, the plurality of different hotspot categories include one or more members selected from the group consisting of: (i) bone hotspots, the others of which are determined (eg, by a second machine learning model) group) to represent lesions localized in bone; (ii) lymphatic hotspots, which were determined (eg, by a second machine learning module) to represent lesions localized in lymph nodes; and (iii) prostate hotspots, which Equals are determined (eg, by a second machine learning module) to represent lesions localized in the prostate.

In another aspect, the invention relates to a method for automatically processing 3D images of a subject to identify and/or characterize (eg, grade; eg, classify; eg, such as representing a particular lesion) via an adaptive bounding method type) a system of cancerous lesions in the subject, the system comprising: a processor of a computing device; and a memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the Processor: (a) Receive (eg, and/or access) a 3D functional image of the subject obtained using a functional imaging modality [eg, Positron Emission Tomography (PET); eg, Single Photon Emission computed tomography (SPECT)]; (b) receiving (eg, and/or accessing) a preliminary 3D heat map identifying one or more preliminary hot spot volumes within the 3D functional image; (c) determining one or more reference values, each reference value based on a measure of the intensity of the voxels of the 3D functional image positioned within a specific reference volume corresponding to a specific reference tissue region (eg, a blood pool reference value based on an intensity corresponding to the subject's aorta) (d) by targeting at least the intensities identified by the preliminary 3D heatmap A portion of each particular preliminary hotspot volume of the one or more preliminary hotspot volumes creates a refined 3D heat map based on the preliminary hotspot volumes and using adaptive bound-based segmentation: the intensity of the voxel to determine the corresponding hotspot intensity [eg, where the hotspot intensity is the maximum value of the intensities of the voxels (eg, representing SUVs) within the particular preliminary hotspot volume]; and for the particular preliminary hotspot volume based on (i ) the corresponding hotspot intensity and (ii) at least one of the one or more reference values to determine a hotspot-specific threshold; segment at least a portion of the 3D functional image using a threshold-based segmentation algorithm (eg, approximately in a particular preliminary hotspot volume) sub-volume), the bound-based segmentation algorithm performs image segmentation using a hotspot-specific threshold value for a specific preliminary hotspot determination [eg, and identifying the maximum intensity volume that has an intensity above the hotspot-specific threshold and includes the preliminary hotspot voxel clusters of voxels (e.g., 3D clusters of voxels connected to each other in n-connected components (e.g., where n = 6, n = 18, etc.))] to determine the refinement, analyzing the segmented hot spot volume; and including the refined hot spot volume in the refined 3D heat map; and (e) storing and/or providing the refined 3D heat map for display and/or further processing.

In some embodiments, the hotspot specific threshold is determined using a specific threshold function selected from a plurality of threshold functions, the specific threshold function being selected based on a comparison of the corresponding hotspot intensity with the at least one reference value [For example, where each of the plurality of bounding functions is associated with a particular range of intensity (eg, SUV) values, and the particular bounding function falls within the hotspot intensity and/or its (eg, predetermined) percentage (eg, and wherein each particular range of intensity values is at least partially bounded by a multiple of the at least one reference value)].

In some embodiments, the hotspot specific threshold is determined (eg, by the specific threshold function) as a variable percentage of the corresponding hotspot intensity, wherein the variable percentage decreases as the hotspot intensity increases [ For example, where the variable percentage itself corresponds to a function of hot spot intensity (eg, a decreasing function)].

In another aspect, the invention relates to a method for automatically processing a 3D image of a subject to identify and/or characterize (eg, grade; eg, classify; eg, as represent a particular lesion type) the subject A system of cancerous lesions within, the system comprising: a processor of a computing device; and a memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receiving (eg, and/or accessing) a 3D image of the subject obtained using an anatomical imaging modality [eg, x-ray computed tomography (CT); eg, magnetic resonance imaging (MRI); eg, ultrasound] an anatomical image, wherein the 3D anatomical image includes a graphical representation of tissue (eg, soft tissue and/or bone) within the subject; (b) automatically segment the 3D anatomical image to create a 3D segmentation map that identifies the A plurality of volumes of interest (VOIs) in the 3D anatomical image, including the liver volume corresponding to the subject's liver and the aortic volume corresponding to the aortic portion (eg, thoracic and/or abdominal portion); (c) receiving (eg, and/or access) a 3D image of the subject obtained using a functional imaging modality [eg, positron emission tomography (PET); eg, single photon emission computed tomography (SPECT)] Functional image [eg, wherein the 3D functional image includes a plurality of voxels, each voxel representing a specific physical volume within the subject and having an intensity value representing detected radiation emitted from the specific physical volume, wherein the 3D functional image At least a portion of the plurality of voxels of the image represents a solid volume within the target tissue region]; (d) automatically segment one or more hot spots within the 3D functional image, each segmented hot spot corresponding to a local area of increased intensity relative to its surroundings region and represent (eg, indicate) a potential cancerous lesion within the subject, thereby identifying one or more auto-segmented hot spot volumes; (e) causing a transfer of the one or more auto-segmented hot spot volumes a graphical representation for display within an interactive graphical user interface (GUI) (eg, a quality control and reporting GUI); (f) receiving via the interactive GUI a volume that includes at least a portion of the one or more automatically segmented hotspot volumes User selection of the final set of hotspots (eg, up to all); (g) each hotspot volume for the final set is based on (i) the functional image corresponding to the hotspot volume (eg, positioned within the hotspot volume) and (ii) use one or more reference values determined by the intensities of the voxels of the functional image corresponding to the liver volume and the aortic volume to determine a lesion index value; and (e) store and/or The final set of hotspots and/or lesion index values are provided for display and/or further processing.

In certain embodiments, at step (b), the instructions cause the processor to segment the anatomical image such that the 3D segmentation map identifies one or more bone volumes corresponding to one or more bones of the subject, and at Step (d), the instructions cause the processor to identify a bone volume within the functional image using the one or more bone volumes and segment one or more bone hot spot volumes located within the bone volume (eg, by applying a Gaussian filter) one or more of the differences and bound to the bone volume).

In certain embodiments, at step (b), the instructions cause the processor to segment the anatomical image such that the 3D segmentation map identifies soft tissue organs (eg, left/right lungs, left/right gluteus maximus) corresponding to the subject , urinary bladder, liver, left/right kidneys, gallbladder, spleen, thoracic and abdominal aorta, and as needed (eg, prostate for patients who have not undergone radical prostatectomy) one or more organ volumes, and in Step (d), the instructions cause the processor to identify soft tissue (eg, lymph and optionally prostate) volumes within the functional image using the one or more segmented organ volumes and segment one or more localized within the soft tissue volume Lymphatic and/or prostate hot spot volume (eg, by applying one or more Laplacian operators of Gaussian filters and qualifying the soft tissue volume).

In certain embodiments, at step (d), the instructions cause the processor to, prior to segmenting the one or more lymphatic and/or prostate hot spot volumes, adjust the intensity of the functional image to suppress the intensity of the functional image from one or more high uptake tissues The strength of the region (eg, using one or more of the suppression methods described herein).

In some embodiments, at step (g), the instructions cause the processor to use the intensities of the voxels of the functional image corresponding to the liver volume to determine a liver reference value.

In certain embodiments, the instructions cause the processor to: fit a two-component Gaussian mixture model to a histogram of intensities corresponding to functional image voxels of the liver volume, using the two-component Gaussian mixture model fit to automatically The liver volume identifies and excludes voxels with intensities associated with regions of abnormally low uptake, and uses the intensities of the remaining (eg, not excluded) voxels to determine liver reference values.

Features of embodiments described with respect to one aspect of the invention may be applied with respect to another aspect of the invention.

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims US Provisional Patent Application Serial No. 63/048,436, filed July 6, 2020, and US Non-Provisional Patent Application No. 17/008,411, filed August 31, 2020 , the priority and rights of U.S. Provisional Patent Application No. 63/127,666, filed on December 18, 2020, and U.S. Provisional Patent Application No. 63/209,317, filed on June 10, 2021, in all cases The contents are incorporated herein by reference.

The systems, devices, methods, and programs of the claimed invention are contemplated to encompass variations and adaptations developed using information from the embodiments described herein. Adaptations and/or variations of the systems, devices, methods, and procedures described herein can be performed by those of ordinary skill in the relevant art.

Throughout the description in which articles, devices and systems are described as having, comprising or including particular components or in which procedures and methods are described as having, comprising or including particular steps, it is contemplated that there is additionally that the invention essentially consists of such Articles, devices and systems consisting of or consisting of the recited components, and in addition there are programs and methods according to the present invention that consist essentially of or consist of the recited processing steps.

It should be understood that the order of steps, or order for performing a particular action, is immaterial as long as the invention remains operable. Furthermore, two or more steps or actions may be performed simultaneously.

Reference herein to any publication (eg, in the [PRIOR ART] paragraph) is not an admission that such publication is prior art with respect to any of the claims made herein. The [PRIOR ART] paragraph is presented for clarity and is not intended to be a description of the prior art with respect to any claim.

The headers are provided as a convenience to the reader, and their presence and/or placement are not intended to limit the scope of the subject matter described herein.

In this application, unless otherwise clear from the context, (i) the term "a" can be read to mean "at least one"; (ii) the term "or" can be read to mean "and/or"; (iii) the terms "comprising" and "comprising" can be understood to encompass the recited elements or steps, whether presented alone or in combination with one or more additional elements or steps; and (iv) the terms "about" and "about" "Approximate" is to be understood as allowing standard variations as understood by those of ordinary skill; and (v) where ranges are provided, inclusive of the endpoints.

In certain embodiments, the term "about" when used herein for a reference value refers to a value that is, in context, similar to the reference value. In general, those skilled in the art familiar with the context will understand the relative degree of variation encompassed by "about" in that context. For example, in some embodiments, the term "about" can encompass 25%, 20%, 19%, 18%, 17%, 16%, 15%, 14%, 13%, 12%, 11% of the referenced value %, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1% or less. A. Nuclear Medicine Imaging

Nuclear medicine images are obtained using nuclear imaging modalities such as bone scan imaging, positron emission tomography (PET) imaging, and single photon emission computed tomography (SPECT) imaging.

As used herein, "image" (eg, a 3-D image of a mammal) includes any visual representation, such as a photo, video frame, streaming video, and any electronic, digital, or electronic, digital representation of a photo, video frame, or streaming video or mathematical simulations. In certain embodiments, any apparatus described herein includes a display for displaying an image or any other result produced by a processor. In certain embodiments, any method described herein includes the step of displaying an image or any other result produced by the method.

As used herein, "3-D" or "three-dimensional" with respect to "image" is meant to convey information about three dimensions. The 3-D imagery can be converted into a three-dimensional dataset and/or can be displayed as a collection of two-dimensional representations, or displayed as a three-dimensional representation.

In certain embodiments, nuclear medicine imaging uses imaging agents including radiopharmaceuticals. Nuclear medicine images are obtained after administration of a radiopharmaceutical to a patient (eg, a human subject) and provide information about the distribution of the radiopharmaceutical within the patient. Radiopharmaceuticals are compounds that include radionuclides.

As used herein, "administering" an agent means introducing a substance (eg, an imaging agent) into a subject. In general, any route of administration can be utilized, including, for example, parenteral (eg, intravenous), oral, topical, subcutaneous, peritoneal, intraarterial, inhalation, vaginal, rectal, nasal, introduction into cerebrospinal fluid, or instillation into the body compartment.

As used herein, "radionuclide" refers to a portion that includes a radioisotope of at least one element. Exemplary suitable radionuclides include, but are not limited to, those described herein. In some embodiments, the radionuclide is one used in positron emission tomography (PET). In some embodiments, the radionuclide is one used in single photon emission computed tomography (SPECT). In some embodiments, the non-limiting list of radionuclides includes99mTc, ^111In , ^64Cu , ^67Ga , ^68Ga , ^186Re , ^188Re , ^153Sm , ^177Lu , ^67Cu , ^{123I , 124I} , ¹²⁵ I, ¹²⁶ I, ¹³¹ I, ¹¹ C, ¹³ N, ¹⁵ O, ¹⁸ F, ¹⁵³ Sm, ¹⁶⁶ Ho, ¹⁷⁷ Lu, ¹⁴⁹ Pm, ⁹⁰ Y, ²¹³ Bi, ¹⁰³ Pd, ¹⁰⁹ Pd, ¹⁵⁹ Gd, ¹⁴⁰ La, ¹⁹⁸ Au, ¹⁹⁹ Au, ¹⁶⁹ Yb, ¹⁷⁵ Yb, ¹⁶⁵ Dy, ¹⁶⁶ Dy, ¹⁰⁵ Rh, ¹¹¹ Ag, ⁸⁹ Zr, ²²⁵ Ac, ⁸² Rb, ⁷⁵ Br, ⁷⁶ Br, ⁷⁷ Br, ⁸⁰ Br, ^80m Br, ⁸² Br, ⁸³ Br, ²¹¹ At and ¹⁹² Ir.

As used herein, a "radiopharmaceutical" refers to a compound that includes a radionuclide. In certain embodiments, radiopharmaceuticals are used for diagnostic and/or therapeutic purposes. In certain embodiments, the radiopharmaceutical comprises a small molecule labeled with one or more radionuclides, an antibody labeled with one or more radionuclides, and an antigen-binding portion of an antibody labeled with one or more radionuclides.

Nuclear medicine images (eg, PET scans; eg, SPECT scans; eg, whole body bone scans; eg, synthetic PET-CT images; eg, synthetic SPECT-CT images) detect radiation emitted from radionuclides of radiopharmaceuticals to form image. The distribution of a particular radiopharmaceutical in a patient can be determined by biological mechanisms, such as blood flow or perfusion, and by specific enzyme or receptor binding interactions. Different radiopharmaceuticals can be designed to take advantage of different biological mechanisms and/or specific specific enzyme or receptor binding interactions and thus, when administered to a patient, selectively concentrate within a specific type of tissue and/or region within the patient . Greater amounts of radiation are emitted from areas within the patient that have higher radiopharmaceutical concentrations than other areas, making these areas appear brighter in nuclear medicine images. Thus, intensity variations within nuclear medicine images can be used to map the distribution of radiopharmaceuticals within a patient. This mapped distribution of radiopharmaceuticals within a patient can be used, for example, to infer the presence of cancerous tissue within various regions of the patient's body.

For example, after administration to a patient, Xun99m methylene bisphosphonate ( ^99m Tc MDP) selectively accumulates in the patient's skeletal areas, especially at sites with abnormal osteogenesis associated with malignant bone lesions . The selective concentration of radiopharmaceuticals at these sites creates identifiable hot spots - localized areas of high intensity in nuclear medicine imaging. Thus, the presence of malignant bone lesions associated with metastatic prostate cancer can be inferred by identifying these hot spots within the patient's whole body scan. As described below, an automated analysis of intensity changes in whole body scans obtained after administration ^of99mTc MDP to a patient can be calculated with patient overall survival and other prognostic measures indicative of disease state, progression, treatment efficacy, and the like. associated risk index. In certain embodiments, other radiopharmaceuticals can also be used in a manner similar ^to99mTc MDP.

In certain embodiments, the specific radiopharmaceutical used depends on the specific nuclear medicine imaging modality used. For example, 18F sodium fluoride (NaF) also accumulates in bone lesions similar ^to99mTc MDP, but can be used with PET imaging. In certain embodiments, PET imaging may also utilize a radioactive form of vitamin choline that is readily taken up by prostate cancer cells.

In certain embodiments, radiopharmaceuticals that selectively bind to specific proteins or receptors of interest, particularly those whose expression is increased in cancerous tissue, can be used. These proteins or receptors of interest include, but are not limited to, tumor antigens such as CEA expressed in colorectal cancer; Her2/neu expressed in various cancers; BRCA 1 and BRCA expressed in breast and ovarian cancers 2; and TRP-1 and TRP-2 expressed in melanoma.

For example, human prostate-specific membrane antigen (PSMA) is upregulated in prostate cancer, including metastatic disease. PSMA is expressed by nearly all prostate cancers and its expression is further increased in poorly differentiated, metastatic, and hormone-refractory cancers. Thus, radiopharmaceuticals corresponding to PSMA-binding agents (eg, compounds with high affinity for PSMA) labeled with one or more radionuclides can be used to obtain nuclear medicine images of patients from which various regions of the patient can be assessed ( For example, including but not limited to the presence and/or status of prostate cancer within the skeletal region. In certain embodiments, nuclear medicine imaging obtained using a PSMA binding agent is used to identify the presence of cancerous tissue within the prostate when the disease is in a localized state. In certain embodiments, nuclear medicine images obtained using radiopharmaceuticals including PSMA binding agents are used to identify the presence of cancerous tissue in various regions, including not only the prostate, but other organ and tissue regions such as the lungs , lymph nodes, and bone), as relevant when the disease is metastatic.

Specifically, after administration to a patient, radionuclide-labeled PSMA-binding agents selectively accumulate in cancerous tissue based on their equivalent affinity for PSMA. In a manner similar to that described above for ^the99mTc MDP, the selective concentration of radionuclide-labeled PSMA binders at specific sites within the patient produces detectable hot spots in nuclear medicine imaging. When the PSMA-binding agent is localized in various cancerous tissues and regions of the body expressing PSMA, localized cancer in the prostate of a patient and/or metastatic cancer in various regions of the patient's body can be detected and assessed. Correlations to patient overall survival and other prognostic measures indicative of disease state, progression, treatment efficacy, and the like can be calculated based on automated analysis of intensity changes in nuclear medicine images obtained after administration of a PSMA binder radiopharmaceutical to a patient. risk index.

Various radionuclide-labeled PSMA binders are useful as radiopharmaceutical imaging agents for nuclear medicine imaging to detect and assess prostate cancer. In certain embodiments, the particular radionuclide-labeled PSMA-binding agent used depends on factors such as the particular imaging modality (eg, PET; eg, SPECT) and the particular region (eg, organ) of the patient being imaged . For example, certain radionuclide-labeled PSMA binders are suitable for PET imaging, while others are suitable for SPECT imaging. For example, certain radionuclide-labeled PSMA binders facilitate imaging of a patient's prostate and are primarily used when disease is localized, while others facilitate imaging of organs and regions throughout a patient's body and are useful for evaluating metastatic prostate cancer of.

Various PSMA binding agents and radionuclide-labeled versions thereof are described in US Pat. Nos. 8,778,305, 8,211,401, and 8,962,799, the entire contents of each of which are incorporated herein by reference. Certain PSMA binders and radionuclide-labeled versions thereof are also described in PCT application PCT/US2017/058418 (PCT publication WO 2018/081354) filed on October 26, 2017, which is incorporated by reference in its entirety manner is incorporated into this article. Section J below also describes several exemplary PSMA binders and their radionuclide-labeled versions. B. Automated Lesion Detection and Analysis i. Automated Lesion Detection

In certain embodiments, the systems and methods described herein utilize machine learning techniques for automated image segmentation and detection of hot spots that correspond to and indicate likely cancerous lesions within a subject.

In certain embodiments, the systems and methods described herein may be implemented in a cloud-based platform, eg, as in PCT/US2017/058418 (PCT publication WO 2018/081354) filed on October 26, 2017 ), which is incorporated by reference in its entirety.

In certain embodiments, as described herein, the machine learning module implements one or more machine learning techniques, such as random forest classifiers, artificial neural networks (ANNs), convoluted neural networks (CNNs), and the like . In some embodiments, for example, manually segmented and/or labeled images are used to train a machine learning module implementing machine learning techniques to identify and/or classify portions of the images. This training can be used to determine various parameters of the machine learning algorithm implemented by the machine learning module, such as the weights associated with the layers in the neural network. In some embodiments, once a machine learning module is trained (eg,) to accomplish a particular task (such as identifying a particular target region within an image), the value of the determined parameter is fixed and the (eg, invariant, static) The machine learning module is used to process new data (eg, different from the training data) and complete its training task without further updating its parameters (eg, the machine learning module does not receive feedback and/or updates). In some embodiments, the machine learning module may receive feedback, eg, based on a user's view of accuracy, and this feedback may be used as additional training data to dynamically update the machine learning module. In some embodiments, the trained machine learning module is a classification algorithm with adjustable and/or fixed (eg, locked) parameters, such as a random forest classifier.

In some embodiments, machine learning techniques are used to automatically segment anatomical structures in anatomical images (such as CT, MRI, ultrasound, etc. images) in order to identify structures that correspond to specific organs such as prostate, lymph node regions, kidneys, liver, The volume of interest for specific target tissue regions of the bladder, aortic portion) and bone. In this way, machine learning modules can be used to generate segmentation masks and/or anatomical backgrounds that can be mapped to (eg, projected onto) functional images, such as PET or SPECT images, to provide an assessment of intensity fluctuations therein. Or a segmentation map (eg, comprising a plurality of segmentation masks, each segmentation mask corresponding to and identifying a specific target tissue region). Methods for segmenting images and using the obtained anatomical context for analyzing nuclear medicine images are described in further detail, for example, in PCT/US2019/012486 (PCT publication WO 2019/136349) filed on Jan. 7, 2019, and In PCT/EP2020/050132 (PCT Publication WO 2020/144134) filed on January 6, 2020, the entire contents of each case are incorporated herein by reference.

In certain embodiments, underlying lesions are detected as localized regions of high intensity in functional images, such as PET images. These localized regions of increased intensity (also known as hotspots) can be detected using image processing techniques that do not necessarily involve machine learning, such as filtering and delimiting, and segmented using methods such as fast marching methods. Anatomical information created from segmentation of anatomical images allows anatomical marking of detected hot spots representing potential lesions. Anatomical background can also be used to allow different detection and segmentation techniques to be used for hot spot detection in different anatomical regions, which can improve sensitivity and performance.

In some embodiments, the automatically detected hotspots may be presented to the user via an interactive graphical user interface (GUI). In some embodiments, to account for target lesions detected by the user (eg, physician) but missed or poorly segmented by the system, manual segmentation tools are included in the GUI to allow the user to manually "draw" what they think corresponds to The area of the image of a lesion of any shape and size. These manually segmented lesions can then be included in a subsequently generated report along with the selected automatically detected target lesions. ii. AI - based lesion detection

In certain embodiments, the systems and methods described herein utilize one or more machine learning modules to analyze the intensity of 3D functional images and detect hot spots that represent potential lesions. For example, by collecting a dataset of PET/CT images in which hot spots representing lesions have been manually detected and segmented, training material for AI-based lesion detection algorithms can be obtained. These manually labeled images can be used to train one or more machine learning algorithms to automatically analyze functional images (eg, PET images) to accurately detect and segment hot spots corresponding to cancerous lesions.

1A shows an example process 100a for automated lesion detection and/or segmentation using machine learning modules implementing machine learning algorithms such as ANNs, CNNs, and the like. As shown in FIG. 1A , a 3D functional image 102 (such as a PET or SPECT image) is received 106 and used as input to a machine learning module 110 . Figure 1A shows an example PET image obtained using PyL ^™ as the radiopharmaceutical 102a. PET image 102a is shown overlaid on CT image (eg, as a PET/CT image), but machine learning module 110 may receive PET (eg, or other functional image) itself (eg, without CT, or other anatomical image) as input. In some embodiments, as described below, anatomical images may also be received as input. The machine learning module automatically detects and/or segments hot spots 120 that are determined (by the machine learning module) to represent potential cancerous lesions. An example image showing hot spots appearing in PET image 120b is also shown in FIG. 1A. Accordingly, the machine learning module generates one or both of (i) a hotspot list 130 and (ii) a hotspot map 132 as outputs. In some embodiments, the hotspot list identifies the locations (eg, centroids) of detected hotspots. In some embodiments, the heat map identifies the 3D volume of detected hot spots as determined through image segmentation performed by the machine learning module 110 and/or delineates the 3D boundaries of the detected hot spots. A list of hot spots and/or heat maps (eg, to other software modules) may be stored and/or provided for display and/or further processing 140.

In certain embodiments, machine learning based lesion detection algorithms may be trained not only on functional image information (eg, from PET images) but also on anatomical information, and such machine learning based lesion detection algorithms may utilize not only Functional image information (eg, from PET images) and utilize anatomical information. For example, in some embodiments, two channels (a first channel corresponding to a portion of a PET image and a second channel corresponding to a portion of the CT image) may be trained for one or more of lesion detection and segmentation A machine learning module, and the one or more machine learning modules can receive the two channels as input. In some embodiments, information derived from anatomical (eg, CT) images may also be used as input to a machine learning module for lesion detection and/or segmentation. For example, in some embodiments, 3D segmentation maps that identify various tissue regions within anatomical and/or functional images may also be used (eg, by one or more machine learning modules receiving such 3D segmentation maps as input, For example, as a separate input channel) to provide anatomical context.

1B shows an example procedure 100b in which both 3D anatomical images 104 (such as CT or MR images) and 3D functional images 102 are received 108 and used as input to a machine learning module 112, which is based on Information (eg, voxel intensities) from both the 3D anatomical image and the 3D functional image 102 performs hot spot detection and/or segmentation 122, as described herein. Hotspot list 130 and/or heatmap 132 may be generated as output from a machine learning module and stored/provided for further processing (eg, graph re-layout for display, subsequent manipulation by other software modules, etc.) 140 .

