JP7574181B2 - DIAGNOSIS SUPPORT DEVICE, ESTIMATION DEVICE, DIAGNOSIS SUPPORT SYSTEM, DIAGNOSIS SUPPORT METHOD, DIAGNOSIS SUPPORT PROGRAM, AND TRAINED MODEL - Google Patents

Description

The present invention relates to a diagnostic support device, an inference device, a diagnostic support system, a diagnostic support method, a diagnostic support program, and a trained model.

Previously, research has been conducted using machine learning technology to predict the risk of dementia based on brain MRI (Magnetic Resonance Imaging) images, etc.

For example, the following Patent Document 1 describes a diagnostic support device that predicts whether a subject with mild cognitive impairment will develop Alzheimer's disease within a specified period of time. The diagnostic support device divides a brain image acquired from the subject into gray matter, white matter, and cerebrospinal fluid, sets multiple regions of interest in each divided region, calculates t-values and p-values for the volume of each region of interest, calculates z-values for each region of interest based on the t-values and p-values, and predicts whether the subject will develop Alzheimer's disease within a specified period of time based on the z-values in accordance with a machine-learned prediction algorithm.

The following Patent Document 2 describes a medical image processing device for accurately acquiring the amount of change between brain images of the same patient. The medical image processing device aligns a first brain image of the subject with a standard brain image as a reference, then divides it into multiple regions, and aligns a second brain image of the subject with the first brain image as a reference to acquire the amount of change for at least one region. As a division method, a method of dividing the brain into regions that control each function such as movement, language, perception, memory, vision, and hearing based on Brodmann's brain map, and a method of dividing the brain into six types of regions, namely, the cerebrum, diencephalon, midbrain, hindbrain, cerebellum, and medulla oblongata, are described. The same document also describes the creation of a discriminator by machine learning the amount of change for each region, i.e., the atrophy rate, as teacher data for multiple past patients.

The following non-patent document 1 shows the distribution of SUVR values using the VBM (Voxel Based Morphometry) method, which divides brain images into voxels and evaluates them using PET scans. Figure 2 in the same document shows a map of the disease contribution rate of the SUVR value for each region divided into the orbitofrontal cortex, middle frontal cortex, middle temporal cortex, temporo-occipital junction, posterior cingulate cortex, angular gyrus, precuneus, putamen, and nucleus accumbens using logistic regression analysis.

The following non-patent document 2 describes how differential diagnosis of dementia is performed using physical examinations, neurological examinations, blood tests, imaging tests, etc.

Patent No. 6483890 International Publication No. WO2019/003749

MATHOTAARACHCHI, Sulantha et al., “Identifying incipient dementia individuals using machine learning and amyloid imaging”, Neurobiology of Aging, November 2017, Vol. 59, p. 80-90 "Dementia Disease Treatment Guidelines 2017", Igaku Shoin Co., Ltd., August 1, 2017, pp. 36-37

Using machine learning technology, the risk of dementia is predicted based on MRI images, etc., and the results are used by doctors to make diagnoses.

However, simply presenting a speculation result does not make it clear why the speculation was made. This makes it difficult for doctors to judge the validity of the speculation result, which may result in the speculation result being less likely to be adopted.

In addition, there are cases where it is difficult to make an accurate diagnosis using conventional methods that evaluate only certain areas of the brain, such as when there is no atrophy in the hippocampus but the condition does not correspond to AD (Alzheimer's Disease), or when a PET scan detects significant accumulation of amyloid beta protein but the condition does not correspond to MCI (Mild Cognitive Impairment), a precursor state of AD.

Therefore, the present invention provides a diagnostic support device, an inference device, a diagnostic support system, a diagnostic support method, a diagnostic support program, and a trained model that make it easy to confirm the validity of the inference results based on the trained model and that enables improved diagnostic accuracy.

A diagnostic support device according to one aspect of the present invention includes a first acquisition unit that acquires an image of the subject's whole brain, a second acquisition unit that inputs the image into a trained model and acquires an inference result based on which an inference related to the subject's dementia is made, and a display unit that displays the inference result and information indicating the region of interest of the whole brain in the image that is the basis for the inference result.

According to this aspect, by displaying the results of inferences made using a trained model related to a subject's dementia together with information showing the region of interest of the entire brain in the image on which the inference results are based, it becomes easier to confirm the validity of the inference results made by the trained model.

Here, "whole brain image" refers to an image of all areas surrounded by the skull (in the case of a cross-sectional image, all areas surrounded by the skull in that cross-section). Typically, a whole brain image of a horizontal cross-section (axial section, axial plane) of the brain includes the frontal, temporal and occipital lobes, etc., a vertical cross-section parallel to the anterior-posterior direction (sagittal section, sagittal plane) includes the frontal, parietal and occipital lobes and medulla oblongata, and a vertical cross-section parallel to the lateral direction (coronal section, coronal plane) includes the temporal, frontal or parietal lobes and medulla oblongata.

In addition, the "whole-brain region of interest in the image that serves as the basis for the inference result" refers to a region that, if the image data of that region is changed, will result in a different inference result. For example, if the image data of the region of interest displayed in the image data of a whole-brain image that has resulted in a prediction result that there is a high possibility of dementia is changed and input into a trained model, it refers to a region that will result in a prediction result that there is a low possibility of dementia.

"Whole brain region of interest" means the region of interest within all areas enclosed by the skull, and "region of interest" means the region giving rise to a dementia-related inferred result.

In the above aspect, the first acquisition unit may acquire a three-dimensional image of the entire brain.

According to this aspect, by using a three-dimensional image of the entire brain, it is possible to capture the three-dimensional characteristics of the entire brain and make inferences related to dementia.

In the above aspect, the display unit may display the region of interest of the entire brain in the three-dimensional image that serves as the basis for the inference result in a manner that allows the three-dimensional position to be identified.

According to this embodiment, the region of interest of the entire brain in the image on which the inference results are based can be captured in three dimensions, making it easier to confirm the validity of the inference results based on the trained model.

In the above aspect, the second acquisition unit may use the trained model to acquire a prediction result for predicting the risk of the subject developing dementia within a specified period based on the image.

According to this aspect, the validity of a prediction result regarding a subject's risk of developing dementia within a specified period of time can be confirmed by information indicating the region of interest of the entire brain in the image on which the prediction result is based.

In the above aspect, the second acquisition unit may use the trained model to acquire an inference result that infers the accumulation of dementia-causing proteins, such as amyloid beta protein and tau protein, throughout the entire brain based on images.

According to this embodiment, the validity of the inferred results of the accumulation of dementia-causing proteins, such as amyloid beta protein and tau protein, in the subject's entire brain can be confirmed by information indicating the region of interest in the entire brain in the image that serves as the basis for the inferred results.

