WO2023167157A1

WO2023167157A1 - Computer program, information processing device, and information processing method

Info

Publication number: WO2023167157A1
Application number: PCT/JP2023/007187
Authority: WO
Inventors: 大毅藤林; フェリックスユリアンブランデンブルグ
Original assignee: 株式会社Splink
Priority date: 2022-03-01
Filing date: 2023-02-28
Publication date: 2023-09-07

Abstract

Provided is a computer program with which it is possible to provide brain diagnosis information using a minimally-invasive MRI image, and without performing a PET scan. Also provided are an information processing device and an information processing method.　This computer program causes a computer to execute a process for acquiring an MRI image of a subject, specifying prescribed brain diagnosis information on the basis of the acquired MRI image, and newly generating and displaying the specified brain diagnosis information.

Description

Computer program, information processing device and information processing method

The present invention relates to a computer program, an information processing device, and an information processing method.

In recent years, as the population ages, the number of dementia patients and dementia reserves (mild cognitive impairment) is increasing. The four major diseases that cause dementia include Alzheimer's disease (AD), dementia with Lewy bodies (DLB), cerebrovascular dementia, and frontotemporal lobar degeneration (FTLD). Various neurological symptoms including dementia also appear in sexual sclerosis (MS). Although the cause of Alzheimer's dementia has not yet been elucidated, characteristic lesions are observed in the brain as the disease progresses. For example, in Alzheimer's dementia and frontotemporal lobar degeneration, deposition of senile plaques due to amyloid β and accumulation of abnormal proteins, mainly tau, are known outside of nerve cells. Deposition of senile plaques has been shown to occur from the earliest stages of the onset of Alzheimer's dementia, beginning long before clinical symptoms appear (eg, over a decade ago). In addition, α-synuclein, a protein of unknown function, is expressed in brain neurons, and is believed to be the cause of neurodegenerative diseases such as Parkinson's disease.

MRI (Magnetic Resonance Imaging) and PET (Positron Emission Tomography) are used for brain examination. Since MRI detects signals in a magnetic field, there is no exposure to radiation, and it is possible to measure "shape" and "size" and detect even small lesions. On the other hand, PET can examine tissue activity by taking in radioactive glucose. Patent Literature 1 discloses an apparatus in which a drug that binds to amyloid β in brain tissue is injected into a subject and a PET image representing the concentration distribution of the drug on a cut plane across the brain is used.

International Publication No. WO2014/034724

As mentioned above, it is desirable to examine the shape and function of the brain with both MRI and PET. However, taking a PET image using a drug requires the body to take in a radioactive substance, even if it is in a small amount, and there is a risk of radiation exposure, which is highly invasive and imposes a burden on the patient. Also, it may not be possible to administer drugs to patients suffering from certain diseases, such as kidney disease. Furthermore, there is also the problem that the cost of PET examination is generally high. The time burden and trouble of having to undergo both modalities is also a problem.

The present invention has been made in view of such circumstances, and provides a computer program, an information processing apparatus, and an information processing method that can provide brain diagnostic information using less invasive MRI images without performing a PET examination. for the purpose.

The present application includes a plurality of means for solving the above problems. To give one example, a computer program acquires an MRI image of a subject in a computer and outputs predetermined brain diagnostic information based on the acquired MRI image. A process of specifying, newly generating and displaying the specified brain diagnosis information is executed.

According to the present invention, brain diagnostic information can be provided by less invasive MRI images.

It is a figure which shows an example of a structure of the information processing system of this embodiment. It is a figure which shows an example of a structure of an information processing apparatus. It is a figure which shows an example of the process of a 1st learning model. FIG. 10 is a diagram showing a first example of processing of the second learning model; FIG. 10 is a diagram showing a second example of processing of the second learning model; It is a figure which shows an example of the process of a 3rd learning model. FIG. 10 is a diagram showing an example of a processing procedure when using a first learning model; FIG. 10 is a diagram showing an example of a processing procedure when using a second learning model; It is a figure which shows an example of the processing procedure in the case of using a 3rd learning model. FIG. 4 is a diagram showing an example of a processing procedure for calculating the degree of iron accumulation or the degree of amyloid accumulation; FIG. 4 is a diagram showing a first display example of brain diagnosis information; FIG. 10 is a diagram showing a second display example of brain diagnosis information; FIG. 11 is a diagram showing a third display example of brain diagnosis information; FIG. 11 is a diagram showing a fourth display example of brain diagnosis information; FIG. 11 is a diagram showing a fifth display example of brain diagnosis information;

Hereinafter, embodiments of the present invention will be described based on the drawings. FIG. 1 is a diagram showing an example of the configuration of an information processing system according to this embodiment. The information processing system includes an information processing device 50 . An input device 20 and a display device 30 are connected to the information processing device 50 . An MRI apparatus 10 and an image data server 100 are connected to the information processing apparatus 50 via the communication network 1 .

The MRI apparatus 10 is an apparatus that can capture a tomographic image using a magnetic resonance phenomenon, and can obtain an MRI image (also referred to as an MR image). By selecting imaging conditions, an MRI image reflecting tissue density, relaxation time (longitudinal relaxation time T1, transverse relaxation time T2), blood flow, amount of hydrogen atoms (proton density), and the like can be obtained. An MRI image can be generated by performing reconstruction processing on an MRI signal containing position information.