In certain embodiments, automated lesion detection and analysis (eg, for inclusion in reports) includes three tasks: (i) detecting hot spots corresponding to lesions; (ii) segmenting detected hot spots (eg, to identify the 3D volume corresponding to each lesion within the functional image); (iii) classify the detected hot spots as having a high or low probability (eg, and therefore suitable or unsuitable) corresponding to the actual lesion within the subject included in the radiologist's report). In some embodiments, one or more machine learning modules may be used, for example, to accomplish these three tasks one after the other (eg, sequentially) or in combination. For example, in some embodiments, a first machine learning module is trained to detect hotspots and identify hotspot locations, a second machine learning module is trained to segment hotspots, and a third machine learning module is trained to, for example, use automated The information obtained by the other two machine learning modules classifies the detected hotspots.

For example, as shown in the example process 100c of FIG. 1C, a 3D functional image 102 may be received 106 and used as input to a first machine learning module 114 that performs automated hotspot detection. The first machine learning module 114 automatically detects one or more hot spots 124 in the 3D functional image and generates a hot spot list 130 as an output. The second machine learning module 116 may receive the hotspot list 130 along with the 3D functional image as input, and perform automated hotspot segmentation 126 to generate the heatmap 132 . Heatmap 132 and hotspot list 130 may be stored and/or provided for further processing 140, as previously described.

In some embodiments, a single machine learning module is trained to directly segment hotspots within an image (eg, a 3D functional image; eg, to generate a 3D heatmap identifying volumes corresponding to detected hotspots) to combine detection Two steps before detecting and segmenting hotspots. The detected hotspots can then be classified using a second machine learning module, eg, based on the previously determined segmented hotspots. In some embodiments, a single machine learning module can be trained to accomplish all three tasks—detection, segmentation, and classification—in a single step. iii. Lesion index value

In certain embodiments, lesion index values are calculated for detected hot spots to provide, for example, a measure of relative uptake within the corresponding solid lesion and/or the size of the corresponding solid lesion. In some embodiments, a reference value operation for a particular hot spot is based on (i) a measure of the intensity of the hot spot and (ii) a measure of intensity corresponding to one or more reference volumes (each corresponding to a particular reference tissue region) Lesion index value. For example, in some embodiments, the reference value includes an aortic reference value (also referred to as a blood pool reference) that measures the intensity within the aortic volume corresponding to a portion of the aorta and a measurement corresponding to the subject's The liver reference value for the intensity within the liver volume of the liver. In certain embodiments, the intensities of voxels in nuclear medicine images (eg, PET images) represent standard uptake values (SUVs) (eg, calibrated for injected radiopharmaceutical dose and/or patient weight), and the intensity of the hot spot The metric and/or reference value metric is the SUV value. The use of these reference values for calculating lesion index values is described in further detail, eg, in PCT/EP2020/050132, filed on January 6, 2020, which is incorporated herein by reference in its entirety.

In some embodiments, segmentation masks are used to identify specific reference volumes in, for example, PET images. For a particular reference volume, a segmentation mask identifying the reference volume may be obtained through segmentation of anatomical (eg, CT) images. For example, in some embodiments (eg, as described in PCT/EP2020/050132), segmentation of a 3D anatomical image may be performed to generate a segmentation map including a plurality of segmentation masks each identifying regions of interest for a particular tissue. Thus, one or more segmentation masks of the segmentation map generated in this way can be used to identify one or more reference volumes.

In some embodiments, to identify the voxels to be used to compute the reference volume corresponding to the reference value, the mask may be eroded a fixed distance (eg, at least one voxel) to establish an identification corresponding to a voxel that is completely within the reference tissue region. The reference organ mask for the reference volume of the solid region. For example, erosion distances of 3 mm and 9 mm can be used for aortic and liver reference volumes, respectively. Other erosion distances may also be used. Additional mask refinement may also be performed (eg, to select a particular, desired set of voxels for computing reference values), eg, as described below with respect to liver reference volumes.

Various measures of intensity within the reference volume can be used. For example, in some embodiments, a robust average of voxel intensities inside a reference volume (eg, as defined by the reference volume segmentation mask after erosion) may be determined to be within the interquartile range of voxel intensities The mean of the values (IQR _mean ). Other metrics (such as peak, maximum, median, etc.) may also be determined. In certain embodiments, the aortic reference value is determined as a robust average of SUVs from voxels inside the aortic mask. The robust mean is computed as the mean of the values within the interquartile range, IQR _mean .

In certain embodiments, a subset of voxels within the reference volume are selected to avoid influences from reference tissue regions that may have abnormally low radiopharmaceutical uptake. Although the automated segmentation techniques described and referenced herein can provide accurate contours (eg, identification) of image regions corresponding to specific tissue regions, areas of abnormally low uptake often exist in the liver, which should be calculated from reference values exclude. For example, liver reference values (eg, liver SUV values) can be calculated in order to avoid regions in the liver with very low tracer (radiopharmaceutical) activity that may occur, eg, attributable to tumors without tracer uptake Impact. In certain embodiments, to account for the effect of abnormally low uptake in a reference tissue region, the intensities of the voxels corresponding to the liver (eg, voxels within the identified liver reference volume) are calculated and analyzed for the liver reference value. The histogram also removes (eg, excludes) intensities if they form a second histogram peak of lower intensity, thereby only including intensities associated with higher intensity value peaks.

For example, for liver, the reference SUV can be computed as a two-component Gaussian of a histogram of SUVs fitted to voxels within the liver reference volume (eg, as identified by the liver segmentation mask, eg, after the erosion procedure described above) The mean SUV of the principal components in the mixture model (also referred to as "modes", eg, referred to as in "primary modes"). In certain embodiments, if the secondary component has an average SUV greater than the primary component, and the secondary component has a weight of at least 0.33, an error is returned and any reference value for the liver is not determined. In certain embodiments, if the secondary component has a greater mean than the primary peak, the liver reference mask remains as-is. Otherwise, compute the separation SUV threshold value. In certain embodiments, the separation threshold is defined such that the probability of belonging to a major component of an SUV at the threshold or greater and the probability of belonging to a minor component of an SUV at the separation threshold or less same. The reference liver mask is then refined by removing voxels with SUVs less than the separation threshold. The liver reference value can then be determined as a measure of the intensity (eg, SUV) of the voxels identified by the liver reference mask, eg, as described herein for the aortic reference. 2A depicts an example liver reference operation showing a histogram of liver SUV values with Gaussian mixture components shown in red (major component 244 and minor component 246) and separation threshold 242 marked in green .

Figure 2B shows the resulting portion of the liver volume used to calculate the liver reference value, with voxels corresponding to lower value peaks excluded from the reference value calculation. Contours 252a and 252b of the refined liver volume mask are shown on each image in FIG. 2B, with voxels corresponding to lower value peaks (eg, having intensities below the separation threshold 242) excluded. As shown in the figure, regions of lower intensity towards the base of the liver, as well as regions near the edges of the liver, have been excluded.

2C shows an example procedure 200 in which a multi-component mixture model is used to avoid effects from regions with low tracer uptake, as described herein for liver reference volume calculations. The procedures shown in FIG. 2C and described herein with respect to the liver can also be similarly applied to compute intensity measures of regions of interest for other organs and tissues, such as the aorta (eg, aortic portions such as thoracic aortic portions or abdominal aorta), parotid gland, gluteal muscle. As shown, in FIG. 2C and described herein, in a first step, a 3D functional image 202 is received, and a reference volume corresponding to a particular reference tissue region (eg, liver, aorta, parotid gland) is identified 208 therein. The multicomponent mixture model 210 is then fitted to the distribution intensities (eg, intensity histograms) of a reference volume (eg, within), and the dominant mode 212 of the mixture model is identified. A measure of the intensity associated with the major mode (eg, and excluding proportions from intensities associated with other, minor modes) is determined 214 and used as a reference intensity value for the identified reference volume. In certain embodiments, as described herein, a measure of intensity associated with a dominant mode is determined by identifying a separation threshold such that intensities above the separation threshold are determined to be associated with a dominant mode , and intensities below the separation threshold were determined to be associated with secondary modes. Voxels with intensities above the separation threshold are used to determine the reference intensity value, while voxels with intensities below the separation threshold are excluded from the reference intensity value calculation.

In certain embodiments, detecting 216 hotspots and the reference intensity value determined in this way can be used, for example, via methods such as PCT/US2019/012486 filed on Jan. 7, 2019 and PCT filed on Jan. 6, 2020 /EP2020/050132, the entire contents of which are incorporated herein by reference, to determine the lesion index value 218 of the detected hot spots. iv. Inhibition of intensity exudation associated with normal uptake in high uptake organs

In certain embodiments, the intensities of voxels of the functional image are adjusted in order to suppress/correct the intensity exudation associated with specific organs where high uptake normally occurs. This method can be used, for example, on organs such as the kidney, liver, and urinary bladder. In certain embodiments, correcting for intensity exudation associated with multiple organs is performed one organ at a time in a stepwise fashion. For example, in certain embodiments, renal uptake is inhibited first, followed by inhibition of hepatic uptake, followed by inhibition of urinary bladder uptake. Thus, the input to liver suppression is the image in which renal uptake has been corrected (eg, and the input to bladder suppression is the image in which kidney and liver uptake have been corrected).

3 shows an example procedure 300 for correcting for intensity exudation from areas of high uptake tissue. As shown in FIG. 3, a 3D functional image is received 304 and a high intensity volume corresponding to a high uptake tissue region is identified 306. In another step, inhibition volumes outside the high intensity volume are identified 308. In certain embodiments, as described herein, the suppression volume may be determined to be the volume enclosing an area outside the high-intensity volume but within a predetermined distance from the high-intensity volume. In another step, the decision 310 is determined 310, eg, by assigning voxels within the high-intensity volume intensities based on intensity determinations outside the high-intensity volume (eg, within the suppression volume), eg, via interpolation (eg, using convolutions). Background image. In another step, the estimated image is determined 312 by subtracting the background image from the 3D functional image (eg, via voxel-wise intensity subtraction). In another step, a decision 314 suppresses the map. As described herein, in certain embodiments, the estimated image is used to determine the suppression map by extrapolating the intensity values of voxels within the high-intensity volume to locations outside the high-intensity volume. In some embodiments, the intensities are only extrapolated to locations within the suppression volume, and the intensities of voxels outside the suppression volume are set to zero. The suppression map is then used to adjust the intensity 316 of the 3D functional image, eg, by subtracting the suppression map from the 3D functional image (eg, performing voxel-wise intensity subtraction).

An example procedure for suppressing/correcting for intensity exudation from specific organs (in some embodiments, the kidneys are treated together) for PET/CT composite images is as follows: 1. Adjust the projected CT organ mask segmentation to the high-intensity area of the PET image to handle PET/CT misalignment. If the PET-adjusted organ mask is less than 10 pixels, the organ is not suppressed. 2. Compute the "background image" to replace all high uptakes with interpolated background uptakes within the attenuation distance from the PET adjusted organ mask. This is done using the spin of a Gaussian core. 3. Calculate the intensity that should be considered when estimating suppression as the difference between the input PET and the background image. This "estimated image" has high intensity inside the given organ and zero intensity at locations more than the attenuation distance from the given organ. 4. Estimate the suppression map from the estimated image using the exponential model. The suppression map is non-zero only in regions within the attenuation distance of the PET-adjusted organ segmentation. 5. Subtract the suppression map from the original PET image.

As described above, these five steps can be repeated in a sequential manner for each of the collections of multiple organs. v. Anatomical markers of detected lesions

In certain embodiments, detected hot spots, for example, are automatically assigned anatomical markers that identify specific anatomical regions and/or groups of regions where the lesions represented by the detected markers are determined to be located. For example, as shown in the example process 400 of FIG. 4, a 3D-enabled image can be received 404 and used, for example, to automatically detect hot spots 406 via any of the methods described herein. Once the hotspots are detected, the anatomical classification of each hotspot can be automatically determined 408 and each hotspot can be marked with the determined anatomical classification. Automated anatomical marking can be performed, for example, using automatically determined locations of detected hot spots in conjunction with anatomical information provided by, for example, identifying 3D segmentation maps and/or anatomical images of image regions corresponding to particular tissue regions. Hot spots and anatomical markers for each hot spot can be stored and/or provided for further processing 410 .

For example, detected hotspots can be automatically classified into the following five categories: ● T (prostate tumor) ● N (pelvic lymph nodes) ● Ma (non-pelvic lymph) ● Mb (bone metastases) ● Mc (soft tissue metastases not located in prostate or lymph nodes)

Table 1 below lists the organizational areas associated with each of the five categories. Hotspots corresponding to locations within each of the organizational areas associated with a particular category may be automatically assigned to that category accordingly. Table 1. List of tissue regions corresponding to the five categories in the lesion anatomical labeling method Bone Mb Lymph node Ma Pelvic lymph nodes N Prostate T soft tissue Mc skull cervix template right prostate brain Thorax supraclavicular template left neck lumbar spine armpit Presacral lung thoracic mediastinum other, pelvis esophagus pelvis hilum liver limbs mesentery gallbladder elbow spleen popliteal pancreas Peri-aortic/para-aortic adrenal glands Other, non-pelvic kidney bladder skin muscle other vi. Graphical User Interface and Quality Control and Reporting

In certain embodiments, detected hotspots and associated information (such as computed lesion index values and anatomical landmarks) are displayed with an interactive graphical user interface (GUI) to allow medical professionals (such as physicians, radiologists, physicians, technicians, etc.) to check. Thus, a medical professional can use the GUI to view and confirm the accuracy of detected hot spots and corresponding index values and/or anatomical landmarks. In some embodiments, the GUI may also allow the user to identify and segment (eg, manually) additional hot spots within the medical image, thereby allowing the medical professional to identify additional potential lesions that he/she believes may be missed by the automated detection process. Once identified, lesion index values and/or anatomical landmarks may also be determined for these manually identified and segmented lesions. For example, as indicated in FIG. 5B, the user may view the location determined for each hot spot, as well as anatomical landmarks, such as (eg, automatically determined) miTNM classifications. The miTNM classification scheme is described in further detail in "Prostate Cancer Molecular Imaging Standardized Evaluation (PROMISE): Proposed miTNM Classification for the Interpretation of Eiber et al., J. Nucl. Med. Vol. 59, pp. 469-78 (2018)) PSMA-Ligand PET/CT", which is incorporated by reference in its entirety. Once users are satisfied with the collection of detected hotspots and the information computed from them, they can confirm their approval and generate a final, signed report that can be viewed and used for discussion with patients Outcomes and diagnosis, and assess prognosis and treatment options.

For example, as shown in FIG. 5A, in an example process 500 for interactive hotspot viewing and detection, a 3D-enabled image is received 504 and automatically, for example, using any of the automated detection methods described herein Detect 506 hotspots. A set 508 of automatically detected hotspots is graphically represented and listed within the interactive GUI for viewing by the user. The user may select at least some (eg, up to all) of the automatically detected hotspots for inclusion 510 in a final set of hotspots, which may then be used for further calculations 512 (eg, to determine risk to the patient) index value.

5B shows an example workflow 520 for user viewing of detected lesions and lesion index values for quality control and reporting. This example workflow allows users of segmented lesions to view liver and aorta segmentations for use in calculating lesion index values, as described herein. For example, in a first step, the user reviews the quality 522 of the image (eg, CT image) and the accuracy 524 of the automated segmentation for obtaining liver and blood pool (eg, aorta) reference values. As shown in Figures 6A and 6B, the GUI allows the user to evaluate the image and segment the overlay to ensure that the automated segmentation of the liver (602, purple in Figure 6A) is within healthy liver tissue and the automated segmentation of the blood pool (aortic portion) 604, shown as salmon in Figure 6B) within the aorta and left ventricle.

In another step 526, the user validates the automatically detected hotspots and/or identifies additional hotspots, eg, to create a final set of hotspots corresponding to lesions for inclusion in the generated report. As shown in Figure 6C, the user may select automatically identified hotspots (eg, as overlays and/or overlays on PET and/or CT images) by hovering the mouse over the graphical representation of the hotspot displayed within the GUI. / or marked area). To facilitate hotspot selection, the particular hotspot selected may be indicated to the user via a color change (eg, turning green). The user can then click on the hotspot to select it, which can be visually confirmed to the user via another color change. For example, as shown in Figure 6C, after selection, the hot spot turns pink. Following user selection, quantitatively determined values (such as lesion index and/or anatomical landmarks) may be displayed to the user to allow them to verify the automatically determined values 528 .

In some embodiments, the GUI allows the user to select hotspots from a set of (automatically) pre-identified hotspots to confirm that they indeed represent lesions 526a and also to identify additional hotspots 526b corresponding to lesions that have not been automatically detected.

As shown in Figures 6D and 6E, a user may use a GUI tool to draw on tiles of an image (eg, a PET image and/or a CT image; eg, a PET image overlaid on a CT image) to mark labels corresponding to New, manually identified areas of lesions. Quantitative information, such as lesion index and/or anatomical markers, may be determined automatically for manually identified lesions, or may be entered manually by the user.

In another step, for example, once the user selects and/or manually identifies all lesions, the GUI displays a quality control checklist for the user to review 530, as shown in FIG. Once the user has viewed and completed the checklist, they can click "Create Report" to sign and generate the final report 532. An example of the generated report is shown in FIG. 8 . C. Example Machine Learning Network Architecture for Lesion Segmentation i. Machine Learning Module Input and Architecture

Referring to FIG. 9, which shows an example hotspot detection and segmentation process 900 in some embodiments, hotspot detection and/or segmentation by a machine learning model that receives functional images 902 and anatomical images 904 and segmentation maps 906 as input Group 908 performs, segmentation map 906 provides, for example, segmentation of various tissue regions such as soft tissue and bone and various organs as described herein.

The functional image 902 may be a PET image. As described herein, the intensity voxels of functional image 902 may be scaled to represent SUV values. In certain embodiments, other functional images as described herein may also be used. The anatomical image 904 may be a CT image. In some embodiments, the voxel intensities of the CT image 904 are scaled to represent Hounsfield units. In certain embodiments, other anatomical images as described herein may be used.

In some embodiments, the machine learning module 908 implements a machine learning algorithm using the U-Net architecture. In some embodiments, the machine learning module 908 implements a machine learning algorithm using a Feature Pyramid Network (FPN) architecture. In some embodiments, various other machine learning architectures may be used to detect and/or segment lesions. In certain embodiments, a machine learning module as described herein performs semantic segmentation. In certain embodiments, a machine learning module as described herein performs instance segmentation, eg, to distinguish one lesion from another.

In some embodiments, the three-dimensional segmentation map 906 received as input by the machine learning module identifies as corresponding to a particular organ such as (eg, prostate, liver, aorta, bladder, various other organs described herein, etc.) and Various volumes (eg, via a plurality of 3D segmentation masks) in the received 3D anatomical and/or functional images of regions of interest for a particular tissue of bone. Additionally or alternatively, the machine learning module may receive a 3D segmentation map 906 that identifies groups of tissue regions. For example, in some embodiments, a 3D segmentation map that identifies soft tissue regions, bone, and then background regions may be used. In some embodiments, the 3D segmentation map can identify groups of high uptake organs in which high levels of radiopharmaceutical uptake occur. For example, a high uptake organ group can include liver, spleen, kidney, and urinary bladder. In some embodiments, the 3D segmentation map identifies groups of high uptake organs as well as one or more other organs, such as the aorta (eg, low uptake soft tissue organs). Other organizational area groups may also be used.

The functional images, anatomical images, and segmentation map inputs to the machine learning module 908 can be of various sizes and dimensions. For example, in some embodiments, each of the functional image, anatomical image, and segmentation map is a patch of a three-dimensional image (eg, represented by a three-dimensional matrix). In some embodiments, each of the patches is the same size, eg, a [32 x 32 x 32] or [64 x 64 x 64] patch of each input line voxel.

The machine learning module 908 segments the hotspots and generates a 3D heatmap 910 that identifies one or more hotspot volumes. For example, the 3D heat map 910 may include one or more masks of the same size as one or more of the functional image, anatomical image, or segmentation map inputs and identifying one or more hot spot volumes. In this way, the 3D heat map 910 can be used to identify volumes within functional images, anatomical images, or segmentation maps that correspond to hot spots and thus solid lesions.

In some embodiments, the machine learning module 908 segments the hotspot volume to distinguish background (ie, non-hotspot) regions from the hotspot volume. For example, the machine learning module 908 may be a binary classifier that classifies voxels as background or belonging to a single hotspot class. Thus, the machine learning module 908 can generate as output a class-agnostic (eg, or "single-class") 3D heatmap that identifies hotspot volumes but does not distinguish between the different anatomical locations a particular hotspot volume can represent and /or type of lesion (eg, bone metastases, lymph nodes, localized prostate). In some embodiments, the machine learning module 908 segments the hotspot volume and also classifies the hotspots according to a plurality of hotspot categories, each hotspot category representing a particular anatomical location and/or lesion type represented by the hotspot. In this manner, the machine learning module 908 can directly generate a multi-class 3D heat map that identifies one or more hotspot volumes and labels each hotspot volume as belonging to a particular one of the plurality of hotspot classes. For example, detected hot spots can be classified as bone metastases, lymph nodes, or prostate lesions. In some embodiments, other soft tissue classifications may be included.

In addition to or instead of classifying hotspots according to their likelihood to represent true lesions, this classification may be performed, as described herein in, for example, Section B.ii. ii. Lesion classification post-processing and / or output

10A and 10B, in some embodiments, after hot spot detection and/or segmentation in an image by one or more machine learning modules (with solid lesions as hot spots appearing, for example, in functional images) Understanding that the terms "lesion" and "hot spot" are used interchangeably in Figures 9-12), post-processing 1000 is performed to mark hot spots as belonging to a particular hot spot category. For example, detected hot spots can be classified as bone metastases, lymph nodes, or prostate lesions. In some embodiments, the labeling scheme in Table 1 can be used. In some embodiments, this labeling may be performed by a machine learning module, which may be the same machine learning module used to perform segmentation and/or detection of hotspots, or may be performed individually or in conjunction with other Inputs (such as 3D functional images, 3D anatomical images, and/or segmentation maps, as described herein) are received along with a list of detected hot spots (eg, identifying their locations) and/or a 3D heat map (eg, depicting as The splitting module that takes the hotspot boundary determined by splitting) as input. As shown in FIG. 10B , in some embodiments, the segmentation map 906 used as input to the machine learning module 908 to perform lesion detection and/or segmentation can also be used to classify lesions, eg, according to anatomical location. In some embodiments, other (eg, different) segmentation maps may be used (eg, not necessarily the same segmentation maps that are fed into the machine learning module as input). iii. Parallel organ-specific lesion detection module

Referring to FIGS. 11A and 11B , in some embodiments, one or more machine learning modules include one or more organ-specific modules that perform detection and/or segmentation of hotspots located in corresponding organs. For example, as shown in the example procedures 1100 and 1120 of FIGS. 11A and 11B , respectively, the prostate module 1108a may be used to perform detection and/or segmentation in the prostate region. In some embodiments, one or more organ-specific modules are used in conjunction with a whole-body module 1108b that detects and/or segments hot spots throughout the subject's entire body. In some embodiments, results 1110a from one or more organ-specific modules are combined with results 1110b from the whole body module to form a final hotspot list and/or hotspot map 1112. In some embodiments, merging may include combining results (eg, hotspot lists and/or 3D heatmaps) 1110a and 1110b with other outputs (such as 3D heatmaps 1114 created by segmenting hotspots using other methods, which may include using other machine learning modules and/or techniques and other segmentation methods) combinations. In some embodiments, after hotspots are detected and/or segmented by one or more machine learning modules, additional segmentation methods may be performed. This additional segmentation step may use, for example, hotspot segmentation and/or detection results obtained from one or more machine learning modules as input. In certain embodiments, as shown in FIG. 11B , an analytical segmentation method 1122 as described herein, eg, in Section C.iv. below, may be used in conjunction with an organ-specific lesion detection module. Analytical segmentation 1122 uses results 1110b and 1110a from upstream machine learning modules 1108b and 1108a, and PET image 1102 to segment hot spots using analytical segmentation techniques (eg, which do not utilize machine learning) and create an analytically segmented 3D heat map 1124 . iv. Analytical segmentation

Referring to Figure 12, in some embodiments, machine learning techniques may be used to perform hot spot detection and/or initial segmentation, and, for example, as a subsequent step, an analytical model is used to perform final segmentation of each hot spot.

As used herein, the terms "analytical model" and "analytical segmentation" refer to segmentation methods based on (eg, using) predetermined rules and/or functions (eg, mathematical functions). For example, in some embodiments, the analysis segmentation method may use one or more predetermined rules (such as an ordered sequence of image processing steps, applying one or more mathematical functions to the image, conditional logic branches, and the like) to segment hot spots . Analytical segmentation methods may include, but are not limited to, bound-based methods (eg, including an image bounding step), level-setting methods (eg, fast-moving methods), graph cutting methods (eg, watershed segmentation), or active contour models . In some embodiments, the analysis segmentation method does not rely on a training step. In contrast, in some embodiments, the machine learning model will use instances that have been automatically trained to use a set of training data (eg, including, for example, images and hotspots manually segmented by, for example, a radiologist or other practitioner). ) models pre-segmented hotspots to segment hotspots and are designed to mimic the segmentation behavior in the training set.