In the above aspect, the second acquisition unit may use the trained model to acquire an inference result for making an inference related to the subject's dementia based on information including at least one or a combination of the following data over time: test scores on the subject's cognitive function, the subject's age, the subject's sex, the subject's height, the subject's weight, the subject's medical history, the volume of each part of the subject's whole brain, and a score representing the degree of atrophy of each part of the subject's whole brain, and data over time on the subject's whole brain.

According to this aspect, the trained model can be used to make inferences related to the subject's dementia based on information about the subject and an image of the subject's whole brain, thereby improving the accuracy of predictions.

A prediction device according to another aspect of the present invention includes a prediction unit that inputs an image of the subject's whole brain into a trained model and makes a prediction related to the subject's dementia, an estimation unit that uses the trained model to estimate a region of interest in the whole brain in the image that serves as the basis for the prediction result, and a providing unit that provides the result of the prediction and information indicating the region of interest in the whole brain in the image that serves as the basis for the prediction result to a diagnostic support device.

According to this aspect, a trained model is used to calculate inferences related to the subject's dementia and information indicating the entire brain region of interest in the image on which the inferences are based, and these are provided to a diagnostic support device, making it easier for a user of the diagnostic support device to confirm the validity of the inferences made by the trained model.

A diagnostic support system according to another aspect of the present invention is a diagnostic support system comprising a diagnostic support device and an inference device storing a trained model, the diagnostic support device having a first acquisition unit that acquires an image of the subject's whole brain, a second acquisition unit that inputs the image into the trained model and acquires an inference result based on which an inference related to the subject's dementia is made, and a display unit that displays the inference result and information indicating the region of interest of the whole brain in the image that is the basis for the inference result.

According to this aspect, by displaying the results of inferences made using a trained model related to a subject's dementia together with information showing the region of interest of the entire brain in the image on which the inference results are based, it becomes easier to confirm the validity of the inference results made by the trained model.

In the above aspect, the prediction device may have an prediction unit that uses a trained model to make a prediction related to the subject's dementia based on an image, and an estimation unit that uses the trained model to estimate a region of interest in the entire brain in the image that serves as the basis for the prediction result.

According to this aspect, the prediction device can use the trained model to calculate prediction results related to the subject's dementia, as well as information indicating the region of interest of the entire brain in an image that serves as the basis for the prediction results, and provide these to the diagnostic support device.

In the above aspect, the estimation device may have a generation unit that executes a learning process for a learning model using learning data including images of the whole brains of multiple subjects associated with actual measurement data related to dementia of the multiple subjects, and generates a learned model.

According to this aspect, a trained model that can make appropriate inferences can be generated using whole-brain images of multiple subjects associated with actual measurement data related to dementia of the multiple subjects.

In the above aspect, the learning data further includes subject data including information on the test scores on the subject's cognitive function, the subject's age, the subject's sex, the subject's height, the subject's weight, the subject's medical history, the volume of each part of the subject's whole brain and the degree of atrophy of each part of the subject's whole brain, and at least one or a combination of these data over time, and the generation unit may generate a learned model based on the images of the whole brains of multiple subjects and the subject data.

According to this aspect, by generating a trained model based on subject data on multiple subjects and images of the subjects' whole brains, a trained model with higher prediction accuracy related to the subjects' dementia can be generated.

In the above aspect, the generation unit may re-train the trained model based on an image of the subject's whole brain and a diagnosis result related to the subject's dementia.

According to this aspect, the trained model can be improved using accumulated images and diagnostic results of the subject's whole brain.

A diagnostic assistance method according to another aspect of the present invention includes obtaining an image of a subject's whole brain, inputting the image into a trained model to obtain an inference result based on which an inference is made related to the subject's dementia, and displaying the inference result and information indicating the region of interest of the whole brain in the image on which the inference result is based.

According to this aspect, by displaying the results of inferences made using a trained model related to a subject's dementia together with information showing the region of interest of the entire brain in the image on which the inference results are based, it becomes easier to confirm the validity of the inference results made by the trained model.

A diagnostic assistance program according to another aspect of the present invention causes a diagnostic assistance device to acquire an image of a subject's whole brain, input the image into a trained model to acquire an inference result based on which an inference related to the subject's dementia is made, and display the inference result and information indicating the region of interest of the whole brain in the image on which the inference result is based.

According to this aspect, by displaying the results of inferences made using a trained model related to a subject's dementia together with information showing the region of interest of the entire brain in the image on which the inference results are based, it becomes easier to confirm the validity of the inference results made by the trained model.

In another aspect of the present invention, a trained model is trained to input an image of the subject's whole brain into a training process of a training model using training data including images of the whole brains of multiple subjects associated with actual measurement data related to dementia of the multiple subjects, make an inference related to the dementia of the subject, and estimate a region of interest of the whole brain in the image that serves as the basis for the inference result.

According to this aspect, the results of an inference made regarding the subject's dementia and information indicating the region of interest of the entire brain in the image that serves as the basis for the inference results can be calculated and provided to a diagnostic support device.

A diagnostic support device according to another aspect of the present invention includes a first acquisition unit that acquires a first image of the subject's entire brain, a second acquisition unit that inputs the first image into a trained model and acquires a first inference result that makes an inference related to the subject's dementia, a setting unit that acquires a partial region from which a second inference result different from the first inference result is obtained when a second image, in which only a partial region of the first image is different data, is input into the trained model, and a display unit that displays the first inference result and the partial region.

A diagnostic assistance method according to another aspect of the present invention includes obtaining a first image of a subject's entire brain, inputting the first image into a trained model to obtain a first inference result that makes an inference related to the subject's dementia, setting a partial region from which a second inference result different from the first inference result is obtained when a second image, which is data that differs only from a partial region of the first image, is input into the trained model, and displaying the first inference result and the partial region.

Here, the partial area corresponds to the area on which the inference result is based, and differs for each subject. In other words, the partial area for one subject is different from the partial area for a different subject.

According to the present invention, it is possible to provide a diagnostic support device, an inference device, a diagnostic support system, a diagnostic support method, a diagnostic support program, and a trained model that make it easy to confirm the validity of the inference results based on a trained model and that enables improved diagnostic accuracy.

1 is a diagram showing a network configuration of a diagnosis support system according to an embodiment of the present invention. FIG. 2 is a diagram showing functional blocks of a diagnosis support device and an estimation device according to the present embodiment. FIG. 1 is a diagram illustrating a physical configuration of a diagnosis support apparatus according to an embodiment of the present invention. 13 is a diagram showing an example of displaying a dementia prediction result and its basis by the prediction device according to the present embodiment on the diagnosis support device. FIG. FIG. 13 is a diagram showing the accuracy of dementia prediction by the prediction device according to the present embodiment. 11 is a flowchart showing a process for displaying a prediction regarding dementia and the basis thereof by the diagnosis support system according to this embodiment. 13 is a flowchart of a learning process of a first trained model by the estimation device according to this embodiment. 11 is a diagram showing an example of displaying the results of prediction regarding accumulation of amyloid β protein by the prediction device according to the present embodiment and the basis thereof on the diagnosis support device. FIG. A figure showing the prediction accuracy regarding the accumulation of amyloid β protein by the prediction device according to this embodiment. 11 is a flowchart showing an inference regarding accumulation of amyloid β protein and display of the basis thereof by the diagnosis support system according to this embodiment.

