WO2021139835A2 - 一种spect成像预测模型创建方法、装置、设备及存储介质 - Google Patents

一种spect成像预测模型创建方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2021139835A2
WO2021139835A2 PCT/CN2021/083206 CN2021083206W WO2021139835A2 WO 2021139835 A2 WO2021139835 A2 WO 2021139835A2 CN 2021083206 W CN2021083206 W CN 2021083206W WO 2021139835 A2 WO2021139835 A2 WO 2021139835A2
Authority
WO
WIPO (PCT)
Prior art keywords
spect
acquisition duration
image
network
prediction model
Prior art date
Application number
PCT/CN2021/083206
Other languages
English (en)
French (fr)
Other versions
WO2021139835A3 (zh
Inventor
龚南杰
项磊
潘博洋
Original Assignee
苏州深透智能科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 苏州深透智能科技有限公司 filed Critical 苏州深透智能科技有限公司
Priority to US18/551,876 priority Critical patent/US20240177377A1/en
Publication of WO2021139835A2 publication Critical patent/WO2021139835A2/zh
Publication of WO2021139835A3 publication Critical patent/WO2021139835A3/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G06T11/006—Inverse problem, transformation from projection-space into object-space, e.g. transform methods, back-projection, algebraic methods
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/02—Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
    • A61B6/03—Computed tomography [CT]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/02—Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
    • A61B6/03—Computed tomography [CT]
    • A61B6/037—Emission tomography
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
    • A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
    • A61B6/5229—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data combining image data of a patient, e.g. combining a functional image with an anatomical image
    • A61B6/5235—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data combining image data of a patient, e.g. combining a functional image with an anatomical image combining images from the same or different ionising radiation imaging techniques, e.g. PET and CT
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00—Indexing scheme for image generation or computer graphics
    • G06T2210/41—Medical
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography
    • G06T2211/421—Filtered back projection [FBP]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography
    • G06T2211/441—AI-based methods, deep learning or artificial neural networks

