JP2024007851A - Information processing device, information processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a technology that, without presence of experts in cardiac pathology, enables cardiomyopathy to be diagnosed with a same proficiency level as the experts.

SOLUTION: A diagnosis information acquisition unit 51 acquires, from a user terminal 2, diagnosis data that includes cardiac muscle tissue images of the diagnosis object. An AI diagnosis unit 52 outputs a probability corresponding to each of three kinds which are, DCM, HCM, and normal, using the diagnosis object data and an AI model 71, and outputs diagnosis results regarding the diagnosis object on the basis of the probability. A diagnosis result information generation unit 53 generates information that indicates diagnosis results by the AI diagnosis unit 52 as diagnosis result information, and presents the diagnosis result information to users such as doctors via the user terminal 2.

SELECTED DRAWING: Figure 3

COPYRIGHT: (C)2024,JPO&INPIT

Description

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

Conventionally, image diagnosis such as echocardiography and MRI (for example, Patent Document 1) has been important for diagnosing cardiomyopathy, but due to limited resolution, pathologists and cardiologists specializing in cardiac pathology ( In many cases, a diagnosis is only reached after an examination by a person (hereinafter referred to as an "expert").

Japanese Patent Application Publication No. 2021-130610

However, there are very few specialists in cardiac pathology, which is the deciding factor in diagnosing cardiomyopathies, nationwide, and as a result, there are only a limited number of facilities that actively perform cardiac pathology diagnosis.
For this reason, there is a need for a technology that can diagnose cardiomyopathy in the same manner as an expert, even without a cardiac pathology expert.

The present invention has been made in view of this situation, and it is an object of the present invention to realize a technique that allows diagnosis of cardiomyopathy equivalent to that of an expert, even in the absence of a cardiac pathology expert.

To achieve the above object, an information processing device according to one embodiment of the present invention includes:
An information processing device that supports diagnosis of cardiomyopathy,
a diagnostic target acquisition means for acquiring diagnostic target data including a myocardial tissue image of the diagnostic target;
Based on the diagnosis target data, output the probability of corresponding to each of three types: DCM which is dilated cardiomyopathy, HCM which is hypertrophic cardiomyopathy, and normal, and based on the probabilities, the diagnosis result for the diagnosis target is output. A cardiomyopathy diagnostic means to output,
Equipped with.

An information processing method and a program according to one embodiment of the present invention are respectively a method and a program corresponding to an information processing apparatus according to one embodiment of the present invention.

According to the present invention, it is possible to realize a technique capable of diagnosing cardiomyopathy as well as an expert, even without a cardiac pathology expert.

1 is a diagram showing an example of the configuration of an information processing system to which a diagnostic device according to an embodiment of the information processing device of the present invention is applied. 2 is a block diagram showing an example of the hardware configuration of a diagnostic device according to an embodiment of the information processing device of the present invention in the information processing system of FIG. 1. FIG. 3 is a block diagram illustrating an example of a functional configuration of a diagnostic device having the hardware configuration of FIG. 2. FIG. 4 is a flowchart illustrating an example of a diagnostic process executed by an AI diagnostic unit of the diagnostic apparatus in FIG. 3. FIG. 4 is a diagram illustrating an example of a two-dimensional degeneration/fibrosis plot generated when DCM is diagnosed by the AI diagnosis unit of the diagnostic device of FIG. 3. FIG. 4 shows an example of diagnostic result information generated in the form of an image by the diagnostic result information generation unit of the diagnostic apparatus of FIG. 3. 7 is an enlarged view of graph information among the diagnosis result information of FIG. 6. FIG.

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing an example of the configuration of an information processing system to which a diagnostic device according to an embodiment of the information processing device of the present invention is applied.
The information processing system shown in FIG. 1 is configured to include a diagnostic device 1 and a user terminal 2. The information processing system shown in FIG.
The diagnostic device 1 and the user terminal 2 are connected to each other via a predetermined network such as the Internet.

The diagnostic device 1 is an information processing device that includes a server and the like that supports the diagnosis of cardiomyopathy.
The user terminal 2 is an information processing terminal configured with a personal computer, a tablet, a smartphone, etc., and is operated by a user such as a doctor in charge of diagnosing cardiomyopathy.