In this specification, MRI images are, for example, T1-weighted images, T2-weighted images, T2 ^* -weighted images, FLAIR (Fluid-Attenuated Inversion Recovery) images, SWI (susceptibility-weighted imaging) images, and PADRE (Phase Difference Enhanced Imaging). ) shall contain images. SWI images can be generated from T2 ^* -weighted images and are qualitatively susceptibility-enhanced images. The PADRE image uses tissue magnetic information contained in the MRI phase image information, and uses a method of emphasizing the tissue contrast in the brain by emphasizing the phase difference (phase-difference-enhanced imaging method). be. By selecting and enhancing phase information so as to selectively depict lesions and tissues to be observed, it is possible to visualize nerves running in the brain, making it possible to diagnose neurodegenerative diseases. Become. A QSM (Quantitative Susceptibility Mapping) image can be generated by performing a predetermined operation on a T2 ^* weighted image. It is also possible to generate an R2 ^* (R2 star) image by performing a predetermined operation on the T2 ^* weighted image. QSM images and R2* images are also collectively referred to as susceptibility images. In a broader sense, the MRI image may include a magnetic susceptibility image. A predetermined arithmetic processing function required for conversion from an MRI image to a magnetic susceptibility image may be provided in the MRI apparatus 10, the information processing apparatus 50, or the image data server 100, or It may be provided in another device (not shown) connected to the communication network 1 . The MRI apparatus 10 is installed, for example, in a medical institution such as a hospital. MRI images obtained by the MRI apparatus 10 are accumulated in the image data server 100 . MRI images are also referred to as MR images.

The image data server 100 records MRI images for each patient. For example, for each patient, the imaging date (examination date) when the MRI image was captured, the imaging conditions, the medication history such as the presence or absence of medication at the time of imaging and the number of medications, the name and amount of the therapeutic drug at the time of medication, the patient's medical history, etc. , are recorded in association with the MRI images.

QSM is a method of calculating local magnetic susceptibility from MRI phase images, and the QSM imaging method captures 3D-GRE (Gradient Echo) intensity images and phase images with multi-echo. Since the magnetic susceptibility is a physical property value unique to a substance, the magnetic susceptibility can be used to estimate the substance information in the voxel. A QSM image is an image that quantitatively represents magnetic susceptibility. In QSM, paramagnetic substances with high magnetic susceptibility (for example, hemosiderin and deoxyhemoglobin in bleeding) are displayed in white, and diamagnetic substances with low magnetic susceptibility are displayed in black. Is displayed.

The display device 30 includes a liquid crystal display panel, an organic EL display panel, or the like, and can display the processing results of the information processing device 50 . For example, the display device 30 can display brain diagnostic information obtained from MRI images. Here, brain diagnostic information means information that may be used by doctors to diagnose the state of the brain. It also includes sensitive information.

The input device 20 is an input interface such as a keyboard and a mouse that accepts operations of the information processing device 50 . The input device 20 may be a touch panel, soft keys, hard keys, or the like provided on the display device 30 . A medical practitioner such as a doctor can operate the input device 20 to input information such as a reference region and a region of interest (ROI) for obtaining brain diagnostic information. Also, a medical worker such as a doctor can operate the input device 20 to display the processing result of the information processing device 50 on the display device 30 . Note that the input device 20 and the display device 30 may be incorporated in the information processing device 50 .

Although not shown, a client device (personal computer or the like) used by a medical practitioner such as a doctor is connected to the communication network 1, and the information processing device 50 serving as a server is sent from the client device to the information processing device 50. be accessible to The medical staff acquires the MRI image from the MRI apparatus 10 into the client apparatus and uploads it to the information processing apparatus 50 . The information processing device 50 may perform processing described later, transmit brain diagnosis information (processing results) to the client device, and display the brain diagnosis information on the client device.

FIG. 2 is a diagram showing an example of the configuration of the information processing device 50. As shown in FIG. The information processing apparatus 50 includes a control unit 51 that controls the entire apparatus, a communication unit 52, a memory 53, an interface unit 54, an image processing unit 55, a display control unit 56, a diagnostic information generation unit 57, a storage unit 58, and a recording medium reading unit. A portion 64 is provided. The storage unit 58 stores a computer program 59, a first learning model 61, a second learning model 62, a third learning model 63, and required information.

The image processing unit 55, the display control unit 56, and the diagnostic information generation unit 57 may be configured by hardware, may be realized by software (computer program), or may be implemented by both hardware and software. may be configured with The information processing device 50 may be configured by distributing functions among a plurality of information processing devices.

The control unit 51 can be configured with a CPU (Central Processing Unit), MPU (Micro-Processing Unit), GPU (Graphics Processing Unit), or the like. The control unit 51 can execute processing defined by a computer program 59 . That is, the processing by the control unit 51 is also the processing by the computer program 59 .

The communication unit 52 includes, for example, a communication module, and has a function of communicating with the MRI apparatus 10 and the image data server 100 via the communication network 1. The communication unit 52 can acquire an MRI image or a magnetic susceptibility image from the MRI apparatus 10 or the image data server 100 . The communication unit 52 can acquire information such as a reference region and a region of interest for obtaining brain diagnosis information.

The memory 53 can be composed of semiconductor memory such as SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), and flash memory. The computer program 59 can be developed in the memory 53 and the control unit 51 can execute the computer program 59 . A recording medium M on which the computer program 59 is recorded can be read by a recording medium reading section 64 .

The interface unit 54 provides an interface function between the input device 20 and the display device 30. The information processing device 50 (control section 51 ) can exchange data and information with the input device 20 and the display device 30 through the interface section 54 .

The storage unit 58 can be composed of, for example, a hard disk or a semiconductor memory.