Definitions

  • the present invention relates to the field of SPECT technology, in particular to a method, device, equipment and storage medium for creating a SPECT imaging prediction model.
  • Single-Photon Emission Computed Tomography is a method of injecting radioisotope-labeled compounds into organisms to collect high-energy gamma particles produced by decay in different directions and at the same time during biological metabolism.
  • Medical imaging technology that obtains projection signals and performs imaging through computational reconstruction. Since the signal collection depends on the radioisotope half-life and the number of signal collection angles, SPECT images face long imaging time in clinical applications and have certain radiation damage to the human body. However, if the imaging angle is reduced or the imaging time is shortened, the imaging results will be caused. It has a series of shortcomings that damage the image quality, such as low signal-to-noise ratio and easy artifacts. Therefore, how to generate high-quality SPECT images in a short acquisition time is a problem that needs to be solved urgently.
  • the purpose of the present invention is to provide a SPECT imaging prediction model creation method, device, equipment and medium, which can greatly reduce the SPECT imaging time and improve the SPECT imaging of subjects while preserving the imaging quality of medical images. Comfort, reduce motion artifacts.
  • the specific plan is as follows:
  • this application discloses a method for creating a SPECT imaging prediction model, including:
  • each scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other;
  • the short acquisition duration SPECT images in the training set are used as the input-side training data, and the standard acquisition duration SPECT images in the training set are used as the output-side training data, and the network to be trained is trained to obtain a SPECT imaging prediction model, so that The SPECT imaging prediction model is used to predict and obtain a SPECT prediction image of a short acquisition duration SPECT image under a standard acquisition duration.
  • the scan image group includes the standard acquisition duration SPECT image, the short acquisition duration SPECT image, and the CT image corresponding to each other; wherein, the CT image is used as the input side training data of the network to be trained .
  • the acquiring a training set containing multiple scanned image groups includes:
  • the scan image group is obtained.
  • the obtaining of the standard acquisition duration gamma particle signal and the short acquisition duration gamma particle signal under the same acquisition condition by the single-photon emission computed tomography imaging device includes:
  • the standard acquisition duration ⁇ particle signal and the short acquisition duration ⁇ particle signal under the same acquisition condition are collected;
  • the single-photon emission computed tomography imaging device uses the single-photon emission computed tomography imaging device to acquire the standard acquisition duration gamma particle signal according to the standard acquisition duration, and then perform down-collection of the standard acquisition duration gamma particle signal to obtain the corresponding short acquisition duration gamma particle signal.
  • the collection conditions include the start time of collection, the person to be collected, the radioisotope drug measurement, and the collection angle.
  • the reconstruction algorithm includes any one of a filtered back-projection reconstruction algorithm, an algebraic reconstruction algorithm, and a fast subset conjugate gradient reconstruction algorithm.
  • the network construction based on the deep convolutional neural network to obtain the network to be trained includes:
  • the network is constructed based on the U2-Net network structure to obtain the network to be trained.
  • this application discloses a SPECT imaging prediction model creation device, including:
  • the training set acquisition module is used to acquire a training set containing multiple scan image groups; wherein each scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other;
  • the network building module is used to construct the network based on the deep convolutional neural network to obtain the network to be trained;
  • the model training module is configured to use the short acquisition duration SPECT images in the training set as the input side training data, and the standard acquisition duration SPECT images in the training set as the output side training data, and train the network to be trained to obtain SPECT imaging prediction model, so as to use the SPECT imaging prediction model to predict and obtain a SPECT prediction image of a short acquisition duration SPECT image under a standard acquisition duration.
  • this application discloses a SPECT imaging prediction method, including:
  • the short acquisition duration SPECT image to be predicted is input to the SPECT imaging prediction model to predict the SPECT prediction image of the short acquisition duration SPECT image to be predicted under the standard acquisition duration; wherein, the SPECT imaging prediction model uses A training set of multiple scan image groups is a model obtained by training a network to be trained based on a deep convolutional neural network; wherein the scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other.
  • a SPECT imaging prediction device including:
  • the image acquisition module is used to acquire a SPECT image with a short acquisition duration to be predicted
  • the image prediction module is used to input the short acquisition duration SPECT image to be predicted into the SPECT imaging prediction model to predict the SPECT prediction image of the short acquisition duration SPECT image to be predicted under the standard acquisition duration; wherein, the SPECT
  • the imaging prediction model is a model obtained by using a training set containing multiple scanned image groups to train a network to be trained based on a deep convolutional neural network; wherein the scanned image group includes mutually corresponding standard acquisition time SPECT images and Short acquisition duration SPECT images.
  • this application discloses an electronic device, including:
  • Memory used to store computer programs
  • the processor is used to execute the computer program to realize the aforementioned SPECT imaging prediction model creation method.
  • the present application discloses a computer-readable storage medium for storing a computer program; wherein the computer program is executed by a processor to realize the aforementioned SPECT imaging prediction model creation method.
  • a training set containing multiple scan image groups is acquired; wherein each scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other; the network is constructed based on a deep convolutional neural network , Obtain the network to be trained; use the short acquisition duration SPECT images in the training set as the input side training data, and use the standard acquisition duration SPECT images in the training set as the output side training data, and train the network to be trained to obtain SPECT imaging prediction model, so as to use the SPECT imaging prediction model to predict and obtain a SPECT prediction image of a short acquisition duration SPECT image under a standard acquisition duration.
  • the SPECT imaging prediction model is trained, and then the short acquisition time is predicted by the SPECT imaging prediction model. Acquisition time SPECT image of the SPECT prediction image under the standard acquisition time. It can greatly reduce the SPECT imaging time while preserving the imaging quality of medical images, improve the comfort of SPECT imaging of subjects, and reduce motion artifacts.
  • Figure 1 is a flow chart of a method for creating a SPECT imaging prediction model provided by this application
  • FIG. 2 is a schematic diagram of the structure of a SPECT imaging prediction model provided by this application.
  • FIG. 3 is a flowchart of a specific method for creating a SPECT imaging prediction model provided by this application;
  • FIG. 4 is a schematic diagram of a SPECT imaging provided by this application.
  • Fig. 5 is a flow chart of a SPECT imaging prediction method provided by this application.
  • FIG. 6 is a schematic structural diagram of a SPECT imaging prediction model creation device provided by this application.
  • FIG. 7 is a schematic structural diagram of a SPECT imaging prediction device provided by this application.
  • FIG. 8 is a structural diagram of an electronic device provided by this application.
  • this application proposes a method for creating a SPECT imaging prediction model.
  • the embodiment of the application discloses a method for creating a SPECT imaging prediction model. As shown in FIG. 1, the method may include the following steps:
  • Step S11 Obtain a training set containing a plurality of scan image groups; wherein each scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other.
  • a plurality of scan image groups are acquired as a training set, where each scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other.
  • the standard acquisition duration SPECT image and the short acquisition duration SPECT image are the images acquired and reconstructed by the single-photon emission computed tomography scanner according to the standard acquisition strategy and the fast scanning strategy under the same acquisition conditions, and the SPECT image It can be acquired based on SPECT equipment or SPECT+CT equipment.
  • the scan image group may include the standard acquisition duration SPECT image, the short acquisition duration SPECT image, and the CT image that correspond to each other; wherein, the CT image serves as the input side of the network to be trained Training data.
  • the aforementioned scan image group may also include CT images corresponding to the aforementioned standard acquisition duration SPECT image and short acquisition duration SPECT image, that is, CT images are also included.
  • the above-mentioned CT image can be acquired by a SPECT+CT device or can be acquired by a CT device.
  • CT images have the advantage of clear anatomical structure
  • SPECT images have the characteristics of reflecting the physiology, metabolism and function of organs.
  • Step S12 Construct the network based on the deep convolutional neural network to obtain the network to be trained.
  • the network is constructed based on the deep convolutional neural network, and the network to be trained is obtained. Specifically, sub-encoders and sub-decoders are constructed based on the convolutional layer, the BN (Batch Normalization) layer and the Relu layer, and then pooling The layers or sampling layers are sequentially connected with different preset numbers of sub-encoders and sub-decoders to construct encoders and decoders of different sizes, and finally the network to be trained is obtained by connecting the encoders and decoders of different sizes in series .
  • the aforementioned deep convolutional neural network may be an image reconstruction convolutional network composed of N convolutional neural network units (CNN, Convolutional Neural Networks) in series including a symmetric serially connected encoder-decoder structure, where Each convolutional neural network unit is composed of a preset number of convolutional layers, pooling layers, nonlinear layers, skip connection layers, and corresponding down-sampling and up-sampling layers.
  • CNN convolutional neural network units
  • Each convolutional neural network unit is composed of a preset number of convolutional layers, pooling layers, nonlinear layers, skip connection layers, and corresponding down-sampling and up-sampling layers.
