JP2021189093A - Diagnostic device for autoimmune bullosis - Google Patents

Abstract

To enable accurate and consistent diagnosis of autoimmune bullosis.SOLUTION: A diagnostic device is provided, diagnosing autoimmune bullosis and comprising: an acquisition unit acquiring a fluorescent antibody image obtained by a fluorescent antibody method; and a diagnostic unit which inputs the fluorescent antibody image into a predetermined machine learning model and generates information on a diagnosis result of the autoimmune bullosis as output of the machine learning model.SELECTED DRAWING: Figure 8

Description

The present invention relates to a diagnostic device, method and program for diagnosing autoimmune vesicular disease, and a method, program and device for generating a machine learning model for diagnosing autoimmune vesicular disease.

Autoimmune blister disease is one of the autoimmune diseases in which an antibody (autoantibody) appears in response to the tissue of one's own body and damages one's own tissue. It is a disease that occurs. Diagnosis of autoimmune vesicular disease is mainly made by immunofluorescence. As shown in Non-Patent Document 1, the fluorescent antibody method includes a direct method and an indirect method. The former is a method in which an autoantibody deposited on the collected skin, which is performed by biopsy of the patient's skin, is labeled with a fluorescent dye and observed with a fluorescence microscope. On the other hand, the latter is a method in which a patient's serum is reacted with a section of healthy human skin, the autoantibody in the patient's blood is labeled with a fluorescent dye, and this is observed with a fluorescence microscope.

Hirotsugu Koga, et al., "Insurance-listed test methods for autoimmune vesicular disease: anti-desmoglein 1 antibody, anti-desmoglein 3 antibody, anti-BP180 antibody", Modern Media, Vol. 59, No. 12, pp. 297-302, 2013

However, even if an autoantibody labeled with a fluorescent dye is observed with a fluorescence microscope, it is often difficult for a diagnostician such as a doctor or a researcher who has little experience with the fluorescent antibody method to diagnose autoimmune vesicular disease. In addition, the diagnosis results vary among skilled diagnosticians.

An object of the present invention is to enable an accurate and consistent diagnosis of autoimmune vesicular disease.

The diagnostic device according to the first aspect is a diagnostic device for diagnosing autoimmune vesicular disease, in which an acquisition unit for acquiring a fluorescent antibody image obtained by an immunofluorescent antibody method and the fluorescent antibody image are used as a predetermined machine learning model. It is provided with a diagnostic unit that inputs and generates information on the diagnosis result of the autoimmune vesicular disease as an output of the machine learning model.

The diagnostic device according to the second aspect is the diagnostic device according to the first aspect, and the information of the diagnosis result includes information indicating whether the autoimmune vesicular disease is positive or negative.

The diagnostic device according to the third aspect is the diagnostic device according to the second aspect, and the information of the diagnosis result includes information indicating a positive level defined in multiple stages.

The diagnostic device according to the fourth aspect is the diagnostic device according to the third aspect, and the information of the diagnosis result includes information indicating the probability corresponding to each of the positive levels defined in the multistage. ..

The diagnostic apparatus according to the fifth aspect is the diagnostic apparatus according to any one of the first aspect to the fourth aspect, and the information of the diagnosis result includes information indicating the type of the autoimmune vesicular disease.

The diagnostic program according to the sixth aspect is a diagnostic program for diagnosing autoimmune vesicular disease, in which a fluorescent antibody image obtained by an immunofluorescent antibody method is acquired and the fluorescent antibody image is input to a predetermined machine learning model. Then, the computer is made to generate the information of the diagnosis result of the autoimmune vesicular disease as the output of the machine learning model.

The diagnostic method according to the seventh aspect is a diagnostic method for diagnosing autoimmune vesicular disease, in which a fluorescent antibody image obtained by the fluorescent antibody method is acquired and the fluorescent antibody image is input to a predetermined machine learning model. However, the output of the machine learning model includes generating information on the diagnosis result of the autoimmune vesicular disease.

The method for generating a machine learning model according to the eighth aspect is a method for generating a machine learning model for diagnosing autoimmune vesicular disease, and the fluorescent antibody image obtained by the fluorescent antibody method and the above-mentioned diagnosed by an expert. A machine that acquires a large number of data sets including information on the diagnosis results of autoimmune vesicular disease, uses the large number of data sets as teacher data, inputs the fluorescent antibody image, and outputs the information on the diagnosis results. Includes training a learning model.