FIG. 2 is a block diagram showing an example of the hardware configuration of a diagnostic device in the information processing system shown in FIG.

The diagnostic device 1 acquires diagnostic target data including a myocardial tissue image to be diagnosed from the user terminal 2 or the like.
Based on the diagnosis target data, the diagnostic device 1 outputs the probability of corresponding to each of three types: DCM which is dilated cardiomyopathy, HCM which is hypertrophic cardiomyopathy, and normal, and determines the diagnosis target based on the probability. The diagnostic results are output and presented to the user terminal 2.

The diagnostic device 1 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, and a GPU (Graphics Processing Unit) 14. , bus 15, and input/output interface 16. , an input section 17, an output section 18, a storage section 19, a communication section 20, and a drive 21. However, if the GPU 14 is not provided, the processing can be performed by the CPU 11 instead.

The CPU 11 executes various processes according to programs recorded in the ROM 12 or programs loaded into the RAM 13 from the storage unit 19. When the GPU 14 is provided, the processing of the learning section 55 and the AI diagnosis section 52 can be executed in place of the CPU 11, thereby shortening the processing time.
The RAM 13 also appropriately stores data necessary for the CPU 11 to execute various processes.

The CPU 11, ROM 12, RAM 13, and GPU 14 are interconnected via a bus 15. An input/output interface 16 is also connected to this bus 15. An input section 17 , an output section 18 , a storage section 19 , a communication section 20 , and a drive 21 are connected to the input/output interface 16 .

The input unit 17 includes, for example, a keyboard, and inputs various information.
The output unit 18 includes a display such as a liquid crystal display, a speaker, and the like, and outputs various information as images and sounds.
The storage unit 19 is composed of a DRAM (Dynamic Random Access Memory) or the like, and stores various data.
The communication unit 20 communicates with other devices (for example, the user terminal 2, etc.) via a network NW including the Internet.

A removable medium 30 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is appropriately installed in the drive 21 . The program read from the removable medium 30 by the drive 21 is installed in the storage unit 19 as necessary.
Further, the removable medium 30 can also store various data stored in the storage section 19 in the same manner as the storage section 19.

Through cooperation between various hardware and various software of the diagnostic apparatus 1 shown in FIG. 2, various processes can be executed.
Hereinafter, with reference to FIG. 3, a functional configuration for the diagnostic device 1 of this embodiment to execute such various processes will be described.
FIG. 3 is a block diagram showing an example of the functional configuration of a diagnostic device having the hardware configuration of FIG. 2. As shown in FIG.

As shown in FIG. 3, in the CPU 11 of the diagnostic device 1, a diagnostic information acquisition section 51, an AI diagnosis section 52, a diagnosis result information generation section 53, a learning data generation section 54, and a learning section 55 function. .
Furthermore, an AI model 71 is stored in the storage unit 19 of the diagnostic device 1 .

The diagnostic information acquisition unit 51 acquires diagnostic target data including a myocardial tissue image (digital data) to be diagnosed from the user terminal 2 .
The myocardial tissue image (digital data) to be diagnosed is sufficient as long as it is an image (digital data) obtained as a result of imaging the heart to be diagnosed for cardiomyopathy using a microscope. Not limited. That is, for convenience of explanation, the form of acquisition from the user terminal 2 was adopted here, but for example, it may also be obtained by directly inputting digital data stored in the removable medium 30, or it may be obtained from a cloud (not shown). The digital data may be acquired from the device through communication.

The AI diagnosis unit 52 uses the diagnosis target data and the AI model 71 to output probabilities corresponding to each of the three types of DCM, HCM, and normal, and outputs a diagnosis result for the diagnosis target based on the probabilities. Hereinafter, such processing will be referred to as "diagnosis processing."
The AI diagnosis section 52 is provided with an amyloidosis judgment section 61, a myocarditis judgment section 62, and a DCM/HCM judgment section 63.

Hereinafter, a detailed example of the diagnosis processing of the AI diagnosis section 52 will be explained with reference to the flowchart in FIG. The configuration will also be explained.
FIG. 4 is a flowchart illustrating an example of a diagnostic process executed by the AI diagnostic unit of the diagnostic apparatus shown in FIG.