It may be advantageous to use an analytical segmentation model to determine the final segmentation, for example, because the analytical model may be easier to understand and debug than machine learning methods in certain situations. In some embodiments, these analytical segmentation methods may operate on 3D functional images in conjunction with lesion segmentation generated by machine learning techniques.

For example, as shown in FIG. 12, in an example procedure 1200 for hot spot segmentation using an analytical model, a machine learning module 1208 receives as input a PET image 1202, a CT image 1204, and a segmentation map 1206. The machine learning module 1208 performs segmentation to create a 3D heat map 1210 that identifies one or more hot spot volumes. The analytical segmentation model 1212 uses the 3D heatmap 1210 generated by the machine learning module, and the PET image 1202 to perform segmentation and create a 3D heatmap 1214 that identifies the analyzed segmented hotspot volume. v. Illustrative Hotspot Segmentation

13A and 13B show examples of machine learning module architectures for hotspot detection and/or segmentation. Figure 13A shows an example U-net architecture (the "N=" in parentheses in Figure 13A identifies the number of filters in each layer) and Figure 13B shows an example FPN architecture. 13C shows another example FPN architecture.

14A-14C show example results of hotspot segmentation obtained using a machine learning module implementing the U-Net architecture. Crosshairs and bright spots in the image indicate segmented hot spots 1402 (representing potential lesions). 15A and 15B show example hotspot segmentation results obtained using a machine learning module implementing FPN. In particular, Figure 15A shows an input PET image superimposed on a CT image. 15B shows an example heat map overlaid on a CT image determined using a machine learning module implementing FPN. The overlaid heat map can show the hot spot volume 1502 near the subject's spine in dark red. D. Example Graphical User Interface

In certain embodiments, the lesion detection, segmentation, classification, and related techniques described herein may include facilitating user interaction (eg, with software programs implementing the various methods described herein) and/or result viewing GUI. For example, in some embodiments, GUI portions and windows, among other things, allow users to upload and manage data to be analyzed, visualize images and results generated through the methods described herein, and generate reports summarizing findings. Screenshots of certain example GUI views are shown in Figures 16A-16E.

For example, Figure 16A shows providing for uploading and viewing of studies by a user [eg, image data collected during the same exam and/or scan (eg, according to the Digital Imaging and Communications in Medicine (DICOM) standard), such as collected via PET/CT scans Example GUI windows for PET images and CT images]. In some embodiments, uploaded studies are automatically added to a patient list listing the identifiers of subjects/patients who have uploaded one or more PET/CT images. For each item in the patient list shown in Figure 16, the patient ID and available PET/CT studies for that patient, along with the corresponding report, are displayed. In some embodiments, the team concept allows for the creation of groups (eg, teams) of multiple users who work on a particular subset of uploaded data and are provided access to that particular subset of uploaded data. In some embodiments, a patient list may be associated with and automatically shared with a particular team to provide members of the team access to the patient list.

16B shows an example GUI viewer 1610 that allows a user to view medical image data. In some embodiments, the viewer is a multimodal viewer to allow the user to view multiple imaging modalities, as well as various formats and/or combinations thereof. For example, the viewer shown in Figure 16B allows the user to view PET and/or CT images, and fusions (eg, overlays) thereof. In some embodiments, the viewer allows the user to view 3D medical image data in various formats. For example, the viewer may allow the user to select and view various 2D tiles along a particular (eg, selected) cross-sectional plane of the 3D image. In some embodiments, the viewer allows the user to view the maximum intensity projection (MIP) of the 3D image data. Other ways of visualizing 3D image data may also be provided. In this example, as shown in Figure 16B, a control panel GUI toolset 1612 is provided on the left hand side of the viewer and allows the user to view available study information (such as dates, various patient data and imaging parameters, etc.).

Referring to FIG. 16C, in some embodiments, the GUI viewer includes a lesion selection tool that allows a user to select a location for an image that the user identifies and selects as likely to represent a real underlying solid lesion, for example. Lesion volume of volume of interest (VOI). In certain embodiments, the lesion volume is selected from a collection of hot spot volumes that are automatically identified and segmented, eg, via any of the methods described herein. Selected lesion volumes can be saved for inclusion in a final set of identified lesion volumes that can be used for reporting and/or further quantitative analysis. In certain embodiments, eg, as shown in FIG. 16C , based on user selection of a particular lesion volume, various characteristic/quantitative measures of a particular lesion are displayed 1614 [eg, maximum intensity, peak intensity, mean intensity, volume, lesion Index (LI), anatomical classification (eg, miTNM class, location, etc.), etc.].

Referring to Figure 16D, in addition or alternatively, a GUI viewer may allow a user to view the results of automated segmentation performed in accordance with various embodiments described herein. Segmentation can be performed via automated analysis of CT images as described herein, and can include identification and segmentation of 3D volumes representing the liver and/or aorta. The segmentation results can be overlaid on representations of medical image data, such as CT and/or PET image representations.

16E shows an example report 1620 generated through analysis of medical image data as described herein. In this example, the report 1620 summarizes the results of the viewed study and provides characteristics and quantitative measures that characterize the selected (eg, by the user) lesion volume 1622 . For example, as shown in Figure 16E, for each selected lesion volume, the report includes lesion ID, lesion type (eg, miTNM classification), lesion location, SUV max, SUV peak, SUV mean, volume, and lesion index value. E. Hotspot segmentation and classification using multiple machine learning modules

In some embodiments, multiple machine learning modules are used in parallel to segment and classify hotspots. 17A is a block flow diagram of an example process 1700 for segmenting and classifying hotspots. The example procedure 1700 performs image segmentation on a 3D PET/CT image to segment hot spot volumes and categorize each segmented hot spot volume as a lymphoid, bone or prostate hot spot in particular according to the (automatically) determined anatomical location.

The example procedure 1700 receives 3D PET image 1702 and 3D CT image 1704 as input and operates on them, among others. The CT image 1704 is input to a first, organ segmentation machine learning module 1706, which performs segmentation to identify regions in the CT image that represent a particular tissue region and/or an organ or organs of interest (eg, related) 3D volumes of anatomical groups of tissue regions and/or organs. Thus, the organ segmentation machine learning module 1706 is used to generate a 3D segmentation map 1708 that identifies specific tissue regions and/or anatomical groups of organs of interest or the like within the CT image. For example, in some embodiments, the segmentation map 1708 identifies two volumes of interest corresponding to two anatomical groups of organs, one volume of interest corresponding to the high uptake soft tissue organs including liver, spleen, kidney, and urinary bladder Anatomical groups, and the second volume of interest corresponds to the aorta (eg, thoracic and abdominal portions) as a low-uptake soft tissue organ. In some embodiments, the organ segmentation machine learning module 1706 generates an initial segmentation map that identifies various individual organs, including those individual organs that make up the anatomical groups of the segmentation map 1708, and in some embodiments other individual organs As output, a segmentation map 1708 is built from the initial segmentation map (eg, by assigning the same labels to volumes of individual organs corresponding to the anatomical group). Thus, in certain embodiments, the 3D segmentation map 1708 uses three markers that identify and differentiate between (i) voxels belonging to high uptake soft tissue organs; (ii) low uptake soft tissue organs, i.e. , the aorta; and (iii) other regions as background.

In the example procedure 1700 shown in Figure 17A, the organ segmentation machine learning module 1706 implements a U-Net architecture. Other architectures (eg, FPN) may be used. PET image 1702, CT image 1704 and 3D segmentation map 1708 are used as input to two parallel hotspot segmentation modules.

In some embodiments, the example process 1700 uses two machine learning modules in parallel to segment and classify hotspots in different ways, and then combine their equal results. For example, it was found that the machine learning module performed more accurate segmentation when it only identified a single hotspot category, e.g. whether an image region was identified as a hotspot, rather than multiple—lymphoid, bone, prostate—desired hotspot categories. Therefore, the procedure 1700 utilizes the first, single-class hotspot segmentation module 1712 to perform accurate segmentation and the second, multi-class hotspot segmentation module 1714 to classify the hotspots into the three desired classes.

In particular, the first, single-class hot spot segmentation module 1712 performs segmentation to generate a first, single-class 3D heat map 1716 that identifies the 3D volume representing the lesion (while other image regions are identified as the background). Therefore, the single-class hotspot segmentation module 1712 performs binary classification to label image voxels as belonging to one of two classes (background or single hotspot class). Second, the multi-class hotspot segmentation module 1714 segments hotspots and assigns one of a plurality of hotspot classification labels to the segmented hotspot volume, as opposed to using a single hotspot class. In particular, the multi-class hotspot segmentation module 1714 classifies the segmented hotspot volumes as lymphoid, bone, or prostate hotspots. Thus, the multi-class hotspot segmentation module generates a second, multi-class 3D heat map 1718 that identifies 3D volumes representing hotspots and labels them as lymphoid, bone or prostate (while other image regions are identified as background). In procedure 1700, the single-class hotspot segmentation module and the multi-class hotspot segmentation module each implement the FPN architecture. Other machine learning architectures (eg, U-Net) may be used.

In some embodiments, to generate a final 3D heat map of segmented and classified hot spots 1724, a single-class heat map 1716 and a multi-class heat map 1718 are combined 1722. In particular, each hotspot volume of the single-class heatmap 1716 is compared to the hotspot volumes of the multi-class heatmap 1718 to identify matching hotspot volumes that represent the same physical location and thus the same (potential) physical lesion. Matching hotspot volumes may be identified, for example, based on various measures of spatial overlap (eg, percent volume overlap), proximity (eg, center of gravity within a threshold distance), and the like. The single-class heat map 1716 for which the matching hot-spot volumes from the multi-class heat map 1718 are identified is assigned the label of the matching hot-spot volume—lymph, bone, or prostate. In this way, hotspots are accurately segmented via single-class hotspot segmentation module 1712 and then marked using the results of multi-class hotspot segmentation module 1714.

Referring to Figure 17B, in certain cases, for a particular hotspot volume of single-class heatmap 1716, no matching hotspot volume from multi-class heatmap 1718 was found. These hot spot volumes are marked based on a comparison with the 3D segmentation map 1738, which may be different from the segmentation map 1708, identifying 3D volumes corresponding to lymphatic and bone regions.

In some embodiments, the single-class hotspot segmentation module 1712 may not segment hotspots in the prostate region, such that the single-class heatmap does not include any hotspots in the prostate region. Hotspot volumes from multi-class heatmap 1718 marked as prostate hotspots may be used for inclusion in merged heatmap 1724. In some embodiments, the single-class hotspot segmentation module 1712 may segment some hotspots in the prostate region, but additional hotspots (eg, not identified in the single-class heatmap 1716 ) may be segmented by the multi-class hotspot segmentation module 1714 And it is identified as a prostate hot spot by the multi-class hot spot segmentation module 1714 . These additional hotspot volumes present in multi-class heatmap 1718 may be included in merged heatmap 1724.

Thus, in some embodiments, information from CT image 1704, PET image 1702, 3D organ segmentation map 1738, single-class heat map 1716, and multi-class heat map 1718 are used in the hotspot merging step 1722 to generate segmented and classified The merged 3D heat map of the hot spot volume 1724.

In one example merging method, overlap (eg, between two hotspot volumes) is determined when any two voxels from the hotspot volumes from the multi-class heatmap and the single-class heatmap correspond to/represent the same physical location . If a particular hotspot volume of a single-class heatmap overlaps with only one hotspot volume of a multi-class heatmap (eg, only one matching hotspot volume from a multiclass heatmap is identified), then the overlapping hotspot volume from the multiclass heatmap is identified as belonging to category to label that specific hotspot volume of the single-category heatmap. If a particular hotspot volume overlaps with two or more hotspot volumes of the multiclass heatmap (each identified as belonging to a different hotspot class), then each voxel of the single-class hotspot volume is assigned to overlap with the one from the multiclass heatmap The closest voxel-identical class in the hotspot volume. If a specific hotspot volume of a single-class heatmap does not overlap any hotspot volume of a multi-class heatmap, then assign a hotspot class to that specific hotspot volume based on comparison with a 3D segmentation map identifying soft tissue regions (eg, organs) and/or bone . For example, in some embodiments, a particular hot spot volume may be marked as belonging to the bone category if any of the following statements are true: (i) if more than 20% of the hot spot volume overlaps the rib segmentation; (ii) if the hot spot The volume does not overlap any marker in the organ segmentation and the mean value of the CT in the hot spot mask is greater than 100 Heinz units; (iii) if the location of the SUV _max of the hot spot volume overlaps with the bone marker in the organ segmentation; or (iv) If more than 50% of the hot spot volume overlaps with bone markers in organ segmentation.

In some embodiments, a particular hotspot volume may be identified as lymph if 50% or more of the hotspot volume does not overlap with bone markers in the organ segmentation.

In some embodiments, when all hotspot volumes of the single-class heatmap are classified as lymphoid, bone, or prostate, any remaining prostate hotspots from the multi-class model are overlaid onto the single-class heatmap and included in the merged heatmap .

Figure 17C shows an example computer program 1750 for implementing a hotspot segmentation and classification method according to the embodiment described with reference to Figures 17A and 17B. F. Analytical segmentation via adaptive bounding methods

In certain embodiments, eg, as described herein in Section C.iv, the image analysis techniques described herein utilize an analysis segmentation step to refine hot spots determined by a machine learning module as described herein segmentation. For example, in some embodiments, a 3D heat map generated by machine learning methods as described herein is used as an initial input to an analytical segmentation model that refines and/or performs a completely new segmentation.

In some embodiments, the analysis segmentation model utilizes a defined algorithm whereby anatomical images (eg, CT images, MR images) and/or functional images (eg, SPECT images, PET images) (eg, synthetic The intensity of the voxels in anatomical and functional images, such as PET/CT or SPECT/CT images, and one or more thresholds to segment hot spots.

Referring to Figure 18A, in some embodiments, an adaptive bounding method whereby, for a particular hotspot, the intensity within an initial hotspot volume determined for that particular hotspot, eg, via machine learning methods as described herein, is compared with One or more reference values to determine threshold values for a particular hotspot. This threshold value for the particular hot spot is then used by analyzing the segmentation model to segment the particular hot spot and determine the final hot spot volume.

18A shows an example procedure 1800 for segmenting hotspots via an adaptive bounding method. Process 1800 utilizes an initial 3D heat map 1802, a PET image 1804, and a 3D organ segmentation map 1806 that identify one or more 3D hotspot volumes. The initial 3D heat map 1802 may be determined automatically via various machine learning methods described herein and/or based on user interaction with the GUI. The user may refine the set of automatically determined hotspot volumes, eg, by selecting a subset for inclusion in the 3D heatmap 1802. Additionally or alternatively, the user may manually determine the 3D hotspot volume, eg, by drawing boundaries on the image with the GUI.

In certain embodiments, the 3D organ segmentation map identifies one or more reference volumes corresponding to a particular reference tissue region, such as aortic portions and/or liver. As described herein, for example, in Section B.iii, the intensities of voxels within a particular reference volume can be used to compute associated reference values 1808, and the intensities of identified and segmented hot spots can be compared to those reference values (eg, , acting as a "measuring ruler"). For example, the liver volume can be used to compute the liver reference, and the aortic portion used to compute the aorta or blood pool reference. In routine 1800, the strength of the aortic portion is used to calculate 1808 the blood pool reference value 1810. The blood pool reference value 1810 is used in conjunction with the initial 3D heat map 1802 and the PET image 1804 to determine threshold values for performing bound-based analysis segmentation of the hot spots in the initial 3D heat map 1802.

In particular, for a specific hotspot volume identified in the initial 3D heat map 1802 (which identifies a specific hotspot representing a solid lesion), the intensities of the PET image 1804 voxels located within the specific hotspot volume are used to determine the specificity of the hotspot. Hot spot strength. In some embodiments, the hot spot intensity is the maximum value of the intensity of the voxels located within a particular hot spot volume. For example, for a PET image intensity representing an SUV, the maximum SUV ( _SUVmax ) within a particular hot spot volume is determined. Other measures may be used, such as peak (eg, SUV _peak ), mean, median, interquartile mean (IQR _mean ).

In some embodiments, a hotspot-specific threshold value for a specific hotspot is determined based on a comparison of the hotspot intensity to a blood pool reference value. In some embodiments, the comparison between the hot spot intensity and the blood pool reference value is used to select one of a plurality of (eg, pre-defined) bounding functions and employ the selected bounding function to operate for a particular hotspot The hotspot specific threshold value. In some embodiments, the bounded function computes a hotspot-specific threshold value as a function of the hotspot's intensity (eg, maximum intensity) and/or blood pool reference value for the specific hotspot. For example, a bounding function may operate a hotspot-specific threshold value as the product of (i) a scaling factor and (ii) the hotspot intensity (or other intensity measure) and/or blood pool reference for the specific hotspot. In some embodiments, the scaling factor is a constant. In some embodiments, the scaling factor is determined as an interpolation of a function of the intensity measure of a particular hot spot. In certain embodiments and/or for a particular bounded function, the scaling factor is a constant used to determine the level of the flat line region corresponding to the maximum threshold value, eg, as described in further detail in Section G herein.

For example, pseudocode for an example method of selecting (eg, via conditional logic) and operating various bounded functions is shown below: If 90% of SUVmax ≤ [blood pool reference], then threshold= 90% of SUVmax . Else, if 50% of SUVmax ≥ 2 x [blood pool reference], then threshold = 2 x [blood pool reference]. Else, use linear interpolation to determine a percentage of SUVmax, with the interpolation starting at 90% at [[ blood pool reference] / 0.9] and ending at 50% at [2 x [blood pool reference] / 0.5]. If [interpolated percentage] x SUVmax is below 2 x [blood pool reference], then threshold = [interpolated percentage] x SUVmax. Else, threshold = 2 x [blood pool reference].

18B and 18C illustrate certain example adaptive bounding methods implemented by the above pseudo-code. FIG. 18B plots the threshold value 1832 as a function of the hotspot intensity (in this example, _SUVmax ) of a particular hotspot. Figure 18C plots the variation of the hotspot specific threshold as a ratio of the _SUVmax for a specific hotspot as a function of the _SUVmax for that specific hotspot. The dashed lines in each graph indicate specific values relative to the blood pool reference (with an SUV of 1.5 in the example plots of Figures 18B and 18C), and also indicate 90% and 50% of the SUV _max in Figure 18C.

Referring to Figures 18D-18F, the adaptive thresholding method as described herein addresses the challenges and disadvantages associated with previous thresholding techniques that utilize fixed or relative threshold values. In particular, while limit-based lesion segmentation based on maximum standard uptake value ( _SUVmax ) in certain embodiments provides a transparent and reproducible method for segmenting hotspot volumes for estimating parameters such as uptake volume and SUV _mean . However, the conventional fixed and relative threshold values do not work well under the full dynamic range of the lesion SUV _max . The fixed threshold method uses a single (eg, user-defined) SUV value as the threshold for segmenting hot spots within an image. For example, the user may set the fixed threshold level at a value of 4.5. The relative threshold method uses a specific, constant fraction or percentage, and splits the hotspots using a local threshold for each hotspot set at a specific fraction or percentage of the hotspot's maximum SUV. For example, the user may set the relative threshold value at 40% so that each hotspot is segmented using a threshold value calculated to be 40% of the maximum hotspot SUV value. There are drawbacks to these two approaches (the conventional fixed and relative thresholds). For example, it is difficult to define an appropriate fixed threshold value that applies to all patients. There are also problems with the conventional relative threshold method, because defining the threshold as a fixed fraction of the hotspot maximum or peak intensity results in using a lower threshold to segment hotspots with lower overall intensities. Thus, segmentation using a low threshold value can represent low-intensity hotspots of smaller lesions with relatively low uptake, resulting in a larger volume of identified hotspots than higher-intensity hotspots that actually represent physically larger lesions.

For example, Figures 18D and 18E illustrate splitting two hotspots using a threshold value determined to be 50% of the maximum hotspot intensity (eg, 50% _SUVmax ). Each graph plots intensity as a function of location in a vertical direction, showing lines passing through hotspots. FIG. 18D shows a graph 1840 depicting the variation in intensity of high-intensity hot spots representing large solid lesions 1848. FIG. The hotspot intensity 1842 peaks near the center of the hotspot and the hotspot threshold 1844 is set to 50% of the maximum value of the hotspot intensity 1842 . Segmenting the hotspot using the hotspot threshold 1844 yields a segmented volume that approximately matches the size of the solid lesion (as shown, eg, by comparing the linear dimension 1846 to the depicted lesion 1848). FIG. 18E shows a graph 1850 depicting the variation in intensity of low-intensity hot spots representing small solid lesions 1858. FIG. The hotspot intensity 1852 also peaks near the center of the hotspot and the hotspot threshold 1854 is also set to 50% of the maximum hotspot intensity 1852. However, since the peak of the hotspot intensity 1852 is less sharp and has a lower intensity peak than the hotspot intensity 1842 of the high intensity hotspot, setting the threshold relative to the maximum value of the hotspot intensity results in a much lower absolute threshold value. Thus, bound-based segmentation results in a larger hotspot volume than the hotspot volume of higher intensity hotspots, although the solid lesions represented are smaller, as shown, for example, by comparing the linear dimension 1856 with the depicted Lesion 1858. Therefore, the relative threshold value can produce larger apparent hot spot volumes for smaller solid lesions. This is especially problematic for assessing treatment response, as lower intensity lesions will have lower threshold values and, therefore, lesions that respond to treatment may appear to increase in volume.

In certain embodiments, adaptive thresholding as described herein addresses these drawbacks by utilizing an adaptive threshold value computed as a percentage of hotspot intensity, the percentage (i) varies with hotspot intensity (eg, SUV _max ) increases and decreases and (ii) depends on both hot spot intensity (eg, SUV _max ) and overall physiological uptake (eg, as measured by a reference such as a blood pool reference). Thus, unlike conventional relative bounding methods, the specific fraction/percentage change in hotspot intensity used in the adaptive bounding method described herein is itself a function of hotspot intensity and in some embodiments Physiological uptake is also considered. For example, as shown in the illustrative plot 1860 of FIG. 18F, the threshold value 1864 is set to a higher percentage of the peak hot spot intensity 1852 using a variable, adaptive thresholding method as described herein (eg, as shown in FIG. 18F ). shown in 90%). As depicted in Figure 18F, doing so allows for the identification of hotspot volumes that more accurately reflect the true size of the lesion 1866 represented by the hotspot based on constrained segmentation.

In certain embodiments, confinement is facilitated by first dividing heterogeneous lesions into homogeneous subcomponents using a watershed algorithm, and finally excluding uptake from nearby intensity peaks. As described herein, adaptive delimitation can be applied to manually pre-segmented lesions and automated detection, for example, by deep neural networks implemented through machine learning modules as described herein, to improve reproducibility Reality and robustness and increased interpretability. G. Example Study Comparing Example Defining Functions and Scaling Factors for PYL-PET/CT Imaging

This example describes what was performed to evaluate various parameters for use in adaptive thresholding methods as described herein, for example, in Section F, and to compare fixed and relative thresholds using manually annotated lesions as references Research.

The study of this example used ¹⁸ F-DCFPyL PET/CT scans of 242 patients, with hot spots corresponding to bone, lymphoid and prostate lesions manually segmented by an experienced nuclear medicine reader. A total of 792 hotspot volumes were annotated, involving 167 patients. Two studies were performed to evaluate the bounded algorithm. In the first study, manually annotated hotspots were refined with different bounded algorithms and the degree to which the order of size was preserved, ie, the extent to which smaller hotspot volumes remained smaller than initially larger hotspot volumes after refinement, were estimated. In a second study, refinement by defining suspicious hotspots automatically detected by machine learning methods according to various embodiments described herein was performed and compared to manual annotation.

The PET image intensities in this example are scaled to represent standard uptake values (SUVs) and are referred to in this section as uptake or uptake intensity. The different bounding algorithms compared are as follows: a fixed threshold at SUV=2.5, a relative threshold of 50% of SUV _max , and a variant of an adaptive threshold. The adaptive threshold is defined by the percent drop in SUV _max (with and without the maximum threshold level). The flat line zone level was set higher than the normal uptake intensity in the zone corresponding to healthy tissue. Two supporting investigation studies were performed to select the appropriate horizon level: one study selected normal uptake intensities in the aorta, and one study selected normal uptake intensities in the prostate. Among other things, delimitation methods were evaluated based on their order of magnitude preserved compared to annotations performed by nuclear medicine readers. For example, if a nuclear medicine reader manually segments hotspots, and the manually segmented hotspot volumes are ordered by size, then the preservation of size order refers to those generated by segmenting the same hotspot using an automated bound method (eg, which does not include user interaction) The extent to which the hotspot volumes will be sorted in the same way according to their equal size. According to a weighted rank correlation measure, two embodiments of the adaptive bound method achieve the best performance in terms of size order preservation. Both of these adaptive delimiting methods utilize approaches that start at 90% of the SUV _max of low-intensity lesions and reach the plateau at twice the blood pool reference value (eg, 2 x [aortic reference uptake]). limit. The first method (referred to as "P9050-sat") reached the flat-line area at the flat-line area level at 50% of SUV _max , and the other method (referred to as "P9040-sat") was in the flat-line area of SUV _max 40% of the time to reach the flat line area.

It was also found that using bounded refinement of automatically detected and segmented hot spots alters the precision-recall trade-off. While raw, automatically detected and segmented hotspots have high recall and low precision, refining the segmentation using the P9050-sat bound method yields more balanced performance in terms of precision and recall.