The method for generating the machine learning model according to the ninth aspect is the generation method according to the eighth aspect, and the diagnosis diagnosed by the first expert with respect to the same fluorescent antibody image in the large number of data sets. It includes a dataset containing information on the results and another dataset containing information on the results of the diagnosis diagnosed by a second expert.

The machine learning model generation program according to the tenth aspect is a machine learning model generation program for diagnosing autoimmune vesicular disease, and the fluorescent antibody image obtained by the fluorescent antibody method and the above-mentioned diagnosed by an expert. A machine that acquires a large number of data sets including information on the diagnosis results of autoimmune vesicular disease, uses the large number of data sets as teacher data, inputs the fluorescent antibody image, and outputs the information on the diagnosis results. Let the computer do the training of the learning model.

The machine learning model generator according to the eleventh aspect is a machine learning model generator for diagnosing autoimmune vesicular disease, and the fluorescent antibody image obtained by the fluorescent antibody method and the above-mentioned diagnosed by an expert. An acquisition unit that acquires a large number of data sets including information on the diagnosis results of autoimmune vesicular disease, an acquisition unit that uses the large number of data sets as teacher data, an immunofluorescent antibody image as input, and information on the diagnosis results as output. It is equipped with a learning unit that trains machine learning models.

From the above perspective, an accurate and consistent diagnosis of autoimmune vesicular disease is possible.

The whole block diagram of the diagnostic system including the computer as the diagnostic apparatus of autoimmune vesicular disease which concerns on one Embodiment of this invention. Computer functional block diagram. Fluorescent antibody image in the case of positive anti-epidermal cell membrane antibody obtained by the direct method. Fluorescent antibody image in the case of positive anti-epidermal basement membrane antibody obtained by the direct method. Fluorescent antibody image in the case of positive anti-epidermal cell membrane antibody obtained by indirect method (non-saline exfoliation method). Fluorescent antibody image in the case of positive anti-epidermal basement membrane antibody obtained by indirect method (non-saline exfoliation method). Fluorescent antibody image in the case of positive epidermal side of anti-epidermal basement membrane antibody obtained by indirect method (saline stripping method). Fluorescent antibody image in the case of positive dermis side of anti-epidermal basement membrane antibody obtained by indirect method (saline stripping method). Fluorescent antibody image in the case of anti-epidermal cell membrane antibody negative and anti-epidermal basement membrane antibody negative obtained by the indirect method (non-saline stripping method). Fluorescent antibody image in the case of anti-epidermal cell membrane antibody negative and anti-epidermal basement membrane antibody negative obtained by the indirect method (saline stripping method). A flowchart showing the flow of learning processing of a machine learning model. A flowchart showing the flow of diagnostic processing based on a machine learning model. The figure which shows the fluorescent antibody image and the corresponding heat map. The graph which shows the concordance rate between the diagnosis result by the machine learning model which concerns on an Example, and the diagnosis result (correct answer data) by a skilled expert.

Hereinafter, with reference to the drawings, a diagnostic device, method and program for autoimmune blister disease according to an embodiment of the present invention, and a method, program and method for generating a machine learning model for diagnosing autoimmune blister disease. The device will be described.

<1. Overall configuration of the diagnostic system for autoimmune vesicular disease>
FIG. 1 shows an overall configuration diagram of a diagnostic system 100 including a computer 1 as a diagnostic device for diagnosing autoimmune vesicular disease according to the present embodiment. The computer 1 is a device for diagnosing autoimmune vesicular disease based on an image obtained by the fluorescent antibody method (hereinafter referred to as a fluorescent antibody image), and as shown in FIG. 1, a diagnostic system 100 is provided together with a fluorescence microscope 2. Configure. The fluorescence microscope 2 is used to take a fluorescent antibody image.