An embodiment of the present invention will be described with reference to the attached drawings. In each drawing, the same reference numerals denote the same or similar configurations.

Figure 1 is a diagram showing the network configuration of a diagnosis support system 100 according to an embodiment of the present invention. The diagnosis support system 100 comprises a diagnosis support device 10, an estimation device 20 storing a trained model, an image management server 30, and an MRI scanner 50. The diagnosis support device 10, the estimation device 20, and the image management server 30 are communicatively connected via a communication network N such as the Internet or a LAN (Local Area Network). The image management server 30 and the MRI scanner are communicatively connected via DICOM (Digital Imaging and Communications in Medicine) or the like.

The diagnostic support device 10 is composed of a general-purpose computer and is used, for example, by a doctor. The diagnostic support device 10 displays the inference results of an inference made by the inference device 20 related to the dementia of the subject, and information indicating the brain region of interest in the MRI image that is the basis for the inference results. A user such as a doctor using the diagnostic support device 10 can make a diagnosis related to the dementia of the subject using the displayed inference results and the information indicating the brain region of interest that is the basis for the inference results as an aid in diagnosis.

The inference device 20 is composed of a general-purpose computer, and uses a trained model to make inferences related to the subject's dementia based on images of the subject's brain obtained from the image management server 30. The inference device 20 also uses the trained model to estimate a brain region of interest in the image that forms the basis of the inference result. The inference result calculated by the inference device 20 and information related to the brain region of interest that forms the basis of the inference result are provided to the diagnosis support device 10.

The image management server 30 is composed of a general-purpose computer and has a database of MRI images of the subject's brain measured by the MRI scanner 50. The image management server 30 may store learning data including images of the brains of multiple subjects, which are used to generate a trained model of the estimation device 20. The image management server 30 may store medical images of the brain, such as CT images of the subject's brain measured by a CT (Computed Tomography) scanner and PET images of the subject's brain measured by a PET (Positron Emission Tomography) scanner, as well as MRI images of the subject's brain measured by the MRI scanner 50. However, in consideration of the ease of examination, it is preferable to use MRI images of the brain.

The MRI scanner 50 uses nuclear magnetic resonance to capture images of the inside of the subject's body. In particular, the MRI scanner 50 captures images of the inside of the subject's head, i.e., images of the brain. The captured MRI images are stored in the image management server 30 and are acquired by the inference device 20 and the diagnosis support device 10.

Figure 2 is a diagram showing functional blocks of the diagnosis support device 10 and the estimation device 20 according to this embodiment. The diagnosis support device 10 includes a first acquisition unit 11, a second acquisition unit 12, and a display unit 10f. The estimation device 20 includes a memory unit 21, a generation unit 22, an estimation unit 23, an estimation unit 24, and a provision unit 25.

The first acquisition unit 11 acquires an image of the subject's brain from the image management server 30. The first acquisition unit 11 may acquire a three-dimensional image of the subject's brain. Here, the three-dimensional image may be an MRI image of the brain measured by an MRI scanner 50. By using the three-dimensional image of the brain, it is possible to capture the three-dimensional characteristics of the brain and make inferences related to dementia.

The second acquisition unit 12 acquires, from the inference device 20, an inference result obtained by making an inference related to the subject's dementia based on the brain image using the first trained model 21b or the second trained model 21c. The second acquisition unit 12 also acquires, from the inference device 20, information indicating the brain region of interest in the image that is the basis for the inference result.

The second acquisition unit 12 may use the first trained model 21b to acquire a prediction result of predicting the risk of the subject developing dementia within a predetermined period based on the brain image. The prediction result of predicting the risk of the subject developing dementia within a predetermined period may be, for example, a result of calculating the probability that the subject will develop dementia within two years.

The second acquisition unit 12 may acquire a result of inferring the accumulation of amyloid β protein in the brain based on the brain image using the second trained model 21c. The result of inferring the accumulation of amyloid β protein in the brain may be a result of inferring the amount of accumulated amyloid β protein, or a result of inferring whether the amount of accumulated amyloid β protein is equal to or greater than a threshold (presence or absence of accumulation of amyloid β protein).

The display unit 10f displays the inference results of an inference related to the subject's dementia and information indicating the brain region of interest in the image that serves as the basis for the inference results. In this way, by displaying the inference results of an inference related to the subject's dementia using the trained model together with information indicating the brain region of interest in the image that serves as the basis for the inference results, it becomes easier to confirm the validity of the inference results from the trained model.

The display unit 10f may display the brain region of interest in the three-dimensional image that is the basis of the inference result in a manner that allows the three-dimensional position to be identified. The display unit 10f may display the brain region of interest in the three-dimensional image that is the basis of the inference result, for example, by a heat map superimposed on an MRI image of the brain. The display unit 10f may also display a three-dimensional image of the brain that is rotatable, and display the brain region of interest in the three-dimensional image that is the basis of the inference result by a heat map superimposed on the three-dimensional image. This allows the brain region of interest in the image that is the basis of the inference result to be captured in three dimensions, making it easier to confirm the validity of the inference result by the trained model.

The display unit 10f may display the result of a prediction of the risk that the subject will develop dementia within a specified period of time and information indicating the brain region of interest in the image that serves as the basis for the prediction result. This makes it possible to confirm whether the prediction result of the risk that the subject will develop dementia within a specified period of time is valid based on the information indicating the brain region of interest in the image that serves as the basis for the prediction result.

The display unit 10f may also display information indicating the inference result of inferring the accumulation of amyloid β protein in the brain and the region of interest in the brain in the image that is the basis for the inference result. This allows the validity of the inference result of inferring the accumulation of amyloid β protein in the subject's brain to be confirmed by the information indicating the region of interest in the brain in the image that is the basis for the inference result. According to the diagnosis support system 100 of this embodiment, the accumulation of amyloid β protein can be inferred by MRI images, and the risk of dementia can be inferred by a relatively simple method of taking brain images with the MRI scanner 50 without performing measurements with a PET scanner or collecting cerebrospinal fluid.