In FIG. 4, diagnostic target data is shown as diagnostic information consisting of pathological images and test data. A pathological image is a myocardial tissue image (digital data) about the heart of a patient to be diagnosed. The test data is data indicating the results of a test performed on the patient by a user (such as a doctor) operating the user terminal 2.

In step S1, the amyloidosis determining unit 61 determines whether or not amyloidosis is present based on the diagnosis target data.
Here, as shown in FIG. 6, which will be described later, the myocardial tissue image is divided into 7×5 unit images (hereinafter referred to as “patches”). It is assumed that the AI model 71 has been trained to be able to determine whether typical image features are included for each patch.
Therefore, the amyloidosis determining unit 61 uses the AI model 71 to determine whether typical image features are included for each patch.
If the number (ratio) of patches that include typical image features exceeds the threshold, the amyloidosis determination unit 61 diagnoses amyloidosis in step S2. This ends the diagnostic process.
On the other hand, if the number (ratio) of patches that include typical image features is less than or equal to the threshold, it is not amyloidosis, and the process proceeds to step S3.

In step S3, the myocarditis determining unit 62 determines whether or not it is myocarditis based on the diagnosis target data.
Here, it is assumed that the AI model 71 has been trained to be able to determine whether or not each patch has myocarditis.
Therefore, the myocarditis determining unit 62 uses the AI model 71 to determine whether or not each patch has myocarditis.
If the number (ratio) of patches determined to be myocarditis exceeds the threshold, in step S4, the myocarditis determining unit 62 diagnoses the patch to be myocarditis. This ends the diagnostic process.
On the other hand, if the number (ratio) of patches determined to be myocarditis is less than or equal to the threshold, it is not myocarditis, and the process proceeds to step S5.

Here, it is assumed that the AI model 71 used in step S5 has been trained to output probabilities (certainty degrees) corresponding to each of the three types, DCM, HCM, and normal, for each patch. Note that the details of learning will be described later as an explanation of the functional configuration of the learning section 55.
In step S5, the DCM/HCM determining unit 63 classifies each of the 35 patches into the type (case) with the highest probability among DCM, HCM, and normal. That is, the type (case) into which a predetermined patch is classified is the diagnosis result of the predetermined patch.
The DCM/HCM determination unit 63 calculates the ratio (to the total of 35 patches) of the number of patches classified into three types (three cases) of DCM, HCM, and normal.
The DCM/HCM determining unit 63 outputs the predetermined type (predetermined case) with the highest ratio of the number of patches as the determination result in step S5.

In step S5, if the predetermined type (predetermined case) with the highest proportion of patches is DCM and the proportion exceeds the threshold, the process proceeds to step S6.
In step S6, the DCM/HCM determining unit 63 diagnoses DCM and advances the process to step S7.
In step S7, the DCM/HCM determining unit 63 determines the degree of myocardial cell degeneration and interstitial fibrosis (by a certain threshold, myocardial cell degeneration is divided into low and high values, and by a certain threshold, interstitial fibrosis is divided into low and high values). The subtype is determined as shown in FIG. 5. This ends the diagnostic process.

FIG. 5 is a diagram showing an example of a two-dimensional degeneration/fibrosis plot generated when DCM is diagnosed.
Since the degeneration/fibrosis two-dimensional plot can visually confirm which position the DCM corresponds to, it is included in the diagnosis result information described later, transmitted to the user terminal 2, and presented to the user such as a doctor. .

Returning to FIG. 4, in step S5, if the predetermined type (predetermined case) with the highest proportion of patches is HCM and the proportion exceeds the threshold, the process proceeds to step S8.
In step S8, the DCM/HCM determining unit 63 determines the presence or absence of vacuolar degeneration.
If it is determined in step S8 that vacuolar degeneration is present, in step S9, the DCM/HCM determining unit 63 diagnoses HCM as a storage disease requiring diagnosis. This ends the diagnostic process.
On the other hand, if it is determined in step S8 that there is no vacuolar degeneration, the DCM/HCM determining unit 63 diagnoses HCM in step S10. This ends the diagnostic process.