When acquiring an MRI image via the communication unit 52, the image processing unit 55 converts the acquired MRI image into a magnetic susceptibility image. The image processing unit 55 standardizes the susceptibility values in the whole brain by standardizing the susceptibility images. The normalization process is a scaling method in which the minimum value of magnetic susceptibility is 0 and the maximum value is 1. The value x' after scaling can be calculated by the formula x'=(x-x _min )/(x _max -x _min ). Here, x indicates the magnetic susceptibility value before scaling, x _max indicates the maximum possible value of x, and x _min indicates the minimum possible value of x. This can reduce variations.

The display control unit 56 controls information displayed on the display device 30 via the interface unit 54 . The display control unit 56 can cause the display device 30 to display brain diagnosis information (processing results) by the information processing device 50, for example.

The diagnostic information generation unit 57 generates brain diagnostic information based on the results of processing by the control unit 51 and the results of processing using the first learning model 61, the second learning model 62, and the third learning model 63. Details of the brain diagnosis information will be described later.

The control unit 51 acquires an MRI image or magnetic susceptibility image of a patient (subject), specifies predetermined brain diagnostic information based on the acquired MRI image or magnetic susceptibility image, and displays the specified brain diagnostic information on the display device 30. indicate. Thereby, brain diagnostic information can be provided by a less invasive MRI image or a magnetic susceptibility image that can be converted from the MRI image without performing a PET examination. A specific description will be given below. In addition, although the QSM image is used as an example of the magnetic susceptibility image, the magnetic susceptibility image is not limited to the QSM image.

A first specific example of brain diagnostic information is that the first learning model 61 predicts a predicted PET image based on an MRI image or a QSM image, and based on the predicted predicted PET image, amyloid β, tau protein, α-synuclein, etc. It specifies brain diagnostic information that indicates the degree of accumulation of aggregates. In a second example, the second learning model 62 identifies brain diagnostic information indicating the degree of accumulation of aggregates such as amyloid β, tau protein, or α-synuclein based on MRI images or QSM images. In a third example, the third learning model 63 identifies brain diagnostic information such as amyloid β positive/negative or tau protein positive/negative based on MRI images or QSM images. A fourth example is to specify brain diagnosis information indicating iron concentration and magnetic susceptibility based on MRI images or QSM images. It is said that the function of brain nerve cells is inhibited when there are many fibrous aggregates, and it is said that it leads to cognitive impairment and movement disorder. , the degree or risk of cerebral neurodegeneration can be estimated, and is highly valuable as diagnostic aid information. The fibrillar aggregates in the present invention are mainly composed of amyloid β, tau, and α-synuclein, but also aggregates having fibrous aggregate morphology, such as complexes with other intracerebral proteins and aggregations due to heteromolecular species. include. Hereinafter, methods for identifying brain diagnostic information will be described in order.

(first example)
FIG. 3 is a diagram showing an example of processing of the first learning model 61. As shown in FIG. The control unit 51 can read the first learning model 61 from the storage unit 58 and perform the processing shown in FIG. 3 using the read first learning model 61 . The first learning model 61 can be composed of, for example, a neural network (eg, CNN: Convolutional Neural Network), U-net, GAN (Generative Adversarial Network), or a combination thereof. When a patient's MRI image or QSM image is input, the first learning model 61 outputs a predicted PET image corresponding to the input MRI image or QSM image.

The predicted PET image output by the first learning model 61 is a predicted PET image (predicted image) predicted from an MRI image or a QSM image, not a PET image obtained by a PET examination using a drug, and the invasiveness is low. The predicted PET images output by the first learning model 61 may include an amyloid predicted PET image, a tau predicted PET image, an α-synuclein aggregate image, and CL (centroid) values. In amyloid PET, since the PET value differs depending on the PET drug even if the degree of amyloid accumulation is the same, the centroid is standardized by a drug-independent value. For example, the degree of cerebral amyloid accumulation can be expressed as a numerical value from 0 to 100, where 0 is the average of healthy young subjects and 100 is the average of definite AD. The predicted PET image output by the first learning model 61 visualizes the distribution of, for example, amyloid β, tau protein, or α-synuclein in the brain, similar to the PET image using a drug. In this specification, the predicted PET image is the predicted PET image output by the first learning model 61 unless otherwise specified.

The first learning model 61 can be generated, for example, as follows. First, as a first step, the control unit 51 acquires first training data including MRI images and voxel values of the MRI images. The first training data may be acquired from the image data server 100 . Based on the acquired first training data, the control unit 51 trains the first learning model 61 to output the voxel values of the MRI image when the MRI image is input. Next, as a second step, the control unit 51 acquires second training data including a magnetic susceptibility image (or an MRI image may be used) and a predicted PET image corresponding to the magnetic susceptibility image. The second training data may be acquired from the image data server 100 . Based on the acquired second training data, the control unit 51 generates the first learning model 61 so as to output a predicted PET image when a magnetic susceptibility image (or an MRI image) is input.

As described above, in the first stage, by training the first learning model 61 using the first training data including the MRI images, in the second stage, the first learning model 61 can be trained in a small amount of data and in a short time. Learning is possible, and even if the second training data is small, the prediction accuracy of the predicted PET image can be improved.

The control unit 51 (diagnostic information generating unit 57) acquires the predicted PET image output by the first learning model 61, and based on the acquired predicted PET image, at least one fibrous aggregate (for example, amyloid At least one of β, tau protein and α-synuclein) can be identified as brain diagnostic information indicative of the degree of accumulation.