The memory unit 21 of the estimation device 20 stores the learning data 21a, the first learned model 21b, and the second learned model 21c. The learning data 21a includes images of the brains of multiple subjects associated with actual measurement data related to dementia of the multiple subjects. The learning data 21a may be MRI images of the brains of multiple subjects associated with actual measurement data indicating whether the cognitive function of the multiple subjects has deteriorated within two years, or MRI images of the brains of multiple subjects associated with actual measurement data indicating accumulation of amyloid beta protein in the brains of the multiple subjects (accumulation amount or presence or absence of accumulation).

The generation unit 22 executes a learning process of the learning model using the learning data 21a, and generates a first learned model 21b and a second learned model 21c. The generation unit 22 executes a learning process of the learning model configured with a convolutional neural network using the learning data 21a including MRI images of the brains of multiple subjects associated with actual measurement data indicating whether the cognitive function of the multiple subjects has deteriorated within two years, and generates the first learned model 21b. The generation unit 22 also executes a learning process of the learning model configured with a convolutional neural network using the learning data 21a including MRI images of the brains of multiple subjects associated with actual measurement data indicating the accumulation of amyloid β protein in the brains of the multiple subjects, and generates the second learned model 21c. The learning process may be, for example, a process of updating parameters of the neural network by an error backpropagation method so as to minimize a loss function that evaluates the error between the prediction result and the correct answer. The first trained model 21b and the second trained model 21c may be configured, for example, by a 3D CNN (Convolutional Neural Network) that receives voxel data of brain images as input. Note that the first trained model 21b and the second trained model 21c may both be configured by convolutional neural networks, but the network structures may be different. In this way, a trained model that can make appropriate inferences can be generated using brain images of multiple subjects associated with actual measurement data related to dementia of multiple subjects.

The prediction unit 23 uses the first trained model 21b or the second trained model 21c to make a prediction related to dementia of the subject based on the brain image. The prediction unit 23 may use the first trained model 21b to predict the risk of the subject developing dementia within a predetermined period based on the brain image, or may use the second trained model 21c to predict the accumulation of amyloid β protein in the brain based on the brain image.

The learning data may further include subject data including the subject's cognitive function test score, the subject's age, the subject's sex, the subject's height, the subject's weight, the subject's medical history, the subject's brain volume for each part, and a score representing the subject's brain atrophy, as well as at least one or a combination of these data over time. The subject data may be transmitted from the diagnosis support device 10 to the estimation device 20 and stored in the memory unit 21, or a portion of the subject data may be imported into the estimation device 20 from an external database. Here, the brain volume for each part may include, for example, the volume of the basal forebrain, the volume of the posterior cingulate gyrus, and the volume of the medial temporal lobe. The score representing the brain atrophy may be expressed as a Z-score (a value converted so that the mean is 0 and the standard deviation is 1). The generation unit 22 may then convert the subject data into numerical values, synthesize the feature map of the brain image calculated by the convolutional neural network with the numerical values representing the subject data, and perform a calculation using, for example, a fully connected layer to generate the first trained model 21b and the second trained model 21c so as to make an inference regarding the subject's dementia. In this way, by generating trained models based on subject data regarding multiple subjects and brain images of the subjects, it is possible to generate trained models with higher inference accuracy related to the subject's dementia.

In this case, the prediction unit 23 makes a prediction related to the dementia of the subject based on the brain image and the subject data. The prediction unit 23 uses the first trained model 21b or the second trained model 21c to make a prediction related to the dementia of the subject based on information on at least one or a combination of the following data over time: the subject's cognitive function test score, the subject's age, the subject's sex, the subject's height, the subject's weight, the subject's medical history, the volume of each part of the subject's brain, the score representing the subject's brain atrophy degree, and data on the subject over time, and the brain image. In this way, by using the trained model to make a prediction related to the dementia of the subject based on information about the subject and the subject's brain image, the prediction accuracy can be improved.

The estimation unit 24 estimates the brain region of interest in the image on which the inference result is based, using the first trained model 21b or the second trained model 21c. When the first trained model 21b or the second trained model 21c is configured by a neural network, the estimation unit 24 may estimate the brain region of interest in the image on which the inference result is based, for example, using LRP (Layer-wise Relevance Propagation).

The generation unit 22 may re-train the first trained model 21b or the second trained model 21c based on the subject's brain image and the diagnosis result related to the subject's dementia. The subject's brain image and the diagnosis result related to the subject's dementia are accumulated as a doctor diagnoses the subject. The generation unit 22 may add a newly measured image of the subject's brain to the training data and re-train the first trained model 21b or the second trained model 21c. This makes it possible to improve the trained model using the accumulated brain images and diagnosis results of the subject.

The providing unit 25 provides the inference result calculated by the estimation unit 23 and information indicating the brain region of interest in the image that is the basis for the inference result estimated by the estimation unit 24 to the diagnostic support device 10. The providing unit 25 may transmit the inference result and the information indicating the brain region of interest in the image that is the basis for the inference result to the diagnostic support device 10 via the communication network N.

In this way, according to the prediction device 20 of this embodiment, a learned model can be used to calculate a prediction result related to the subject's dementia and information indicating the brain region of interest in the image that serves as the basis for the prediction result, and the results can be provided to the diagnostic support device 10.

Figure 3 is a diagram showing the physical configuration of the diagnosis support device 10 according to this embodiment. The diagnosis support device 10 has a CPU (Central Processing Unit) 10a corresponding to a calculation unit, a RAM (Random Access Memory) 10b corresponding to a storage unit, a ROM (Read only Memory) 10c corresponding to a storage unit, a communication unit 10d, an input unit 10e, and a display unit 10f. These components are connected to each other via a bus so that data can be transmitted and received. In this example, the diagnosis support device 10 is configured by one computer, but the diagnosis support device 10 may be realized by combining multiple computers. In addition, the configuration shown in Figure 3 is an example, and the diagnosis support device 10 may have other configurations than these, or may not have some of these configurations.

The CPU 10a is a control unit that controls the execution of programs stored in the RAM 10b or ROM 10c and calculates and processes data. The CPU 10a is a calculation unit that executes a program (diagnostic assistance program) that displays the results of inferences made related to the subject's dementia and information indicating the brain region of interest in the image that is the basis for the inference results. The CPU 10a receives various data from the input unit 10e and communication unit 10d, and displays the results of the data calculations on the display unit 10f or stores them in the RAM 10b.

RAM 10b is a memory unit in which data can be rewritten, and may be composed of, for example, a semiconductor memory element. RAM 10b may store data such as a program executed by CPU 10a and an image of the subject's brain. Note that these are examples, and RAM 10b may store data other than these, or some of these may not be stored.

ROM 10c is a memory from which data can be read and may be composed of, for example, a semiconductor memory element. ROM 10c may store, for example, a diagnostic assistance program or data that is not rewritten.

The communication unit 10d is an interface that connects the diagnostic support device 10 to other devices. The communication unit 10d may be connected to a communication network N such as the Internet.