Autoimmune blister disease is one of the autoimmune diseases in which an antibody (autoantibody) appears in response to the tissue of one's own body and damages one's own tissue. It is a disease that occurs. There are various types of autoimmune vesicular disease, which are roughly classified into the pemphigus group and the pemphigoid group. The pemphigus group includes mucosal-dominant and mucocutaneous pemphigus vulgaris (self-antibody IgG), deciduous pemphigus (self-antibody IgG), and IgA pemphigus (self-antibody IgA). Bullous pemphigoid (self-antibody IgG), gestational eczema (self-antibody IgG), mucosal pemphigoid (self-antibody IgG, self-antibody IgA), Lamina lucida type and Sublamina densa type lines are included in the pemphigoid group. A type of IgA bullous pemphigoid (self-antibody IgA) and acquired epidermal pemphigoid (self-antibody IgG) is included.

Diagnosis of autoimmune vesicular disease can be made by immunofluorescence. The fluorescent antibody method includes a direct method and an indirect method. The former is a method in which an autoantibody deposited on the collected skin, which is performed by biopsy of the patient's skin, is labeled with a fluorescent dye and observed with a fluorescence microscope. On the other hand, the latter is a method in which a patient's serum is reacted with a section of healthy human skin, the autoantibody in the patient's blood is labeled with a fluorescent dye, and this is observed with a fluorescence microscope.

3A and 3B are examples of fluorescent antibody images obtained by the direct method, FIG. 3A is an image in the case of positive pemphigus (positive anti-epidermal cell membrane antibody), and FIG. 3B is Pemphigoid. Is a positive image (anti-epidermal basement membrane antibody positive). 4A and 4B are examples of fluorescent antibody images obtained by the indirect method, FIG. 4A is an image in the case of positive anti-epidermal cell membrane antibody, and FIG. 4B is a case of positive anti-epidermal basement membrane antibody. It is an image of.

Pemphigus is a disease in which autoantibodies destroy substances that bind cells of the epidermis located on the surface of the skin, resulting in blisters and erosions on the skin and mucous membranes. It is observed that the antibody is labeled with a fluorescent dye. Pemphigoid, on the other hand, is a disease in which the protein in the basement membrane that binds the epidermis to the dermis located below the epidermis is destroyed by self-antibodies, causing blisters, erosions, and erythema on the skin and mucous membranes. And in FIG. 4B, it is observed that such an autoantibody is labeled with a fluorescent dye.

Indirect methods include a method of immersing healthy human skin in a saline solution and then reacting the patient's serum with the section (hereinafter, referred to as a saline stripping method) and a method of immersing the skin in the saline solution. There is a method of omitting the above (hereinafter, may be referred to as a non-saline stripping method). 5A and 5B are fluorescent antibody images by the saline stripping method, and FIGS. 4A and 4B are fluorescent antibody images by the non-saline stripping method. When human skin is immersed in saline solution, the epidermis and dermis are exfoliated, which is observed in FIGS. 5A and 5B. In Pemphigoid, there are cases where autoantibodies to substances on the epidermis side of the basement membrane are generated and cases where autoantibodies to substances on the dermis side are generated. Can be observed.

FIG. 6A is a fluorescent antibody image observed by the non-saline stripping method, in the case where pemphigus is negative (anti-epidermal cell membrane antibody negative) and pemphigus is also negative (anti-epidermal basement membrane antibody negative). It is an image. FIG. 6B is a fluorescent antibody image observed by the saline stripping method, which is an image in the case of negative anti-epidermal cell membrane antibody and negative anti-epidermal basement membrane antibody.

The computer 1 diagnoses autoimmune vesicular disease based on the above fluorescent antibody image. The diagnosis in the present embodiment includes a diagnosis for determining whether autoimmune vesicular disease is positive or negative, and a diagnosis for determining the above-mentioned type of autoimmune vesicular disease. In addition, the positive diagnosis of autoimmune vesicular disease in the present embodiment includes a diagnosis for discriminating the positive level defined in multiple stages. In addition, the diagnosis of the type of autoimmune vesicular disease includes a diagnosis for determining whether the patient corresponds to the pemphigus group or the pemphigus group. The diagnosis here is performed using the machine learning model 8 that inputs the fluorescent antibody image and outputs the information of the diagnosis result as described above. The machine learning model 8 is constructed by pre-learning.

Here, it is assumed that the diagnostic system 100 performs not only the diagnosis based on the machine learning model 8 but also the learning of the machine learning model 8 (generation of the machine learning model 8). That is, the computer 1 according to the present embodiment also functions as a generation device for the machine learning model 8 according to the present embodiment. However, such diagnostic processing and pre-learning processing may be performed by separate hardware systems.