In addition, in step S5, for the predetermined type (predetermined case) with the highest proportion of patches, if the proportion is less than the threshold, that is, which of the three types (three cases) of DCM, HCM, and normal If the ratio is also less than the threshold, the process proceeds to step S11.
In step S11, the DCM/HCM determining unit 63 determines the presence or absence of vacuolar degeneration.
If it is determined in step S11 that vacuolar degeneration is present, the DCM/HCM determining unit 63 determines that it is impossible to determine and that vacuolar degeneration is present in step S12. This ends the diagnostic process.
On the other hand, if it is determined in step S11 that there is no vacuolar degeneration, in step S13, the DCM/HCM determining unit 63 diagnoses that it cannot be determined and that there is no vacuolar degeneration. This ends the diagnostic process.

Further, in step S5, if the predetermined type (predetermined case) with the highest proportion of patches is normal and the proportion exceeds the threshold, the process proceeds to step S14.
In step S14, the DCM/HCM determining unit 63 diagnoses that it is normal. This ends the diagnostic process.

Returning to FIG. 3, when the AI diagnosis unit 52 performs the diagnosis process shown in FIG. 2) and presents it to a user such as a doctor via the user terminal 2.
Although the form in which the diagnosis result information is presented to the user is not particularly limited, it is assumed here that the form of the image shown in FIG. 6 is adopted.
FIG. 6 shows an example of diagnostic result information generated in the form of an image by the diagnostic result information generation unit of the diagnostic apparatus shown in FIG.
Diagnosis result information GS in the example of FIG. 6 includes image information GT and graph information GB.

In the image information GT, frames indicating 35 patches are superimposed on the myocardial tissue image to be diagnosed, and each of the 35 patches further includes patch diagnosis results (DCM, HCM, normal 3). Information indicating the type (three cases), which type (diagnosis name) such as amyloidosis or myocarditis has been classified, and the confidence level of the classified type (diagnosis name) is displayed.

Details of the graph information GB are shown in FIG.
As shown in FIG. 7, the graph information GB includes a bar graph obtained by arranging 35 blocks representing each of the 35 patches in a display mode representing the diagnostic results of the patches described above. . This bar graph allows the percentage of each patch for each diagnosis name (type) to be easily recognized.
Here, the "display mode showing the diagnosis result of the patch described above" is not particularly limited, and each patch is classified into DCM, HCM, three types (three cases) of normal, amyloidosis, and myocarditis. There is no particular limitation as long as it is a distinguishable form.
In other words, in FIG. 7, for convenience of explanation, the forms are distinguished by "patches," but in reality, there are three types (three cases) of DCM, HCM, normal, amyloidosis, and myocarditis. A unique color is assigned in advance to each patch, and a block representing each patch is painted in the corresponding color.
Furthermore, in order to be able to see the confidence level, in reality, the block representing each patch is painted in the corresponding color so that the lower the confidence level, the deeper the degree of translucency. .

In addition, in the graph information GB or information not shown, a sentence of a pattern selected from M (M is an integer value of 2 or more independent of N) patterns is presented as a sentence indicating the diagnosis result. You can.
Here, the number of types of patterns M and the information content of each pattern are not particularly limited, but for example, the following M=3 types are adopted here, and the diagnosis result information GS is selected by a user such as a doctor. It is possible to present it by including it in
That is, the first pattern is a pattern containing only the final diagnosis name, and is, for example, text information such as "HCM, please also differentiate storage disease."
The second pattern is a pattern of the percentage of patches of each classification (3 cases), such as DCM, HCM, and normal, and includes text information such as "DCM 15%, Normal 15% HCM 70%, without vacuum". It is.
The third pattern is the confidence ratio of each classification such as DCM, HCM, and normal three types (three cases), and is, for example, text information such as "DCM 13%, Normal 18% HCM 69%, without vacuum". .

Furthermore, when the diagnosis result is DCM, as described above, the two-dimensional plot of degeneration/fibrosis as shown in FIG. 5 is also included in the diagnosis result information GS.

Returning to FIG. 3, the AI model 71 used in the diagnostic processing for generating the above-described diagnostic result information GS is generated or updated by the learning data generation section 54 and the learning section 55. Here, updating means re-learning using new learning data.