The input unit 10e accepts data input from a user and may include, for example, a keyboard and a touch panel.

The display unit 10f visually displays the results of the calculations performed by the CPU 10a and may be configured, for example, with an LCD (Liquid Crystal Display). The display unit 10f may display the results of an inference made regarding the subject's dementia and information indicating the brain region of interest in the image that is the basis for the inference results.

The diagnostic assistance program may be provided by being stored in a computer-readable storage medium such as RAM 10b or ROM 10c, or may be provided via a communication network connected by communication unit 10d. In diagnostic assistance device 10, CPU 10a executes the diagnostic assistance program to realize the various operations described using FIG. 2. Note that these physical configurations are merely examples and do not necessarily have to be independent configurations. For example, diagnostic assistance device 10 may include an LSI (Large-Scale Integration) in which CPU 10a is integrated with RAM 10b and ROM 10c.

The physical configuration of the inference device 20 may be the same as that of the diagnosis support device 10, but may also include a GPU (Graphical Processing Unit) and need not necessarily be the same. The trained model stored in the inference device 20 is generated by a training process of a training model using training data including brain images of multiple subjects associated with actual measurement data related to dementia of the multiple subjects. The trained model is trained to perform the following: making a prediction related to the dementia of the subject based on the brain image of the subject, and estimating a brain region of interest in the image that is the basis for the prediction result. The trained model can calculate the prediction result of the prediction related to the dementia of the subject and information indicating the brain region of interest in the image that is the basis for the prediction result, and provide them to the diagnosis support device 10.

Figure 4 is a diagram showing an example of displaying the dementia prediction result and its basis by the prediction device 20 according to this embodiment on the diagnostic support device 10. The figure shows the first image IMG1, the second image IMG2, and the first prediction result PD1 displayed on the display unit 10f of the diagnostic support device 10.

The first image IMG1 and the second image IMG2 are images showing an MRI image of the subject's brain and the brain region of interest in the image that is the basis for the prediction result by the first trained model 21b. The first image IMG1 is a cross-sectional image of the brain in the sagittal plane, with the first region R1 and the second region R2 shown as the region of interest. The second image IMG2 is a cross-sectional image of the brain in the coronal plane, with the third region R3 shown as the region of interest. In this example, the region of interest is displayed using a heat map in which pixels that contribute more to the prediction result calculated by the first trained model 21b are shown brighter.

The first prediction result PD1 is "Probability (prediction) of developing dementia within two years: 90%" and shows the result of calculating the probability of developing dementia using the first trained model 21b based on the first image IMG1 and the second image IMG2.

A doctor using the diagnostic support device 10 can refer to the first image IMG1 and the second image IMG2 to confirm the validity of the first prediction result PD1 based on the first trained model 21b, and can use the first prediction result PD1 for diagnosis.

In this example, the region of interest is displayed using a heat map in which pixels that contribute more to the prediction result calculated by the first trained model 21b are shown brighter, but the region of interest may also be displayed by marking parts of the brain that contribute more to the prediction result calculated by the first trained model 21b, or may be displayed as text information separate from the brain image.

The first image IMG1, the second image IMG2, the first region R1, the second region R2, the third region R3, and the first inference result PD1 may be displayed on a terminal such as a smartphone or a PC (Personal Computer) used by the subject. In this case, the providing unit 25 transmits information indicating the inference result and the brain region of interest in the image on which the inference result is based to the subject's terminal. This allows the subject to check his or her own condition by referring to the first image IMG1, the second image IMG2, the first region R1, the second region R2, the third region R3, and the first inference result PD1. The providing unit 25 may provide only the first inference result PD1 to the subject's terminal, and provide the first image IMG1, the second image IMG2, the first region R1, the second region R2, and the third region R3 to the subject's terminal as necessary.

Figure 5 is a diagram showing the prediction accuracy of dementia by the prediction device 20 according to this embodiment. In the figure, a part of the learning data is separated in advance as test data, and the first trained model 21b is generated by the prediction device 20. The result of calculating the prediction accuracy of the first trained model 21b using the test data is shown. The "prediction result" in the table in the figure shows the result of the first trained model 21b predicting whether the probability of the subject developing dementia within a specified period is equal to or greater than a threshold based on the brain image of the subject. If the probability of the subject developing dementia within a specified period is equal to or greater than the threshold, the subject is counted as the "AD progression group", and if the probability of the subject developing dementia within a specified period is less than the threshold, the subject is counted as the "AD non-progression group". In addition, the "correct answer" shows actual measurement data indicating whether the subject's cognitive function has deteriorated within a specified period.

The "prediction results" of the first trained model 21b shown in Figure 5 are counted when the test scores on the subject's cognitive function, etc. are combined with the feature map of the brain image to make a final judgment.

In the example shown in FIG. 5, the predicted results are 144 for the AD progression group and 77 for the AD non-progression group. Of the 144 cases predicted to be in the AD progression group, 140 were correct, and 4 were false positives (actually in the AD non-progression group). Of the 77 cases predicted to be in the AD non-progression group, 74 were correct, and 3 were false negatives (actually in the AD progression group). Therefore, the accuracy of the first trained model 21b is (140+74)/(140+4+74+3)×100%=96.8%, the sensitivity is 140/(140+3)×100%=97.9%, and the specificity is 74/(74+4)×100%=94.9%. As such, the first trained model 21b of this embodiment shows high performance with respect to all indicators, and can appropriately assess the risk of dementia by keeping the occurrence of false positives and false negatives low.

When the estimation device 20 generates the first trained model 21b, the estimation device 20 may cause the diagnostic support device 10 to display a table showing the estimation accuracy as shown in FIG. 5, or cause the diagnostic support device 10 to display indices such as accuracy, sensitivity, and specificity. Indices such as a receiver operating characteristic (ROC) curve and an area under the curve (AUC) may also be displayed on the diagnostic support device 10.

Figure 6 is a flowchart showing the display of a prediction regarding dementia and the basis for the prediction by the diagnosis support system 100 according to this embodiment. First, a brain image of the subject is taken by the MRI scanner 50 (S10).

Thereafter, the prediction device 20 predicts the progression of the subject's dementia after a predetermined period of time based on the brain images and the subject data using the first trained model 21b (S11). Here, the subject data refers to information on the subject's cognitive function test scores, the subject's age, the subject's sex, the subject's height, the subject's weight, the subject's medical history, the volume of each part of the subject's brain and a score representing the subject's brain atrophy degree, and at least one or a combination of these data over time.

Furthermore, the estimation device 20 uses the first learned model 21b to estimate a region of interest in the brain image that serves as the basis for the estimation result (S12).

Finally, the diagnostic support device 10 displays the brain image, the inference result, and the region of interest in the brain image that is the basis for the inference result (S13).