Hereinafter, the configuration of the computer 1 will be described, and then the learning process of the machine learning model 8 and the diagnostic process based on the machine learning model 8 will be described in order.

<2. Computer configuration>
The hardware configuration of the computer 1 will be described with reference to FIG. The computer 1 is a general-purpose computer, and is realized as, for example, a desktop computer, a laptop computer, a tablet computer, or a smartphone. The computer 1 is manufactured by installing the program 6 on a general-purpose computer from a computer-readable recording medium such as a CD-ROM or a USB memory, or via a network such as the Internet. Program 6 is software for diagnosing autoimmune vesicular disease based on a fluorescent antibody image sent from a fluorescence microscope 2, and causes a computer 1 to perform an operation described later. The program 6 also includes a program module for causing the computer 1 to execute a learning process described later.

The computer 1 includes a display unit 11, an input unit 12, a storage unit 13, a control unit 14, and a communication unit 15. These units 11 to 15 are connected to each other via the bus line 16 and can communicate with each other. The display unit 11 can be configured as a liquid crystal display or the like, and displays the diagnosis result of autoimmune vesicular disease or the like to the user. The term "user" as used herein is a general term for doctors, researchers, patients, and other persons who require a diagnosis result of autoimmune vesicular disease. The input unit 12 can be composed of a mouse, a keyboard, a touch panel, or the like, and receives an operation from the user on the computer 1.

The storage unit 13 can be composed of a hard disk, a flash memory, or the like. The program 6 is stored in the storage unit 13. Further, in the storage unit 13, information that is learned by the learning process described later and defines the machine learning model 8 used in the diagnostic process described later is stored. The control unit 14 can be composed of a CPU, a ROM, a RAM, and the like. The control unit 14 virtually operates as the acquisition unit 14a, the learning unit 14b, the diagnosis unit 14c, and the display control unit 14d by reading and executing the program 6 in the storage unit 13. Details of the operation of each part 14a to 14d will be described later. The communication unit 15 functions as a communication interface that enables wired or wireless communication with an external device, and is connected to a communication interface mounted on the fluorescence microscope 2.

<3. Machine learning model learning (generation) processing>
Hereinafter, the learning process of the machine learning model 8 will be described with reference to FIG. 7. The machine learning model 8 used in the diagnostic process of autoimmune vesicular disease is generated by this learning process. The machine learning model 8 is typically composed of a neural network, inputs a fluorescent antibody image, and outputs information on a diagnosis result of autoimmune vesicular disease. The type of neural network used here is not particularly limited, but in the present embodiment, it is a deep neural network having a multi-layer structure having a plurality of hidden layers particularly corresponding to deep learning.

First, as step S1, teacher data for learning the machine learning model 8 is prepared. The teacher data is a collection of large datasets containing fluorescent antibody images and corresponding information on the diagnosis of autoimmune vesicular disease. That is, immunofluorescent antibodies, including direct and indirect methods (including saline and non-saline stripping), for a large number of patients suffering from various types of autoimmune vesicular disease and a large number of healthy individuals. Prepare the fluorescent antibody image obtained by the method. In addition, based on each of these fluorescent antibody images, information on the diagnosis result of autoimmune vesicular disease diagnosed by a specialist such as a skilled doctor or researcher who is familiar with the diagnosis of autoimmune vesicular disease is acquired.

In the present embodiment, the fluorescent antibody image includes pemphigus (anti-epidermal cell membrane antibody) as information on the diagnosis result corresponding to the image obtained by the direct method or the non-saline stripping method. Information indicating whether it is positive or negative and information indicating whether pemphigus (anti-epidermal basement membrane antibody) is positive or negative are obtained in association with each other. In addition, the positive level is defined in multiple stages, and the information indicating whether each of pemphigus and pemphigus is positive or negative includes information indicating the positive level. In the present embodiment, if the result is negative, the code "0" is given, and if the result is positive, the code "1", "2" or "3" indicating the level is given in three stages. .. That is, each level of pemphigus and pemphigus is evaluated on a scale of 0 to 3, including negative level 0. The numerical values 1 to 3 indicating the positive level indicate that the larger the numerical value is, the more the positive level is progressing.