The learning data generation unit 54 generates one of a plurality of types including DCM, HCM, and normal for each of the N unit images (patches), which are myocardial tissue images of actual cases. Generate learning data that includes myocardial tissue images that have been annotated.
Using the learning data, the learning unit 55 calculates the probability (certainty) of each of the N unit images corresponding to each of the plurality of types when a myocardial tissue image divided into N unit images (patches) is input. Generate or update the AI model 71 to be output.

Specifically, for example, an interface similar to that shown in FIG. A cardiac pathology expert is provided with an interface that can provide any one of a plurality of types of diagnostic results, including the three types. That is, cardiac pathologists use such an interface to annotate myocardial tissue images of actual cases.
K myocardial tissue images of actual cases that have been annotated in this way (K is a large number of integer values, but it is preferable that the values are such that each of multiple types is included in the same ratio), and the learning The learning data generation unit 54 generates the data as data.
Here, one myocardial tissue image was an image of 1920×1440 pixels, and the patch was an image of N=35 in 5 rows and 7 columns, that is, 256 pixels in the vertical and horizontal directions.

Further, for example, the learning unit 55 uses learning data including the K annotated myocardial tissue images to perform learning using the following learning method to develop an AI configured as a deep learning network. Model 71 was generated.
That is, a learning method was adopted to construct a deep learning network that classifies into four classes: DCM, HCM, normal, and background.
That is, a learning method that uses ResNet (a high-accuracy model for ImageNet classification problems) that has been pre-trained on ImageNet (a large-scale dataset composed of natural images, etc.) as an image feature extractor, and retrains all layers. was adopted. A learning method was adopted in which K myocardial tissue images were randomly divided and cross-validated. A learning method was adopted in which learning is performed using an appropriate number of epochs (learning is performed on all learning data as many times as the number of epochs). A learning method was adopted in which the AI classification results for the N=35 patches in the test image were determined by a majority vote among three classes excluding Background, and the class with the highest frequency was used as the predictive diagnosis result for the myocardial tissue image.

Here, regarding the three types of histological classification of DCM, HCM, and normal, the diagnostic accuracy rate is only 77% even for experts with information only from myocardial tissue images without clinical information.
By using the AI model 71 generated as a result of learning by the above-described learning method, the correct answer rate can reach 71%. In the future, by increasing the number of learning data and repeating re-learning, it can be estimated that an AI model 71 with a level of correct answer rate exceeding that of experts will be obtained.
In this way, it is possible to realize a technique that can diagnose cardiomyopathy in the same manner as an expert, even without a cardiac pathology expert.

As explained above, the diagnostic device 1 of the present embodiment (see FIGS. 1 to 4) automatically performs the following operations when cardiac pathological photographs taken with a microscope (myocardial tissue images to be diagnosed) and various test data are input. Pathological diagnosis names can be output as results.
The diagnostic device 1 of the present embodiment (see FIGS. 1 to 4) can use AI to perform a diagnosis equivalent to that of an expert even without a cardiac pathology expert, and the diagnosis results can also be obtained using a unique interface. It is possible to present the information to the user using the following information.
That is, according to the diagnostic device 1 of this embodiment, even if there is no cardiac pathology expert among the doctors, it is possible to diagnose cardiomyopathy as well as an expert.

An information processing system (for example, an information processing system as shown in FIG. 1) including the diagnostic device 1 of this embodiment will be hereinafter referred to as a "cardiac pathology diagnosis system", and the cardiac pathology diagnosis system will be automated and It is expected that this method will not only be standardized, but also contribute to the optimization of medical care by more effectively identifying rare cardiomyopathy.
In addition, by linking the histopathological analysis results from the cardiac pathology diagnostic system with clinical history and genomic information, it is possible to predict the prognosis and severity from histological images, which is important from a clinical perspective in myocardial pathological diagnosis. This results in an increase in gender.
It is expected that currently existing unclassifiable cardiomyopathies will be able to be classified by histopathological diagnosis using an AI-based cardiac pathological diagnosis system, and this will allow us to propose new disease groups including undiagnosed cardiomyopathies. can.
Furthermore, as a result of these results, if the number of facilities performing myocardial biopsies increases, clinical evidence will accumulate, which will ultimately lead to the revision of clinical guidelines, which is expected to contribute to more accurate and optimized treatment of myocardial diseases. .