Figure 7 is a flowchart of the learning process of the first trained model 21b by the estimation device 20 according to this embodiment. First, the estimation device 20 accumulates learning data 21a including brain images of multiple subjects, subject data, and information regarding the progression of dementia after a predetermined period of time (S20).

The prediction device 20 uses the learning data 21a to perform prediction using a learning model (S21), and calculates the error between the prediction result and the correct answer (S22). Here, the correct answer is information regarding the progression of dementia after a predetermined period of time has elapsed, and may be information indicating whether or not dementia has progressed after the predetermined period of time has elapsed.

The estimation device 20 updates the parameters of the learning model so as to reduce the error (S23). If the learning termination condition is not met thereafter (S24: NO), the estimation device 20 executes steps S21 to S23 again. Here, the learning termination condition may be that the error between the estimation result and the correct answer is equal to or less than a predetermined value, or that the number of times steps S21 to S23 are executed is equal to or greater than a predetermined value.

On the other hand, if the learning termination condition is met (S24: YES), the estimation device 20 stores the generated first trained model 21b (S25). This ends the learning process of the first trained model 21b.

The estimation device 20 may perform a learning process for the second trained model 21c using training data 21a including brain images of multiple subjects, subject data, and information regarding the accumulation of amyloid beta protein, by a process similar to that used to generate the first trained model 21b.

Figure 8 is a diagram showing an example of displaying the results of a prediction regarding accumulation of amyloid β protein by the prediction device 20 according to this embodiment and the basis thereof on the diagnostic support device 10. The figure shows the third image IMG3, the fourth image IMG4, and the second prediction result PD2 displayed on the display unit 10f of the diagnostic support device 10.

The third image IMG3 and the fourth image IMG4 are images showing an MRI image of the subject's brain and the brain region of interest in the image that is the basis for the prediction result by the second trained model 21c. The third image IMG3 is a cross-sectional image of the brain in the sagittal plane, with the fourth region R4 shown as the region of interest. The fourth image IMG4 is a cross-sectional image of the brain in the coronal plane, with the fifth region R5 shown as the region of interest. In this example, the region of interest is displayed using a heat map in which pixels that contribute more to the prediction result calculated by the second trained model 21c are shown brighter.

The second estimation result PD2 is "Probability (estimation) that amyloid beta protein has accumulated above the threshold: 95%" and shows the result of calculating the probability of whether or not amyloid beta protein has accumulated by the second trained model 21c based on the third image IMG3 and the fourth image IMG4.

A doctor using the diagnostic support device 10 can refer to the third image IMG3 and the fourth image IMG4 to confirm the validity of the second prediction result PD2 by the second trained model 21c, and can use the second prediction result PD2 for diagnosis.

In this example, the region of interest is displayed using a heat map in which pixels that contribute more to the prediction result calculated by the second trained model 21c are shown brighter, but the region of interest may also be displayed by marking parts of the brain that contribute more to the prediction result calculated by the second trained model 21c, or may be displayed as text information separate from the brain image.

The third image IMG3, the fourth image IMG4, the fourth region R4, the fifth region R5, and the second inference result PD2 may be displayed on a terminal such as a smartphone or PC used by the subject. In this case, the providing unit 25 transmits information indicating the inference result and the brain region of interest in the image on which the inference result is based to the subject's terminal. This allows the subject to check his or her own condition by referring to the third image IMG3, the fourth image IMG4, the fourth region R4, the fifth region R5, and the second inference result PD2. The providing unit 25 may provide only the second inference result PD2 to the subject's terminal, and provide the third image IMG3, the fourth image IMG4, the fourth region R4, and the fifth region R5 to the subject's terminal as necessary.

9 is a diagram showing the prediction accuracy of the amyloid β protein accumulation by the prediction device 20 according to this embodiment. In the figure, a part of the learning data is separated in advance as test data, and the second trained model 21c is generated by the prediction device 20. The result of calculating the prediction accuracy of the second trained model 21c using the test data is shown. The "prediction result" in the table in the figure shows the result of the second trained model 21c predicting whether or not amyloid β protein has accumulated in the brain at a threshold value or more based on the brain image of the subject. If the probability that amyloid β protein has accumulated in the brain of the subject at a threshold value or more is equal to or greater than a predetermined value, it is counted as "Positive", and if the probability that amyloid β protein has accumulated in the brain of the subject at a threshold value or more is less than the predetermined value, it is counted as "Negative". In addition, the "correct answer" shows the result of measuring whether or not amyloid β protein has accumulated in the brain of the subject at a threshold value or more by PET or cerebrospinal fluid collection.

The "prediction results" of the second trained model 21c shown in Figure 9 are counted when the test scores on the subject's cognitive function, etc. are combined with the feature map of the brain image to make a final judgment.

In the example shown in FIG. 9, the prediction results are 865 positive and 467 negative. Of the 865 cases predicted as positive, 807 were correct and 58 were false positives (actually negative). Of the 467 cases predicted as negative, 427 were correct and 40 were false negatives (actually positive). Therefore, the accuracy of the second trained model 21c is (807+427)/(807+58+427+40)×100%=92.6%, the sensitivity is 807/(807+40)×100%=95.3%, and the specificity is 427/(427+58)×100%=88.0%. As such, the second trained model 21c of this embodiment shows high performance with respect to all indicators, and can appropriately evaluate the presence or absence of accumulation of amyloid beta protein while keeping the occurrence of false positives and false negatives low.

When the estimation device 20 generates the second trained model 21c, the estimation device 20 may cause the diagnostic support device 10 to display a table showing the estimation accuracy as shown in FIG. 9, or cause the diagnostic support device 10 to display indices such as accuracy, sensitivity, and specificity. Indices such as ROC curves and AUC may also be displayed on the diagnostic support device 10.

Figure 10 is a flowchart showing the prediction of amyloid β protein accumulation and the display of the basis thereof by the diagnostic support system 100 according to this embodiment. First, a brain image of the subject is taken by the MRI scanner 50 (S30).

Then, the estimation device 20 estimates the accumulation of amyloid β protein in the subject's brain based on the brain image and the subject data using the second trained model 21c (S31). Here, the subject data refers to information on the subject's cognitive function test score, the subject's age, the subject's sex, the subject's height, the subject's weight, the subject's medical history, the volume of each part of the subject's brain, a score representing the subject's brain atrophy degree, and at least one or a combination of these data over time.

Furthermore, the estimation device 20 uses the second learned model 21c to estimate the region of interest in the brain image that serves as the basis for the estimation result (S32).