  • the network construction based on the deep convolutional neural network to obtain the network to be trained may include: network construction based on the U2-Net network structure to obtain the network to be trained. It is understandable that a network to be trained including multiple encoders and decoders connected in series is constructed based on the U2-Net network structure. Specifically, the structure diagram of the network to be trained is shown in Figure 2. The encoder and decoder Each stage of the device structure is composed of RSU units (residual U-block), and each RSU is composed of sub-encoders and sub-decoders connected in series with 3-5 stages.
  • the sub-encoder and sub-decoder are composed of several 3x3 convolutional layers, BN layers and Relu nonlinear layers, and the sub-encoder and sub-decoder are connected by a 2x2 pooling layer or an up-sampling layer. And through the Dilation Conv contained in the RSU, receptive fields of different sizes are mixed, so regardless of the resolution, the model can capture local and global information from the shallow and deep layers, without significantly increasing the amount of calculation Down also increases the depth of the entire network. At the same time, Skip Connection is used to introduce the feature information on the corresponding scale into the up-sampling process, and solve the problem of gradient explosion and gradient disappearance in the training process in the deep network.
  • Step S13 Use the short acquisition duration SPECT images in the training set as input side training data, and use the standard acquisition duration SPECT images in the training set as output side training data, and train the network to be trained to obtain SPECT imaging predictions Model, so as to use the SPECT imaging prediction model to predict and obtain a SPECT prediction image of a short acquisition duration SPECT image under a standard acquisition duration.
  • the short acquisition duration SPECT images in the above training set are used as the input side training data
  • the standard acquisition duration SPECT images in the above training set are used as the output side training data.
  • the constructed network to be trained is trained to obtain a SPECT imaging prediction model, so that the SPECT imaging prediction model can be used to predict and obtain a SPECT prediction image of a short acquisition time SPECT image under a standard acquisition time.
  • the above-mentioned training set can be divided into training data and test data, the model is trained through the training data, and the performance of the model is tested through the test data to obtain a SPECT imaging prediction model that meets the prediction standard.
  • a training set containing multiple scan image groups is acquired; wherein each scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other; a deep convolutional neural network is used as The network is constructed based on the foundation to obtain the network to be trained; the short acquisition duration SPECT image in the training set is used as the input side training data, and the standard acquisition duration SPECT image in the training set is used as the output side training data. Training is performed to obtain a SPECT imaging prediction model, so as to use the SPECT imaging prediction model to predict and obtain a SPECT prediction image of a SPECT image with a short acquisition duration under a standard acquisition duration.
  • the SPECT imaging prediction model is trained, and then the short acquisition time is predicted by the SPECT imaging prediction model. Acquisition time SPECT image of the SPECT prediction image under the standard acquisition time. It can greatly reduce the SPECT imaging time while preserving the imaging quality of medical images, improve the comfort of SPECT imaging of subjects, and reduce motion artifacts.
  • the embodiment of the present application discloses a specific method for creating a SPECT imaging prediction model. As shown in FIG. 3, the method may include the following steps:
  • Step S21 Use the single-photon emission computed tomography imaging device to acquire the standard acquisition time ⁇ particle signal and the short acquisition time ⁇ particle signal under the same acquisition condition according to the standard acquisition time length and the short acquisition time length.
  • a single-photon emission computer tomography imaging device specifically a SPECT/CT scanner can be used to acquire the standard acquisition time ⁇ particle signal and the short acquisition time ⁇ particle under the same acquisition conditions according to the standard acquisition time and short acquisition time. signal.
  • the collection conditions may include the start time of collection, the person to be collected, the radioisotope drug measurement, and the collection angle.
  • the single-photon emission computed tomography equipment is used to collect more than 10 sets of gamma particle signals with a standard collection duration and a short collection duration under the same collection conditions, that is, the same patient is injected with the same isotope-labeled compound at the same dose at the same time , And collect the projection signal groups of the same number of angles, that is, two signals with different numbers of gamma particles collected under the same projection angle are obtained.
  • the above-mentioned short acquisition time length ⁇ particle signal can be 1/7 of the standard acquisition time length.
  • the SPECT/CT scanner Siemens-Symbia-Intevo is used to collect 20 subjects with whole-body quantitative bone imaging.
  • the injection dose can be Two scanning protocols are set for the subjects at 25 ⁇ 30mci, one is standard scanning, 20 seconds per frame, which will get the ⁇ particle signal of standard acquisition time, and the other is fast scanning, 3 seconds per frame to get 1 /7 Gamma particle signal of acquisition time.
  • Other sampling parameters are: 60 frames, single probe rotation 180°, single rotation 6°.
  • the ordered subset conjugate gradient (OSCG) algorithm is used to reconstruct the SPECT projection data to obtain corresponding standard acquisition duration SPECT images and short acquisition duration SPECT images. It should be noted that the collection scope of this embodiment includes, but is not limited to, brain, bone, heart and other parts.
  • the single-photon emission computed tomography imaging device can also be used to acquire the standard acquisition duration ⁇ particle signal according to the standard acquisition duration, and then the standard acquisition duration ⁇ particle signal is reduced to acquire the corresponding short acquisition duration ⁇ Particle signal.
  • the short acquisition duration ⁇ particle signal can be acquired through acquisition, or it can be acquired by down-sampling the standard acquisition duration ⁇ particle signal.
  • Step S22 Use a reconstruction algorithm to reconstruct the standard acquisition duration ⁇ particle signal and the short acquisition duration ⁇ particle signal to obtain corresponding standard acquisition duration SPECT images and short acquisition duration SPECT images.
  • a reconstruction algorithm is used to reconstruct the acquired standard acquisition duration gamma particle signals and short acquisition duration gamma particle signals to obtain corresponding standard acquisition duration SPECT images and short acquisition duration SPECT images.
  • the reconstruction algorithm includes, but is not limited to, the Filter Back Projection (FBP) reconstruction algorithm, the Algebraic Reconstruction Technique (ART), and the Fast Subset Conjugate Gradient Reconstruction Algorithm (OSCG, Ordered Subset). Conjugate Gradiental).
  • Step S23 Obtain the scan image group based on the standard acquisition duration SPECT image and the short acquisition duration SPECT image to obtain a training set containing multiple scan image groups.
  • a scan image group is obtained according to the standard acquisition duration SPECT image and the short acquisition duration SPECT image obtained above, so as to obtain a training set containing multiple scan image groups.
  • Step S24 Construct a network based on the deep convolutional neural network to obtain a network to be trained.
  • Step S25 Use the short acquisition duration SPECT images in the training set as input side training data, and use the standard acquisition duration SPECT images in the training set as output side training data, and train the network to be trained to obtain SPECT imaging predictions Model, so as to use the SPECT imaging prediction model to predict and obtain a SPECT prediction image of a short acquisition duration SPECT image under a standard acquisition duration.
  • the SPECT imaging prediction model obtained by training, predicts and obtains the SPECT prediction image of the 1/7 SPECT image under the standard acquisition time according to the input 1/7 SPECT image.
  • the model can be evaluated qualitatively and quantitatively by using structural similarity index (SSIM, Structural SIMilarity) and peak signal-to-noise ratio (PSNR, Peak Signal to Noise Ratio) indicators to detect model effects.
  • structural similarity index SSIM, Structural SIMilarity
  • PSNR Peak Signal to Noise Ratio
  • the 1/7 SPECT image shown in Figure 4 the SPECT predicted image obtained by deep learning reconstruction, and the SPECT image under standard acquisition, it can be seen that the quality of the predicted SPECT image is equivalent to that of the standard SPECT image, far better than the original 1/
  • 7SPECT images it is relatively difficult to identify the lesion area on the 1/7SPECT image, but it is easier to identify the lesion area on the predicted SPECT image.
  • step S24 and step S25 please refer to the corresponding content disclosed in the foregoing embodiment, which will not be repeated here.
  • the single-photon emission computed tomography imaging device is used to collect the standard acquisition time ⁇ particle signal and the short acquisition time ⁇ particle signal under the same acquisition conditions according to the standard acquisition time and short acquisition time, and then use the reconstruction algorithm to analyze the standard
  • the acquisition duration ⁇ particle signal and the short acquisition duration ⁇ particle signal are reconstructed to obtain corresponding standard acquisition duration SPECT images and short acquisition duration SPECT images; finally based on the standard acquisition duration SPECT images and the short acquisition duration SPECT images,
  • the scanned image group is obtained to obtain a training set containing a plurality of scanned image groups.
  • the short acquisition duration SPECT image is used as the input side training data
  • the standard acquisition duration SPECT image is used as the output side training data.
  • the network to be trained is trained to obtain the SPECT imaging prediction model, which can be predicted based on the short acquisition duration SPECT image.
  • this embodiment can greatly reduce the imaging time while preserving the image quality, and improve the SPECT imaging comfort of the subject, and can improve the signal-to-noise ratio. Motion artifacts.
  • the embodiment of the present application discloses a SPECT imaging prediction method. As shown in FIG. 5, the method may include the following steps:
  • Step S31 Acquire a short acquisition duration SPECT image to be predicted.
  • the short acquisition duration SPECT image to be predicted is acquired first.
  • Step S32 Input the short acquisition duration SPECT image to be predicted into the SPECT imaging prediction model to predict the SPECT prediction image of the short acquisition duration SPECT image to be predicted under the standard acquisition duration; wherein, the SPECT imaging prediction model In order to use a training set containing multiple scanned image groups to train a network to be trained based on a deep convolutional neural network; wherein the scanned image group includes mutually corresponding standard acquisition duration SPECT images and short acquisition duration SPECT image.
  • the acquired short acquisition duration SPECT image to be predicted is input to the SPECT imaging prediction model to predict and obtain the SPECT prediction image of the aforementioned short acquisition duration SPECT image to be predicted under the standard acquisition duration; wherein, the aforementioned SPECT imaging prediction model In order to use a training set containing multiple scan image groups to train a network to be trained based on a deep convolutional neural network; wherein the scan image group includes mutually corresponding standard acquisition duration SPECT images and short acquisition duration SPECT images image.