Although one embodiment of the present invention has been described above, the present invention is not limited to the above-described embodiment, and modifications, improvements, etc. within the range that can achieve the purpose of the present invention are included in the present invention. regarded as.

For example, the AI model 71 is not particularly limited to one that constructs a deep learning network that classifies into the four classes of DCM, HCM, normal, and Background. A model that outputs some probability of the three types is sufficient.
Further, in the above embodiment, an example in which one AI model 71 is stored in the storage unit 19 has been described, but a plurality of AI models 71 may be provided for each disease, and the number of AI models 71 is not limited to that in the embodiment.
In the above embodiment, when the learning unit 55 receives a myocardial tissue image divided into N unit images using learning data, it outputs the probability that each of the N unit images corresponds to each of a plurality of types. Although the explanation has been given as an example of generating or updating the AI model 71 (see FIG. 3), in addition to this, for example, the learning unit 55 has a function of only generating the AI model 71, and the function of the AI model 71 generated by the learning unit 55 is Evaluation means (evaluation section) for evaluating and updating performance may be provided separately.

Further, the system configuration shown in FIG. 1 and the hardware configuration of the diagnostic device 1 shown in FIG. 2 are merely examples for achieving the object of the present invention, and are not particularly limited.

Further, the functional block diagram shown in FIG. 3 is merely an example and is not particularly limited. In other words, it is sufficient that the information processing system is equipped with a function that can execute the above-mentioned series of processes as a whole, and what kind of functional blocks are used to realize this function is not particularly limited to the example shown in FIG. 3. .

Further, the location of the functional blocks is not limited to that shown in FIG. 3, and may be arbitrary.
For example, in the example of FIG. 3, the above-mentioned processing is performed on the diagnostic device 1 side, but the configuration is not limited thereto. Some may be done.
For example, the functional blocks required for learning are provided in the diagnostic device 1, but this is merely an example, and a learning device (not shown) may be provided.
Similarly, the location of the AI model 71 is not limited to that shown in FIG. 3, and may be any location such as on a cloud (not shown).

Furthermore, the series of processes described above can be executed by hardware or by software.
Further, one functional block may be configured by a single piece of hardware, a single piece of software, or a combination thereof.

When a series of processes is executed by software, a program constituting the software is installed on a computer or the like from a network or a recording medium.
The computer may be a computer built into dedicated hardware.
Further, the computer may be a computer that can execute various functions by installing various programs, such as a server, a general-purpose smartphone, or a personal computer.

A recording medium containing such a program is not only composed of a removable medium (not shown) distributed separately from the device itself, but also a recording medium provided pre-installed in the device body. .

Note that in this specification, the step of writing a program to be recorded on a recording medium is not only a process that is performed chronologically in accordance with the order, but also a process that is not necessarily performed chronologically but in parallel or individually. It also includes the processing to be executed.
Furthermore, in this specification, the term system refers to an overall device composed of a plurality of devices, a plurality of means, and the like.

In summary, the information processing system to which the present invention is applied only needs to have the following configuration, and can take various embodiments.

That is, the information processing device to which the present invention is applied (for example, the diagnostic device 1 in FIGS. 1 to 3),
An information processing device that supports diagnosis of cardiomyopathy,
A diagnostic target acquisition means (for example, the diagnostic information acquisition unit 51 in FIG. 3) that acquires diagnostic target data including a myocardial tissue image of the diagnostic target;
Based on the diagnosis target data, output the probability of corresponding to each of three types: DCM which is dilated cardiomyopathy, HCM which is hypertrophic cardiomyopathy, and normal, and based on the probabilities, the diagnosis result for the diagnosis target is output. Cardiomyopathy diagnostic means to output (for example, AI diagnostic unit 52 in FIG. 3),
Equipped with.

The diagnosis target acquisition means acquires the myocardial tissue image divided into N unit images (N is an integer value of 2 or more) (for example, the above-mentioned N=35 patches),
The cardiomyopathy diagnosing means outputs a probability corresponding to each of the three types for each of the N unit images, and outputs a diagnosis result for the diagnosis target based on the probability.
be able to.