The improved relative size retention indicates that the assessment of treatment response will be improved/more accurate because the algorithm better captures the size order of the nuclear medicine readers' annotations. The trade-off between over-segmentation and under-segmentation can be decoupled from the self-detection step by introducing a separation-defining approach, i.e., in addition to automated hotspot detection and segmentation methods performed using machine learning methods as described herein , also using analytical, adaptive segmentation methods as described herein.

The example supporting studies described herein were used to determine the scaling factor used to compute the flat line region value corresponding to the maximum threshold value. For example, as described herein, in this example, these scaling factors are determined based on intensities in normal, healthy tissue in various reference regions. For example, multiplying a blood pool reference based on intensity in the aortic region by a factor of 1.6 yields levels typically higher than 95% of intensity values in the aorta but lower than typical normal uptake in the prostate. Therefore, in certain example bound functions, higher values are used. In particular, a factor of 2 was determined to achieve a level of intensity that is also generally higher than most intensities in normal prostate tissue. This value was determined manually based on a survey of histograms and image projections in the sagittal, coronal and transverse planes of PET image voxels within the prostate volume, excluding any portion corresponding to tumor uptake. Example image tiles and corresponding histograms showing scaling factors are shown in Figure 18G. i. Introduction

Defining lesion volume in PET/CT can be a subjective procedure, as lesions appear as hot spots in PET and often do not have well-defined boundaries. Sometimes the lesion volume can be segmented based on its anatomical extent in the CT image, however, this approach would result in ignoring specific information about tracer uptake as complete uptake would not be covered. In addition, certain lesions can be seen on functional, PET images, but not on CT images. This section describes an example study designed to accurately identify hot spot volumes that reflect physiological uptake volumes (ie, volumes where uptake is above background). To perform segmentation and identify hotspot volumes in this manner, threshold values are chosen to balance the risk of including background with the risk of not segmenting a sufficiently large hotspot volume to reflect the full ingested volume.

This risk trade-off is usually addressed by choosing 50% or 40% of the SUV _max value determined for the expected hot spot volume as the threshold value. The rationale for this approach is that for high-uptake lesions (eg, corresponding to high-intensity hotspots in PET images), the threshold can be set higher than for low-uptake lesions (eg, corresponding to lower-intensity hotspots), At the same time, the same risk level as the hot spot volume is maintained for the volume representing the complete intake volume without segmentation. However, for low signal-to-noise ratio hotspots, using a threshold value of 50% of SUV _max will result in the inclusion of background in the segmentation. To avoid this, a percentage drop in SUV _max can be used, eg, starting at, eg, 90% or 75% for low intensity hot spots. Furthermore, the risk of including background is low once the threshold value is sufficiently above the background level, which occurs for threshold values well below 50% of the SUV _max for high uptake lesions. Therefore, the threshold value can be above the flat line level of typical background intensity as an upper limit.

Mean liver uptake was used as a reference system for uptake intensity levels well above typical background uptake intensity levels. Other reference levels may be required based on the actual background uptake intensity. Background uptake intensity differed in bone, lymph and prostate, with bone having the lowest background uptake intensity and prostate having the highest background uptake intensity. Using the same qualifying method regardless of tissue is advantageous/preferable because it allows the same segmentation method to be used regardless of the location and/or classification of a particular lesion. Therefore, the study of this example used the same delimiting parameters to assess threshold values for lesions in all three tissue types. The adaptive delimiting variants evaluated in this example included a variant that reached a plateau in hepatic uptake intensity, a variant that reached a plateau at a level estimated to be higher than aortic uptake, and a Several variants were estimated to reach the plateau above the level of prostate uptake intensity.

Certain prior methods have determined the level to be mediastinal blood calculated as the mean of blood pool uptake intensity plus twice the standard deviation of blood pool uptake intensity (eg, mean blood pool uptake intensity + 2 x SD). A function of pool uptake intensity. However, this approach, which relies on standard deviation estimates, can lead to undesired errors and noise susceptibility. In particular, the estimated standard deviation is much less robust than the estimated mean, and can be affected by noise, small segmentation errors, or PET/CT misalignment. A more robust way to estimate levels above blood uptake intensity uses a fixed factor multiplied by the mean or reference aortic value. To find suitable factors, the distribution of uptake intensity in the aorta was studied and described in this example. Normal prostate uptake intensity is also studied to determine appropriate factors that can be applied to reference aortic uptake to calculate levels generally higher than normal prostate intensity. ii. The method is limited to manual annotations

This study used a subset of data that contained only lesions in which at least one other lesion of the same type was in the same patient. This resulted in a dataset with 684 manually segmented lesion uptake volumes (278 in bone, 357 in lymph nodes, and 49 in prostate) involving 92 patients. Perform automatic refinement by bounds and compare the output to the original volume. Performance is measured as a weighted average of the rank correlation between refined volumes and raw volumes within a patient and tissue type, where the weight is given by the number of segmented hotspot volumes in that patient. This efficacy metric indicates whether relative sizes between segmented hotspot volumes are preserved, but absolute sizes (which etc. are subjectively defined, since ingested volumes do not have clear boundaries). However, for a particular patient and tissue type, the same nuclear medicine reader makes all the annotations, and it can therefore be assumed that these annotations are done in a systematic manner, where smaller lesion annotations actually reflect smaller uptake volumes than larger lesion annotations. Defining automatically detected lesions

This study used a subset of data that had not yet been used to train a machine learning module for hotspot detection and segmentation, resulting in 285 manually segmented lesion uptake volumes involving 67 patients (104 in bone, 129 in lymph, A dataset of prostate 52). Precision and recall (sensitivity 90% to 91% for bone and 92% to 93% for lymph) were measured between automatically detected volumes matching the refined (and unrefined) of manually segmented lesions %, and 94% to 98% for the prostate). This equivalent energy quantifies the similarity between automatically detected and possibly refined hotspots and manually annotated hotspots. blood uptake

For 242 patients, a deep learning pipeline was used to segment the thoracic portion of the aorta in the CT component. The segmented aortic volume was projected into the PET space and eroded by 3 mm to minimize the risk of the aortic volume containing areas outside the aorta or in the vessel wall, while preserving as much uptake inside the aorta as possible. For residual uptake intensity, in each patient, the quotient q = (aortaMEAN + 2 x aortaSD) / aortaMEAN was calculated. Prostate uptake

In 29 patients, normal uptake in the prostate was studied. The study was performed using segmented prostate volumes determined by a machine learning module. Uptake intensities in manually annotated prostate lesions were excluded. Residual uptake intensities normalized to aortic reference uptake intensities were visualized by histograms and maximum intensity projections in the axial, sagittal and coronal planes, see example in Figure 18G. The purpose of the maximum projection is to find an explanation for the outlier intensities in the histogram, especially those associated with bladder uptake that are higher than the maximum uptake intensities in healthy tissue. Qualified method

Two baseline methods (fixed threshold at SUV=2.5 and relative threshold at 50% of SUV _max ) were compared with six variants of adaptive thresholds. The adaptive threshold is defined using three bounding functions each associated with a specific range of SUV _max values. Specifically: (1) Low range bound function: The first bound function is used to calculate the threshold value for the SUV _max value in the low range. The first bound function calculates the threshold value as a fixed (high) percentage of SUV _max , (2) Intermediate range bound function: The second bound function is used to calculate the approximate value of SUV _max for the intermediate range limit. The second bound function computes the threshold value as a linear percentage decrease of SUV _max , with the maximum threshold value equal to the threshold value of the upper limit of the range as the upper limit, and (3) high-range bound function: high-range bound The function is used to calculate the threshold value of the SUV _max value in the high range. The high range bound function sets the threshold value at a fixed maximum threshold value (saturation threshold value) or a fixed (low) percentage of SUV _max (non-saturation threshold value).

The exact parameters and ranges of the above three bounding functions vary between various adaptive bounding algorithms and are listed in Table 2 below. Table 2: Ranges and parameters of bounded functions for adaptive bounded algorithms. adaptive threshold low range The middle range * is the upper limit of the right end of the range high range P9050-sat 90% of SUV _max <aorta => 90% of SUV _max 90% SUV _max = aorta, to 50% SUV _max = 2 x aorta => Interp. perc. of SUV _max * 50% SUV _max > 2 x aorta => 2 x aorta P9040-sat 90% of SUV _max <aorta => 90% of SUV _max 90% SUV _max = aorta, to 40% SUV _max = 2 x aorta => Interp. perc. of SUV _max * 40% SUV _max > 2 x aorta => 2 x aorta P7540-sat 75% of SUV _max < aorta => 75% of SUV _max 75% SUV _max = aorta, to 40% SUV _max = 2 x aorta => Interp. perc. of SUV _max * 40% SUV _max > 2 x aorta => 2 x aorta P9050-non-sat 90% of SUV _max <aorta => 90% of SUV _max 90% SUV _max = aorta, to 50% SUV _max = 2 x aorta => Interp. perc. of SUV _max * 50% SUV _max > 2 x aorta => 50 % of SUV _max A9050-sat 90% of SUV _max <aorta => 90% of SUV _max 90% SUV _max = aorta, to 50% SUV _max = 1.6 x aorta => Interp. perc. of SUV _max * 50% SUV _max > 1.6 x aorta => 1.6 x aorta L9050s-sat 90% of SUV _max <aorta => 90% of SUV _max 90% SUV _max = aorta, to 50% SUV _max = liver => Interp. perc. of SUV _max * 50% SUV _max > Liver => Liver

， 其中SUV_high 之50%等於2 x [主動脈攝取強度]，且SUV_low 之90%等於主動脈攝取強度。The interpolated percentage used in the intermediate SUV _max range for the P9050-sat operation as follows:

, where 50% of SUV _high equals 2 x [aortic uptake intensity] and 90% of SUV _low equals aortic uptake intensity.

且對於其他自適應定限演算法類似地。iii. 結果 對經手動註解之病變定限 Interpolated percentages used in other adaptive bounding algorithms are computed similarly. The threshold value in the middle range is then:

And similarly for other adaptive finite algorithms. iii. Results are limited to manually annotated lesions

The highest weighted rank correlation (0.81) was obtained by the P9050-sat and 9040-sat methods, with P7540-sat, A9050-sat and L9050-sat also providing high values. Relative 50% of SUV _max (0.37) and P9050-non-sat (0.61) resulted in the lowest weighted rank correlation. A fixed threshold at SUV=2.5 resulted in a rank correlation between (0.74) lower than most adaptive bound methods. The weighted rank correlation results for each of the qualifying methods are summarized in Table 3 below. Table 3. Weighted Means of Rank Correlation for Evaluated Delimitation Methods limit strategy weighted average of rank correlations Fixed, SUV = 2.5 0.74 Relative 50% of SUV _max 0.37 P9050-sat 0.81 P9040-sat 0.81 P7540-sat 0.79 P9050-non-sat 0.61 A9050-sat 0.80 L9050s-sat 0.78 Defining automatically detected lesions

Without refinement, automatic hotspot detection has low precision (0.31 to 0.47) but high recall (0.83 to 0.92), indicating over-segmentation. Refinement using a bounded algorithm relative to 50% of SUV _max improves precision (0.70 to 0.77), but reduces recall to about 50% (0.44 to 0.58). Refinement with P9050-sat improves precision (0.51 to 0.84) with less drop in recall (0.61 to 0.89) to indicate a balance of less over-segmentation but more under-segmentation. P9040-sat performs similarly to P9050-sat in these respects, while L9050-sat has the highest precision (0.85 to 0.95) but the lowest recall (0.31 to 0.56). Table 4a to Table 4e show the complete results for precision and recall. Table 4a. Precision and recall values without analysis segmentation refinement No refinement Accuracy recall _ bone hot spots 0.38 0.92 lymphatic hotspots 0.47 0.83 Prostate Hotspot 0.31 0.93 Table 4b. Precision and recall values with refinement via relative 50% bound method of SUV _max Relative 50% of SUV _max Accuracy recall _ bone hot spots 0.74 0.58 lymphatic hotspots 0.77 0.44 Prostate Hotspot 0.70 0.51 Table 4c. Precision and recall values with adaptive segmentation using the P9050-sat implementation P9050-sat Accuracy recall bone hot spots 0.84 0.61 lymphatic hotspots 0.70 0.67 Prostate Hotspot 0.51 0.89 Table 4d. Precision and recall values with adaptive segmentation using the P9040-sat implementation P9040-sat Accuracy recall bone hot spots 0.84 0.59 lymphatic hotspots 0.71 0.66 Prostate Hotspot 0.52 0.89 Table 4e. Precision and recall values with adaptive segmentation using the L9050-sat implementation L9050-sat Accuracy recall bone hot spots 0.95 0.31 lymphatic hotspots 0.91 0.39 Prostate Hotspot 0.85 0.56 Support limited method: blood intake

For the resulting quotient, qMEAN + 2 x qSD is 1.54, so using a factor of 1.6 was determined to be a good candidate for achieving a threshold level above most blood uptake intensity values. In the example study, only three patients had aortaMEAN + 2 x aortaSD above 1.6 x aortaMEAN. Three outlier patients had q=1.64, 1.92, and 1.61, where the patient with a factor of 1.92 had erroneous aortic segmentation spilling into the spleen, and the others had quotients close to 1.6. Qualified Method of Support: Prostate Uptake

Based on a manual review taking into account the histogram of normal prostate intensities and projections in the axial, sagittal and coronal planes, a value of 2.0 would be the level to be applied to the aortic reference to obtain higher than typical uptake intensities in the prostate appropriate scaling factor for . H. Example: Using AI -Based Hotspot Segmentation vs. Using Only Throttling

In this example, hotspot detection and segmentation performed using an AI-based approach to segmenting and classifying hotspots utilizing machine learning modules as described herein is compared to conventional methods that utilize only threshold-based segmentation.

19A shows a conventional hotspot segmentation method 1900 that does not utilize machine learning techniques. Instead, hot spot segmentation is performed by the user based on a manual delineation of the hot spot, followed by intensity (eg, SUV) based qualification 1904 . The user manually masks placement 1922 a circular marker indicating a region of interest (ROI) 1924 within the image 1920. Once the ROI is placed, a fixed or relatively defined approach can be used to segment 1926 hot spots within the manually placed ROI. Individually, relative bounding methods set the threshold for a particular ROI as a fixed percentage of the maximum SUV within the ROI, and SUV-based bounding methods are used to segment each user-identified hotspot to refine the initial used boundaries drawn by the author. Since this conventional method relies on the user to manually identify and draw the boundaries of the hotspots, it can be time-consuming, and furthermore, the segmentation results and downstream quantification 1906 (eg, computation of hotspot metrics) can vary from user to user. Furthermore, as conceptually depicted in images 1928 and 1930, different thresholds may result in different hotspot segmentations 1929, 1931, depending on the particular threshold. Furthermore, although the SUV threshold level can be tuned to detect early disease, doing so often results in a large number of false positives that distract from true positives. Finally, for example, as explained herein, conventional fixed or relative SUV based bounding methods suffer from overestimation and/or underestimation of lesion size.

Referring to Figure 19B, an AI-based method 1950 according to certain embodiments described herein utilizes one or more machine learning modules to automatically analyze (eg, synthesize PET/CT) CT 1954 and PET images 1952 to detect hot spots Detect, segment, and classify 1956 instead of using manual, user-based selection combined with SUV-based delimitation of ROIs containing hotspots. As described in further detail herein, machine learning-based hotspot segmentation and classification can be used to create an initial 3D heat map, which can then be used to analyze segmentation methods 1958 (such as described herein, for example, in Sections F and G) Adaptive bounding technique) input. Among other things, using machine learning methods reduces user subjectivity and time required to view images (eg, by a medical practitioner, such as a radiologist). In addition, the AI model is capable of performing complex tasks and can identify early-stage lesions as well as high-burden metastatic disease, while keeping false-positive rates low. Improved hotspot segmentation in this way increases the accuracy of downstream quantifications 1960 associated with measures that can be used to assess disease severity, prognosis, treatment response, and the like.

Figure 20 demonstrates the improved performance of the machine learning based segmentation method compared to conventional bound methods. In a machine learning-based approach, hotspots are detected and segmented by first using a machine learning module as described herein (eg, in Section E), and using the adaptive bounding described in implementing Sections F and G The analytical model of the technical version is refined to perform hot spot segmentation. Conventional bound methods are performed using a fixed bound to segment clusters of voxels with intensities above a fixed threshold. As shown in Figure 20, while the conventional bounding method produced false positives 2002a and 2002b due to uptake of radiopharmaceuticals in the urethra, the machine learning segmentation technique correctly ignored urethral uptake and only segmented prostate lesions 2004 and 2006.

Figures 21A-21I compare the results of hotspot segmentation in the abdominal region performed by a conventional bounding method (left-hand image) with the results of hotspot segmentation (right-hand image) according to a machine learning method according to some embodiments described herein . Figures 21A-21I show a series of 2D tiles of a 3D image moving in a vertical direction in the abdominal region, where the hotspot regions identified by each method are superimposed. The results presented in the figures demonstrate that abdominal uptake is a problem with conventional limit-of-limit methods, where a large false-positive area appears in the left-hand side image. This may be caused by large uptake in the kidneys and bladder. Conventional segmentation methods require sophisticated methods to suppress this uptake and limit these false positives. In contrast, the machine learning model shown in Figures 21A-21I for segmenting images does not rely on any such suppression, and instead learns to ignore such uptake. I. Example CAD Device Implementations

This section describes example CAD device implementations according to certain embodiments described herein. The CAD device described in this example is called "aPROMISE" and uses a number of machine learning modules to perform automated organ segmentation. An example CAD device implementation uses an analytical model to perform hot spot detection and segmentation.

The example implementation of aPROMISE ( Automated Prostate-Specific Membrane Antigen Imaging Segmentation ) described in this example utilizes a cloud-based software platform with a web interface where users can upload PSMA in the form of a DICOM file Body scans of PET/CT imaging data, review patient studies and share study assessments within the team. The software is compliant with the Digital Imaging and Communications in Medicine (DICOM) 3 standard. Multiple scans can be uploaded for each patient, and the system provides individual viewing of each study. The software includes a GUI that provides a viewing page that displays and allows the user to view the study in a 4-panel view showing PET, CT, PET/CT fusion, and maximum intensity projection (MIP) simultaneously, and includes the option to display each view separately . The device is used to view the entire patient study to enable the user to identify and mark regions of interest (ROI) using image visualization and analysis tools. When viewing image data, the user can freely select by pre-defined hotspots that highlight when hovering over the segmented area with the mouse pointer, or by manually drawing (ie, selecting individual voxels in the image tile to include as the hotspot) to mark the ROI. Quantitative analysis is performed automatically on selected or (manually) drawn hotspots. The user can view the results of this quantitative analysis and decide which hot spots should be reported as suspicious lesions. In aPROMISE, a region of interest (ROI) is defined as a contiguous subsection of the image; hotspots are defined as ROIs with high local intensity (eg, indicative of high uptake) (eg, relative to surrounding regions), and lesions are defined as User-defined or user-selected ROI for suspected disease.

To create the report, the software of the example implementation requires the signing user to confirm quality control and electronically sign the report preview. The signed report is saved on the device and can be exported as a JPG or DICOM file.

aPROMISE devices are implemented in a microservices architecture, as described in further detail herein and shown in Figures 29A and 29B. i. Workflow

Figure 22 depicts the workflow of an aPROMISE device from uploading a DICOM file to exporting an electronically signed report. When logged in, users can import DICOM files into aPROMISE. The imported DICOM file is uploaded to the patient list, where the user can click on the patient to display the corresponding study available for viewing. The layout principle of the patient list is shown in FIG. 23 .

This view 2300 lists all patients within the team that have uploaded studies and displays patient information (name, ID, and gender), date of last study upload, and study status. Study status indicates whether each patient's study is ready for review (blue symbols, 2302), studies with errors (red symbols, 2304), calculated studies (orange symbols, 2306), and studies with available reports (black symbols) symbol, 2308). The number in the upper right corner of the status symbol indicates the number of studies with a particular status for each patient. Viewing of the study was initiated by clicking on the patient, selecting the study, and identifying whether the patient had undergone prostatectomy. Research data will be made public and displayed in the viewing window.

FIG. 24 shows a viewing window 2400 in which the user can review PET/CT image data. Lesions are manually marked and reported by the user, either from predefined hotspots segmented by software, or from user-defined ones made by using a drawing tool for selecting voxels for inclusion in the program as hotspots hotspots to choose from. Predefined hot spots (regions of interest with high local intensity uptake) are automatically segmented using specific methods for soft tissue (prostate and lymph nodes) and bone, and are highlighted when hovering over segmented regions with the mouse pointer. The user can choose to turn on the split display option to visually present the splits of the predefined hotspots at the same time. The selected or plotted hot spots are the subject of automated quantitative analysis and are detailed in panels 2402, 2422, and 2442.

The retractable panel 2402 on the left summarizes patient and study information extracted from DICOM data. Panel 2402 also displays and lists quantitative information about hotspots selected by the user. Manual verification of hotspot location and type—T: localized in primary tumor; N: regional metastatic disease; Ma/b/c: distant metastatic disease (lymph nodes, bone, and soft tissue). The device displays automated quantitative analysis—SUV max, SUV peak, SUV mean, lesion volume, lesion index (LI) for user-selected hotspots to allow the user to view and decide which hotspots are reported as lesions in a standardized report.

Middle panel 2422 contains a four-panel view display of DICOM image data. The CT image is shown in the upper left corner, the PET/CT fusion view is shown in the upper right corner, the PET image is shown in the lower left corner and the MIP is shown in the lower right corner.

MIP is a visualization method for volumetric data that displays 2D projections of 3D image volumes from various perspectives. The MIP imaging system is described in Wallis JW, Miller TR, Lerner CA, Kleerup EC. 3D Display in Nuclear Medicine. IEEE Trans Med Imaging. 1989;8(4):297-30. doi: 10.1109/42.41482. PMID: 18230529.

The retractable right panel 2442 includes the following visual controls for optimizing image viewing and their shortcut keys for manipulating the image for viewing purposes:

Viewport: ● presence or absence of reticle ● Fading option for PET/CT fusion images ● Which standard nuclear medicine colormap to choose to visualize PET tracer uptake intensity.

SUV and CT windows: ● Windowing the image, also known as contrast expansion, histogram modification, or contrast enhancement, where the image is manipulated by intensity to change the appearance of the image to highlight specific structures. ● In SUV window, the open window preset for SUV intensity can be adjusted by slider or shortcut key. ● In the CT window, the window preset for Heinz Intensity can be selected from the drop-down list using shortcut keys or by clicking and dragging the input, where the brightness of the image is adjusted by the window level and the contrast is by the window width Adjustment.

segmentation ● To enable or disable the visualization of reference organ segmentation or whole-body segmentation visualization of organ segmentation display options. ● User can select which panel views to display organ segmentation. ● Hotspot split display option to turn on or off the presentation of predefined hotspots in the selected area; pelvis area, in bone or all hotspots.

Viewer Gestures ● Shortcut keys and combinations for zooming, panning, changing blocks and hiding the hotspots of the viewing window in the CT window.

To create a report, the signature user clicks the create report button 2462. Before the report will be created, the user must confirm the following quality control items: ● Image quality is acceptable ● PET and CT images are properly aligned ● Patient study data are correct ● Reference values (blood pool, liver) are acceptable ● The study is not a super bone scan

After confirming the quality control items, a preview of the report is displayed for the user to electronically sign. The report contains a patient overview, total quantitative lesion burden, and quantitative assessment of individual lesions from user-selected hotspots for user confirmation as lesions.

FIG. 25 shows an example generated report 2500. Report 2500 includes three sections 2502, 2522, and 2542.

Section 2502 of report 2500 provides an overview of patient data obtained from the DICOM tag. It contains: a summary of the patient; patient name, patient ID, age and weight, and a summary of study data; study date, injected dose at the time of injection, radiopharmaceutical imaging tracer used and its half-life, and The time between the tracer and the acquisition of image data.

Section 2522 of report 2500 provides summarized quantitative information from hot spots selected by the user to be included as lesions. Summarized quantitative information showing total lesion burden per lesion type (primary prostate tumor (T), local/regional pelvic lymph nodes (N) and distant metastases - lymph nodes, bone or soft tissue organs (Ma/b/c)) . Summary section 2522 also shows the quantitative uptake (SUV mean) observed in the reference organ.

Section 2542 of report 2500 is a detailed quantitative assessment and location of each lesion from selected hotspots identified by the user. When viewing the report, the user must electronically sign his/her patient study review results (including selected hotspots and quantified as lesions). The report is then saved on the device and can be exported as a JPG or DICOM file. ii. Image processing to preprocess DICOM input data

Image input data is presented in DICOM format, which is a rich data representation. DICOM data includes strength data as well as meta data and communication structure. To optimize data for use by aPROMISE, data is passed through a microservice that re-encodes, compresses, and removes unnecessary or sensitive information. Intensity data is also collected from a separate DICOM series and encoded into a single lossless PNG file with an associated JSON metadata file.