Further, in the present embodiment, if the fluorescent antibody image is an image obtained by the saline stripping method, is the epidermal side of the anti-epidermal basement membrane antibody positive as information on the diagnosis result corresponding to this? Information indicating whether the antibody is negative and information indicating whether the dermis side of the anti-epidermal basement membrane antibody is positive or negative are obtained in association with each other. Again, the positive level is defined in multiple stages, and the information indicating whether each of the epidermal side and the dermis side of the anti-epidermal basement membrane antibody is positive or negative includes information indicating the positive level. In the present embodiment, if the result is negative, the code "0" is given, and if the result is positive, the code "1", "2" or "3" indicating the level is given in three stages. .. That is, the respective levels of the anti-epidermal basement membrane antibody on the epidermal side and the dermis side are evaluated on a scale of 0 to 3 including a negative level 0. The numerical values 1 to 3 indicating the positive level indicate that the larger the numerical value is, the more the positive level is progressing.

In the present embodiment, each data set included in the teacher data includes 16 values as the information of the diagnosis result as described above. More specifically, each dataset includes levels 0-3 for pemphigus (anti-epidermal cell membrane antibody), levels 0-3 for pemphigoid (anti-epidermal basement membrane antibody), and anti-epidermal basement membrane antibody. Information indicating the probability of corresponding to each level of each type is included for levels 0 to 3 on the epidermal side and levels 0 to 3 on the dermis side of the anti-epidermal basement membrane antibody. That is, here, values indicating 16 probabilities of 4 types × 4 levels are acquired. In this embodiment, the value indicating the probability included in the teacher data is represented by two values of 0 or 1.

Further, the teacher data prepared in the present embodiment includes a plurality of data sets including information on diagnosis results diagnosed by a plurality of experts for the same one fluorescent antibody image. Therefore, for example, the teacher data includes a data set including information on the diagnosis result diagnosed by the first expert and information on the diagnosis result diagnosed by the second expert for the same one fluorescent antibody image. Contains another dataset. The diagnostic results by the first and second experts may be the same or different. That is, when a person makes a diagnosis of autoimmune vesicular disease, the diagnosis results are not always the same even if all are skilled diagnosticians, and the diagnosis results may vary. In the present embodiment, in consideration of such variations, the teacher data includes diagnostic results by different experts in order to obtain one accurate diagnostic result. In addition, the teacher data here may include information on the diagnosis result diagnosed by only one expert for the same one fluorescent antibody image, or diagnosed by three or more experts. Information on diagnostic results may also be included.

The teacher data acquired as described above will include not only information indicating whether autoimmune vesicular disease is positive or negative, but also information indicating its type and positive level. In step S1, such teacher data is input to the computer 1. That is, the acquisition unit 14a acquires such teacher data and stores it in the storage unit 13.

In the following step S2, the learning unit 14b trains the machine learning model 8 based on the teacher data acquired in step S1. More specifically, the learning unit 14b fetches the data sets included in the teacher data one by one from the storage unit 13, and inputs the fluorescent antibody image contained in the data set into the current machine learning model 8. In the present embodiment, the machine learning model 8 has the same level as the diagnostic result information contained in the teacher data, that is, the level of pemphigus (anti-epidermal cell membrane antibody) 0 to 3, and the level of pemphigus (anti-epidermal basement membrane antibody). Information indicating the probability of corresponding to each level of each type for 0 to 3, levels 0 to 3 on the epidermal side of the anti-epidermal basement membrane antibody, and levels 0 to 3 on the dermis side of the anti-epidermal basement membrane antibody. It is configured to output. That is, from the machine learning model 8, values indicating 16 probabilities of 4 types × 4 levels are output. In the present embodiment, the value indicating the probability output from the machine learning model 8 is not represented as a binary value of 0 or 1, but as a numerical value having a predetermined number of digits between 0 and 1. Subsequently, the learning unit 14b has the information of the diagnosis result output from the machine learning model 8 at this time (value representing 16 probabilities of 4 types × 4 levels) and the information of the diagnosis result included in the same data set (values representing 16 probabilities of 4 types × 4 levels). The parameters of the machine learning model 8 are updated so as to minimize the error (output value representing 16 probabilities of 4 types × 4 levels). Then, the machine learning model 8 is optimized while applying a large number of data sets included in the teacher data acquired in step S1 one after another.