SUVR (Standardized Uptake Value Ratio) can be used as an indicator of the degree of accumulation. SUVR can be calculated by the formula SUVR=SUV in region of interest/SUV in reference region. SUV (Standardized Uptake Value) can be expressed by the formula SUV=PET value/medicated radioactivity/body weight. Regions of interest include, but are not limited to, diagnostic target regions such as, for example, the frontal lobe, occipital lobe, parietal lobe, posterior cingulate gyrus, and striatum. Reference regions include, for example, but are not limited to, the entire cerebellum, the entire cerebellum plus the brainstem, the cerebellar gray matter, the pons, and the like. SUV of amyloid β, tau protein, or α-synuclein, for example, is obtained by counting the number of voxels in which the value indicating the brightness of each voxel constituting the region of interest and the reference region (required site) is equal to or higher than a predetermined threshold. be able to. The degree of integration (for example, OO%, etc.) can be calculated from the ratio of the count value to the total number of voxels of the desired site.

In addition, CL (centroid) may be used as an index representing the degree of accumulation. CL can be calculated for each image based on SUVR. However, it is necessary to calculate the CL value using an appropriate conversion formula corresponding to the drug given to the patient (subject).

(Second example)
FIG. 4 is a diagram showing a first example of processing of the second learning model 62. As shown in FIG. The control unit 51 can read the second learning model 62 from the storage unit 58 and perform the processing shown in FIG. 4 using the read second learning model 62 . The second learning model 62 can be composed of, for example, a neural network. The second learning model 62 outputs SUVR values (or CL values for MRI or QSM images) for each voxel when input with information about the patient's MRI or QSM images and the reference region. The CL value may be calculated for one image such as an MRI image or a QSM image, and calculated from the SUVR value of each voxel output by the second learning model 62 using a required conversion formula. The control unit 51 identifies the SUVR value (or the CL value for the MRI image or the QSM image) output by the second learning model 62 as brain diagnosis information.

FIG. 5 is a diagram showing a second example of processing of the second learning model 62. FIG. The second learning model 62 outputs the SUVR value (or the CL value for the MRI image or QSM image) of each voxel in the region of interest when the patient's MRI image or QSM image and information about the reference region and the region of interest are input. do. The CL value may be calculated for one image such as an MRI image or a QSM image, and calculated from the SUVR value of the region of interest output by the second learning model 62 using a required conversion formula. The control unit 51 identifies the SUVR value (or the CL value for the MRI image or the QSM image) output by the second learning model 62 as brain diagnosis information.

Similarly to the first learning model 61, the second learning model 62 can be generated, for example, as follows. First, as a first step, the control unit 51 acquires first training data including MRI images and voxel values of the MRI images. The first training data may be acquired from the image data server 100 . Based on the acquired first training data, the control unit 51 trains the second learning model 62 to output the voxel values of the MRI image when the MRI image is input. Next, as a second step, the control unit 51 accumulates a magnetic susceptibility image (or an MRI image) and at least one of fibrous aggregates (for example, at least one of amyloid β and tau protein). Second training data is obtained that includes brain diagnostic information indicative of the degree of strength. The second training data may be acquired from the image data server 100 . When a magnetic susceptibility image (or MRI image) is input to the control unit 51 based on the acquired second training data, at least one of fibrous aggregates (for example, at least one of amyloid β and tau protein ), the second learning model 62 is generated so as to output brain diagnosis information indicating the degree of accumulation.

As described above, in the first stage, by training the second learning model 62 using the first training data including the MRI images, in the second stage, the second learning model 62 can be trained in a small amount of data and in a short time. Learning is possible, and even if the second training data is small, the prediction accuracy of brain diagnosis information can be improved.

(Third example)
FIG. 6 is a diagram showing an example of processing of the third learning model 63. As shown in FIG. The control unit 51 can read the third learning model 63 from the storage unit 58 and perform the processing shown in FIG. 6 using the read third learning model 63 . The third learning model 63 can be composed of, for example, a neural network. When the patient's MRI image or QSM image and information about the region of interest are input, the third learning model 63 detects at least one of amyloid β positive, amyloid β negative, tau protein positive, and tau protein negative in the region of interest. Output brain diagnostic information shown. Amyloid β-positive means that abnormal accumulation of amyloid β is present, and amyloid β-negative means that even if amyloid β accumulation is present, it is not abnormal accumulation. The same is true for tau protein positive/negative. If the information of the region of interest is not input, amyloid β positive/negative and tau protein positive/negative are predicted from the entire input QSM image or MRI image.

Similarly to the first learning model 61, the third learning model 63 can be generated, for example, as follows. First, as a first step, the control unit 51 acquires first training data including MRI images and voxel values of the MRI images. The first training data may be acquired from the image data server 100 . Based on the acquired first training data, the control unit 51 trains the third learning model 63 so that when an MRI image is input, the voxel values of the MRI image are output. Next, as a second step, the control unit 51 generates second training data including a magnetic susceptibility image (or an MRI image) and brain diagnostic information indicating at least one of amyloid β positive/negative and tau protein positive/negative. to get The second training data may be acquired from the image data server 100 . Based on the obtained second training data, when a magnetic susceptibility image (or an MRI image may be used) is input, the control unit 51 generates brain diagnosis information indicating at least one of amyloid β positive/negative and tau protein positive/negative. A third learning model 63 is generated to output.

As described above, in the first stage, by training the third learning model 63 using the first training data including the MRI images, in the second stage, the third learning model 63 can be trained in a small amount of data and in a short time. Learning is possible, and even if the second training data is small, the prediction accuracy of brain diagnosis information can be improved.