Finally, the diagnosis support device 10 displays the brain image, the inference result, and the region of interest in the brain image that is the basis for the inference result (S33).
The above-described embodiments are intended to facilitate understanding of the present invention, and are not intended to limit the present invention. The elements of the embodiments, as well as their arrangements, materials, conditions, shapes, sizes, etc., are not limited to those shown as examples and can be changed as appropriate. In addition, configurations shown in different embodiments can be partially substituted or combined with each other.

In the above-described diagnostic support device 10, the first image IMG1 (Fig. 4) which is a sagittal plane of the subject shows brain images of all areas surrounded by the skull, and therefore corresponds to an image of the whole brain. Also, the second image IMG2 (Fig. 4) which is a coronal plane of the subject shows brain images of all areas surrounded by the skull, and therefore corresponds to an image of the whole brain.

Therefore, the learning data 21a stored in the memory unit 21 of the estimation device 20 includes images of the whole brains of multiple subjects. The generation unit 22 generates the first learned model 21b and the second learned model 21c by executing a learning process of a learning model composed of a convolutional neural network that receives as input the learning data 21a including MRI images of the whole brains of multiple subjects. The estimation unit 23 receives as input an image of the subject's whole brain and makes a prediction related to dementia, and the estimation unit 24 estimates a region of interest in the whole brain in the image that is the basis for the prediction result. Here, the first region R1 of the first image IMG1 indicates a region of the whole brain near the occipital lobe, the second region R2 indicates a region near the frontal lobe, and the third image IMG3 indicates a region near the temporal lobe. Therefore, the estimation unit 24 is configured to be able to estimate a region of interest in all regions surrounded by the skull. Furthermore, the provision unit 25 provides the estimated region of interest in the whole brain to the diagnosis support device 10.

Furthermore, the first acquisition unit 11 of the diagnosis support device 10 acquires an image of the subject's whole brain as an input. The second acquisition unit 12 acquires an inference result of an inference made related to the subject's dementia based on the image of the subject's whole brain from the inference device 20. Furthermore, the display unit 10f displays the inference result of the inference made related to the subject's dementia and information indicating the region of interest of the whole brain in the image that is the basis for the inference result.

According to this configuration, by acquiring an image of the whole brain of the subject and inputting this image into the trained model, it is possible to obtain a result of inference made on dementia of the subject. Therefore, it is possible to perform an evaluation based on the correlation of multiple regions, which was difficult to evaluate by a method of dividing the brain image into voxels. For example, when there is a certain amount of accumulation of dementia-causing proteins in each of the frontal lobe, occipital lobe, and temporal lobe, even if the method of dividing the brain image and evaluating it gives an evaluation result that the possibility of dementia is low because the amount of accumulation of dementia-causing proteins in each region is small, etc., according to the diagnosis support device 10 of the present invention, it is possible to perform an evaluation based on the correlation of multiple regions or an evaluation caused by a cause that crosses multiple regions. Therefore, for example, it is possible to obtain an inference result that there is a high possibility of dementia due to the accumulation of dementia-causing proteins to a certain extent in each of the frontal lobe, occipital lobe, and temporal lobe. Furthermore, it is possible to expect the discovery of new brain regions related to dementia or new correlations that were not recognized in the conventional technical common sense. Furthermore, even doctors who do not have sufficient specialized knowledge regarding the identification of individual parts of the brain, such as doctors who specialize in brain surgery, will be able to make a diagnosis easily.

Furthermore, the estimation unit 24 estimates the region of interest within all the regions surrounded by the skull, and the display unit 10f displays the region of interest within all the regions surrounded by the skull. For example, the display unit 10f displays related regions across multiple regions, such as the frontal lobe, the occipital lobe, and the temporal lobe, as shown in the first image IMG1 to the third image IMG3. Therefore, the user of the diagnostic support device 10 can obtain new knowledge that the relationship between multiple regions may be the cause of dementia. The effect brought about by such a configuration is difficult to obtain with VBM (Voxel Based Morphometry), which divides and evaluates brain images by voxels, or with PET examinations, which display only the areas that absorb drugs.

In addition, the region of interest of the whole brain displayed by the display unit 10f is the basis for the inference result, and therefore usually differs depending on the subject. For example, in a subject with frontotemporal dementia, the frontal lobe or temporal lobe may be displayed as the region of interest, while in a subject with vascular dementia, a site with damaged blood vessels may be displayed as the region of interest. The region of interest is expressed by highlighting the target pixel, for example, by increasing the brightness of the target pixel. Therefore, even in the same frontal lobe, different regions of interest are usually displayed depending on the subject. This configuration is completely different from simply highlighting a region (e.g., the frontal lobe) that is known in advance to be highly likely to cause dementia.

As a method for extracting the region of interest, that is, the region of the whole brain that is the basis of the inference result, the above-mentioned LRP is a method for calculating the influence of the input value on the output value. For example, after propagating a neural network such as CNN to a certain layer, it is possible to identify the part with the greatest influence by setting values other than the part to be examined to zero and performing back propagation. In addition, other than LRP, a method is conceivable in which the input information is changed to create multiple pieces of input information, which are input into the same trained model, and the results are compared to obtain the region that is the basis of the inference result. Specifically, first, the inference unit 23 inputs an image of the subject's whole brain (an example of a "first image") into the first trained model 21b or the second trained model 21c to obtain an inference result (an example of a "first inference result"). Next, the estimation unit 24 generates multiple whole-brain images (an example of a "second image") in which only a part of the first image is different. For example, a second image in which some pixel values of the frontal lobe are masked, a second image in which some pixel values of the occipital lobe are masked and changed to different values, etc. are generated. Then, the estimation unit 23 inputs the generated second images into the first trained model 21b or the second trained model 21c to obtain a plurality of estimation results (an example of a "second estimation result"). Then, the estimation unit 24 compares the first estimation result with the second estimation result to determine a second image from which a different estimation result is obtained. The estimation unit 24 acquires a masked region in the second image that has yielded a second estimation result different from the first estimation result as a region of interest that is the basis for the estimation result, and provides it to the providing unit 25. The display unit 10f increases the brightness of pixels corresponding to the region of interest and displays the pixels superimposed on the image of the subject's whole brain. Note that when three or more types of estimation results are selected, the region of interest can be configured to include two or more types of regions (a first region of interest having a basis and a second region of interest having a stronger basis).

With the above-described configuration, it is possible to provide a diagnostic support device, an inference device, a diagnostic support system, a diagnostic support method, a diagnostic support program, and a trained model that make it easy to confirm the validity of the inference results based on the trained model and that enables improved diagnostic accuracy.