Data processing of the PET image data includes estimating SUV (normalized uptake value) factors, which are included in the JSON metadata file. The SUV factor is a scalar used to convert image intensities into SUV values. The SUV factor was calculated according to the QIBA guidelines (Quantitative Imaging Biomarkers Consortium). Algorithmic Image Processing

FIG. 26 shows an example image processing workflow (procedure) 2600.

aPROMISE used a CNN (Circular Neural Network) model to segment 2602 patient bones and selected organs. Organ segmentation 2602 allows automated calculation of standard uptake value (SUV) references 2604 in the patient's aorta and liver. The SUV references for the aorta and liver were then used as reference values when determining certain quantitative indices based on SUV values, such as the Lesion Index (LI) and Intensity Weighted Tissue Lesion Volume (ITLV). A detailed description of the quantitative index is provided in Table 6 below.

Lesions are manually marked and reported 2608 by a user who selects 2608a from predefined hotspots segmented by software, or from a user made by using a drawing tool for selecting voxels for inclusion in the GUI as hotspots Defined hotspot to select 2608b. Predefined hot spots (regions of interest with high local intensity uptake) are automatically segmented using some specific method for soft tissue (prostate and lymph nodes) and bone (eg, as shown in Figure 28, a specific method for bone can be used. segmentation method and another specific segmentation method for soft tissue regions). Based on organ segmentation, the software determines the type and location of selected hotspots in the prostate, lymphatic or bony regions. The determined type and location are displayed in a list of selected hotspots shown in panel 2502 of viewer 2400. Types and locations of selected hotspots in other regions (eg, not located in prostate, lymphoid, or bone regions) are manually added by the user. The user may add and edit the types and locations of all hotspots at any time, if applicable, during the hotspot selection. Hot spot types were determined using the miTNM system, a clinical standard and notation system for reporting cancer spread. In this method, individual hotspots are assigned types according to letter-based codes that indicate specific entity characteristics as follows: ● T indicates primary tumor ● N indicates nearby lymph nodes affected by the primary tumor ● M indicates distant metastasis

For distant metastases, localization was grouped into a/b/c systems corresponding to additional pelvic lymph nodes (a), bone (b), and soft tissue organs (c).

For all hotspots selected to be included as lesions, 2610 SUV values and indices were calculated and displayed in the report. Organ segmentation in CT

Organ segmentation 2602 is performed using the CT image as input. Starting with two coarse segmentations from the full image, smaller image segments are extracted and selected to contain a given set of organs. Perform fine segmentation of organs on each image segment. Finally, all segmented organs from all image segments were assembled into the complete image segmentation displayed in aPROMISE. The successfully completed segmentation identified 52 different bones and 13 soft tissue organs as visualized in Figure 27 and presented in Table 5. Both the coarse segmentation procedure and the fine segmentation procedure consist of three steps: 1. Preprocessing CT images, 2. CNN segmentation, and 3. Post-processing segmentation.

Preprocessing CT images prior to coarse segmentation involves three steps: (1) removing image tiles that represent air only (eg, having <= 0 Heinz units); (2) resampling the image to a fixed size; and (3) ) normalized images based on the mean and standard deviation of the training data, as described below.

The CNN model performs semantic segmentation, where each pixel in the input image is assigned a label corresponding to the background or its segmented organ, resulting in a label map of the same size as the input data.

Post-processing is performed after segmentation and includes the following steps: - Absorb clusters of adjacent pixels at a time. - Absorb adjacent pixel clusters until no such cluster exists. - Remove all clusters that are not the largest of each marker. - Self-segmentation discards skeletal parts; some segmentation models segment skeletal parts as reference points when segmenting soft tissue. After the segmentation is complete, it is intended to remove the skeletal parts in these models.

Two different coarse segmentation neural networks and ten different fine segmentation neural networks were used, including segmentation of the prostate. Prostates were not segmented if the patient had undergone prostatectomy prior to examination (information provided by the user when verifying the patient's study background prior to opening the study for review). The combinations of fine and coarse segmentation and which body part each combination provides are presented in Table 5. Table 5: Overview of how to combine coarse and fine segmentation networks to segment different body parts Organ / Bone Coarse Segmentation Neural Network finely segmented neural network right lung Rough segmentation-02 Fine Segmentation - Right Lung - 01 left lung Rough segmentation-02 Fine Segmentation - Left Lung - 01 Left/Right Femur Left/Right Gluteus Maximus Rough segmentation-04 Thin Segmentation-Legs-01 Left/Right Hip Sacrum and Coccyx Urinary Bladder Rough segmentation-04 Subdivision-Pelvis-Non-Prostate-01 Liver Left/Right Kidney Gallbladder Spleen Rough segmentation-02 Fine Segmentation - Abdomen - 02 Right Ribs 1-12 Right Scapula Right Clavicle Rough segmentation-02 Fine segmentation - upper right body bone - 02 Left Rib-12 Left Scapula Left Clavicle Rough segmentation-02 Fine segmentation-left upper body bone-02 cervical spine Rough segmentation-02 Fine Segmentation-Spine-02 Thoracic vertebrae 1-12 Lumbar vertebrae 1-5 Sternal aorta, thoracic part of the aorta, abdominal part Rough segmentation-02 Fine Segmentation-Aorta-01 prostate* Rough segmentation-02 Subdivision - Pelvic Region - Mixed *Additional split network is only available for patients with prostate.

Training a CNN model involves an iterative minimization problem, where the training algorithm updates the model parameters to reduce segmentation errors. Segmentation error is defined as the deviation from perfect overlap between manual segmentation and CNN model segmentation. Each neural network used for organ segmentation is trained to configure optimal parameters and weights. As described above, the training data used to develop the neural network for aPROMISE consisted of low-dose CT images with manually segmented and labeled body parts. The CT images used to train the segmentation network were obtained as part of the NIMSA project (http://nimsa.se/) and are available at clinicaltrials.gov (https://www.clinicaltrials.gov/ct2/show/NCT01667536?term=99mTc -MIP-1404&draw=2&rank=5) during the Phase II clinical trial of the registered drug candidate 99mTc-MIP-1404. The NIMSA project consisted of 184 patients and the 99mTc-MIP-1404 profile consisted of 62 patients. Calculating the reference material in PSMA PET (SUV reference )

Reference values are used when assessing physiological uptake of PSMA tracers. Current clinical practice uses the SUV intensity in the identified volume corresponding to the blood pool or in the liver or both tissues as a reference value. For PSMA tracer intensity, blood pools were measured in the aortic volume.

In aPROMISE, the SUV intensity in the volume corresponding to the thoracic portion of the aorta and the liver was used as a reference value. The uptake shown in the PET images and the organ segmentation of the aortic and liver volumes were the basis for calculating the SUV references in the respective organs.

aorta . To ensure that the portion of the image corresponding to the vessel wall is not included in the volume used to calculate the SUV reference for the aortic region, the segmented aorta volume was reduced. The segmentation reduction (3 mm) was heuristically chosen to balance the trade-off of maintaining as much aortic volume as possible while excluding the vessel wall region. The reference SUV for the blood pool is a robust average of the SUVs from the pixels inside the reduced segmentation mask identifying the aortic volume. The robust mean is computed as the mean of the values within the interquartile range.

liver . When measuring the reference value in the liver volume, the segmentation is reduced along the edges to create a buffer to adjust for possible misalignment between the PET image and the CT image. The reduction (9 mm) was determined heuristically using manual observation of images with PET/CT misalignment.

Cysts or malignancies in the liver can lead to areas of low tracer uptake in the liver. To reduce the effect of these local differences in tracer uptake on the calculation of the SUV reference, a two-component Gaussian mixture model approach according to the embodiment described above in Section B.iii with respect to Figure 2A was used. In particular, a two-component Gaussian mixture model was fitted to voxels from inside the reference organ mask and the SUVs of the primary and secondary components of the identified distributions. The SUV reference for liver volume was initially calculated as the mean SUV of the principal components from a Gaussian mixture model. If the secondary component is judged to have an average SUV greater than the primary component, the liver reference organ mask remains unchanged unless the weight of the secondary component exceeds 0.33, in which case when the weight of the secondary component exceeds 0.33 At 0.33, an error is thrown and the liver reference will not be calculated.

If the minor component has an average SUV less than the major component, then a separation threshold is calculated, eg, as shown in Figure 2A. The separation threshold is defined such that: ● Probability of being a major component of an SUV at a threshold value or greater, and; ● Probability of being a minor component of an SUV at a threshold value or less; are the same.

The reference mask is then refined by removing pixels below the separation threshold. Predefinition of Hot Spots in PSMA PET

Referring to Figure 28, in the aPROMISE implementation of this example, segmentation by aPROMISE in PSMA PET is performed by an analysis model 2800 based on input from a PET image 2802 and an organ segmentation map 2804 determined from a CT image and projected in PET space. regions of high local intensity (so-called predefined hotspots). For the soft body to segment hot spots in bone, raw PET images are used 2802, and to segment hot spots in lymph and prostate, the PET images are processed by suppressing normal PET tracer uptake 2806. A graphical overview of the analytical model used in this example implementation is presented in FIG. 28 . The analysis method, as explained further below, was designed to find regions of high local uptake intensity that could represent the ROI without excessive irrelevant regions or PET tracer background noise. The analysis method was developed from a labeled dataset including PSMA PET/CT images.

Inhibition of normal PSMA tracer uptake intensity 2806 was performed at that time in a high uptake organ. First, the intensity of uptake in the kidney is suppressed, then the liver and finally the bladder. Suppression is performed by applying the estimated suppression map to high intensity regions of the PET. The suppression map was built using the organ map previously segmented in CT and projected and adjusted to the PET image, creating a PET adjusted organ mask.

Adjustment corrects small misalignments between PET images and CT images. Using the adjusted map, calculate the background image. This background image is subtracted from the original PET image and produces an uptake estimate image. An inhibition map is then estimated from the captured estimated image using an exponential function that depends on the Euclidean distance from the out-of-segmentation voxels to the PET-adjusted organ mask. Because the uptake intensity decreases exponentially with distance from the organ, an exponential function is used. Finally, the suppression map was subtracted from the original PET image, thereby suppressing the intensity associated with high normal uptake in the organ.

After suppressing normal PSMA tracer uptake intensity, the organ segmentation mask 2804 and the suppressed PET image 2808 produced by the suppression step 2806 are used to segment hot spots 2812 in the prostate and lymph. Prostate hot spots are not segmented for patients who have undergone prostatectomy. Bone and lymphatic hotspot segmentation is suitable for all patients. Each hot spot is segmented using the fast-travel method, where the underlying PET image is used as the velocity map and the volume of the input region determines the travel time. This input area is also used as an initial segmentation mask to identify the volume of interest for the fast-running method and is established differently depending on whether hot spot segmentation is performed in bone or soft tissue. Bone hot spots are segmented using a fast marching method and Difference of Gaussian (DoG) filtering method 2810 and lymph and, where applicable, prostate hot spots are segmented using a fast marching method and Laplacian of Gaussian (LoG) filtering method 2812.

For detection and segmentation of bone hotspots, a bone region mask is created to identify bone volumes in which bone hotspots can be detected. The Bone Region Mask includes the following bone regions: Thoracic (1-12), Lumbar (1-5), Clavicle (L+R), Scapula (L+R), Sternal Ribs (L+R, 1-12), Hip Bone (L+R), Femur (L+R), Sacrum and Coccyx. Normalizing the masked image based on the mean intensity of healthy bone tissue in the PET image was performed by iteratively normalizing the image using DoG filtering. The filter sizes used in the DoG are 3 mm pitch and 5 mm pitch. DoG filtering acts as a bandpass filter on the image, attenuating signals further away from the center of the frequency band, which emphasizes clusters of voxels with higher intensities relative to their surroundings. Confining the normalized image obtained in this way produces a voxel cluster that is distinguishable from the background and thus segmented, creating a 3D segmentation map 2814 that identifies the hot spot volume located in the bone region.

For the detection and segmentation of lymphatic hotspots, a lymphatic region mask is created in which hotspots corresponding to potential lymph nodes can be detected. The lymphatic region mask includes voxels within the bounding box enclosing all segmented bone and organ regions, but excludes voxels within the segmented organ itself (with the exception of lung volumes, which are preserved). Another, prostate region mask, is created in which hot spots corresponding to potential prostate tumors can be detected. This prostate region mask is a single voxel expansion of the prostate volume determined from the organ segmentation procedure described herein. Applying a lymphatic region mask to a PET image yields a masked image that includes voxels within the lymphatic region (eg, and excluding other voxels) and similarly, applying a prostate region mask to a PET image yields a masked image that includes voxels within the prostate volume. Masked image of voxels.

Soft tissue hotspots (ie, lymphatic and prostate hotspots) were obtained by applying three different sized LoG filters (one with 4 mm/spacing XYZ, one with 8 mm/spacing XYZ and one with 12 mm/spacing XYZ) to detect, resulting in three LoG filtered images for each of the two soft tissue types (prostate and lymph). For each soft tissue type, three corresponding Log-filtered images were bounded using a value of minus 70% of the aortic SUV reference and a local minimum was found using a 3x3x3 minimum filter. This method produces three filtered images that each include clusters of voxels corresponding to hotspots. The three filtered images are combined by taking the union of local minima from the three images to generate a hotspot region mask. The components in the hotspot area mask are segmented using a level setting method to determine one or more hotspot volumes. This segmentation method is performed both for the prostate and for lymphatic hotspots, thereby automatically segmenting hotspots in the prostate and lymphatic regions. iii. Quantification

SUV平均數 經計算為表示熱點之所有體素之平均攝取。

SUV峰值 經計算為具有在SUV最大值所定位之體素之中點之5 mm內之中點的所有體素之平均數。

體積 計算體素之數目乘以以(ml)為單位顯示之體素體積。

LI 病變指數—基於主動脈(亦被稱為血池)及肝臟之SUV參考值計算。病變指數係基於與以下跨度中之線性內插相關之病變之SUV平均數之在0與3之間的實數：

若無法計算肝臟或主動脈之SUV參考，或若主動脈值高於肝臟值，則將不計算病變指數且將其顯示為「-」。 在患者層級針對各病變類型報告之值： ILTV 強度加權之組織病變體積—對於各病變類型，計算ITLV。ITLV係針對特定類型之病變體積之經加總總和，其中權重係病變指數。

iv. 基於網路之平台架構 Table 6 identifies the values calculated by the software and displayed for each hotspot after user selection. ITLV is a summary value and only displayed in the report. All calculations are from the SUV variant of PSMA PET/CT. Table 6. Values calculated by aPROMISE Values reported for each selected hot spot and suspicious lesion: SUV max Represents the highest uptake in one voxel of the hotspot.

Average SUV Calculated as the average uptake of all voxels representing hot spots.

SUV peak Calculated as the mean of all voxels with a midpoint within 5 mm of the midpoint of the voxel where the SUV maximum is located.

volume Calculate the number of voxels multiplied by the voxel volume displayed in (ml).

LI Lesion Index - Calculated based on SUV reference values for the aorta (also known as the blood pool) and liver. The Lesion Index is a real number between 0 and 3 based on the mean of the SUVs of the lesions associated with linear interpolation in the following spans:

If the liver or aorta SUV reference cannot be calculated, or if the aortic value is higher than the liver value, the lesion index will not be calculated and will be displayed as "-". Values reported at the patient level for each lesion type: ILTV Intensity-Weighted Tissue Lesion Volume—For each lesion type, ITLV was calculated. ITLV is the aggregated sum of lesion volumes for a specific type, where the weight is the lesion index.

iv. Web-based platform architecture

aPROMISE utilizes a microservices architecture. Deployment to AWS is handled in a cloud-formed command file found in the AWS code repository. The aPROMISE cloud architecture is provided in Figure 29A and the microservice communication design diagram is provided in Figure 29B. J. Imaging agents i. PET imaging radionuclide-labeled PSMA binders

In certain embodiments, the radionuclide-labeled PSMA-binding agent is a radionuclide-labeled PSMA-binding agent suitable for use in PET imaging.

[18F]DCFPyL， 或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent includes [18F]DCFPyL (also known as PyL ^™ ; also known as DCFPyL-18F):

[18F]DCFPyL, or a pharmaceutically acceptable salt thereof.

[18F]DCFBC， 或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises [18F]DCFBC:

[18F] DCFBC, or a pharmaceutically acceptable salt thereof.

⁶⁸ Ga-PSMA-HBED-CC， 或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises68Ga-PSMA-HBED-CC (also known ^as68Ga - ^PSMA -11):

^68Ga -PSMA-HBED-CC, or a pharmaceutically acceptable salt thereof.

PSMA-617 或其之藥學上可接受之鹽。在某些實施例中，放射性核素標記之PSMA結合劑包括⁶⁸ Ga-PSMA-617 (其係用⁶⁸ Ga標記之PSMA-617)，或其之藥學上可接受之鹽。在某些實施例中，放射性核素標記之PSMA結合劑包括¹⁷⁷ Lu-PSMA-617 (其係用¹⁷⁷ Lu標記之PSMA-617)，或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises PSMA-617:

PSMA-617 or a pharmaceutically acceptable salt thereof. In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises68Ga-PSMA-617 (which is ^PSMA -617 labeled ^with68Ga ), or a pharmaceutically acceptable salt thereof. In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises177Lu-PSMA-617 (which is ^PSMA -617 labeled ^with177Lu ), or a pharmaceutically acceptable salt thereof.

PSMA-I&T 或其之藥學上可接受之鹽。在某些實施例中，放射性核素標記之PSMA結合劑包括⁶⁸ Ga-PSMA-I&T (其係用⁶⁸ Ga標記之PSMA-I&T)，或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises PSMA-I&T:

PSMA-I&T or a pharmaceutically acceptable salt thereof. In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises68Ga-PSMA-I&T (which is ^PSMA -I&T labeled ^with68Ga ), or a pharmaceutically acceptable salt thereof.

PSMA-1007， 或其之藥學上可接受之鹽。在某些實施例中，放射性核素標記之PSMA結合劑包括¹⁸ F-PSMA-1007 (其係用¹⁸ F標記之PSMA-1007)，或其之藥學上可接受之鹽。ii.SPECT 成像放射性核素標記之 PSMA 結合劑 In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises PSMA-1007:

PSMA-1007, or a pharmaceutically acceptable salt thereof. In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises18F-PSMA-1007 (which is ^PSMA -1007 labeled ^with18F ), or a pharmaceutically acceptable salt thereof. ii. SPECT imaging radionuclide-labeled PSMA binders

In certain embodiments, the radionuclide-labeled PSMA-binding agent is a radionuclide-labeled PSMA-binding agent suitable for use in SPECT imaging.

1404， 或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent includes 1404 (also known as MIP-1404):

1404, or a pharmaceutically acceptable salt thereof.

1405， 或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises 1405 (also known as MIP-1405):

1405, or a pharmaceutically acceptable salt thereof.

1427， 或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent includes 1427 (also known as MIP-1427):

1427, or a pharmaceutically acceptable salt thereof.

1428， 或其之藥學上可接受之鹽。In certain embodiments, the radionuclide-labeled PSMA-binding agent includes 1428 (also known as MIP-1428):

1428, or a pharmaceutically acceptable salt thereof.

In certain embodiments, PSMA binding agents are prepared by chelating a radionuclide to a radioisotope of a metal [eg, a radioisotope of Xun (Tc) (e.g., Xun99m ( ^99mTc )); for example, rhenium (Re ) radioisotopes (for example, rhenium 188 ( ¹⁸⁸ Re); for example, rhenium 186 ( ¹⁸⁶ Re)); for example, yttrium (Y) radioisotopes (for example, ⁹⁰ Y); for example, lutetium (Lu) radioisotopes ( For example, ¹⁷⁷ Lu); for example, a radioisotope of gallium (Ga) (for example, ⁶⁸ Ga; for example, ⁶⁷ Ga); for example, a radioisotope of indium (for example, ¹¹¹ In); for example, a radioisotope of copper (Cu) ( For example, ⁶⁷ Cu)] and label with the radionuclide.

^99m Tc-MIP-1404， 或其之藥學上可接受之鹽。在某些實施例中，1404可螯合至其他金屬放射性同位素[例如，錸(Re)之放射性同位素(例如，錸188 (¹⁸⁸ Re)；例如，錸186 (¹⁸⁶ Re))；例如，釔(Y)之放射性同位素(例如，⁹⁰ Y)；例如，鎦(Lu)之放射性同位素(例如，¹⁷⁷ Lu)；例如，鎵(Ga)之放射性同位素(例如，⁶⁸ Ga；例如，⁶⁷ Ga)；例如，銦之放射性同位素(例如，¹¹¹ In)；例如，銅(Cu)之放射性同位素(例如，⁶⁷ Cu)]以形成具有類似於上文針對^99m Tc-MIP-1404所展示之結構之結構的化合物(其中用另一金屬放射性同位素代替^99m Tc)。In certain embodiments, 1404 is labeled with a radionuclide (eg, of a radioisotope chelated to a metal). In certain embodiments, radionuclide-labeled PSMA-binding agents ^include99mTc -MIP-1404, which is 1404 labeled (eg, chelated ^to99mTc ) ^with99mTc :

^99m Tc-MIP-1404, or a pharmaceutically acceptable salt thereof. In certain embodiments, 1404 can chelate to other metal radioisotopes [eg, radioisotopes of rhenium (Re) (eg, rhenium 188 ( ¹⁸⁸ Re); eg, rhenium 186 ( ¹⁸⁶ Re)); eg, yttrium ( Y) radioisotopes (eg, ⁹⁰ Y); eg, radioisotopes of titanium (Lu) (eg, ¹⁷⁷ Lu); eg, radioisotopes of gallium (Ga) (eg, ⁶⁸ Ga; eg, ⁶⁷ Ga); eg , radioactive isotopes of indium (eg, ^111In ); eg, radioactive isotopes of copper (Cu) (eg, ^67Cu )] to form compounds with structures similar to those shown above ^for99mTc -MIP-1404 (in ^which99mTc is replaced by another metal radioisotope).

^99m Tc-MIP-1405， 或其之藥學上可接受之鹽。在某些實施例中，1405可螯合至其他金屬放射性同位素[例如，錸(Re)之放射性同位素(例如，錸188 (¹⁸⁸ Re)；例如，錸186 (¹⁸⁶ Re))；例如，釔(Y)之放射性同位素(例如，⁹⁰ Y)；例如，鎦(Lu)之放射性同位素(例如，¹⁷⁷ Lu)；例如，鎵(Ga)之放射性同位素(例如，⁶⁸ Ga；例如，⁶⁷ Ga)；例如，銦之放射性同位素(例如，¹¹¹ In)；例如，銅(Cu)之放射性同位素(例如，⁶⁷ Cu)]以形成具有類似於上文針對^99m Tc-MIP-1405所展示之結構之結構的化合物(其中用另一金屬放射性同位素代替^99m Tc)。In certain embodiments, 1405 is labeled with a radionuclide (eg, of a radioisotope chelated to a metal). In certain embodiments, radionuclide-labeled PSMA-binding agents ^include99mTc -MIP-1405, which is 1405 labeled (eg, chelated ^to99mTc ) ^with99mTc :

^99m Tc-MIP-1405, or a pharmaceutically acceptable salt thereof. In certain embodiments, 1405 can chelate to other metal radioisotopes [eg, radioisotopes of rhenium (Re) (eg, rhenium 188 ( ¹⁸⁸ Re); eg, rhenium 186 ( ¹⁸⁶ Re)); eg, yttrium ( Y) radioisotopes (eg, ⁹⁰ Y); eg, radioisotopes of titanium (Lu) (eg, ¹⁷⁷ Lu); eg, radioisotopes of gallium (Ga) (eg, ⁶⁸ Ga; eg, ⁶⁷ Ga); eg , a radioactive isotope of indium (eg, ^111In ); eg, a radioactive isotope of copper (Cu) (eg, ^67Cu )] to form compounds with structures similar to those shown above ^for99mTc -MIP-1405 (in ^which99mTc is replaced by another metal radioisotope).

螯合至金屬之1427， 或其之藥學上可接受之鹽，其中M係用其標記1427之金屬放射性同位素[例如，鍀(Tc)之放射性同位素(例如，鍀99m (^99m Tc))；例如，錸(Re)之放射性同位素(例如，錸188 (¹⁸⁸ Re)；例如，錸186 (¹⁸⁶ Re))；例如，釔(Y)之放射性同位素(例如，⁹⁰ Y)；例如，鎦(Lu)之放射性同位素(例如，¹⁷⁷ Lu)；例如，鎵(Ga)之放射性同位素(例如，⁶⁸ Ga；例如，⁶⁷ Ga)；例如，銦之放射性同位素(例如，¹¹¹ In)；例如，銅(Cu)之放射性同位素(例如，⁶⁷ Cu)]。In certain embodiments, 1427 is labeled with a metal radioisotope (eg, chelated to a metal radioisotope) to form a compound according to the formula:

1427 chelated to a metal, or a pharmaceutically acceptable salt thereof, wherein M is a metal radioisotope with which 1427 is labeled [for example, a radioisotope of Xun (Tc) (for example, Xun 99m ( ^99m Tc)); for example , a radioisotope of rhenium (Re) (eg, rhenium 188 ( ¹⁸⁸ Re); eg, rhenium 186 ( ¹⁸⁶ Re)); eg, a radioisotope of yttrium (Y) (eg, ⁹⁰ Y); eg, titanium (Lu) for example, ^{radioisotopes} of gallium (Ga) (for example, ⁶⁸ Ga; for example, ⁶⁷ Ga); for example, radioisotopes of indium (for example, ¹¹¹ In); for example, copper (Cu) radioactive isotopes (eg, ⁶⁷ Cu)].