(Fourth example)
The control unit 51 obtains a QSM image of the patient, and based on the obtained QSM image, identifies brain diagnostic information indicating the degree of iron accumulation compared with that of a healthy subject. According to pathological studies, a course such as amyloid accumulation→phosphorylated tau accumulation→inflammation in the brain→nerve displacement→disease is observed. Iron accumulation may occur in areas of inflammation in the brain. Brain diagnostic information indicating the degree of iron accumulation enables estimation of amyloid accumulation and evaluation of brain atrophy. Further, when obtaining an MRI image of a patient, the control unit 51 may convert the obtained MRI image into a QSM image and obtain the QSM image. The control unit 51 may also acquire information about the region of interest, and specify brain diagnostic information indicating the degree of iron accumulation or the degree of amyloid accumulation in the region of interest compared with that of a healthy subject. A specific description will be given below.

First, collect data from healthy subjects in advance and create a database of magnetic susceptibility distribution in the data from healthy subjects. Then, a QSM image of the normal brain of a healthy subject is created. A database of healthy subjects is constructed, and a z-score (brain diagnostic information) for each region of interest of the patient is calculated based on the magnetic susceptibility distribution for each region of interest. That is, the QSM image of the patient is image-processed in units of voxels, which are three-dimensional pixels (VBM: Voxel Based Morphometry). A typical statistical process is to generate a z-score map.

The z-score can be calculated as follows. Calculate the average value and standard deviation of the magnetic susceptibility for each voxel from the QSM image of the normal brain of a healthy subject, and calculate the z-score based on the calculated average value and standard deviation and the magnetic susceptibility of the QSM image of the patient. . The z-score is
z-score = (M (x, y, z) - I (x, y, z)) / SD (x, y, z)
It can be calculated by the formula M is the average value of magnetic susceptibility of healthy subjects, SD is the standard deviation of magnetic susceptibility of healthy subjects, and I is the magnetic susceptibility of patients. The z-score indicates an index of how many times the standard deviation of the magnetic susceptibility distribution of a healthy subject's normal brain is apart. By using the z-score map, the patient's QSM image can be compared with that of a healthy person (normal standard brain) to quantitatively analyze what kind of change occurs in which part. For example, voxels with positive values on the z-score map indicate regions with atrophy compared to normal brains, and larger values can be interpreted as statistically greater divergence. For example, if the z-score is "2", it means that it exceeds twice the standard deviation from the mean value, and it is evaluated that there is a statistically significant difference with a risk of about 5%, and atrophy in the region of interest can be evaluated quantitatively.

Next, the processing of the information processing device 50 will be described.

FIG. 7 is a diagram showing an example of a processing procedure when using the first learning model 61. FIG. In the following description, for the sake of convenience, the main body of processing is assumed to be the control unit 51 . The control unit 51 acquires an MRI image of a subject (patient) (S11), and receives setting of a region of interest and a reference region (S12). The control unit 51 converts the acquired MRI image into a QSM image (S13), and standardizes the converted QSM image (S14). Note that the process of step S13 is unnecessary when the QSM image is directly acquired.

The control unit 51 inputs the standardized QSM image to the first learning model 61 and acquires the predicted PET image output by the first learning model 61 (S15). Based on the obtained predicted PET image, the control unit 51 calculates amyloid β, SUVR indicating the degree of tau protein accumulation, and CL (centroid) for the QSM image in the region of interest (S16). The control unit 51 outputs the brain diagnosis information (S17) and terminates the process.

FIG. 8 is a diagram showing an example of a processing procedure when using the second learning model 62. FIG. The control unit 51 acquires an MRI image of a subject (patient) (S21), and receives setting of a region of interest and a reference region (S22). The control unit 51 converts the acquired MRI image into a QSM image (S23), and standardizes the converted QSM image (S24). Note that the process of step S23 is unnecessary when the QSM image is directly acquired.

The control unit 51 inputs the set region of interest, the reference region, and the standardized QSM image to the second learning model 62, and the second learning model 62 outputs the amyloid β in the region of interest, the degree of accumulation of tau protein CL (centroid) for the SUVR and QSM images showing is acquired (S25). The control unit 51 outputs the brain diagnosis information (S26) and terminates the process.

FIG. 9 is a diagram showing an example of a processing procedure when using the third learning model 63. FIG. The control unit 51 acquires an MRI image of a subject (patient) (S31), and receives setting of a region of interest (S32). The control unit 51 converts the acquired MRI image into a QSM image (S33), and standardizes the converted QSM image (S34). Note that the process of step S33 is unnecessary when the QSM image is directly acquired.

The control unit 51 inputs the standardized QSM image to the third learning model 63, and outputs the amyloid β positive/negative (+/-), tau protein positive/negative (+ /-) is acquired (S35). The control unit 51 outputs the brain diagnosis information (S36) and ends the process.

FIG. 10 is a diagram showing an example of a processing procedure for calculating the degree of iron accumulation or the degree of amyloid accumulation. The control unit 51 acquires an MRI image of a subject (patient) (S41), and receives setting of a region of interest and a reference region (S42). The control unit 51 converts the acquired MRI image into a QSM image (S43), and standardizes the converted QSM image (S44). Note that the process of step S43 is unnecessary when the QSM image is directly acquired.

The control unit 51 refers to the healthy subject DB and generates a QSM image of the normal brain of the healthy subject group (S45). The control unit 51 calculates a Z-score (z-score) indicating the degree of iron accumulation or amyloid accumulation in the region of interest based on the QSM image of the subject and the QSM image of the healthy subject group. Based on the QSM image of the subject, the control unit 51 calculates SUVR indicating the degree of iron accumulation or amyloid accumulation in the region of interest and the centroid (CL) for the QSM image (S47). The control unit 51 outputs the brain diagnosis information (S48) and terminates the process.

The control unit 51 may perform all the processes of the first to fourth examples described above and output brain diagnostic information, or may perform required processes of the first to fourth examples. good. For example, only the first example and the fourth example may be performed to output the brain diagnosis information.