  • the acquisition process of the training set containing multiple scanned image groups includes: obtaining the standard acquisition duration gamma particle signal and the short acquisition duration gamma particle signal under the same acquisition condition through a single photon emission computed tomography imaging device; using a reconstruction algorithm Reconstruct the standard acquisition duration ⁇ particle signal and the short acquisition duration ⁇ particle signal to obtain corresponding standard acquisition duration SPECT images and short acquisition duration SPECT images; based on the standard acquisition duration SPECT images and the short acquisition duration SPECT images to obtain the scan image group.
  • obtaining the standard acquisition duration ⁇ particle signal and the short acquisition duration ⁇ particle signal under the same acquisition condition through the single photon emission computed tomography imaging device may include: using the single photon emission computed tomography imaging device according to the standard acquisition duration and short duration.
  • the acquisition time, the standard acquisition time ⁇ particle signal and the short acquisition time ⁇ particle signal under the same acquisition conditions are collected; or, the single photon emission computer tomography imaging device is used to acquire the standard acquisition time ⁇ particle signal according to the standard acquisition time, and then The ⁇ particle signal of the standard acquisition duration is reduced to obtain the corresponding ⁇ particle signal of the short acquisition duration.
  • the above-mentioned collection conditions include the start time of collection, the person to be collected, the radioisotope drug measurement, and the collection angle.
  • the aforementioned reconstruction algorithms include, but are not limited to, filtered back-projection reconstruction algorithms, algebraic reconstruction algorithms, and fast subset conjugate gradient reconstruction algorithms.
  • the aforementioned network to be trained may be a network to be trained obtained by constructing a network based on the U2-Net network structure.
  • the SPECT imaging prediction model is used to predict the acquired short acquisition duration SPECT images, and the SPECT prediction images under the standard acquisition duration corresponding to the short acquisition duration SPECT images are obtained.
  • the SPECT imaging time can be greatly reduced while the imaging quality of the medical image is preserved, and the SPECT imaging comfort of the subject can be improved, and motion artifacts can be reduced.
  • an embodiment of the present application also discloses a SPECT imaging prediction model creation device. As shown in FIG. 6, the device includes:
  • the training set acquisition module 11 is used to acquire a training set containing multiple scan image groups; wherein each scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other;
  • the network construction module 12 is used to construct the network based on the deep convolutional neural network to obtain the network to be trained;
  • the model training module 13 is configured to use the short acquisition duration SPECT images in the training set as the input side training data, and use the standard acquisition duration SPECT images in the training set as the output side training data to train the network to be trained, A SPECT imaging prediction model is obtained, so as to use the SPECT imaging prediction model to predict a SPECT prediction image of a SPECT image with a short acquisition duration under a standard acquisition duration.
  • a training set containing multiple scan image groups is acquired; wherein each scan image group includes a standard acquisition duration SPECT image and a short acquisition duration SPECT image corresponding to each other; a deep convolutional neural network is used as The network is constructed based on the foundation to obtain the network to be trained; the short acquisition duration SPECT image in the training set is used as the input side training data, and the standard acquisition duration SPECT image in the training set is used as the output side training data. Training is performed to obtain a SPECT imaging prediction model, so as to use the SPECT imaging prediction model to predict and obtain a SPECT prediction image of a SPECT image with a short acquisition duration under a standard acquisition duration.
  • the SPECT imaging prediction model is trained, and then the short acquisition time is predicted by the SPECT imaging prediction model. Acquisition time SPECT image of the SPECT prediction image under the standard acquisition time. It can greatly reduce the SPECT imaging time while preserving the imaging quality of medical images, improve the comfort of SPECT imaging of subjects, and reduce motion artifacts.
  • the scan image group may specifically include the standard acquisition duration SPECT image, the short acquisition duration SPECT image, and the CT image that correspond to each other; wherein, the CT image is used as the network to be trained. Input side training data.
  • the training set acquisition module 11 may specifically include:
  • the ⁇ particle signal acquisition unit is used to obtain the standard acquisition duration ⁇ particle signal and the short acquisition duration ⁇ particle signal under the same acquisition condition through the single-photon emission computed tomography imaging device;
  • An image reconstruction unit configured to use a reconstruction algorithm to reconstruct the standard acquisition duration ⁇ particle signal and the short acquisition duration ⁇ particle signal to obtain corresponding standard acquisition duration SPECT images and short acquisition duration SPECT images;
  • the scanning image group determining unit is configured to obtain the scanning image group based on the standard acquisition duration SPECT image and the short acquisition duration SPECT image.
  • the gamma particle signal collection unit may specifically include:
  • the first acquisition unit is configured to use the single-photon emission computed tomography imaging device to acquire the standard acquisition time ⁇ particle signal and the short acquisition time ⁇ particle signal under the same acquisition condition according to the standard acquisition time length and the short acquisition time length;
  • the second acquisition unit is configured to use the single-photon emission computed tomography imaging device to acquire the standard acquisition duration ⁇ particle signal according to the standard acquisition duration, and then perform down-collection of the standard acquisition duration ⁇ particle signal to obtain the corresponding short acquisition duration ⁇ Particle signal.
  • the collection conditions may specifically include the start time of collection, the person to be collected, the radioisotope drug measurement, and the collection angle.
  • the reconstruction algorithm may specifically include any one of a filtered back-projection reconstruction algorithm, an algebraic reconstruction algorithm, and a fast subset conjugate gradient reconstruction algorithm.
  • the network construction module 12 may specifically include:
  • the network construction unit is used to construct the network based on the U2-Net network structure to obtain the network to be trained.
  • the embodiment of the present application also discloses a SPECT imaging prediction device.
  • the device includes:
  • the image acquisition module 21 is used to acquire a short acquisition duration SPECT image to be predicted
  • the image prediction module 22 is configured to input the short acquisition duration SPECT image to be predicted into the SPECT imaging prediction model to predict and obtain the SPECT prediction image of the short acquisition duration SPECT image to be predicted under the standard acquisition duration; wherein, the The SPECT imaging prediction model is a model obtained by training a network to be trained based on a deep convolutional neural network using a training set containing multiple scan image groups; wherein the scan image group includes mutually corresponding standard acquisition time SPECT images And short acquisition duration SPECT images.
  • the SPECT imaging prediction model is used to predict the acquired short acquisition duration SPECT images, and the SPECT prediction images under the standard acquisition duration corresponding to the short acquisition duration SPECT images are obtained.
  • the SPECT imaging time can be greatly reduced while the imaging quality of the medical image is preserved, and the SPECT imaging comfort of the subject can be improved, and motion artifacts can be reduced.
  • the embodiment of the present application also discloses an electronic device. As shown in FIG. 8, the content in the figure cannot be regarded as any limitation on the scope of use of the present application.
  • FIG. 8 is a schematic structural diagram of an electronic device 30 provided by an embodiment of this application.
  • the electronic device 30 may specifically include: at least one processor 31, at least one memory 32, a power supply 33, a communication interface 34, an input/output interface 35, and a communication bus 36.
  • the memory 32 is used to store a computer program, and the computer program is loaded and executed by the processor 31 to implement the relevant steps in the SPECT imaging prediction model creation method disclosed in any of the foregoing embodiments.
  • the power supply 33 is used to provide working voltages for each hardware device on the electronic device 30; the communication interface 34 can create a data transmission channel for the electronic device 30 with external devices, and the communication protocol it follows is applicable Any communication protocol in the technical solution of this application is not specifically limited here; the input and output interface 35 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs. There is no specific limitation.
  • the memory 32 as a resource storage carrier, can be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc.
  • the resources stored on it include an operating system 321, a computer program 322, and data 323 including scanned image groups.
  • the storage method can be short-term storage or permanent storage.
  • the operating system 321 is used to manage and control the various hardware devices and computer programs 322 on the electronic device 30 to realize the operation and processing of the massive data 323 in the memory 32 by the processor 31. It can be Windows Server, Netware, Unix, Linux etc.
  • the computer program 322 may further include a computer program that can be used to complete other specific tasks.
  • an embodiment of the present application also discloses a computer storage medium, the computer storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, the disclosure of any of the foregoing embodiments is realized.
  • SPECT imaging prediction model creation method steps are provided.
  • the steps of the method or algorithm described in the embodiments disclosed in this document can be directly implemented by hardware, a software module executed by a processor, or a combination of the two.
  • the software module can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or all areas in the technical field. Any other known storage media.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Physics & Mathematics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Surgery (AREA)
  • Public Health (AREA)
  • Optics & Photonics (AREA)
  • Radiology & Medical Imaging (AREA)
  • Animal Behavior & Ethology (AREA)
  • Pathology (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Veterinary Medicine (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Algebra (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Nuclear Medicine (AREA)
  • Image Processing (AREA)