螯合至金屬之1428， 或其之藥學上可接受之鹽，其中M係用其標記1428之金屬放射性同位素[例如，鍀(Tc)之放射性同位素(例如，鍀99m (^99m Tc))；例如，錸(Re)之放射性同位素(例如，錸188 (¹⁸⁸ Re)；例如，錸186 (¹⁸⁶ Re))；例如，釔(Y)之放射性同位素(例如，⁹⁰ Y)；例如，鎦(Lu)之放射性同位素(例如，¹⁷⁷ Lu)；例如，鎵(Ga)之放射性同位素(例如，⁶⁸ Ga；例如，⁶⁷ Ga)；例如，銦之放射性同位素(例如，¹¹¹ In)；例如，銅(Cu)之放射性同位素(例如，⁶⁷ Cu)]。In certain embodiments, 1428 is labeled with a metal radioisotope (eg, chelated to a metal radioisotope) to form a compound according to the formula:

1428 chelated to a metal, or a pharmaceutically acceptable salt thereof, wherein M is a metal radioisotope with which 1428 is labeled [for example, a radioisotope of Xun (Tc) (for example, Xun 99m ( ^99m Tc)); for example , a radioisotope of rhenium (Re) (eg, rhenium 188 ( ¹⁸⁸ Re); eg, rhenium 186 ( ¹⁸⁶ Re)); eg, a radioisotope of yttrium (Y) (eg, ⁹⁰ Y); eg, titanium (Lu) for example, ^{radioisotopes} of gallium (Ga) (for example, ⁶⁸ Ga; for example, ⁶⁷ Ga); for example, radioisotopes of indium (for example, ¹¹¹ In); for example, copper (Cu) radioactive isotopes (eg, ⁶⁷ Cu)].

PSMA I&S， 或其之藥學上可接受之鹽。在某些實施例中，放射性核素標記之PSMA結合劑包括^99m Tc-PSMA I&S (其係用^99m Tc標記之PSMA I&S)，或其之藥學上可接受之鹽。 K. 電腦系統及網路架構 In certain embodiments, the radionuclide-labeled PSMA binding agent comprises PSMA I&S:

PSMA I&S, or a pharmaceutically acceptable salt thereof. In certain embodiments, the radionuclide-labeled PSMA-binding agent comprises99mTc-PSMA I&S (which is ^PSMA I&S labeled ^with99mTc ), or a pharmaceutically acceptable salt thereof. K. Computer System and Network Architecture

As shown in FIG. 30, an implementation of a network environment 3000 for providing the systems, methods, and architectures described herein is shown and described. In a brief overview, referring now to FIG. 30, a block diagram of an exemplary cloud computing environment 3000 is shown and described. Cloud computing environment 3000 may include one or more resource providers 3002a, 3002b, 3002c (collectively 3002). Each resource provider 3002 may contain computing resources. In some implementations, computing resources may include any hardware and/or software used to process data. For example, computing resources may include hardware and/or software capable of executing algorithms, computer programs, and/or computer applications. In some implementations, exemplary computing resources may include application servers and/or databases with storage and retrieval capabilities. Each resource provider 3002 can be connected to any other resource provider 3002 in the cloud computing environment 3000 . In some embodiments, resource providers 3002 may be connected via computer network 3008 . Each resource provider 3002 may be connected via a computer network 3008 to one or more computing devices 3004a, 3004b, 3004c (collectively 3004).

Cloud computing environment 3000 may include resource manager 3006 . The resource manager 3006 can be connected to the resource provider 3002 and the computing device 3004 via the computer network 3008 . In some implementations, the resource manager 3006 can facilitate the provisioning of computing resources by the one or more resource providers 3002 to the one or more computing devices 3004. Resource manager 3006 may receive requests for computing resources from particular computing devices 3004 . Resource manager 3006 can identify one or more resource providers 3002 capable of providing computing resources requested by computing device 3004 . The resource manager 3006 can select a resource provider 3002 that provides computing resources. Resource manager 3006 may facilitate connections between resource providers 3002 and particular computing devices 3004. In some implementations, the resource manager 3006 can establish a connection between a particular resource provider 3002 and a particular computing device 3004. In some implementations, the resource manager 3006 can redirect a particular computing device 3004 to a particular resource provider 3002 having the requested computing resource.

31 shows an example of a computing device 3100 and a mobile computing device 3150 that may be used to implement the techniques described in this disclosure. Computing device 3100 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blades, mainframes, and other suitable computers. Mobile computing device 3150 is intended to represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are intended to be illustrative only, and are not intended to be limiting.

The computing device 3100 includes a processor 3102, a memory 3104, a storage device 3106, a high-speed interface 3108 connected to the memory 3104 and a plurality of high-speed expansion ports 3110, and a low-speed interface 3112 connected to the low-speed expansion port 3114 and the storage device 3106. Each of the processor 3102, memory 3104, storage 3106, high-speed interface 3108, high-speed expansion ports 3110, and low-speed interface 3112 are interconnected using various bus bars and may be mounted on a common motherboard or otherwise as appropriate . Processor 3102 may process instructions executed within computing device 3100, including graphics stored in memory 3104 or on storage device 3106 for displaying GUIs on external input/output devices, such as display 3116 coupled to high-speed interface 3108 Information instructions. In other implementations, multiple processors and/or multiple buses may be used as appropriate, along with multiple memories and multiple types of memory. Also, multiple computing devices may be connected to each device that provides some of the required operations, such as, for example, a server bank, a group of blade-type servers, or a multiprocessor system. Thus, when terms are used herein, where a plurality of functions are described as being performed by a "processor," this includes any number of processors ( one or more) to perform an embodiment of the plurality of functions. Furthermore, where a function is described as being performed by a "processor," this includes any number of processors therein by (eg, in a distributed computing system) any number of computing device(s) An embodiment(s) that perform the function.

The memory 3104 stores information in the computing device 3100 . In some implementations, memory 3104 is one (or several) volatile memory cells. In some implementations, memory 3104 is one (or several) non-volatile memory cells. Memory 3104 may also be another form of computer-readable medium, such as a magnetic disk or optical disk.

The storage device 3106 can provide mass storage for the computing device 3100 . In some implementations, the storage device 3106 may be or contain a computer-readable medium, such as a floppy disk device, a hard disk device, an optical or magnetic tape device, a flash memory or other similar solid state memory device, or A device array of devices included in a storage area network or other configuration. Instructions may be stored in an information carrier. The instructions, when executed by one or more processing devices (eg, processor 3102), perform one or more methods, such as those described above. Instructions may also be stored by one or more storage devices, such as computer-readable or machine-readable media (eg, memory 3104, storage device 3106, or memory on processor 3102).

High-speed interface 3108 manages bandwidth-intensive operations for computing device 3100, while low-speed interface 3112 manages less bandwidth-intensive operations. These functional assignments are for illustration only. In some implementations, high-speed interface 3108 is coupled to memory 3104, display 3116 (eg, through a graphics processor or accelerator), and to high-speed expansion port 3110 that accepts various expansion cards (not shown). In an implementation, the low-speed interface 3112 is coupled to the storage device 3106 and the low-speed expansion port 3114 . Low-speed expansion port 3114, which may include various communication ports (eg, USB, Bluetooth®, Ethernet, wireless Ethernet), may be coupled to one or more input/output devices (such as keyboards, pointing devices, scanners) , or coupled (eg, through a network adapter) to a network connection device such as a switch or router.

Computing device 3100 may be implemented in many different forms, as shown in the figures. For example, it may be implemented as a standard server 3120 or multiple times in a group of such servers. Additionally, it may be implemented in a personal computer such as laptop 3122. It can also be implemented as part of a rack server system 3124. Alternatively, components from computing device 3100 may be combined with other components (not shown) in a mobile device, such as mobile computing device 3150. Each of these devices may include one or more of a computing device 3100 and a mobile computing device 3150, and the overall system may be made up of multiple computing devices in communication with each other.

Mobile computing device 3150 includes processor 3152, memory 3164, input/output devices (such as display 3154), communication interface 3166, and transceiver 3168, among other components. The mobile computing device 3150 may also have a storage device (such as a micro-hard drive or other device) to provide additional storage. Each of the processor 3152, memory 3164, display 3154, communication interface 3166, and transceiver 3168 are interconnected using various bus bars, and some of these components may be mounted on a common motherboard or otherwise as appropriate .

Processor 3152 can execute instructions within mobile computing device 3150 , including instructions stored in memory 3164 . The processor 3152 may be implemented as a chip set including chips for discrete and multiple analog and digital processors. The processor 3152 may provide, for example, coordination of other components of the mobile computing device 3150 , such as control of the user interface, applications run by the mobile computing device 3150 , and wireless communication by the mobile computing device 3150 .

The processor 3152 can communicate with a user through a control interface 3158 and a display interface 3156 coupled to the display 3154. Display 3154 may be, for example, a TFT (Thin Film Transistor Liquid Crystal Display) display or an OLED (Organic Light Emitting Diode) display or other suitable display technology. Display interface 3156 may include appropriate circuitry for driving display 3154 to present graphics and other information to the user. The control interface 3158 can receive commands from the user and convert the commands for submission to the processor 3152. In addition, the external interface 3162 may provide communication with the processor 3152 to enable near area communication between the mobile computing device 3150 and other devices. External interface 3162 may provide, for example, wired communication in some implementations, or wireless communication in other implementations, and multiple interfaces may also be used.

The memory 3164 stores information in the mobile computing device 3150 . Memory 3164 may be implemented as one or more of a computer-readable medium (or several), one (or several) volatile memory cells, or one (or several) non-volatile memory cells. Expansion memory 3174 may also be provided and connected to mobile computing device 3150 through expansion interface 3172, which may include, for example, a SIMM (Single Row In-Line Memory Module) card interface. The expansion memory 3174 can provide additional storage space for the mobile computing device 3150 , or can also store applications or other information for the mobile computing device 3150 . Specifically, expansion memory 3174 may contain instructions to perform or supplement the procedures described above, and may also contain security information. Thus, for example, expansion memory 3174 may be provided as a security module for mobile computing device 3150 and may be programmed with instructions that permit secure use of mobile computing device 3150. Additionally, the security application may be provided via the SIMM card along with additional information (such as placing identifying information on the SIMM card in an unhackable manner).

Memory may include, for example, flash memory and/or NVRAM memory (non-volatile random access memory), as discussed below. In some implementations, instructions are stored in an information carrier and the instructions, when executed by one or more processing devices (eg, processor 3152), perform one or more methods, such as those described above. Instructions may also be stored by one or more storage devices, such as one or more computer-readable or machine-readable media (eg, memory 3164, expansion memory 3174, or memory on processor 3152). In some implementations, instructions may be received in propagated signals, eg, via transceiver 3168 or external interface 3162.

The mobile computing device 3150 can communicate wirelessly through the communication interface 3166, which can include digital signal processing circuitry if necessary. The communication interface 3166 may provide communication under various modes or protocols such as GSM voice telephony (Global System for Mobile Communications), SMS (Short Messaging Service), EMS (Enhanced Messaging Service) or MMS Messaging (Multimedia Messaging) delivery service), CDMA (Code Division Multiple Access), TDMA (Time Division Multiple Access), PDC (Personal Digital Cellular Phone), WCDMA (Wideband Code Division Multiple Access), CDMA2000 or GPRS (General Packet Radio Service) Wait. This communication may occur using radio frequency, for example, through transceiver 3168. Additionally, short-range communication may occur, such as using Bluetooth®, Wi-Fi™, or other such transceivers (not shown). Additionally, the GPS (Global Positioning System) receiver module 3170 can provide additional navigation-related and location-related wireless data to the mobile computing device 3150 that can be appropriately used by applications running on the mobile computing device 3150.

Mobile computing device 3150 may also communicate audibly using audio codec 3160, which may receive spoken information from the user and convert it into usable digital information. The audio codec 3160 can also produce audible sound to the user, such as through a speaker in the handset of the mobile computing device 3150, for example. This sound may include sound from a voice call, may include recorded sound (eg, voice messages, music files, etc.) and may also include sound generated by applications operating on the mobile computing device 3150.

Mobile computing device 3150 may be implemented in many different forms, as shown in the figures. For example, it may be implemented as a cellular phone 3180. It can also be implemented as part of a smart phone 3182, a personal digital assistant or other similar mobile devices.

Various implementations of the systems and techniques described herein can be implemented in digital electronic circuitry, integrated circuits, specially designed ASICs (application-specific integrated circuits), computer hardware, firmware, software, and/or the like in the combination. These various implementations may include implementations in one or more computer programs, which may include at least one programmable processor (which may be special purpose or general purpose, coupled to a self-storage system) Execute and/or interpret on a programmable system that receives and transmits data and instructions to the storage system), at least one input device and at least one output device.

Such computer programs (also known as programs, software, software applications, or code) contain machine instructions for programmable processors, and can be written in high-level procedural and/or object-oriented programming languages, and/or in combinations of language/machine language implementation. As used herein, the terms machine-readable medium and computer-readable medium refer to a machine-readable medium for providing machine instructions and/or data to a programmable processor (which includes a machine-readable medium that receives machine instructions as machine-readable signals) any computer program product, equipment and/or device (eg, disk, optical disk, memory, Programmable Logic Device (PLD)). The term machine-readable signal refers to any signal used to provide machine instructions and/or data to a programmable processor.

To provide interaction with a user, the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) for displaying information to the user) monitor) and the keyboard and pointing devices (eg, mouse or trackball) by which the user can provide input to the computer. Other types of devices may also be used to provide interaction with the user; for example, feedback provided to the user may be any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and input from the user Can be received in any form, including sound, voice or tactile input.

The systems and techniques described herein may be implemented in a computing system that includes back-end components (eg, as a data server), or includes intermediate software components (eg, an application server), or includes front-end components (eg, as a data server) , a client computer with a graphical user interface or web browser through which users can interact with implementations of the systems and techniques described herein), or any combination of these backend, middleware, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

The computing system may include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The client-server relationship occurs by computer programs running on the respective computers and having a client-server relationship to each other.

In some implementations, the various modules described herein can be separated, combined, or incorporated into a single or combined module. The modules depicted in the figures are not intended to limit the systems described herein to the software architecture shown therein.

Elements of different implementations described herein can be combined to form other implementations not expressly set forth above. Elements may be excluded from the programs, computer programs, databases, etc. described herein without adversely affecting their operation. Additionally, the logic flows depicted in the figures do not require the particular order shown, or sequential order, to achieve desirable results. Various discrete elements may be combined into one or more individual elements to perform the functions described herein.

Throughout descriptions in which apparatuses and systems are described as having, comprising or including particular components or in which procedures and methods are described as having, comprising or including particular steps, it is contemplated that there will be additionally that the invention consists essentially of or consists of such recited components Apparatus and systems consisting of the recited components, and additionally there are programs and methods that consist essentially of or consist of the recited processing steps in accordance with the present invention.

It should be understood that the order of steps, or order for performing a particular action, is immaterial as long as the invention remains operable. Furthermore, two or more steps or actions may be performed simultaneously.

While the invention has been shown and described with particular reference to certain preferred embodiments, those skilled in the art will understand that the invention can be used in Various changes in form and detail were made in the preferred embodiments.

100a: Procedures 100b: Procedure 100c: Procedures 102: 3D functional images 102a: Radiopharmaceuticals 102b: Positron Emission Tomography (PET) Imaging 104: 3D Anatomy Imaging 106: Steps 108: Steps 110: Machine Learning Modules 112: Machine Learning Modules 114: The first machine learning module 116: Second Machine Learning Module 120: Steps 122: Steps 124: Steps 126: Steps 130: Hotspot List 132: Heatmap 140: Steps 200: Program 202: 3D Functional Imagery 208: Steps 210: Steps 212: Steps 214: Steps 216: Steps 218: Steps 242: Separation Threshold 244: Main Components 246: Minor components 252a: Outline 252b: Contour 300: Procedure 304: Step 306: Steps 308: Steps 310: Steps 312: Steps 314: Steps 316: Steps 400: Procedure 404: Step 406: Step 408: Step 410: Steps 500: Procedure 504: Step 506: Steps 508: Steps 510: Steps 512: Steps 520: Workflow 522: Steps 524: Steps 526: Steps 526a: Steps 526b: Steps 528: Steps 530: Steps 532: Steps 902: Functional Image 904: Anatomical Imaging/Computed Tomography (CT) Imaging 906: Segmentation Map/3D Segmentation Map/3D Segmentation Map 908: Machine Learning Modules 910: 3D heat map 1000: Postprocessing 1100: Program 1102: Positron Emission Tomography (PET) Imaging 1108a: Prostate Module/Machine Learning Module 1108b: Full Body Mods/Machine Learning Mods 1110a: Results 1110b: Results 1112: Final Hotspot List and/or Heatmap 1114: 3D Heat Maps 1120: Procedure 1122: Analytical Segmentation Methods / Analytical Segmentation 1124: Analytically segmented 3D heat map 1200: Procedure 1202: Positron Emission Tomography (PET) Imaging 1204: Computed Tomography (CT) Imaging 1206: Split Map 1208: Machine Learning Modules 1210: 3D Heat Map 1212: Analytical Segmentation Models 1214: 3D Heat Maps 1402: Hotspot 1502: Hotspot Volume 1610: Graphical User Interface (GUI) Viewer 1612: Control Panel GUI Toolset 1614: Various features/quantitative measures showing specific lesions 1620: Report 1622: Lesion volume 1700: Procedure 1702: 3D Positron Emission Tomography (PET) Imaging/Positron Emission Tomography (PET) Imaging 1704: 3D Computed Tomography (CT) Imaging/Computed Tomography (CT) Imaging 1706: First Organ Segmentation Machine Learning Module/Organ Segmentation Machine Learning Module 1708: 3D Segmentation Map / Segmentation Map 1712: First, single-category hotspot segmentation module/single-category hotspot segmentation module 1714: Second, multi-category hotspot segmentation module/multi-category hotspot segmentation module 1716: First, single-class 3D heat map / single-class heat map 1718: Second, multi-class 3D heat map / multi-class heat map 1722: Merged heatmap/hotspot merge step 1724: Segmented and Classified Hotspot/Segmented and Classified Hotspot Volume 1738: 3D Segmentation Map/3D Organ Segmentation Map 1800: Program 1802: Initial 3D Heatmap/3D Heatmap 1804: Positron Emission Tomography (PET) Imaging 1806: 3D Organ Segmentation Map 1808: Steps 1810: Blood pool reference value 1832: Threshold value 1840: Graphs 1842: Hotspot Strength 1844: Hot Spot Threshold 1846: Linear Dimensions 1848: Large solid lesions/lesions 1850: Graphs 1852: Hotspot Intensity/Peak Hotspot Intensity/Maximum Hotspot Intensity 1854: Hot Spot Threshold 1856: Linear Dimensions 1858: Small solid lesions/lesions 1860: Drawing 1864: Threshold value 1866: Lesion 1900: Known hotspot segmentation method 1904: Limits based on strength 1906: Downstream Quantization 1920: Video 1922: Steps 1924: Region of Interest (ROI) 1926: Steps 1928: Video 1929: Hotspot segmentation 1930: Video 1931: Hotspot segmentation 1950: Methods based on artificial intelligence (AI) 1952: Positron emission tomography (PET) imaging 1954: Computed tomography (CT) imaging 1956: Detection, segmentation and classification of hot spots 1958: Analytical segmentation methods 1960: Downstream Quantization 2002a: False Positives 2002b: False Positives 2004: Prostate lesions 2006: Prostate lesions 2300: View 2302: blue symbol 2304: red symbol 2306: Orange symbol 2308: Black Symbol 2400: View Window/Viewer 2402: Panel/Retractable Panel 2422: Panel/Middle Panel 2442: Panel/Retractable Right Panel 2462: Build report button 2500: Report 2502: Section/Panel 2522: Section/Overview Section 2542: Segment 2600: Image Processing Workflow 2602: Steps/Organ Segmentation 2604: Steps 2608: Steps 2608a: Procedure 2608b: Steps 2610: Steps 2800: Analytical Models 2802: Positron Emission Tomography (PET) Imaging 2804: Organ Segmentation Map/Organ Segmentation Mask 2806: Suppression step 2808: Suppressed Positron Emission Tomography (PET) Imaging 2810: Difference of Gaussian (DoG) filtering method 2812: Step/Laplacian of Gaussian (LoG) filtering method 2814: 3D segmentation map 3000: Network Environment/Cloud Computing Environment 3002a: Resource Provider 3002b: Resource Provider 3002c: Resource Providers 3004a: Computing Devices 3004b: Computing Devices 3004c: Computing Devices 3006: Resource Manager 3008: Computer Network 3100: Computing Device 3102: Processor 3104: Memory 3106: Storage Device 3108: High-speed interface 3110: High-speed expansion port 3112: Low speed interface 3114: Low speed expansion port 3116: Display 3120: Standard Server 3124: Rack Server Systems 3150: Mobile Computing Devices 3152: Processor 3154: Display 3156: Display interface 3158: Control interface 3160: Audio Codec 3162: External interface 3164: Memory 3166: Communication interface 3168: Transceiver 3170: Global Positioning System (GPS) Receiver Module 3172: Extension interface 3174: Extended memory 3180: Cellular Phone 3182: Smartphone

The foregoing and other objects, aspects, features and advantages of the present invention will become more apparent and better understood by reference to the following description taken in conjunction with the accompanying drawings, wherein:

1A is a block flow diagram of an example procedure for artificial intelligence (AI)-based lesion detection, according to an illustrative embodiment.

IB is a block flow diagram of an example procedure for AI-based lesion detection, according to an illustrative embodiment.

1C is a block flow diagram of an example procedure for AI-based lesion detection, according to an illustrative embodiment.

2A is a graph showing a histogram of liver SUV values overlaid with a two-component Gaussian mixture model according to an illustrative embodiment.

2B is a PET image superimposed on a CT image showing the portion of the liver volume used to calculate the liver reference value, according to an illustrative embodiment.

2C is a block flow diagram of an example procedure for calculating reference intensity values to avoid/reduce effects from tissue regions associated with low radiopharmaceutical uptake, according to an illustrative embodiment.

3 is a block flow diagram of an example procedure for correcting for intensity exudation from one or more tissue regions associated with high radiopharmaceutical uptake, according to an illustrative embodiment.

4 is a block flow diagram of an example procedure for automatically marking hot spots corresponding to detected lesions, according to an illustrative embodiment.

5A is a block flow diagram of an example process for interactive lesion detection that allows user feedback and viewing via a graphical user interface (GUI), according to an illustrative embodiment.

5B is an example procedure for user viewing, quality control, and reporting of automatically detected lesions, according to an illustrative embodiment.

6A is a screenshot of a GUI for confirming accurate segmentation of a liver reference volume, according to an illustrative embodiment.

6B is a screenshot of a GUI for confirming accurate segmentation of aortic section (blood pool) reference volumes, according to an illustrative embodiment.

6C is a screenshot of a GUI for user selection and/or validation of automatically segmented hotspots corresponding to detected lesions within a subject, according to an illustrative embodiment.

6D is a screen shot of a portion of a GUI that allows a user to manually identify lesions within an image, according to an illustrative embodiment.

6E is a screen shot of another portion of a GUI that allows a user to manually identify lesions within an image, according to an illustrative embodiment.

7 is a screen shot of a portion of a GUI showing a quality control checklist, according to an illustrative embodiment.

8 is a screen shot of a report generated by a user using an embodiment of the automated lesion detection tool described herein, according to an illustrative embodiment.

9 is a block flow diagram showing an example architecture for hot spot (lesion) segmentation via a machine learning module that receives 3D anatomical images, 3D functional images, and 3D segmentation maps as input, according to an illustrative embodiment.

10A is a block flow diagram showing an example procedure in which lesion type mapping is performed after hotspot segmentation, according to an illustrative embodiment.

10B is another block flow diagram showing the use of a 3D segmentation map, showing an example procedure in which lesion type mapping is performed after hotspot segmentation, according to an illustrative embodiment.

11A is a block flow diagram showing a procedure for detecting and/or segmenting hotspots representing lesions using a whole-body network and a prostate-specific network, according to an illustrative embodiment.

11B is a block flow diagram showing a procedure for detecting and/or segmenting hotspots representing lesions using a whole-body network and a prostate-specific network, according to an illustrative embodiment.

12 is a block flow diagram showing the use of the analysis segmentation step following AI-based hotspot segmentation, according to an illustrative embodiment.

13A is a block diagram showing an example U-net architecture for hotspot segmentation, according to an illustrative embodiment.

13B and 13C are block diagrams showing an example FPN architecture for hotspot segmentation, according to an illustrative embodiment.

14A, 14B, and 14C show example images of hotspot segmentation using an example U-net architecture, according to an illustrative embodiment.

15A and 15B show example images of hotspot segmentation using an FPN architecture according to an illustrative embodiment.

16A, 16B, 16C, 16D, and 16E are screenshots of example GUIs for uploading, analyzing, and generating reports from medical image data, according to an illustrative embodiment.

17A and 17B are block flow diagrams of an example process for segmenting and classifying hotspots using two parallel machine learning modules, according to an illustrative embodiment.