10...diagnosis support device, 10a...CPU, 10b...RAM, 10c...ROM, 10d...communication unit, 10e...input unit, 10f...display unit, 11...first acquisition unit, 12...second acquisition unit, 20...estimation device, 21...storage unit, 21a...learning data, 21b...first trained model, 21c...second trained model, 22...generation unit, 23...estimation unit, 24...estimation unit, 25...provision unit, 30...image management server, 50...MRI scanner, 100...diagnosis support system

Claims (15)

A first acquisition unit that acquires an image of the subject's whole brain;
a second acquisition unit that inputs the image of the whole brain of the subject into a trained model generated based on training data including subject data associated with actual measurement data of dementia of the multiple subjects, images of the whole brain of the multiple subjects, and information on test scores on cognitive function of the subjects, the age of the subject, the sex of the subject, the height of the subject, the weight of the subject, the medical history of the subject, the volume of each part of the whole brain of the subject, a score representing the atrophy degree of the whole brain of the subject, and at least any one or a combination of a plurality of these pieces of data over time , and acquires an inference result of making an inference related to dementia of the subject;
a display unit that displays the inference result and information indicating a region of interest of the whole brain in an image of the subject that is the basis of the inference result;
A diagnostic support device comprising: The first acquisition unit acquires a three-dimensional image of the whole brain.
The diagnosis support device according to claim 1 . The display unit displays the region of interest of the whole brain in the three-dimensional image that is the basis of the inference result in a manner that allows the three-dimensional position to be identified.
The diagnostic support device according to claim 2 . The second acquisition unit acquires an inference result of inferring a risk of the subject developing dementia within a predetermined period of time based on an image of the subject using the trained model.
The diagnosis support device according to claim 1 . The second acquisition unit acquires an inference result of inferring the accumulation of dementia-causing proteins in the whole brain based on an image of the subject by using the trained model.
The diagnosis support device according to claim 1 . The second acquisition unit acquires an inference result of inferring accumulation of amyloid beta protein in the entire brain based on an image of the subject by using the trained model.
The diagnosis support device according to claim 1 . The second acquisition unit acquires the inference result by using the trained model to make an inference related to dementia of the subject based on information of at least one or a combination of the following data over time: a test score on the cognitive function of the subject, the age of the subject, the sex of the subject, the height of the subject, the weight of the subject, the medical history of the subject, the volume of the subject's whole brain for each part, a score representing the degree of atrophy of the subject's whole brain for each part, and data over time of the subject, and an image of the subject .
The diagnosis support device according to claim 1 . an inference unit that inputs an image of the whole brain of a subject into a trained model that is generated based on training data including subject data associated with actual measurement data of dementia of a plurality of subjects, images of the whole brain of the plurality of subjects, and information on test scores on cognitive function of the subjects, the age of the subject, the sex of the subject, the height of the subject, the weight of the subject, the medical history of the subject, the volume of each part of the whole brain of the subject, a score representing the atrophy degree of the whole brain of the subject, and at least any one or a combination of a plurality of these pieces of data over time , and makes an inference related to dementia of the subject;
An estimation unit that estimates a region of interest of the entire brain in an image of the subject that is the basis for the result of the estimation using the trained model;
a providing unit that provides a diagnosis support device with information indicating a result of the inference and a region of interest of the entire brain in the image of the subject that is the basis of the result of the inference;
An estimation device comprising: 1. A diagnostic support system comprising: a diagnostic support device; and an estimation device that stores a trained model generated based on training data including subject data associated with actual measurement data of dementia of a plurality of subjects, and including images of the whole brains of the plurality of subjects, and test scores on cognitive functions of the subjects, ages of the subjects, sexes of the subjects, heights of the subjects, weights of the subjects, medical histories of the subjects, volumes of the whole brains of the subjects for each site, scores representing atrophy of the whole brains of the subjects, and information on at least one or a combination of a plurality of these pieces of aging data ,
The diagnosis support device includes:
A first acquisition unit that acquires an image of the subject's whole brain;
A second acquisition unit that inputs an image of the subject's whole brain into the trained model and acquires an inference result related to dementia of the subject;
A display unit that displays the inference result and information indicating a region of interest of the entire brain in the image of the subject that is the basis of the inference result.
Diagnostic support system. The estimation device includes:
A prediction unit that uses the trained model to make a prediction related to dementia of the subject based on an image of the subject ;
An estimation unit that estimates a region of interest of the entire brain in an image of the subject that is the basis for the result of the estimation using the trained model,
The diagnosis support system according to claim 9. The estimation device includes:
A generation unit that generates the trained model,
The diagnosis support system according to claim 10. The generation unit re-learns the trained model based on an image of the subject's whole brain and a diagnosis result related to dementia of the subject.
The diagnosis support system according to claim 11. acquiring an image of the subject's whole brain by a first acquisition unit of the diagnosis support device;
inputting the image of the whole brain of the subject into a trained model generated based on training data including subject data associated with actual measurement data of dementia of the multiple subjects, images of the whole brain of the multiple subjects, and information on test scores on cognitive function of the subjects, the age of the subject, the sex of the subject, the height of the subject, the weight of the subject, the medical history of the subject, the volume of each part of the whole brain of the subject, a score representing the atrophy degree of the whole brain of the subject, and at least any one or a combination of a plurality of these pieces of data over time ; and acquiring an inference result of making an inference related to dementia of the subject by a second acquisition unit of the diagnosis support device;
displaying, by a display unit of the diagnosis support device, the inference result and information indicating a region of interest of the entire brain in an image of the subject that is the basis of the inference result;
A diagnostic support method comprising: Diagnostic support devices:
obtaining an image of the subject's whole brain;
inputting the whole brain image of the subject into a trained model generated based on training data including subject data associated with actual measurement data of dementia of the multiple subjects, images of the whole brain of the multiple subjects, and information on test scores on cognitive function of the subjects, the age of the subject, the sex of the subject, the height of the subject, the weight of the subject, medical history of the subject, volume of each part of the whole brain of the subject, a score representing the atrophy degree of the whole brain of the subject, and at least any one or a combination of a plurality of these pieces of data over time , and obtaining an inference result for making an inference related to dementia of the subject;
Displaying the inference result and information indicating a region of interest of the whole brain in an image of the subject that is the basis of the inference result;
A diagnostic support program that executes the above. A trained model for causing a calculation unit of a computer to perform calculations,
A learning process of a learning model using learning data including subject data in which actual measurement data of dementia of a plurality of subjects is associated and which includes images of the whole brains of the plurality of subjects, test scores on the cognitive functions of the subjects, the ages of the subjects, the sexes of the subjects, the heights of the subjects, the weights of the subjects, the medical history of the subjects, the volumes of the regions of the whole brain of the subjects, scores representing the degree of atrophy of the whole brain of the subjects, and at least any one or a combination of a plurality of these pieces of data over time,
An image of the whole brain of a subject is input, and an arithmetic unit of a computer makes an inference related to dementia of the subject;
A calculation unit of the computer estimates a region of interest of the whole brain of the subject in the image on which the result of the estimation is based;
A trained model that has been trained to perform the following:

Images