Abstract

一种SPECT成像预测模型创建方法、装置、设备及存储介质。该方法包括：获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像；以深度卷积神经网络为基础进行网络构建，得到待训练网络；将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。能在保留医学影像的成像质量的情况下大幅度降低SPECT成像时间。

Description

一种SPECT成像预测模型创建方法、装置、设备及存储介质
本申请要求于2021年03月24日提交中国专利局、申请号为202110311613.7、发明名称为“一种SPECT成像预测模型创建方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及SPECT技术领域，特别涉及一种SPECT成像预测模型创建方法、装置、设备及存储介质。
背景技术
单光子发射计算机断层成像(Single-Photon Emission Computed Tomo graphy，SPECT)是一种通过向生物体内注射放射性同位素标记的化合物，在生物代谢过程中，在不同方向相同时间内收集衰变产生的高能γ粒子以获得投影信号，通过计算重建进行成像的医学影像技术。由于信号收集依赖于放射性同位素半衰期和信号收集角度数量，SPECT图像在临床应用中面临成像时间长，对人体有一定辐射损害等问题，但如果减少成像角度或缩短成像时间，则会造成成像的结果具有信噪比低，易产生伪影等一系列损害图像质量的缺点。因此如何在短采集时长下生成高质量的SPECT图像是目前亟需解决的问题。
发明内容
有鉴于此，本发明的目的在于提供一种SPECT成像预测模型创建方法、装置、设备及介质，能够，在保留医学影像的成像质量的情况下大幅度降低SPECT成像时间，提高受试者SPECT成像舒适度，减小运动伪影。其具体方案如下：
第一方面，本申请公开了一种SPECT成像预测模型创建方法，包括：
获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像；
以深度卷积神经网络为基础进行网络构建，得到待训练网络；
将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。
可选的，所述扫描图像组中包括相互对应的所述标准采集时长SPECT图像、所述短采集时长SPECT图像和CT图像；其中，所述CT图像作为所述待训练网络的输入侧训练数据。
可选的，所述获取含有多个扫描图像组的训练集，包括：
通过单光子发射计算机断层成像设备得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号；
利用重建算法对所述标准采集时长γ粒子信号和所述短采集时长γ粒子信号进行重建，得到对应的标准采集时长SPECT图像和短采集时长SPECT图像；
基于所述标准采集时长SPECT图像和所述短采集时长SPECT图像，得到所述扫描图像组。
可选的，所述通过单光子发射计算机断层成像设备得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号，包括：
利用所述单光子发射计算机断层成像设备按照标准采集时长和短采集时长，采集得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号；
或，利用所述单光子发射计算机断层成像设备按照标准采集时长采集得到标准采集时长γ粒子信号，然后对所述标准采集时长γ粒子信号进行降采集得到对应的短采集时长γ粒子信号。
可选的，所述采集条件包括采集开始时刻、被采集者、放射性同位素药物计量和采集角度。
可选的，所述重建算法包括滤波反投影重建算法、代数重建算法和快速子集共轭梯度重建算法中的任意一项。
可选的，所述以深度卷积神经网络为基础进行网络构建，得到待训练网络，包括：
以U2-Net网络结构为基础进行网络构建，得到所述待训练网络。
第二方面，本申请公开了一种SPECT成像预测模型创建装置，包括：
训练集获取模块，用于获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像；
网络构建模块，用于以深度卷积神经网络为基础进行网络构建，得到待训练网络；
模型训练模块，用于将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。
第三方面，本申请公开了一种SPECT成像预测方法，包括：
获取待预测短采集时长SPECT图像；
将所述待预测短采集时长SPECT图像输入至SPECT成像预测模型，以预测得到所述待预测短采集时长SPECT图像在标准采集时长下的SPECT预测图像；其中，所述SPECT成像预测模型为利用含有多个扫描图像组的训练集对基于深度卷积神经网络构建的待训练网络进行训练得到的模型；其中，所述扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。
第四方面，本申请公开了一种SPECT成像预测装置，包括：
图像获取模块，用于获取待预测短采集时长SPECT图像；
图像预测模块，用于将所述待预测短采集时长SPECT图像输入至SPECT成像预测模型，以预测得到所述待预测短采集时长SPECT图像在标准采集时长下的SPECT预测图像；其中，所述SPECT成像预测模型为利用含有多个扫描图像组的训练集对基于深度卷积神经网络构建的待训练网络进行训练得到的模型；其中，所述扫描图像组中包括相互对应的标准采集 时长SPECT图像和短采集时长SPECT图像。
第五方面，本申请公开了一种电子设备，包括：
存储器，用于保存计算机程序；
处理器，用于执行所述计算机程序，以实现前述的SPECT成像预测模型创建方法。
第六方面，本申请公开了一种计算机可读存储介质，用于存储计算机程序；其中计算机程序被处理器执行时实现前述的SPECT成像预测模型创建方法。
本申请中，获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像；以深度卷积神经网络为基础进行网络构建，得到待训练网络；将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。可见，通过将训练集中的短采集时长SPECT图像作为输入侧训练数据，并将训练集中的标准采集时长SPECT图像作为输出侧训练数据，训练得到SPECT成像预测模型，然后利用SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。能在保留医学影像的成像质量的情况下大幅度降低SPECT成像时间，并提高受试者SPECT成像舒适度，减小运动伪影。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据提供的附图获得其他的附图。
图1为本申请提供的一种SPECT成像预测模型创建方法流程图；
图2为本申请提供的一种SPECT成像预测模型结构示意图；
图3为本申请提供的一种具体的SPECT成像预测模型创建方法流程图；
图4为本申请提供的一种SPECT成像示意图；
图5为本申请提供的一种SPECT成像预测方法流程图；
图6为本申请提供的一种SPECT成像预测模型创建装置结构示意图；
图7为本申请提供的一种SPECT成像预测装置结构示意图；
图8为本申请提供的一种电子设备结构图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
现有技术中，SPECT图像在临床应用中面临成像时间长，对人体有一定辐射损害等问题，但如果减少成像角度或缩短成像时间，则会造成成像的结果具有信噪比低，易产生伪影等一系列损害图像质量的缺点。为克服上述技术问题，本申请提出一种SPECT成像预测模型创建方法，
本申请实施例公开了一种SPECT成像预测模型创建方法，参见图1所示，该方法可以包括以下步骤：
步骤S11：获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。
本实施例中，首先获取含有多个扫描图像组作为训练集，其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。可以理解的是，标准采集时长SPECT图像和短采集时长SPECT图像是在相同采集条件下，由单光子发射计算机断层成像扫描仪根据标准采集策略和快速扫描策略采集并重建得到的图像，并且SPECT图像可以为基于SPECT设备或者SPECT+CT设备获取的。
本实施例中，所述扫描图像组中可以包括相互对应的所述标准采集时长SPECT图像、所述短采集时长SPECT图像和CT图像；其中，所述CT图像作为所述待训练网络的输入侧训练数据。可以理解的是，上述扫描图像组 中除了标准采集时长SPECT图像和短采集时长SPECT图像之外，还可以包括与上述标准采集时长SPECT图像和短采集时长SPECT图像对应的CT图像，即CT图像也为针对该采集者的CT图像，具体的，上述CT图像可以通过SPECT+CT设备获取也可以通过CT设备获取。可以理解的是，CT图像具有解剖结构清晰的优势，SPECT图像具有反映器官的生理、代谢和功能的特点，通过组建含有标准采集时长SPECT图像、短采集时长SPECT图像和CT图像的训练集，以便后续训练利用各类图像特点提高训练效果。