The cardiomyopathy diagnostic means includes:
Classifying each of the N unit images into the type with the highest probability among the three types (for example, steps S6, S8, and S11 in FIG. 4);
Among the three types, if the proportion of the predetermined type exceeds a threshold for the predetermined type with the highest proportion of the number of classified unit images, the predetermined type is output as the diagnosis result (for example, Steps S7, S9, S10, S14 of FIG. 4), if the proportion of the predetermined type is below the threshold value, output undeterminable as the diagnosis result (for example, steps S12, S13 of FIG. 4);
be able to.

When outputting the DCM, the HCM, or the indeterminable as the diagnosis result, the cardiomyopathy diagnostic means additionally verifies the presence or absence of vacuolar degeneration (for example, steps S6, S8, and S11 in FIG. 4). ,
be able to.

Before outputting the probabilities corresponding to each of the three types, the cardiomyopathy diagnostic means further:
Based on the myocardial tissue image, it is determined whether or not it is amyloidosis (for example, step S1 in FIG. 4), and then whether or not it is myocarditis (for example, step S3 in FIG. 4),
If it is determined that the amyloidosis is applicable, output the amyloidosis as the diagnosis result (for example, step S2 in FIG. 4);
If it is determined that the myocarditis corresponds to the myocarditis, the myocarditis is outputted as the diagnosis result (for example, step S4 in FIG. 4);
If it is determined that neither the amyloidosis nor the myocarditis applies, the probabilities corresponding to each of the three types are output, and the diagnosis result is output based on the probabilities (for example, from step S5 in FIG. 4 S14),
be able to.

Learning data including myocardial tissue images of actual cases, in which each of the N unit images has been annotated with one of a plurality of types including at least the three types. A learning data generating means (for example, the learning data generating unit 54 in FIG. 3) to generate,
Using the learning data, when a myocardial tissue image divided into the N unit images is input, a model that outputs the probability corresponding to each of the plurality of types for each of the N unit images (for example, in FIG. A learning means (for example, the learning unit 55 in FIG. 3) that generates or updates the AI model 71);
Furthermore,
The cardiomyopathy diagnosing means inputs the myocardial tissue image of the diagnosis target to the model, and outputs the diagnosis result based on the output of the model.
be able to.

Diagnosis by the cardiomyopathy diagnosing means together with the myocardial tissue image of the diagnosis target divided into the N unit images to which the type with the highest probability among the three types outputted by the cardiomyopathy diagnosing means is respectively assigned. Diagnosis result presentation means (for example, the diagnosis result information generation unit 53 in FIG. 3) that presents the results (for example, the image information GT in FIG. 6)
It is possible to further include the following.

The diagnostic result presentation means further includes:
A bar graph obtained by arranging N blocks representing each of the N unit images in a display mode indicating which of the diagnosis results including at least the three types has been classified (for example, FIG. 6 and FIG. 7 graph information GB),
be able to.

The diagnostic result presentation means includes:
When presenting the DCM as the diagnosis result, further,
Presenting a degeneration/fibrosis two-dimensional plot (see e.g. Figure 5),
be able to.

The diagnostic result presentation means includes:
Presenting a sentence of a pattern selected from M (M is an integer value of 2 or more independent of N) patterns as a sentence indicating the diagnosis result;
be able to.

DESCRIPTION OF SYMBOLS 1...Diagnostic device, 11...CPU, 19...Storage part, 21...Drive, 30...Removable media, 51...Diagnostic information acquisition part, 52...AI diagnosis part, 53... Diagnosis result information generation section, 54... Learning data generation section, 55... Learning section, 61... Amyloidosis judgment section, 62... Myocarditis judgment section, 63... DCM/HCM Judgment department

Claims (12)