The diagnostic information generation unit 57 generates brain diagnostic information to be output, and the display control unit 56 performs control processing for displaying the brain diagnostic information on the display device 30 . A display example of brain diagnosis information will be described below.

FIG. 11 is a diagram showing a first display example of brain diagnosis information. The diagnosis information screen 210 includes a patient information area 211 that displays patient information, an image area 214 that displays patient images, a numerical area 217 that displays indexes and numerical values of brain diagnosis information, and a similarity area that displays similar scores. A score area 216 and a recommended inspection area 218 displaying recommended inspection items are displayed.

A patient information area 211 displays information such as a patient ID (which may include a name) for selecting a patient, date of birth, age, sex, examination date such as MRI examination, medication history, and medical history. . The patient ID may be selectable from among multiple patients. Also, if there are multiple inspection dates, the inspection date may be selectable.

By selecting the "predicted image" tab 212, the predicted PET image output by the first learning model 61 is displayed in the image area 214. The predicted PET images are displayed in the form of axial, sagittal, and coronal cross-sectional images. In addition, a horizontal bar and a cursor 215 for designating which slice image to display among the plurality of tomographic images are displayed below each cross-sectional image (or on the horizontal side), and by moving the cursor 215 , the desired cross-sectional image can be displayed. Note that the QSM image of the patient may also be displayed in the image area 214 at the same time. By selecting the "input image" tab 213, the image area 214 can display the input QSM image.

Numerical area 217 displays a reference region setting window (in the figure, the entire cerebellum is set) for selecting a reference region, and SUVR values indicating the degree of accumulation of amyloid β or tau protein for each region of interest. be done. The value of SUVR may use information output by the second learning model 62 or may be calculated based on the predicted PET image output by the first learning model 61 . In addition, as shown in the figure, the centroid (CL) value in the entire image may be displayed, and as a comprehensive judgment, amyloid β positive/negative (+/-), or tau protein positive/negative (+/-) ) may be displayed.

In the similarity score area 216, the degree of similarity with the positive image is displayed in the range of 0 to 1 (in the example of the figure, the degree of similarity is displayed as 0.35). The recommended examination area 218 displays examination items that can be recommended for the patient based on the brain diagnosis information. For example, the diagnostic information generation unit 57 may output the recommended inspection items on a rule basis.

As a result, when the doctor observes the QSM image and makes his own judgment, brain diagnosis information is automatically provided, so it is possible to support the doctor's diagnosis and reduce the burden on the doctor at the time of diagnosis. In addition, since brain diagnosis information is automatically provided, it is possible to reduce variations in diagnosis due to the experience of individual doctors.

FIG. 12 is a diagram showing a second display example of brain diagnosis information. The diagnostic information screen 220 displays a patient information area 211 that displays patient information, an image area 221 that displays patient images, and a numerical value area 222 that displays indices and numerical values of brain diagnosis information. The patient information area 211 is the same as in the first display example shown in FIG.

A predicted PET image output by the first learning model 61 is displayed in the image area 221 . It is possible to select any one of axial section (Axial), sagittal section (Sagittal), and coronal section (Coronal) as the section image to be displayed in the predicted PET image. In the example shown, an axial section (Axial) is selected. In the image area 221, the input QSM image and the predicted PET image are displayed in a comparable manner. As in the case of the first display example, by moving the cursor 215, a desired cross-sectional image can be displayed.

As in the first display example, the numerical value area 222 includes a reference area setting window for selecting a reference area (in the figure, the entire cerebellum is set), amyloid β or tau protein for each region of interest. A value of SUVR indicating the degree of accumulation of is displayed. In addition, CL values indicating the degree of accumulation of amyloid β or tau protein are displayed for each region of interest. Numerical area 222 displays the determination result of amyloid β positive/negative (+/-) or tau protein positive/negative (+/-) output by third learning model 63 .

FIG. 13 is a diagram showing a third display example of brain diagnosis information. The diagnostic information screen 230 displays a patient information area 211 that displays patient information, an image area 233 that displays patient images, and a numerical value area 234 that displays indexes and numerical values of brain diagnosis information. The patient information area 211 is the same as in the first display example shown in FIG.

By selecting the "slice" tab 231, in the image area 233, a desired number of slice images are selected from the cross-sectional images of the input QSM image, and a Z-score (z- A Z-score map is displayed on which the value of the score) is superimposed. A cross-sectional image of the QSM image can be selected to be axial, sagittal, or coronal. In the example shown, an axial section (Axial) is selected. In addition, the degree of iron accumulation can be visualized by adding gradation to the QSM image according to the value of the Z score. In the example of the figure, the Z-score is visualized in the range of 0 to 4. As a result, it is possible to easily determine to what extent iron accumulation occurs in which part of the brain. The degree of amyloid accumulation may be estimated from the degree of iron accumulation and displayed.

The value of magnetic susceptibility is displayed for each region of interest in the numerical area 234 . By selecting the "result overview" tab 232, a diagnosis information screen 240 of a fourth display example, which will be described later, is displayed.

As described above, when the doctor observes the QSM image and makes a judgment by himself/herself, brain diagnosis information is automatically provided, so that the doctor's diagnosis can be supported, and the burden on the doctor at the time of diagnosis can be reduced. . In addition, since brain diagnosis information is automatically provided, it is possible to reduce variations in diagnosis due to the experience of individual doctors.

FIG. 14 is a diagram showing a fourth display example of brain diagnosis information. The diagnostic information screen 240 displays a patient information area 211 that displays patient information, and a first numerical area 241 and a second numerical area 242 that display indices and numerical values of brain diagnosis information. The patient information area 211 is the same as in the first display example shown in FIG.