步骤S12：以深度卷积神经网络为基础进行网络构建，得到待训练网络。
本实施例中，以深度卷积神经网络为基础进行网络构建，得到待训练网络，具体的基于卷积层、BN(Batch Normalization)层和Relu层构建子编码器和子解码器，然后通过池化层或采样层依次连接不同预设数量的子编码器和子解码器，以构建不同尺寸大小的编码器和解码器，最后通过串联连接所述不同尺寸大小的编码器和解码器，得到待训练网络。具体的，上述深度卷积神经网络可以为一个具有N个卷积神经网络单元(CNN，Convolutional Neural Networks)串联构成的包括对称串接连接的编码器-解码器结构的图像重建卷积网络，其中，每一个卷积神经网络单元由预设数量的卷积层、池化层、非线性层、跳过连接层(skip connection)以及对应的下采样和上采样层组成。
本实施例中，所述以深度卷积神经网络为基础进行网络构建，得到待训练网络，可以包括：以U2-Net网络结构为基础进行网络构建，得到所述待训练网络。可以理解的是，以U2-Net网络结构为基础构建包含多个串联连接的编码器和解码器的待训练网络，具体的，上述待训练网络的结构示意图如图2所示，编码器和解码器结构的每一级由RSU单元(residual U-block)组成，每个RSU由3-5级串接连接的子编码器和子解码器组成。子编码器和子解码器由若干个3x3卷积层、BN层和Relu非线性层组成，子编码器和子解码器间由2x2池化层或者上采样层连接。且通过RSU中包含的空洞卷积(Dilation Conv)，混合不同大小的感受野，因此无论分辨率如何，该模型都能从浅层和深层捕获局部和全局信息，在不显著增加计算 量的情况下也增加了整个网络的深度。同时利用跳跃连接(Skip Connection)将对应尺度上的特征信息引入上采样过程，在深度网络中解决训练过程中梯度爆炸和梯度消失的问题。
步骤S13：将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。
本实施例中，得到训练集并构建完成待训练网络后，将上述训练集中的短采集时长SPECT图像作为输入侧训练数据，将上述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对上述构建的待训练网络进行训练，得到SPECT成像预测模型，以便利用该SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。其中，上述训练集可以分割为训练数据和测试数据，通过训练数据对模型进行训练，再通过测试数据对模型性能进行测试，以得到满足预测标准的SPECT成像预测模型。
由上可见，本实施例中获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像；以深度卷积神经网络为基础进行网络构建，得到待训练网络；将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。可见，通过将训练集中的短采集时长SPECT图像作为输入侧训练数据，并将训练集中的标准采集时长SPECT图像作为输出侧训练数据，训练得到SPECT成像预测模型，然后利用SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。能在保留医学影像的成像质量的情况下大幅度降低SPECT成像时间，并提高受试者SPECT成像舒适度，减小运动伪影。
本申请实施例公开了一种具体的SPECT成像预测模型创建方法，参见图3所示，该方法可以包括以下步骤：
步骤S21：利用单光子发射计算机断层成像设备按照标准采集时长和短采集时长，采集得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号。
本实施例中，利用单光子发射计算机断层成像设备，具体可以利用SPECT/CT扫描仪按照标准采集时长和短采集时长，采集得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号。本实施例中，所述采集条件可以包括采集开始时刻、被采集者、放射性同位素药物计量和采集角度。可以理解的，使用单光子发射计算机断层成像设备收集大于10组同一采集条件下的标准采集时长的γ粒子信号和短采集时长的γ粒子信号，即同一患者同一时间注射相同同位素标记的相同剂量化合物，并采集相同角度数量的投影信号组，即得到相同投影角度下采集到的γ粒子数量不同的两种信号。
其中，上述短采集时长γ粒子信可以为标准采集时长的1/7，例如，利用SPECT/CT扫描仪Siemens-Symbia-Intevo采集20例全身定量骨显像的受试者，其中，注射剂量可以为25～30mci，对受试者设置两种扫描协议，一种是标准扫描，每帧20秒，即得到标准采集时长的γ粒子信号，另一种是快速扫描，每帧3秒，得到1/7采集时长的γ粒子信号。其他采样参数为：60帧，单探头旋转180°，单次旋转6°。采用有序子集共轭梯度(OSCG)算法重建SPECT投影数据得到相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。需要说明的是本实施例采集范围包括但不限于脑、骨、心脏等部位。
本实施例中，还可以利用所述单光子发射计算机断层成像设备按照标准采集时长采集得到标准采集时长γ粒子信号，然后对所述标准采集时长γ粒子信号进行降采集得到对应的短采集时长γ粒子信号。可以理解的是，短采集时长γ粒子信号可以通过采集获取，也可以通过对标准采集时长γ粒子信号进行降采样获取得到。
步骤S22：利用重建算法对所述标准采集时长γ粒子信号和所述短采集时长γ粒子信号进行重建，得到对应的标准采集时长SPECT图像和短采集时长SPECT图像。
本实施例中，利用重建算法对获取的标准采集时长γ粒子信号和短采集时长γ粒子信号进行重建，得到对应的标准采集时长SPECT图像和短采集时长SPECT图像。本实施例中，所述重建算法包括但不限于滤波反投影重建算法(FBP，Filter Back Projection)、代数重建算法(ART，Algebraic Reconstruction Technique)和快速子集共轭梯度重建算法(OSCG，Ordered Subset Conjugate Gradiental)。
步骤S23：基于所述标准采集时长SPECT图像和所述短采集时长SPECT图像，得到所述扫描图像组以获取含有多个扫描图像组的训练集。
本实施例中，根据上述得到的标准采集时长SPECT图像和短采集时长SPECT图像，得到扫描图像组，以获取含有多个扫描图像组的训练集。
步骤S24：以深度卷积神经网络为基础进行网络构建，得到待训练网络。
步骤S25：将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。
本实施例中，通过训练得到的SPECT成像预测模型，根据输入的1/7SPECT图像，预测得到1/7SPECT图像在标准采集时长下的SPECT预测图像。在训练得到SPECT成像预测模型后，可以通过利用结构相似性指数(SSIM，Structural SIMilarity)和峰值信噪比(PSNR，Peak Signal to Noise Ratio)指标对模型进行定性和定量评价，以检测模型效果。例如图4所示的1/7SPECT图像，和深度学习重建得到的SPECT预测图像，以及标准采集下的SPECT图像，可见，预测得到的SPECT图像质量与标准SPECT图像相当，远优于原始的1/7SPECT图像，在1/7SPECT图像上识别病变区域相对困难，而在预测的SPECT图像上识别病变区域较为容易。
其中，关于上述步骤S24、步骤S25的具体过程可以参考前述实施例公 开的相应内容，在此不再进行赘述。
由上可见，利用单光子发射计算机断层成像设备按照标准采集时长和短采集时长，采集得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号，然后利用重建算法对所述标准采集时长γ粒子信号和所述短采集时长γ粒子信号进行重建，得到对应的标准采集时长SPECT图像和短采集时长SPECT图像；最后基于所述标准采集时长SPECT图像和所述短采集时长SPECT图像，得到所述扫描图像组以获取含有多个扫描图像组的训练集。由此一来，将短采集时长SPECT图像作为输入侧训练数据，将标准采集时长SPECT图像作为输出侧训练数据，对待训练网络进行训练得到SPECT成像预测模型，可以根据短采集时长SPECT图像预测出对应标准采集时长SPECT图像，相比于传统SPECT成像方法，本实施例中能在保留图像质量的情况下大幅度减小成像时间，并提高受试者SPECT成像舒适度，能够提高信噪比减小运动伪影。
本申请实施例公开了一种SPECT成像预测方法，参见图5所示，该方法可以包括以下步骤：
步骤S31：获取待预测短采集时长SPECT图像。
本实施例中，首先获取待预测短采集时长SPECT图像。
步骤S32：将所述待预测短采集时长SPECT图像输入至SPECT成像预测模型，以预测得到所述待预测短采集时长SPECT图像在标准采集时长下的SPECT预测图像；其中，所述SPECT成像预测模型为利用含有多个扫描图像组的训练集对基于深度卷积神经网络构建的待训练网络进行训练得到的模型；其中，所述扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。