In an information processing device that supports the diagnosis of cardiomyopathy,
a diagnostic target acquisition means for acquiring diagnostic target data including a myocardial tissue image of the diagnostic target;
Based on the diagnosis target data, output the probability of corresponding to each of three types: DCM which is dilated cardiomyopathy, HCM which is hypertrophic cardiomyopathy, and normal, and based on the probabilities, the diagnosis result for the diagnosis target is output. A cardiomyopathy diagnostic means to output,
An information processing device comprising: The diagnosis target acquisition means acquires the myocardial tissue image divided into N unit images (N is an integer value of 2 or more);
The cardiomyopathy diagnosing means outputs a probability corresponding to each of the three types for each of the N unit images, and outputs a diagnosis result for the diagnosis target based on the probability.
The information processing device according to claim 1. The cardiomyopathy diagnostic means includes:
Classifying each of the N unit images into the type with the highest probability among the three types,
Among the three types, if the proportion of the predetermined type exceeds the threshold for the predetermined type with the highest ratio of the number of classified unit images, the predetermined type is output as the diagnosis result, and the predetermined type is outputted as the diagnosis result. If the percentage of types is less than a threshold, outputting indeterminable as the diagnosis result;
The information processing device according to claim 2. When the cardiomyopathy diagnostic means outputs the DCM, the HCM, or the indeterminable as the diagnosis result, the cardiomyopathy diagnostic means further verifies the presence or absence of vacuolar degeneration.
The information processing device according to claim 3. Before outputting the probabilities corresponding to each of the three types, the cardiomyopathy diagnostic means further:
Based on the myocardial tissue image, determining whether or not amyloidosis is present, and then determining whether or not myocarditis is present, respectively,
If it is determined that the amyloidosis is applicable, output the amyloidosis as the diagnosis result,
If it is determined that the myocarditis corresponds to the myocarditis, output the myocarditis as the diagnosis result,
If it is determined that neither the amyloidosis nor the myocarditis applies, output the probability that each of the three types applies, and output the diagnostic result based on the probability;
The information processing device according to claim 1. Learning data including myocardial tissue images of actual cases, in which each of the N unit images has been annotated with one of a plurality of types including at least the three types. a learning data generating means to generate;
Using the learning data, when a myocardial tissue image divided into the N unit images is input, a model is generated or updated that outputs the probability corresponding to each of the plurality of types for each of the N unit images. learning means and
Furthermore,
The cardiomyopathy diagnosing means inputs the myocardial tissue image of the diagnosis target to the model, and outputs the diagnosis result based on the output of the model.
The information processing device according to claim 2. Diagnosis by the cardiomyopathy diagnosing means together with the myocardial tissue image of the diagnosis target divided into the N unit images to which the type with the highest probability among the three types outputted by the cardiomyopathy diagnosing means is respectively assigned. The information processing apparatus according to claim 3, further comprising: a diagnosis result presentation means for presenting the results. The diagnostic result presentation means further includes:
Presenting a bar graph obtained by arranging N blocks representing each of the N unit images in a display mode indicating which of the diagnostic results including at least the three types has been classified.
The information processing device according to claim 7. The diagnostic result presentation means includes:
When presenting the DCM as the diagnosis result, further,
Presenting a degeneration/fibrosis two-dimensional plot,
The information processing device according to claim 7. The diagnostic result presentation means includes:
Presenting a sentence of a pattern selected from M (M is an integer value of 2 or more independent of N) patterns as a sentence indicating the diagnosis result;
The information processing device according to claim 7. In an information processing method executed by an information processing device that supports diagnosis of cardiomyopathy,
a diagnostic target acquisition step of acquiring diagnostic target data including a myocardial tissue image of the diagnostic target;
Based on the diagnostic target data, the probability of corresponding to each of the three types of DCM, which is dilated cardiomyopathy, HCM, which is hypertrophic cardiomyopathy, and normal, is output, and the diagnostic result for the diagnostic target is determined based on the probability. Cardiomyopathy diagnosis step to output,
Information processing methods including A computer that supports the diagnosis of cardiomyopathy,
a diagnostic target acquisition step of acquiring diagnostic target data including a myocardial tissue image of the diagnostic target;
Based on the diagnosis target data, output the probability of corresponding to each of three types: DCM which is dilated cardiomyopathy, HCM which is hypertrophic cardiomyopathy, and normal, and based on the probabilities, the diagnosis result for the diagnosis target is output. Cardiomyopathy diagnosis step to output,
A program that executes control processing including.

Images