17C depicts the interaction between various software modules (eg, APIs) of an example implementation of a procedure for segmenting and classifying hotspots using two parallel machine learning modules, according to an illustrative embodiment and Block flow diagram of data flow.

18A is a block flow diagram of an example procedure for segmenting hotspots by an analytical model using an adaptive bound method, according to an illustrative embodiment.

18B and 18C are graphs showing the variation of a hotspot specific threshold value as a function of hotspot intensity ( _SUVmax ) used in an adaptive bounding method, according to an illustrative embodiment.

Figures 18D, 18E, and 18F are diagrams illustrating certain definition techniques in accordance with an illustrative embodiment.

18G is a graph showing a histogram of prostate voxel intensities along axial, sagittal and coronal planes, and illustrative settings of prostate voxel intensity values and threshold scaling factors, according to an illustrative embodiment .

19A is a block flow diagram illustrating hotspot segmentation using conventional manual region of interest (ROI) definitions and conventional fixed and/or relative bounds, according to an illustrative embodiment.

19B is a block flow diagram illustrating hot spot segmentation using an AI-based approach in combination with an adaptive bounding approach, according to an illustrative embodiment.

20 is a set of images comparing exemplary segmentation results for delimitation only with segmentation results obtained via an AI-based approach in combination with an adaptive delimiting method, according to an illustrative embodiment.

21A, 21B, 21C, 21D, 21E, 21F, 21G, 21H, and 21I show a sequence of 2D blocks of 3D PET images moved in the vertical direction in the abdominal region. The images compare the results of hot spot segmentation within the abdominal region performed by the bound method only (left hand image) with the results of hot spot segmentation according to the machine learning method according to some embodiments described herein (right hand image), and are shown Hotspot regions overlaid on the PET image tiles identified by each method.

22 is a block flow diagram of a process for uploading and analyzing PET/CT image data using a CAD device that provides automated image analysis, according to certain embodiments described herein.

23 is a screenshot of an example GUI that allows a user to upload image data for viewing and analysis via a CAD device that provides automated image analysis, according to certain embodiments described herein.

24 is a screenshot of an example GUI viewer that allows a user to view and analyze medical image data (eg, 3D PET/CT images) and the results of automated image analysis, according to an illustrative embodiment.

25 is a screen shot of an automatically generated report according to an illustrative embodiment.

26 is a block flow diagram of an example workflow providing automated analysis and user input and viewing for analyzing medical image data, according to an illustrative embodiment.

27 shows three views of CT images with superimposed segmented bone and soft tissue volumes, according to an illustrative embodiment.

28 is a block flow diagram of an analytical model for segmenting hotspots, according to an illustrative embodiment.

Figure 29A is a block diagram of a cloud computing architecture used in some embodiments.

29B is a block diagram of an example microservice communication flow used in certain embodiments.

30 is a block diagram of an exemplary cloud computing environment used in certain embodiments.

31 is a block diagram of an example computing device and an example mobile computing device used in certain embodiments.

The features and advantages of the present invention will become more apparent from the [Embodiments] set forth below when taken in conjunction with the accompanying drawings, wherein like reference numerals identify corresponding elements throughout. In the drawings, identical reference numerals generally indicate identical, functionally similar, and/or structurally similar elements.

100a: Procedures

102: 3D functional images

102a: Radiopharmaceuticals

102b: Positron Emission Tomography (PET) Imaging

106: Steps

110: Machine Learning Modules

120: Steps

130: Hotspot List

132: Heatmap

140: Steps

Claims (148)