The first numerical value area 241 displays Z-score values in whole brain gray matter and whole brain white matter, and the status of iron deposition. In the example of the figure, the Z-score is visualized in the range of 0 to 4. In addition, the second numerical value area 242 displays the Z-score values and the status of iron deposition in regions other than whole brain gray matter and whole brain white matter. In the example shown, the frontal lobe, temporal lobe, occipital lobe, and parietal lobe are displayed, but are not limited to these. The degree of amyloid accumulation may be estimated from the degree of iron accumulation and displayed.

As described above, the Z-score indicating the degree of iron accumulation and the state of iron deposition are displayed for whole brain gray matter, whole brain white matter, and other regions of interest, reducing the burden on doctors at the time of diagnosis. In addition, it is possible to reduce variations in diagnosis due to the experience of individual doctors.

FIG. 15 is a diagram showing a fifth display example of brain diagnosis information. The diagnostic information screen 250 displays a patient information area 211 that displays patient information, an image area 253 that displays patient images, and a numerical value area 254 that displays indices and numerical values of brain diagnostic information. The patient information area 211 is the same as in the first display example shown in FIG.

By selecting the "magnetic susceptibility distribution image" tag 251, the magnetic susceptibility distribution image is displayed in the image area 253. The magnetic susceptibility distribution image can visualize the magnetic susceptibility (iron accumulation) by adding gradation to the QSM image according to the value of the magnetic susceptibility. A cross-sectional image of the QSM image can be selected to be axial, sagittal, or coronal. In the example shown, an axial section (Axial) is selected. As in the case of the first display example, by moving the cursor 215, a desired cross-sectional image can be displayed. The degree of amyloid accumulation may be estimated from the degree of iron accumulation and displayed.

The numerical value area 254 displays the magnetic susceptibility value and the Z score value for each region of interest.

According to this embodiment, since less invasive MRI images are used, the burden on the patient can be reduced compared to the actual PET examination. In addition, since the spread of MRI examination is higher than that of PET examination, more patients can be examined.

In the present embodiment, the information processing device 50 collects information other than the MRI image of the subject, such as test results regarding the cognitive ability of the subject (cognitive ability test, cognitive function test results, etc.), and biomarkers (for example, blood test results). , genetic information, etc.), specified brain diagnostic information is specified based on the acquired information, and the specified brain diagnostic information is newly generated and displayed.

The computer program of the present embodiment causes a computer to acquire an MRI image of a subject, specify predetermined brain diagnostic information based on the acquired MRI image, and newly generate and display the specified brain diagnostic information. to run.

The computer program of the present embodiment acquires a magnetic susceptibility image generated based on the MRI image of the subject to the computer, and outputs a predicted PET image when the magnetic susceptibility image is input. A process of acquiring a predicted PET image by inputting a magnetic susceptibility image and identifying the brain diagnostic information indicating the degree of accumulation of at least one of fibrous aggregates based on the obtained predicted PET image is executed.

The computer program of the present embodiment acquires a magnetic susceptibility image generated based on an MRI image of a subject into a computer, and when the magnetic susceptibility image is input, the degree of accumulation of at least one of the fibrous aggregates is shown. The brain diagnostic information is obtained by inputting the acquired magnetic susceptibility image into a second learning model that outputs brain diagnostic information and acquiring the brain diagnostic information indicating the degree of accumulation of at least one of the fibrous aggregates. Identify, take action.

The computer program of the present embodiment acquires a magnetic susceptibility image generated based on an MRI image of a subject into a computer, and when the magnetic susceptibility image is input, at least one of amyloid β positive/negative and tau protein positive/negative By inputting the acquired magnetic susceptibility image into a third learning model that outputs brain diagnostic information indicating the above-mentioned Identify brain diagnostic information and cause processing to be performed.

The computer program of the present embodiment causes the computer to acquire a magnetic susceptibility image generated based on the MRI image of the subject, and based on the acquired magnetic susceptibility image, the brain diagnosis indicating the degree of iron accumulation compared with that of a healthy subject. Identify information, cause an action to take place.

The computer program of the present embodiment acquires first training data including an MRI image and voxel values of the MRI image in a computer, and based on the acquired first training data, when an MRI image is input, the MRI image Train the first learning model to output voxel values of, acquire second training data including a magnetic susceptibility image and a predicted PET image, and input a magnetic susceptibility image based on the acquired second training data If so, a process of generating the first learning model to output a predicted PET image is executed.

The computer program of the present embodiment acquires first training data including an MRI image and voxel values of the MRI image in a computer, and based on the acquired first training data, when an MRI image is input, the MRI training the second learning model to output image voxel values to obtain second training data comprising magnetic susceptibility images and brain diagnostic information indicative of the degree of accumulation of at least one of fibrous aggregates; A process of generating the second learning model so as to output brain diagnostic information indicating the degree of accumulation of at least one of fibrous aggregates when a magnetic susceptibility image is input based on the obtained second training data. to run.

The computer program of the present embodiment acquires first training data including an MRI image and voxel values of the MRI image in a computer, and based on the acquired first training data, when an MRI image is input, the MRI training the third learning model to output image voxel values to obtain second training data including magnetic susceptibility images and brain diagnostic information indicative of at least one of amyloid-β positive/negative and tau protein positive/negative; and, based on the acquired second training data, when a magnetic susceptibility image is input, the third learning model is configured to output brain diagnostic information indicating at least one of amyloid β positive/negative and tau protein positive/negative. Generate, execute processing.