本实施例中，将获取的待预测短采集时长SPECT图像输入至SPECT成像预测模型，以预测得到上述待预测短采集时长SPECT图像在标准采集时长下的SPECT预测图像；其中，上述SPECT成像预测模型为利用含有多个扫描图像组的训练集对基于深度卷积神经网络构建的待训练网络进行 训练得到的模型；其中，上述扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。
本实施例中，含有多个扫描图像组的训练集的获取过程包括：通过单光子发射计算机断层成像设备得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号；利用重建算法对所述标准采集时长γ粒子信号和所述短采集时长γ粒子信号进行重建，得到对应的标准采集时长SPECT图像和短采集时长SPECT图像；基于所述标准采集时长SPECT图像和所述短采集时长SPECT图像，得到所述扫描图像组。其中，上述通过单光子发射计算机断层成像设备得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号，可以包括：利用所述单光子发射计算机断层成像设备按照标准采集时长和短采集时长，采集得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号；或，利用所述单光子发射计算机断层成像设备按照标准采集时长采集得到标准采集时长γ粒子信号，然后对所述标准采集时长γ粒子信号进行降采集得到对应的短采集时长γ粒子信号。其中，上述采集条件包括采集开始时刻、被采集者、放射性同位素药物计量和采集角度。其中，上述重建算法包括但不限于滤波反投影重建算法、代数重建算法和快速子集共轭梯度重建算法。本实施例中，上述待训练网络可以为以U2-Net网络结构为基础进行网络构建，得到的待训练网络。
由上可见，本实施例中利用SPECT成像预测模型对获取的短采集时长SPECT图像进行预测，得到与短采集时长SPECT图像对应的标准采集时长下的SPECT预测图像。通过这种方式，能在保留医学影像的成像质量的情况下大幅度降低SPECT成像时间，并提高受试者SPECT成像舒适度，减小运动伪影。
相应的，本申请实施例还公开了一种SPECT成像预测模型创建装置，参见图6所示，该装置包括：
训练集获取模块11，用于获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时 长SPECT图像；
网络构建模块12，用于以深度卷积神经网络为基础进行网络构建，得到待训练网络；
模型训练模块13，用于将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。
由上可见，本实施例中获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像；以深度卷积神经网络为基础进行网络构建，得到待训练网络；将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。可见，通过将训练集中的短采集时长SPECT图像作为输入侧训练数据，并将训练集中的标准采集时长SPECT图像作为输出侧训练数据，训练得到SPECT成像预测模型，然后利用SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。能在保留医学影像的成像质量的情况下大幅度降低SPECT成像时间，并提高受试者SPECT成像舒适度，减小运动伪影。
在一些具体实施例中，所述扫描图像组具体可以包括相互对应的所述标准采集时长SPECT图像、所述短采集时长SPECT图像和CT图像；其中，所述CT图像作为所述待训练网络的输入侧训练数据。
在一些具体实施例中，所述训练集获取模块11具体可以包括：
γ粒子信号采集单元，用于通过单光子发射计算机断层成像设备得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号；
图像重建单元，用于利用重建算法对所述标准采集时长γ粒子信号和所述短采集时长γ粒子信号进行重建，得到对应的标准采集时长SPECT 图像和短采集时长SPECT图像；
扫描图像组确定单元，用于基于所述标准采集时长SPECT图像和所述短采集时长SPECT图像，得到所述扫描图像组。
在一些具体实施例中，所述γ粒子信号采集单元具体可以包括：
第一采集单元，用于利用所述单光子发射计算机断层成像设备按照标准采集时长和短采集时长，采集得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号；
第二采集单元，用于利用所述单光子发射计算机断层成像设备按照标准采集时长采集得到标准采集时长γ粒子信号，然后对所述标准采集时长γ粒子信号进行降采集得到对应的短采集时长γ粒子信号。
在一些具体实施例中，所述采集条件具体可以包括采集开始时刻、被采集者、放射性同位素药物计量和采集角度。
在一些具体实施例中，所述重建算法具体可以包括滤波反投影重建算法、代数重建算法和快速子集共轭梯度重建算法中的任意一项。
在一些具体实施例中，所述网络构建模块12具体可以包括：
网络构建单元，用于以U2-Net网络结构为基础进行网络构建，得到所述待训练网络。
相应的，本申请实施例还公开了一种SPECT成像预测装置，参见图7所示，该装置包括：
图像获取模块21，用于获取待预测短采集时长SPECT图像；
图像预测模块22，用于将所述待预测短采集时长SPECT图像输入至SPECT成像预测模型，以预测得到所述待预测短采集时长SPECT图像在标准采集时长下的SPECT预测图像；其中，所述SPECT成像预测模型为利用含有多个扫描图像组的训练集对基于深度卷积神经网络构建的待训练网络进行训练得到的模型；其中，所述扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。
由上可见，本实施例中利用SPECT成像预测模型对获取的短采集时长SPECT图像进行预测，得到与短采集时长SPECT图像对应的标准采集时 长下的SPECT预测图像。通过这种方式，能在保留医学影像的成像质量的情况下大幅度降低SPECT成像时间，并提高受试者SPECT成像舒适度，减小运动伪影。
进一步的，本申请实施例还公开了一种电子设备，参见图8所示，图中的内容不能被认为是对本申请的使用范围的任何限制。
图8为本申请实施例提供的一种电子设备30的结构示意图。该电子设备30，具体可以包括：至少一个处理器31、至少一个存储器32、电源33、通信接口34、输入输出接口35和通信总线36。其中，所述存储器32用于存储计算机程序，所述计算机程序由所述处理器31加载并执行，以实现前述任一实施例公开的SPECT成像预测模型创建方法中的相关步骤。
本实施例中，电源33用于为电子设备30上的各硬件设备提供工作电压；通信接口34能够为电子设备30创建与外界设备之间的数据传输通道，其所遵循的通信协议是能够适用于本申请技术方案的任意通信协议，在此不对其进行具体限定；输入输出接口35，用于获取外界输入数据或向外界输出数据，其具体的接口类型可以根据具体应用需要进行选取，在此不进行具体限定。
另外，存储器32作为资源存储的载体，可以是只读存储器、随机存储器、磁盘或者光盘等，其上所存储的资源包括操作系统321、计算机程序322及包括扫描图像组在内的数据323等，存储方式可以是短暂存储或者永久存储。
其中，操作系统321用于管理与控制电子设备30上的各硬件设备以及计算机程序322，以实现处理器31对存储器32中海量数据323的运算与处理，其可以是Windows Server、Netware、Unix、Linux等。计算机程序322除了包括能够用于完成前述任一实施例公开的由电子设备30执行的SPECT成像预测模型创建方法的计算机程序之外，还可以进一步包括能够用于完成其他特定工作的计算机程序。
进一步的，本申请实施例还公开了一种计算机存储介质，所述计算机 存储介质中存储有计算机可执行指令，所述计算机可执行指令被处理器加载并执行时，实现前述任一实施例公开的SPECT成像预测模型创建方法步骤。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其它实施例的不同之处，各个实施例之间相同或相似部分互相参见即可。对于实施例公开的装置而言，由于其与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
结合本文中所公开的实施例描述的方法或算法的步骤可以直接用硬件、处理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
最后，还需要说明的是，在本文中，诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
以上对本发明所提供的一种SPECT成像预测模型创建方法、装置、设备及介质进行了详细介绍，本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想；同时，对于本领域的一般技术人员，依据本发明的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本发明的限制。