A method for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) by the processor, using a machine learning module to automatically detect one or more hot spots within the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing a region within the subject Potentially cancerous lesions to create one or both of the following (i) and (ii): (i) a list of hotspots that identifies the location of the hotspot for each hotspot, and (ii) a 3D heat map that identifies the hotspot for each hotspot the corresponding 3D hotspot volume within the 3D functional image; and (c) storing and/or providing the hotspot list and/or the 3D heatmap for display and/or further processing. The method of claim 1, wherein the machine learning module receives at least a portion of the 3D-enabled image as input, and automatically detects the one or more hot spots based at least in part on intensities of voxels in the received portion of the 3D-enabled image . The method of claim 1 or 2, wherein the machine learning module receives as input a 3D segmentation map that identifies one or more volumes of interest (VOIs) within the 3D functional image, each VOI corresponding to the subject specific target tissue regions and/or specific anatomical regions within the patient. As in the method of any of the preceding claims, It includes receiving, by the processor, a 3D anatomical image of the subject obtained using an anatomical imaging modality, wherein the 3D anatomical image includes a graphical representation of tissue within the subject, And wherein the machine learning module receives at least two input channels, the input channels include a first input channel corresponding to at least a part of the 3D anatomical image and a second input channel corresponding to at least a part of the 3D functional image. The method of claim 4, wherein the machine learning module receives as input a 3D segmentation map that identifies one or more volumes of interest (VOIs) within the 3D functional image and/or the 3D anatomical image, each VOI Corresponds to specific target tissue regions and/or specific anatomical regions. The method of claim 5, comprising automatically segmenting, by the processor, the 3D anatomical image, thereby creating the 3D segmentation map. The method of any one of the preceding claims, wherein the machine learning module is a region-specific machine learning module that receives the 3D functional image corresponding to one or more specific tissue regions and/or anatomy of the subject A specific part of the area is used as input. The method of any of the preceding claims, wherein the machine learning module generates the hotspot list as output. The method of any of the preceding claims, wherein the machine learning module generates the 3D heat map as output. The method of any one of the preceding claims, comprising: (d) determining, by the processor, for each of at least a portion of the hotspots, a lesion likelihood classification corresponding to the likelihood of the intrasubject lesion represented by the hotspot. The method of claim 10, wherein step (d) includes using the machine learning module to determine the lesion likelihood classification for each of the hot spots in the portion. The method of claim 10, wherein step (d) includes using a second machine learning module to determine the lesion likelihood classification for each hot spot. The method of claim 12, comprising determining, by the processor, for each hotspot a set of one or more hotspot features, and using the set of one or more hotspot features as input to the second machine learning module . The method of any one of claims 10 to 13, comprising: (e) selecting, by the processor, a subset of the one or more hotspots corresponding to hotspots having a high probability corresponding to cancerous lesions based at least in part on the lesion likelihood classifications of the hotspots. The method of any one of the preceding claims, comprising: (f) adjusting, by the processor, the intensities of voxels of the 3D functional image to correct for intensity bleed from one or more high intensity volumes of the 3D functional image, each of the one or more high intensity volumes Corresponds to areas of high uptake tissue within the subject associated with high radiopharmaceutical uptake under normal conditions. The method of claim 15, wherein step (f) comprises correcting for intensity bleed from the plurality of high intensity volumes one at a time in a sequential manner. The method of claim 15 or 16, wherein the one or more high intensity volumes correspond to one or more high uptake tissue regions selected from the group consisting of kidney, liver and bladder. The method of any one of the preceding claims, comprising: (g) determining, by the processor, for at least a portion of each of the one or more hotspots, an indicator indicative of the level of radiopharmaceutical uptake and/or the size of the underlying lesion within the underlying lesion corresponding to the hotspot Corresponding lesion index. 19. The method of claim 18, wherein step (g) comprises comparing the intensity(s) of the one or more voxels associated with the hot spot with one or more reference values, each reference value and a reference corresponding to the reference tissue region The volume is associated with a specific reference tissue region. 19. The method of claim 19, wherein the one or more reference values comprise selected from the group consisting of an aortic reference value associated with an aortic portion of the subject and a liver reference value associated with the subject's liver One or more members of the group. The method of claim 19 or 20, wherein determining the specific reference value for at least one specific reference value associated with the specific reference tissue region comprises fitting intensities of voxels within the specific reference volume corresponding to the specific reference tissue region to into a multicomponent mixture model. The method of any one of claims 18 to 21, comprising computing an overall risk index for the subject that is indicative of the subject's cancer status and/or risk using the determined lesion index values. The method of any preceding claim, comprising determining, by the processor, for each hot spot an anatomical classification corresponding to a particular anatomical region and/or group of anatomical regions within the subject, wherein the hot spot is determined Indicates the location of the potential cancerous lesion. The method of any one of the preceding claims, comprising: (h) causing, by the processor, to transfer at least a portion of the graphical representation of the one or more hotspots for display within a graphical user interface (GUI) for viewing by a user. The method of claim 24, comprising: (i) Receive, by the processor, via the GUI, a user selection by the user to view a subset of one or more hot spots identified as likely to represent potential cancerous lesions in the subject. The method of any of the preceding claims, wherein the 3D functional image comprises a PET or SPECT image obtained after administration of an agent to the subject. The method of claim 26, wherein the agent comprises a PSMA binding agent. The method of claim 26 or ²⁷ , wherein the reagent comprises18F. The method of claim 27 or 28, wherein the agent comprises [18F]DCFPyL. The method of claim 27 or 28, wherein the agent comprises PSMA-11. The method of claim 26 or 27, wherein the reagent comprises one or more members selected from the group consisting ^of99mTc , ^68Ga , ^177Lu , ^225Ac , ^111In , ^123I , ^{124I and131I} . The method of any of the preceding claims, wherein the machine learning module implements a neural network. The method of any one of the preceding claims, wherein the processor is a processor of a cloud-based system. A method for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) receiving, by the processor, a 3D anatomical image of the subject obtained using an anatomical imaging modality, wherein the 3D anatomical image includes a graphical representation of tissue within the subject; (c) automatically detect, by the processor, using a machine learning module, one or more hot spots within the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing a region within the subject Potentially cancerous lesions to create one or both of the following (i) and (ii): (i) a list of hotspots that identifies the location of the hotspot for each hotspot, and (ii) a 3D heat map that identifies the hotspot for each hotspot The corresponding 3D hotspot volume in the 3D functional image, wherein the machine learning module receives at least two input channels, the input channels include a first input channel corresponding to at least a portion of the 3D anatomical image and a second input channel corresponding to at least a portion of the 3D functional image and/or from there and other derived anatomical information; and (d) storing and/or providing the hot spot list and/or the 3D heat map for display and/or further processing. A method for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) automatically detecting, by the processor, using a first machine learning module, one or more hot spots in the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing the subject Potential cancerous lesions within, thereby establishing a hotspot list identifying the location of the hotspot for each hotspot; (c) by the processor, using the second machine learning module and the hotspot list, to automatically determine the corresponding 3D hotspot volume in the 3D functional image for each of the one or more hotspots, thereby creating a 3D heatmap; and (d) storing and/or providing the hot spot list and/or the 3D heat map for display and/or further processing. The method of claim 35, comprising: (e) determining, by the processor, for each of at least a portion of the hotspots, a lesion likelihood classification corresponding to the likelihood that the hotspot represents a lesion within the subject. The method of claim 36, wherein step (e) comprises using a third machine learning module to determine the lesion likelihood classification for each hot spot. The method of any one of claims 35 to 37, comprising: (f) selecting, by the processor, a subset of one or more hotspots corresponding to hotspots having a high likelihood corresponding to cancerous lesions based at least in part on the lesion likelihood classifications of the hotspots. A method of measuring intensity values within a reference volume corresponding to a reference tissue region in order to avoid effects from tissue regions associated with low radiopharmaceutical uptake, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) identifying, by the processor, the reference volume within the 3D functional image; (c) fitting, by the processor, a multicomponent mixture model to the intensities of voxels within the reference volume; (d) identifying, by the processor, the dominant mode of the multicomponent model; (e) determining, by the processor, a measure of intensity corresponding to the principal mode, thereby determining a reference intensity value corresponding to a measure of intensity of voxels that are (i) within the reference tissue volume and ( ii) associated with the primary mode; (f) detecting, by the processor, one or more hot spots in the functional image corresponding to potential cancerous lesions; and (g) determining, by the processor, a lesion index value using at least the reference intensity value for each of at least a portion of the detected hot spots. A method of correcting for intensity exudation from areas of high uptake tissue within a subject associated with normally high radiopharmaceutical uptake, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) identifying, by the processor, a high-intensity volume within the 3D functional image, the high-intensity volume corresponding to a particular high-uptake tissue region in which high radiopharmaceutical uptake occurs under normal conditions; (c) identifying, by the processor, based on the identified high-intensity volume, a suppression volume within the 3D functional image, the suppression volume corresponding to outside the boundaries of the identified high-intensity volume and at a distance from the identified high-intensity volume the volume within the predetermined decay distance of the boundary of the identified high-intensity volume; (d) determining, by the processor, a background image corresponding to the 3D functional image, wherein the volume within the high intensity volume is replaced with an interpolation based on the intensity determination of the voxels of the 3D functional image within the suppression volume the strength of the element; (e) determining, by the processor, an estimated image by subtracting the intensities of the voxels of the background image from the intensities of the voxels from the 3D functional image; (f) By the processor, the suppression map is determined by: extrapolating the intensities of the voxels of the estimated image corresponding to the high-intensity volume to the positions of the voxels within the suppression volume to determine the intensities of the voxels of the suppression map corresponding to the suppression volume; and setting the intensities of voxels of the suppression map corresponding to locations outside the suppression volume to zero; and (g) adjusting, by the processor, the intensity of the voxels of the 3D functional image based on the suppression map to correct for intensity bleed from the high-intensity volume. The method of claim 40, comprising performing steps (b) through (g) for each of a plurality of high-intensity volumes in a sequential manner, thereby correcting for intensity bleed from each of the plurality of high-intensity volumes. The method of claim 41, wherein the plurality of high-intensity volumes comprise one or more members selected from the group consisting of kidney, liver, and bladder. A method for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) automatically detecting, by the processor, one or more hot spots in the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing a potential cancerous lesion in the subject ; (c) causing, by the processor, a graphical representation of the one or more hotspots to be displayed for display within an interactive graphical user interface (GUI); (d) receiving, by the processor, via the interactive GUI, a user selection of a final set of hotspots comprising at least a portion of the one or more automatically detected hotspots; and (e) store and/or provide the final set of hotspots for display and/or further processing. The method of claim 43, comprising: (f) receiving, by the processor, through the GUI, user selections of one or more additional, user-identified hotspots for inclusion in the final set of hotspots; and (g) updating, by the processor, the final set of hotspots to include the one or more additional user-identified hotspots. The method of claim 43 or 44, wherein step (b) comprises using one or more machine learning modules. A method for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) automatically detecting, by the processor, one or more hot spots in the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing a potential cancerous lesion in the subject ; (c) automatically determining, by the processor, for at least a portion of each of the one or more hot spots, an anatomical classification corresponding to a particular anatomical region and/or group of anatomical regions within the subject, wherein the hot spot is determined the location of the indicated potential cancerous lesion; and (d) storing and/or providing the identification of the one or more hotspots and, for each hotspot, an anatomical classification corresponding to the hotspot for display and/or further processing. The method of claim 46, wherein step (b) includes using one or more machine learning modules. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the system comprising: a processor of a computing device; and memory having instructions stored thereon that, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject using a functional imaging modality; (b) using a machine learning module to automatically detect one or more hot spots within the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing a potential cancerous lesion within the subject, thereby One or both of the following (i) and (ii) are created: (i) a list of hotspots, which identifies the location of the hotspot for each hotspot, and (ii) a 3D heatmap, which identifies, for each hotspot, the corresponding to the 3D hotspot volume; and (c) storing and/or providing the hotspot list and/or the 3D heatmap for display and/or further processing. The system of claim 48, wherein the machine learning module receives at least a portion of the 3D-enabled image as input and automatically detects the one or more hot spots based at least in part on intensities of voxels in the received portion of the 3D-enabled image . The system of claim 48 or 49, wherein the machine learning module receives as input a 3D segmentation map that identifies one or more volumes of interest (VOIs) within the 3D functional image, each VOI corresponding to the subject specific target tissue regions and/or specific anatomical regions within the patient. The system of any of claims 48 to 50, wherein the instructions cause the processor to: receiving a 3D anatomical image of the subject obtained using an anatomical imaging modality, wherein the 3D anatomical image includes a graphical representation of tissue within the subject, And wherein the machine learning module receives at least two input channels, the input channels include a first input channel corresponding to at least a part of the 3D anatomical image and a second input channel corresponding to at least a part of the 3D functional image. The system of claim 51, wherein the machine learning module receives as input a 3D segmentation map that identifies one or more volumes of interest (VOIs) within the 3D functional image and/or the 3D anatomical image, each VOI Corresponds to specific target tissue regions and/or specific anatomical regions. The system of claim 52, wherein the instructions cause the processor to automatically segment the 3D anatomical image, thereby creating the 3D segmentation map. The system of any one of claims 48 to 53, wherein the machine learning module is a region-specific machine learning module that receives the 3D functional image corresponding to one or more specific tissue regions of the subject and/or or a specific part of an anatomical region as input. The system of any of claims 48-54, wherein the machine learning module generates the hotspot list as output. The system of any of claims 48 to 55, wherein the machine learning module generates the 3D heat map as output. The system of any of claims 48 to 56, wherein the instructions cause the processor to: (d) For each of at least a portion of the hotspots, determining a lesion likelihood classification corresponding to the hotspot that represents the likelihood of a lesion within the subject. The system of claim 57, wherein in step (d) the instructions cause the processor to use the machine learning module to determine the lesion likelihood classification for each of the hot spots in the portion. The system of claim 57, wherein in step (d) the instructions cause the processor to use a second machine learning module to determine the lesion likelihood classification for each hot spot. The system of claim 59, wherein the instructions cause the processor to determine a set of one or more hotspot features for each hotspot, and use the set of the one or more hotspot features as a feed to the second machine learning module the input. The system of any of claims 57 to 60, wherein the instructions cause the processor to: (e) selecting a subset of the one or more hotspots corresponding to hotspots having a high probability of corresponding to cancerous lesions based at least in part on the lesion likelihood classifications of the hotspots. The system of any of claims 48 to 61, wherein the instructions cause the processor to: (f) adjusting, by the processor, the intensities of voxels of the 3D functional image to correct for intensity bleed from one or more high intensity volumes of the 3D functional image, each of the one or more high intensity volumes Corresponds to areas of high uptake tissue within the subject associated with high radiopharmaceutical uptake under normal conditions. The system of claim 62, wherein in step (f) the instructions cause the processor to correct the intensity bleed from the plurality of high intensity volumes one at a time in a sequential manner. The system of claim 62 or 63, wherein the one or more high intensity volumes correspond to one or more high uptake tissue regions selected from the group consisting of kidney, liver and bladder. The system of any of claims 48 to 64, wherein the instructions cause the processor to: (g) For each of at least a portion of the one or more hotspots, determining a corresponding lesion index indicative of the level of radiopharmaceutical uptake and/or the size of the latent lesion within the underlying lesion to which the hotspot corresponds. The system of claim 65, wherein in step (g) the instructions cause the processor to compare the intensity(s) of one or more voxels associated with the hot spot with one or more reference values, each reference value being equal to A particular reference tissue region within the subject is associated and determined based on the intensity of the reference volume corresponding to the reference tissue region. The system of claim 66, wherein the one or more reference values comprises a reference value selected from the group consisting of an aortic reference value associated with a portion of the aorta of the subject and a liver reference value associated with the liver of the subject One or more members of the group. The system of claim 66 or 67, wherein for at least one particular reference value associated with a particular reference tissue region, the instructions cause the processor to pass voxels within the particular reference volume corresponding to the particular reference tissue region The intensities of are fitted to a multi-component mixed model to determine the specific reference value. The system of any one of claims 65 to 68, wherein the instructions cause the processor to use the determined lesion index values to compute an overall risk index for the subject that is indicative of the subject's cancer status and/or or risk. The system of any one of claims 48 to 69, wherein the instructions cause the processor to determine, for each hot spot, an anatomical classification corresponding to a particular anatomical region and/or group of anatomical regions within the subject, wherein the determination of the The location of a potentially cancerous lesion indicated by the hot spot. The system of any one of claims 48 to 70, wherein the instructions cause the processor to: (h) causing a graphical representation of at least a portion of the one or more hot spots to be displayed in a graphical user interface (GUI) for viewing by a user. The system of claim 71, wherein the instructions cause the processor to: (i) Receive, via the GUI, a user selection of a subset of one or more hotspots identified as likely to represent a potential cancerous lesion within the subject viewed by the user. The system of any one of claims 48 to 72, wherein the 3D functional image comprises a PET or SPECT image obtained after administration of an agent to the subject. The system of claim 73, wherein the agent comprises a PSMA binding agent. ^The system of claim 73 or 74, wherein the reagent comprises18F. The system of claim 74, wherein the reagent comprises [18F]DCFPyL. The system of claim 74 or 75, wherein the agent comprises PSMA-11. The system of claim 73 or 74, wherein the reagent comprises one or more members selected from the group consisting ^of99mTc , ^68Ga , ^177Lu , ^225Ac , ^111In , ^123I , ^{124I and131I} . The system of any one of claims 48 to 78, wherein the machine learning module implements a neural network. The system of any one of claims 48 to 79, wherein the processor is a processor of a cloud-based system. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the system comprising: a processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) receiving a 3D anatomical image of the subject obtained using an anatomical imaging modality, wherein the 3D anatomical image includes a graphical representation of tissue within the subject; (c) using a machine learning module to automatically detect one or more hot spots within the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing a potential cancerous lesion in the subject, One or both of the following (i) and (ii) are thus established: (i) a hotspot list, which identifies the location of the hotspot for each hotspot, and (ii) a 3D heatmap, which identifies, for each hotspot, within the 3D functional image corresponds to the 3D hotspot volume, Wherein the machine learning module receives at least two input channels, the input channels include a first input channel corresponding to at least a part of the 3D anatomical image and a second input channel corresponding to at least a part of the 3D functional image, and/or self- such derived anatomical information; and (d) storing and/or providing the hot spot list and/or the 3D heat map for display and/or further processing. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the system comprising: a processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) using a first machine learning module to automatically detect one or more hot spots in the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings, and representing a potential cancer in the subject lesions, thereby establishing a hotspot list identifying the location of the hotspot for each hotspot; (c) using the second machine learning module and the hotspot list to automatically determine the corresponding 3D hotspot volume within the 3D functional image for each of the one or more hotspots, thereby creating a 3D heatmap; and (d) storing and/or providing the hot spot list and/or the 3D heat map for display and/or further processing. The system of claim 82, wherein the instructions cause the processor to: (e) For each of at least a portion of the hotspots, determining a lesion likelihood classification corresponding to the likelihood that the hotspot represents a lesion within the subject. The system of claim 83, wherein in step (e) the instructions cause the processor to use a third machine learning module to determine the lesion likelihood classification for each hot spot. The system of any one of claims 82 to 84, wherein the instructions cause the processor to: (f) selecting a subset of the one or more hotspots corresponding to hotspots having a high probability of corresponding to cancerous lesions based, at least in part, on the lesion likelihood classifications of the hotspots. A system for measuring intensity values within a reference volume corresponding to a reference tissue region in order to avoid effects from tissue regions associated with low radiopharmaceutical uptake, the system comprising: a processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) identifying the reference volume within the 3D functional image; (c) fitting a multicomponent mixture model to the intensities of the voxels within the reference volume; (d) identify the main modes of the multicomponent model; (e) determine a measure of intensity corresponding to the dominant mode, thereby determining a reference intensity value corresponding to the measure of intensity of voxels that are (i) within the reference tissue volume and (ii) related to the dominant mode Associated; (f) detecting within the 3D functional image one or more hot spots corresponding to potential cancerous lesions; and (g) for each of at least a portion of the detected hot spots, determining a lesion index value using at least the reference intensity value. A system for correcting for intensity exudation from areas of high uptake tissue in a subject associated with normally high radiopharmaceutical uptake, the method comprising: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) identifying high-intensity volumes within the 3D functional image, the high-intensity volumes corresponding to specific regions of high uptake tissue in which high radiopharmaceutical uptake occurs under normal conditions; (c) identifying a suppressed volume within the 3D functional image based on the identified high-intensity volume, the suppressed volume corresponding to the the volume within a predetermined decay distance of the boundary; (d) determining a background image corresponding to the 3D functional image, wherein the intensities of the voxels within the high-intensity volume are replaced by interpolation based on the determination of the intensities of the voxels of the 3D functional image within the suppression volume; (e) Determining an estimated image by subtracting the intensities of the voxels of the background image from the intensities of the voxels from the 3D functional image; (f) Determine the inhibition map by: extrapolating the intensities of the voxels of the estimated image corresponding to the high-intensity volume to the positions of the voxels within the suppression volume to determine the intensities of the voxels of the suppression map corresponding to the suppression volume; and setting the intensities of voxels of the suppression map corresponding to locations outside the suppression volume to zero; and (g) Adjusting the intensity of the voxels of the 3D functional image based on the suppression map to correct for intensity bleed from the high-intensity volume. The system of claim 87, wherein the instructions cause the processor to perform steps (b) through (g) for each of the plurality of high-intensity volumes in a sequential manner, thereby correcting for intensity bleed from each of the plurality of high-intensity volumes . The system of claim 88, wherein the plurality of high intensity volumes comprise one or more members selected from the group consisting of kidney, liver and bladder. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the system comprising: a processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) automatically detect one or more hot spots within the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing a potential cancerous lesion in the subject; (c) cause a graphical representation of the one or more hotspots to be re-listed for display within an interactive graphical user interface (GUI); (d) receiving, via the interactive GUI, a user selection of a final set of hotspots comprising at least a portion of the one or more automatically detected hotspots; and (e) store and/or provide the final set of hotspots for display and/or further processing. The system of claim 90, wherein the instructions cause the processor to: (f) receiving, via the GUI, user selections of one or more additional, user-identified hotspots for inclusion in the final set of hotspots; and (g) updating the final set of hotspots to include the one or more additional user-identified hotspots. The system of claim 90 or 91, wherein in step (b) the instructions cause the processor to use one or more machine learning modules. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the system comprising: a processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) automatically detect one or more hot spots within the 3D functional image, each hot spot corresponding to a local area of increased intensity relative to its surroundings and representing a potential cancerous lesion in the subject; (c) automatically determining, for at least a portion of each of the one or more hot spots, an anatomical classification corresponding to a particular anatomical region and/or group of anatomical regions within the subject, wherein the hot spot is determined to represent a potentially cancerous lesion the location; and (d) storing and/or providing the identification of the one or more hotspots and the anatomical classification for each hotspot corresponding to the hotspot for display and/or further processing. The system of claim 93, wherein the instructions cause the processor to perform step (b) using one or more machine learning modules. A method for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) receiving, by the processor, a 3D anatomical image of the subject obtained using an anatomical imaging modality; (c) receiving, by the processor, a 3D segmentation map identifying one or more specific tissue regions or groups of tissue regions within the 3D functional image and/or within the 3D anatomical image; (d) using one or more machine learning modules to automatically detect and/or segment, by the processor, a set of one or more hotspots within the 3D functional image, each hotspot corresponding to an increased intensity relative to its surroundings local area and represent a potential cancerous lesion within the subject, thereby establishing one or both of the following (i) and (ii): (i) a list of hotspots that identifies the location of the hotspot for each hotspot, and (ii) ) a 3D heat map that identifies the corresponding 3D hotspot volume within the 3D functional image for each hotspot, wherein at least one of the one or more machine learning modules receives as input (i) the 3D functional image, (ii) the 3D anatomical image, and (iii) the 3D segmentation map; and (e) storing and/or providing the hot spot list and/or the 3D heat map for display and/or further processing. The method of claim 95, comprising: receiving, by the processor, an initial 3D segmentation map that identifies one or more specific tissue regions within the 3D anatomical image and/or the 3D functional image; and identifying, by the processor, at least a portion of the one or more specific tissue regions as specific ones belonging to one or more tissue groups, and updating the 3D segmentation map by the processor to indicate the identified specific ones the area belongs to that specific group of organizations; and By the processor, the updated 3D segmentation map is used as an input to at least one of the one or more machine learning modules. The method of claim 96, wherein the one or more tissue groups comprise soft tissue groups, thereby identifying the particular tissue region representing soft tissue as belonging to the soft tissue group. The method of claim 96 or 97, wherein the one or more tissue groups comprise a bone tissue group, thereby identifying the particular tissue region representing bone as belonging to the bone tissue group. The method of any one of claims 96 to 98, wherein the one or more tissue groups comprise a high uptake organ group such that the one or more organs associated with high radiopharmaceutical uptake are identified as belonging to the high uptake organ Ingest groups. The method of any one of claims 95 to 99, comprising for each detected and/or segmented hotspot, determining, by the processor, a classification of the hotspot. The method of claim 100, comprising using at least one of the one or more machine learning modules to determine the classification of the hot spot for each detected and/or segmented lesion. The method of any one of claims 95 to 101, wherein the one or more machine learning modules comprise: (A) a systemic lesion detection module that detects and/or segments hot spots throughout the body; and (B) Prostate lesion module to detect and/or segment hot spots in the prostate. The method of claim 102, comprising generating a hotspot list and/or map using each of (A) and (B), and combining the results. The method of any one of claims 95 to 103, wherein: Step (d) includes: segmenting and classifying the set of one or more hotspots by the following operations to create a labeled 3D heatmap that identifies, for each hotspot, the corresponding 3D hotspot volume within the 3D functional image, and wherein Each hotspot volume is marked as belonging to a particular hotspot class of a plurality of hotspot classes: segmenting a first initial set of one or more hot spots within the 3D functional image using a first machine learning module to create a first initial 3D heat map identifying the first initial set of hot spot volumes, wherein the first machine learning module Segment the hotspots of the 3D functional image according to a single hotspot category; segmenting a second initial set of one or more hot spots within the 3D functional image using a second machine learning module to create a second initial 3D heat map identifying the second initial set of hot spot volumes, wherein the second machine learning module segmenting the 3D functional image according to the plurality of different hotspot categories so that the second initial 3D heatmap is a multi-category 3D heatmap, wherein each hotspot volume is marked as belonging to a specific one of the plurality of different hotspot categories; and By the processor, for at least a portion of the hot spot volumes identified by the first initial 3D heat map, the following operations are performed to merge the first initial 3D heat map and the second initial 3D heat map: identifying a matching hotspot volume of the second initial 3D heatmap that has been marked as belonging to a particular hotspot category of the plurality of different hotspot categories; and marking the specific hotspot volume of the first initial 3D heat map as belonging to the specific hotspot class, thereby creating a merged 3D heatmap that includes the segmented hotspot volume of the first 3D heatmap, which has been Labeling according to the identified class to which the matching hotspot volumes of the second 3D heatmap belong; and Step (e) includes storing and/or providing the merged 3D heat map for display and/or further processing. The method of claim 104, wherein the plurality of different hotspot categories include one or more members selected from the group consisting of: (i) Bone hot spots, which are determined to represent lesions localized in the bone; (ii) lymphatic hot spots, which are determined to represent lesions localized in lymph nodes, and (iii) Prostate hot spots, which are determined to represent lesions localized in the prostate. The method of any one of claims 95 to 105, further comprising: (f) receive and/or access the hotspot list; and (g) For each hotspot in the hotspot list, use an analytical model to segment the hotspot. The method of any one of claims 95 to 105, further comprising: (h) receive and/or access the heat map; and (i) For each hot spot in the heat map, use an analytical model to segment the hot spot. The method of claim 107, wherein the analytical model is an adaptive bounding method, and step (i) comprises: determining one or more reference values, each reference value based on a measure of the intensity of the voxels of the 3D functional image positioned within a specific reference volume corresponding to a specific reference tissue region; and For each specific hotspot volume of this 3D heatmap: determining, by the processor, a corresponding hotspot intensity based on intensities of voxels within the particular hotspot volume; and By the processor, a hotspot specific threshold is determined for the specific hotspot based on (i) the corresponding hotspot intensity and (ii) at least one of the one or more reference values. The method of claim 108, wherein the hotspot specific threshold is determined using a specific threshold function selected from a plurality of threshold functions, the specific threshold function based on a comparison of the corresponding hotspot intensity with the at least one reference value to choose. The method of claim 108 or 109, wherein the hotspot specific threshold is determined as a variable percentage of the corresponding hotspot intensity, wherein the variable percentage decreases as the hotspot intensity increases. A method for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) automatically segmenting, by the processor, a first initial set of one or more hot spots in the 3D functional image using a first machine learning module, thereby establishing a first initial 3D hot spot identifying the first initial hot spot volume set Figure, wherein the first machine learning module segments the hotspots of the 3D functional image according to a single hotspot category; (c) automatically segmenting, by the processor, a second initial set of one or more hotspots within the 3D functional image using a second machine learning module, thereby establishing a second initial 3D hotspot identifying the second initial set of hotspot volumes , wherein the second machine learning module segments the 3D functional image according to a plurality of different hotspot categories, so the second initial 3D heatmap is a multi-category 3D heatmap, wherein each hotspot volume is marked as belonging to the plurality of different hotspots specific ones of the hotspot category; (d) by the processor, for each specific hotspot volume of at least a portion of the first initial hotspot volume set identified by the first initial 3D heatmap, perform the following operations to merge the first initial 3D heatmap and the first initial 3D heatmap Two initial 3D heat maps: identifying a matching hotspot volume of the second initial 3D heatmap that has been marked as belonging to a particular hotspot category of the plurality of different hotspot categories; and marking the specific hotspot volume of the first initial 3D heat map as belonging to the specific hotspot class, thereby creating a merged 3D heatmap that includes the segmented hotspot volume of the first 3D heatmap, which has been The matching hot spots of the second 3D heat map are identified according to the category to which they belong; and (e) store and/or provide the merged 3D heat map for display and/or further processing. The method of claim 111, wherein the plurality of different hotspot categories comprise one or more members selected from the group consisting of: (i) Bone hot spots, which are determined to represent lesions localized in the bone; (ii) lymphatic hot spots, which are determined to represent lesions localized in lymph nodes, and (iii) Prostate hot spots, which are determined to represent lesions localized in the prostate. A method of automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject via an adaptive bounding method, the method comprising: (a) receiving, by the processor of the computing device, a 3D functional image of the subject obtained using a functional imaging modality; (b) receiving, by the processor, a preliminary 3D heat map identifying one or more preliminary hot spot volumes within the 3D functional image; (c) determining, by the processor, one or more reference values, each reference value based on a measure of the intensity of a voxel of the 3D functional image positioned within a specific reference volume corresponding to a specific reference tissue region; (d) performing, by the processor, the following for each specific preliminary hotspot volume of at least a portion of the one or more preliminary hotspot volumes identified by the preliminary 3D heat map, based on the preliminary hotspot volumes and using the self-based Adapt to limited segmentation and create a refined 3D heat map: determining the corresponding hot spot intensity based on the intensities of the voxels within the specific preliminary hot spot volume; for the specific preliminary hotspot volume, determining a hotspot specific threshold based on (i) the corresponding hotspot intensity and (ii) at least one of the one or more reference values; segmenting at least a portion of the 3D functional image using a threshold-based segmentation algorithm that performs image segmentation using the hotspot-specific threshold value for the specific preliminary hotspot volume determination, thereby determining the image corresponding to the specific hotspot volume Refinement of preliminary hotspot volumes, analysis of segmented hotspot volumes; and including the refined hot spot volume in the refined 3D heat map; and (e) storing and/or providing the refined 3D heat map for display and/or further processing. The method of claim 113, wherein the hotspot specific threshold is determined using a specific threshold function selected from a plurality of threshold functions, the specific threshold function based on a comparison of the corresponding hotspot intensity with the at least one reference value to choose. The method of claim 113 or 114, wherein the hotspot specific threshold is determined as a variable percentage of the corresponding hotspot intensity, wherein the variable percentage decreases as the hotspot intensity increases. A method for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the method comprising: (a) receiving, by a processor of the computing device, a 3D anatomical image of the subject obtained using an anatomical imaging modality, wherein the 3D anatomical image includes a graphical representation of tissue within the subject; (b) automatically segmenting the 3D anatomical image by the processor to create a 3D segmentation map that identifies a plurality of volumes of interest (VOIs) in the 3D anatomical image, including the liver corresponding to the subject's liver volume and the volume of the aorta corresponding to the part of the aorta; (c) receiving, by the processor, a 3D functional image of the subject obtained using a functional imaging modality; (d) automatically segmenting, by the processor, one or more hotspots within the 3D functional image, each segmented hotspot corresponding to a localized region of increased intensity relative to its surroundings and representing a potential cancerous lesion within the subject , thereby identifying one or more automatically segmented hotspot volumes; (e) causing, by the processor, a graphical representation of the one or more automatically segmented hotspot volumes to be displayed for display in an interactive graphical user interface (GUI); (f) receiving, by the processor, via the interactive GUI, a user selection of a final hotspot set comprising at least a portion of the one or more automatically segmented hotspot volumes; (g) for each hotspot volume of the final set, by the processor, based on (i) the intensities of the voxels of the functional image corresponding to the hotspot volume and (ii) using the volume corresponding to the liver and the aorta One or more reference values for the determination of the intensity of the voxels of the functional image of the volume to determine the lesion index value; and (e) storing and/or providing the final hotspot set and/or the lesion index values for display and/or further processing. The method of claim 116, wherein: Step (b) includes segmenting the anatomical image such that the 3D segmentation map identifies one or more bone volumes corresponding to one or more bones of the subject, and Step (d) includes, within the functional image, identifying a bone volume using the one or more bone volumes, and segmenting one or more bone hot spot volumes located within the bone volume. A method as in claim 116 or 117, wherein: Step (b) includes segmenting the anatomical image such that the 3D segmentation map identifies one or more organ volumes corresponding to soft tissue organs of the subject, and Step (d) includes identifying, within the functional image, one or more soft tissue volumes using the one or more segmented organ volumes, and segmenting one or more lymphoid and/or prostate hot spot volumes located within the soft tissue volume . The method of claim 118, wherein step (d) further comprises, prior to segmenting the one or more lymphatic and/or prostate hot spot volumes, adjusting the intensity of the functional image to suppress intensity from one or more high uptake tissue regions . The method of any one of claims 116 to 119, wherein step (g) includes determining a liver reference value using the intensities of the voxels of the functional image corresponding to the liver volume. The method of claim 120, comprising fitting a two-component Gaussian mixture model to a histogram of intensities corresponding to functional image voxels of the liver volume, using the two-component Gaussian mixture model Fitting to identify and exclude voxels with intensities associated with regions of abnormally low uptake from the liver volume, and use the intensities of the remaining voxels to determine the liver reference value. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the system comprising: a processor of a computing device; and memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) receiving a 3D anatomical image of the subject obtained using an anatomical imaging modality; (c) receiving a 3D segmentation map identifying one or more specific tissue regions or groups of tissue regions within the 3D functional image and/or within the 3D anatomical image; (d) using one or more machine learning modules to automatically detect and/or segment a set of one or more hotspots within the 3D functional image, each hotspot corresponding to a local area of increased intensity relative to its surroundings, and representing the Potentially cancerous lesions within the subject, thereby establishing one or both of the following (i) and (ii): (i) a list of hotspots that identifies the location of that hotspot for each hotspot, and (ii) a 3D heatmap, It identifies the corresponding 3D hotspot volume in the 3D functional image for each hotspot, wherein at least one of the one or more machine learning modules receives as input (i) the 3D functional image, (ii) the 3D anatomical image, and (iii) the 3D segmentation map; and (e) storing and/or providing the hot spot list and/or the 3D heat map for display and/or further processing. The system of claim 122, wherein the instructions cause the processor to: receiving an initial 3D segmentation map that identifies one or more specific tissue regions within the 3D anatomical image and/or the 3D functional image; identifying at least a portion of the one or more specific tissue regions as specific ones belonging to one or more tissue groups, and updating the 3D segmentation map to indicate that the identified specific regions belong to the specific tissue group; and The updated 3D segmentation map is used as an input to at least one of the one or more machine learning modules. The system of claim 123, wherein the one or more tissue groups comprise soft tissue groups, thereby identifying the particular tissue region representing soft tissue as belonging to the soft tissue group. The system of claim 123 or 124, wherein the one or more tissue groups include a bone tissue group, thereby identifying the particular tissue region representing bone as belonging to the bone tissue group. The system of any one of claims 123 to 125, wherein the one or more tissue groups comprise a high uptake organ group, thereby identifying the one or more organs associated with high radiopharmaceutical uptake as belonging to the high uptake group. The system of any of claims 122-126, wherein the instructions cause the processor to determine, for each detected and/or segmented hotspot, a classification of the hotspot. The system of claim 127, wherein the instructions cause the processor to use at least one of the one or more machine learning modules to determine the classification of the hotspot for each detected and/or segmented hotspot. The system of any one of claims 122 to 128, wherein the one or more machine learning modules comprise: (A) a systemic lesion detection module that detects and/or segments hot spots throughout the body; and (B) Prostate lesion module to detect and/or segment hot spots in the prostate. The system of claim 129, wherein the instructions cause the processor to generate the hotspot list and/or the maps using each of (A) and (B), and to combine the results. The system of any one of claims 122 to 130, wherein: In step (d), the instructions cause the processor to segment and classify the set of one or more hotspots to create a labeled 3D heatmap for each hotspot by the following operations Identifying the corresponding 3D hotspot volume within the 3D functional image, and wherein each hotspot is marked as belonging to a specific hotspot category of the plurality of hotspot categories: segmenting a first initial set of one or more hot spots within the 3D functional image using a first machine learning module to create a first initial 3D heat map identifying the first initial set of hot spot volumes, wherein the first machine learning module Segment the hotspots of the 3D functional image according to a single hotspot category; segmenting a second initial set of one or more hot spots within the 3D functional image using a second machine learning module to create a second initial 3D heat map identifying the second initial set of hot spot volumes, wherein the second machine learning module The 3D functional image is segmented according to the plurality of different hotspot categories, so that the second initial 3D heatmap is a multi-category 3D heatmap, wherein each hotspot volume is marked as belonging to a specific one of the plurality of different hotspot categories; and The first initial 3D heat map and the second initial 3D heat map are merged by performing the following operations on at least a portion of the hot spot volumes identified by the first initial 3D heat map: identifying a matching hotspot volume of the second initial 3D heatmap that has been marked as belonging to a particular hotspot category of the plurality of different hotspot categories; and marking the specific hotspot volume of the first initial 3D heat map as belonging to the specific hotspot class, thereby creating a merged 3D heatmap that includes the segmented hotspot volume of the first 3D heatmap, which has been The matching hotspots of the second 3D heatmap are identified according to the category to which they belong; and In step (e), the instructions cause the processor to store and/or provide the merged 3D heat map for display and/or further processing. The system of claim 131, wherein the plurality of different hotspot categories include one or more members selected from the group consisting of: (i) Bone hot spots, which are determined to represent lesions localized in the bone; (ii) lymphatic hot spots, which are determined to represent lesions localized in lymph nodes, and (iii) Prostate hot spots, which are determined to represent lesions localized in the prostate. The system of any of claims 122-132, wherein the instructions further cause the processor to: (f) receive and/or access the hotspot list; and (g) For each hotspot in the hotspot list, use an analytical model to segment the hotspot. The system of any one of claims 122 to 133, wherein the instructions further cause the processor to: (h) receive and/or access the heat map; (i) For each hot spot in the heat map, use an analytical model to segment the hot spot. The system of claim 134, wherein the analytical model is an adaptive bounding method, and in step (i), the instructions cause the processor to: determining one or more reference values, each reference value based on a measure of the intensity of voxels of the 3D functional image positioned within a specific reference volume corresponding to a specific reference tissue region; and For each specific hotspot volume of this 3D heatmap: Determine the corresponding hot spot intensity based on the intensities of the voxels within the particular hot spot volume; and For the specific hotspot, a hotspot specific threshold is determined based on (i) the corresponding hotspot intensity and (ii) at least one of the one or more reference values. The system of claim 135, wherein the hotspot specific threshold is determined using a specific threshold function selected from a plurality of threshold functions, the specific threshold function based on a comparison of the corresponding hotspot intensity with the at least one reference value to choose. The system of claim 135 or 136, wherein the hotspot specific threshold is determined as a variable percentage of the corresponding hotspot intensity, wherein the variable percentage decreases as the hotspot intensity increases. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the system comprising: a processor of a computing device; and memory having instructions stored thereon that, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) using a first machine learning module to automatically segment a first initial set of one or more hot spots in the 3D functional image, thereby establishing a first initial set of hot spot volumes that identify the corresponding 3D hot spot volumes in the 3D functional image the first initial 3D heat map, wherein the first machine learning module segments the hotspots of the 3D functional image according to a single hotspot category; (c) using a second machine learning module to automatically segment a second initial set of one or more hot spots in the 3D functional image, thereby creating a second initial 3D heat map identifying the second initial set of hot spot volumes, wherein the first Two machine learning modules segment the 3D functional image according to a plurality of different hotspot categories, so the second initial 3D heat map is a multi-category 3D heat map, wherein each hotspot volume is marked as belonging to a specific one of the plurality of different hotspot categories; (d) merging the first initial 3D heat map and the second initial 3D heat map by performing the following operations for each specific hotspot volume of at least a portion of the first initial hotspot volume set identified by the first initial 3D heatmap : identifying a matching hotspot volume of the second initial 3D heatmap that has been marked as belonging to a particular hotspot category of the plurality of different hotspot categories; and marking the specific hotspot volume of the first initial 3D heat map as belonging to the specific hotspot class, thereby creating a merged 3D heatmap that includes the segmented hotspot volume of the first 3D heatmap, which has been The matching hot spots of the second 3D heat map are identified according to the category to which they belong; and (e) store and/or provide the merged 3D heat map for display and/or further processing. The system of claim 138, wherein the plurality of different hotspot categories include one or more members selected from the group consisting of: (i) Bone hot spots, which are determined to represent lesions localized in the bone; (ii) lymphatic hot spots, which are determined to represent lesions localized in lymph nodes, and (iii) Prostate hot spots, which are determined to represent lesions localized in the prostate. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject via an adaptive bounding method, the system comprising: a processor of a computing device; and memory having instructions stored thereon that, when executed by the processor, cause the processor to: (a) receive a 3D functional image of the subject obtained using a functional imaging modality; (b) receiving a preliminary 3D heat map identifying one or more preliminary hot spot volumes within the 3D functional image; (c) determining one or more reference values, each reference value based on a measure of the intensity of the voxels of the 3D functional image positioned within a specific reference volume corresponding to a specific reference tissue region; (d) by doing the following for each particular preliminary hotspot volume of at least a portion of the one or more preliminary hotspot volumes identified by the preliminary 3D heat map, based on the preliminary hotspot volumes and using adaptive bound-based segmentation , to create a refined 3D heat map: determine the corresponding hot spot intensity based on the intensities of the voxels within the particular preliminary hot spot volume; and for the specific preliminary hotspot volume, determining a hotspot specific threshold based on (i) the corresponding hotspot intensity and (ii) at least one of the one or more reference values; Segmenting at least a portion of the 3D functional image using a threshold-based segmentation algorithm that performs image segmentation using the hotspot-specific threshold determined for the specific preliminary hotspot, thereby determining the corresponding specific preliminary hotspot Volume refinement, analysis of segmented hot spot volumes; and including the refined hot spot volume in the refined 3D heat map; and (e) storing and/or providing the refined 3D heat map for display and/or further processing. The system of claim 140, wherein the hotspot specific threshold is determined using a specific threshold function selected from a plurality of threshold functions, the specific threshold function based on a comparison of the corresponding hotspot intensity with the at least one reference value to choose. The system of claim 140 or 141, wherein the hotspot specific threshold is determined as a variable percentage of the corresponding hotspot intensity, wherein the variable percentage decreases as the hotspot intensity increases. A system for automatically processing 3D images of a subject to identify and/or characterize cancerous lesions in the subject, the system comprising: a processor of a computing device; and memory having instructions stored thereon that, when executed by the processor, cause the processor to: (a) receiving a 3D anatomical image of the subject obtained using an anatomical imaging modality, wherein the 3D anatomical image includes a graphical representation of tissue within the subject; (b) automatically segmenting the 3D anatomical image to create a 3D segmentation map that identifies a plurality of volumes of interest (VOIs) in the 3D anatomical image, including liver volumes corresponding to the subject's liver and liver volumes corresponding to the subject's liver Aortic volume of the arterial part; (c) receive a 3D functional image of the subject obtained using a functional imaging modality; (d) automatically segment one or more hotspots within the 3D functional image, each segmented hotspot corresponding to a localized region of increased intensity relative to its surroundings and representing a potential cancerous lesion within the subject, thereby identifying an or Multiple auto-segmented hotspot volumes; (e) cause a graphical representation of the one or more automatically segmented hotspot volumes to be displayed for display in an interactive graphical user interface (GUI); (f) receiving, via the interactive GUI, a user selection of a final hotspot set comprising at least a portion of the one or more automatically segmented hotspot volumes; (g) for each hotspot volume of the final set, based on (i) the intensities of the voxels of the functional image corresponding to the hotspot volume and (ii) using the one or more reference values for the determination of the intensity of the voxel to determine the lesion index value; and (e) storing and/or providing the final hotspot set and/or the lesion index values for display and/or further processing. The system of claim 143, wherein: In step (b), the instructions cause the processor to segment the anatomical image such that the 3D segmentation map identifies one or more bone volumes corresponding to one or more bones of the subject, and In step (d), the instructions cause the processor to identify a bone volume within the functional image using the one or more bone volumes, and segment one or more bone hot spot volumes located within the bone volume. A system as claimed in claim 143 or 144, wherein: In step (b), the instructions cause the processor to segment the anatomical image such that the 3D segmentation map identifies one or more organ volumes corresponding to soft tissue organs of the subject, and In step (d), the instructions cause the processor to identify a soft tissue volume within the functional image using the one or more segmented organ volumes, and segment one or more lymph nodes and/or lymph nodes located within the soft tissue volume Prostate hot spot volume. The system of claim 145, wherein in step (d) the instructions cause the processor to adjust the intensity of the functional image to suppress the intensity of the functional image prior to segmenting the one or more lymphatic and/or prostate hotspot volumes to suppress the Strength of multiple high uptake tissue areas. The system of any one of claims 143 to 146, wherein in step (g) the instructions cause the processor to use intensities of voxels of the functional image corresponding to the liver volume to determine liver reference values. The system of claim 147, wherein the instructions cause the processor to: fitting a two-component Gaussian mixture model to a histogram of the intensities of the functional image voxels corresponding to the liver volume, using the two-component Gaussian mixture model fit to identify and exclude voxels with intensities associated with regions of abnormally low uptake from the liver volume, and The liver reference value is determined using the intensities of the remaining voxels.

Images