The computer program of the present embodiment acquires information including at least one of a test result on cognition of a subject and biomarkers into a computer, and specifies predetermined brain diagnostic information based on the acquired information. A process of newly generating and displaying brain diagnosis information is executed.

The information processing apparatus of this embodiment includes an acquisition unit that acquires an MRI image of a subject, a specification unit that specifies predetermined brain diagnostic information based on the acquired MRI image, and newly generates the specified brain diagnostic information. and a display for displaying.

The information processing method of the present embodiment acquires an MRI image of a subject, specifies predetermined brain diagnostic information based on the acquired MRI image, and newly generates and displays the specified brain diagnostic information.

This embodiment includes dementia, multiple sclerosis, mild cognitive impairment (MCI), mild cognitive impairment due to Alzheimer's disease (MCI due to AD), prodromal AD, and pre-onset Alzheimer's disease. Stage/preclinical AD, Parkinson's disease, insomnia, sleep disorder, cognitive decline, cognitive impairment, amyloid positive/negative disease, movement disorder, motor dysfunction, movement disorder disease, Alzheimer's disease, It can be used to diagnose synucleinopathy, multiple system atrophy, vascular dementia, cerebrovascular disease, dementia with Lewy bodies, other neurodegenerative diseases, and the like.

1 communication network 10 MRI apparatus 20 input device 30 display device 50 information processing device 51 control unit 52 communication unit 53 memory 54 interface unit 55 image processing unit 56 display control unit 57 diagnostic information generation unit 58 storage unit 59 computer program 61 first learning model 62 second learning model 63 third learning model 64 recording medium reading unit 100 image data server

Claims

to the computer,
Acquiring an MRI image of the subject,
Identifying predetermined brain diagnostic information based on the acquired MRI image,
newly generate and display the identified brain diagnostic information;
A computer program that causes a process to be performed.
to the computer,
obtaining a magnetic susceptibility image generated based on the MRI image of the subject;
When a magnetic susceptibility image is input, the obtained magnetic susceptibility image is input to the first learning model that outputs a predicted PET image to obtain a predicted PET image,
identifying the brain diagnostic information indicative of the degree of accumulation of at least one of fibrous aggregates based on the acquired predicted PET image;
2. The computer program according to claim 1, causing a process to be executed.
to the computer,
obtaining a magnetic susceptibility image generated based on the MRI image of the subject;
When a magnetic susceptibility image is input, the obtained magnetic susceptibility image is input to a second learning model that outputs brain diagnostic information indicating the degree of accumulation of at least one of the fibrous aggregates, identifying the brain diagnostic information by acquiring the brain diagnostic information indicating the degree of accumulation of at least one of
2. The computer program according to claim 1, causing a process to be executed.
to the computer,
obtaining a magnetic susceptibility image generated based on the MRI image of the subject;
When a magnetic susceptibility image is input, the acquired magnetic susceptibility image is input to a third learning model that outputs brain diagnostic information indicating at least one of amyloid β positive/negative and tau protein positive/negative, and amyloid β positive/negative. identifying the brain diagnostic information by obtaining the brain diagnostic information indicating at least one of negative and tau protein positive/negative;
2. The computer program according to claim 1, causing a process to be executed.
to the computer,
obtaining a magnetic susceptibility image generated based on the MRI image of the subject;
Identifying the brain diagnostic information indicating the degree of iron accumulation compared with that of a healthy subject based on the acquired magnetic susceptibility image;
5. The computer program according to any one of claims 1 to 4, causing a process to be executed.
to the computer,
obtaining first training data comprising MRI images and voxel values of the MRI images;
training the first learning model to output voxel values of the MRI image when an MRI image is input based on the acquired first training data;
obtaining second training data comprising susceptibility images and predicted PET images;
generating the first learning model so as to output a predicted PET image when a magnetic susceptibility image is input based on the acquired second training data;
3. The computer program according to claim 2, causing a process to be executed.
to the computer,
obtaining first training data including MRI images and voxel values of the MRI images;
training the second learning model to output voxel values of the MRI image when an MRI image is input based on the acquired first training data;
obtaining second training data comprising magnetic susceptibility images and brain diagnostic information indicative of the degree of accumulation of at least one of fibrous aggregates;
generating the second learning model so as to output brain diagnostic information indicating the degree of accumulation of at least one of the fibrous aggregates when a magnetic susceptibility image is input based on the acquired second training data;
4. The computer program according to claim 3, causing a process to be executed.
to the computer,
obtaining first training data including MRI images and voxel values of the MRI images;
training the third learning model to output voxel values of the MRI image when an MRI image is input based on the acquired first training data;
Obtaining second training data comprising magnetic susceptibility images and brain diagnostic information indicative of at least one of amyloid β positive/negative and tau protein positive/negative;
Based on the obtained second training data, the third learning model is generated so as to output brain diagnostic information indicating at least one of amyloid β positive/negative and tau protein positive/negative when a magnetic susceptibility image is input. ,
5. The computer program according to claim 4, causing a process to be executed.
to the computer,
Obtaining information including at least one biomarker and a test result regarding the cognitive ability of the subject;
identifying predetermined brain diagnostic information based on the acquired information;
newly generate and display the identified brain diagnostic information;
5. The computer program according to any one of claims 1 to 4, causing a process to be executed.
an acquisition unit that acquires an MRI image of a subject;
an identification unit that identifies predetermined brain diagnostic information based on the acquired MRI image;
a display unit that newly generates and displays the identified brain diagnostic information,
Information processing equipment.
Acquiring an MRI image of the subject,
Identifying predetermined brain diagnostic information based on the acquired MRI image,
newly generate and display the identified brain diagnostic information;
Information processing methods.