Claims (12)

  1. 一种SPECT成像预测模型创建方法，其特征在于，包括：
    获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像；
    以深度卷积神经网络为基础进行网络构建，得到待训练网络；
    将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。
  2. 根据权利要求1所述的SPECT成像预测模型创建方法，其特征在于，所述扫描图像组中包括相互对应的所述标准采集时长SPECT图像、所述短采集时长SPECT图像和CT图像；其中，所述CT图像作为所述待训练网络的输入侧训练数据。
  3. 根据权利要求1所述的SPECT成像预测模型创建方法，其特征在于，所述获取含有多个扫描图像组的训练集，包括：
    通过单光子发射计算机断层成像设备得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号；
    利用重建算法对所述标准采集时长γ粒子信号和所述短采集时长γ粒子信号进行重建，得到对应的标准采集时长SPECT图像和短采集时长SPECT图像；
    基于所述标准采集时长SPECT图像和所述短采集时长SPECT图像，得到所述扫描图像组。
  4. 根据权利要求3所述的SPECT成像预测模型创建方法，其特征在于，所述通过单光子发射计算机断层成像设备得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号，包括：
    利用所述单光子发射计算机断层成像设备按照标准采集时长和短采集时长，采集得到同一采集条件下的标准采集时长γ粒子信号和短采集时长γ粒子信号；
    或，利用所述单光子发射计算机断层成像设备按照标准采集时长采集得到标准采集时长γ粒子信号，然后对所述标准采集时长γ粒子信号进行降采集得到对应的短采集时长γ粒子信号。
  5. 根据权利要求3所述的SPECT成像预测模型创建方法，其特征在于，所述采集条件包括采集开始时刻、被采集者、放射性同位素药物计量和采集角度。
  6. 根据权利要求3所述的SPECT成像预测模型创建方法，其特征在于，所述重建算法包括滤波反投影重建算法、代数重建算法和快速子集共轭梯度重建算法中的任意一项。
  7. 根据权利要求1至6任一项所述的SPECT成像预测模型创建方法，其特征在于，所述以深度卷积神经网络为基础进行网络构建，得到待训练网络，包括：
    以U2-Net网络结构为基础进行网络构建，得到所述待训练网络。
  8. 一种SPECT成像预测模型创建装置，其特征在于，包括：
    训练集获取模块，用于获取含有多个扫描图像组的训练集；其中，每个扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像；
    网络构建模块，用于以深度卷积神经网络为基础进行网络构建，得到待训练网络；
    模型训练模块，用于将所述训练集中的短采集时长SPECT图像作为输入侧训练数据，将所述训练集中的标准采集时长SPECT图像作为输出侧训练数据，对所述待训练网络进行训练，得到SPECT成像预测模型，以便利用所述SPECT成像预测模型预测得到短采集时长SPECT图像在标准采集时长下的SPECT预测图像。
  9. 一种SPECT成像预测方法，其特征在于，包括：
    获取待预测短采集时长SPECT图像；
    将所述待预测短采集时长SPECT图像输入至SPECT成像预测模型，以预测得到所述待预测短采集时长SPECT图像在标准采集时长下的SPECT预测图像；其中，所述SPECT成像预测模型为利用含有多个扫描 图像组的训练集对基于深度卷积神经网络构建的待训练网络进行训练得到的模型；其中，所述扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。
  10. 一种SPECT成像预测装置，其特征在于，包括：
    图像获取模块，用于获取待预测短采集时长SPECT图像；
    图像预测模块，用于将所述待预测短采集时长SPECT图像输入至SPECT成像预测模型，以预测得到所述待预测短采集时长SPECT图像在标准采集时长下的SPECT预测图像；其中，所述SPECT成像预测模型为利用含有多个扫描图像组的训练集对基于深度卷积神经网络构建的待训练网络进行训练得到的模型；其中，所述扫描图像组中包括相互对应的标准采集时长SPECT图像和短采集时长SPECT图像。
  11. 一种电子设备，其特征在于，包括：
    存储器，用于保存计算机程序；
    处理器，用于执行所述计算机程序，以实现如权利要求1至7任一项所述的SPECT成像预测模型创建方法。
  12. 一种计算机可读存储介质，其特征在于，用于存储计算机程序；其中计算机程序被处理器执行时实现如权利要求1至7任一项所述的SPECT成像预测模型创建方法。
PCT/CN2021/083206 2021-03-24 2021-03-26 一种spect成像预测模型创建方法、装置、设备及存储介质 WO2021139835A2 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/551,876 US20240177377A1 (en) 2021-03-24 2021-03-26 Spect imaging prediction model creation method and apparatus, and device and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110311613.7A CN113012252A (zh) 2021-03-24 2021-03-24 一种spect成像预测模型创建方法、装置、设备及存储介质
CN202110311613.7 2021-03-24

Publications (2)

Publication Number Publication Date
WO2021139835A2 true WO2021139835A2 (zh) 2021-07-15
WO2021139835A3 WO2021139835A3 (zh) 2021-11-11

Family

ID=76405853

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/083206 WO2021139835A2 (zh) 2021-03-24 2021-03-26 一种spect成像预测模型创建方法、装置、设备及存储介质

Country Status (3)

Country Link
US (1) US20240177377A1 (zh)
CN (1) CN113012252A (zh)
WO (1) WO2021139835A2 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114155340A (zh) * 2021-10-20 2022-03-08 清华大学 扫描光场数据的重建方法、装置、电子设备及存储介质

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113436183A (zh) * 2021-07-09 2021-09-24 浙江大学 一种图像的相关性分析装置
CN114494251B (zh) * 2022-04-06 2022-07-15 南昌睿度医疗科技有限公司 Spect图像处理方法以及相关设备
CN117909722A (zh) * 2022-10-09 2024-04-19 深圳先进技术研究院 模型训练方法、光子检测方法、终端设备以及存储介质
CN117593610B (zh) * 2024-01-17 2024-04-26 上海秋葵扩视仪器有限公司 图像识别网络训练及部署、识别方法、装置、设备及介质

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107638188B (zh) * 2017-09-28 2021-03-19 江苏赛诺格兰医疗科技有限公司 图像衰减校正方法及装置
CN108961237B (zh) * 2018-06-28 2020-08-21 安徽工程大学 一种基于卷积神经网络的低剂量ct图像分解方法
CN109215014B (zh) * 2018-07-02 2022-03-04 中国科学院深圳先进技术研究院 Ct图像预测模型的训练方法、装置、设备及存储介质
CN109961435B (zh) * 2019-04-02 2022-02-22 上海联影医疗科技股份有限公司 脑图像获取方法、装置、设备及存储介质
CN112381741B (zh) * 2020-11-24 2021-07-16 佛山读图科技有限公司 基于spect数据采样与噪声特性的断层图像重建方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114155340A (zh) * 2021-10-20 2022-03-08 清华大学 扫描光场数据的重建方法、装置、电子设备及存储介质
CN114155340B (zh) * 2021-10-20 2024-05-24 清华大学 扫描光场数据的重建方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
CN113012252A (zh) 2021-06-22
US20240177377A1 (en) 2024-05-30
WO2021139835A3 (zh) 2021-11-11

Similar Documents

Publication Publication Date Title
WO2021139835A2 (zh) 一种spect成像预测模型创建方法、装置、设备及存储介质
Vaquero et al. Positron emission tomography: current challenges and opportunities for technological advances in clinical and preclinical imaging systems
US11250543B2 (en) Medical imaging using neural networks
Tong et al. Image reconstruction for PET/CT scanners: past achievements and future challenges
Naït-Ali et al. Compression of biomedical images and signals
JP5864658B2 (ja) 生理学的リストモード核医学イメージングにおけるカウントの最適時間サンプリングのための逆データ再構成
Lee et al. Morphology supporting function: attenuation correction for SPECT/CT, PET/CT, and PET/MR imaging
JP2005193018A (ja) 画像処理のシステム及び方法
CN110415311B (zh) Pet图像重建方法、系统、可读存储介质和设备
CN112419303B (zh) 神经网络训练方法、系统、可读存储介质和设备
CN109697740A (zh) 图像重建方法、装置和计算机设备
EP2732437B1 (en) Transmission image reconstruction and imaging using poissonian detector data
US5323007A (en) Method of recovering tomographic signal elements in a projection profile or image by solving linear equations
CN1175783C (zh) 正电子断层扫描中热源高分辨快速图像迭代重建方法
CN111626964A (zh) 扫描图像的优化方法、优化装置和医学扫描系统
CN110264537B (zh) Pet图像重建方法、系统、可读存储介质和设备
Matsutomo et al. Performance of myocardial perfusion imaging using multi-focus fan beam collimator with resolution recovery reconstruction in a comparison with conventional SPECT
Sensakovic et al. ACR testing of a dedicated head SPECT unit
JP2022159648A (ja) 画像処理装置、画像処理方法および断層画像取得システム
US11663758B2 (en) Systems and methods for motion estimation in PET imaging using AI image reconstructions
WO2021153555A1 (ja) 減弱分布画像作成装置、画像処理装置、放射線断層撮影システム、減弱分布画像作成方法、画像処理方法
Price et al. Analytical techniques for image superposition
CN112927132B (zh) 提升pet系统空间分辨率均匀性的pet图像重建方法
Hashimoto et al. Emis2Trans: attenuation correction for brain PET with many types of PET ligands using convolutional neural networks
Sureau et al. Improved clinical parametric imaging using list-mode reconstruction via resolution modeling

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 18551876

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21738482

Country of ref document: EP

Kind code of ref document: A2