<4. Diagnostic processing based on machine learning model>
Next, the diagnostic process of autoimmune vesicular disease based on the machine learning model 8 will be described with reference to FIG.

First, in step S11, one of the direct method and the indirect method, a saline stripping method and a non-saline stripping method, is carried out for a patient suspected of having autoimmune vesicular disease. At this time, the fluorescent antibody image taken by the fluorescence microscope 2 is input to the computer 1, acquired by the acquisition unit 14a, and stored in the storage unit 13.

In the following step S12, the diagnostic unit 14c generates information on the diagnosis result of autoimmune vesicular disease based on the fluorescent antibody image acquired in step S1. More specifically, the diagnostic unit 14c takes out a fluorescent antibody image from the storage unit 13, inputs it to the machine learning model 8, and outputs the diagnostic result information (4 types × 4 levels) as the output of the machine learning model 8. (Values representing the 16 probabilities of) are generated.

In the following step S13, the display control unit 14d outputs the diagnostic result information (values representing 16 probabilities of 4 types × 4 levels) output from the machine learning model 8. More specifically, the display control unit 14d has a level 0 to 3 for pemphigus (anti-epidermal cell membrane antibody), a level 0 to 3 for pemphigus (anti-epidermal basement membrane antibody), and an anti-epidermal basement membrane antibody. A screen is created to display information indicating the probability of corresponding to each level of each type for levels 0 to 3 on the epidermis side and levels 0 to 3 on the dermis side of the anti-epidermal basement membrane antibody, and this is displayed. 11 Displayed on top. The user sees the screen and, based on the probability of falling into each level of each type, the patient has or does not have any of the various types of autoimmune vesicular disease. Judge. Also, if the patient has any type of autoimmune vesicular disease, the level of positivity is also determined.

The information to be output in step S13 is not limited to the above. For example, the diagnostic unit 14c may process the diagnostic result information (values representing 16 probabilities of 4 types × 4 levels) output from the machine learning model 8 and output this processed information. For example, which of the various types of autoimmune vesicular disease the patient suffers from, or any, based on the probability that the diagnostic unit 14c corresponds to each level of each type output from the machine learning model 8. And if you have any type of autoimmune vesicular disease, you may determine its positive level. In this case, the display control unit 14d creates a screen for displaying further diagnosis results (machining information) by the diagnosis unit 14c based on this information in place of or in addition to the information indicating the probability corresponding to each level of each type. , This is displayed on the display unit 11.

Further, the diagnostic unit 14c may create a heat map based on the fluorescent antibody image to be diagnosed, further create a screen for displaying the heat map, and display the heat map on the display unit 11. The heat map referred to here is, for example, an image as shown in FIG. 9, and the region contributes to the diagnosis result output from the machine learning model 8 in each region on the fluorescent antibody image to be diagnosed. It is an image in which dark colors are superimposed according to the contribution rate. In this case, the user can easily grasp the problematic part. In addition, doctors, researchers, and the like who see this can learn what characteristics of the fluorescent antibody image should be seen when diagnosing autoimmune vesicular disease. As in the example of FIG. 9, such a heat map may be displayed on the display unit 11 together with the fluorescent antibody image to be diagnosed.

As described above, accurate diagnostic results can be obtained for autoimmune vesicular disease even when there is no skilled diagnostician. In addition, there is no variation depending on the diagnostician, and one consistent diagnostic result can be obtained. Moreover, since the positive level can be grasped in multiple stages, the recovery process can be easily grasped.

<5. Modification example>
Although one embodiment of the present invention has been described above, the present invention is not limited to the above embodiment, and various modifications can be made without departing from the spirit of the present invention. For example, the following changes are possible. In addition, the gist of the following modifications can be combined as appropriate.

<5-1>
In the above embodiment, transfer learning may be performed in the learning process of the machine learning model 8. That is, in step S2, after the machine learning model 8 is trained using various images not related to autoimmune vesicular disease such as images of nature and animals, the fluorescent antibody image and the information of the diagnosis result are obtained. The included data set may be used to train the machine learning model 8.

<5-2>
The information of the diagnosis result output from the machine learning model 8 is not limited to the above-mentioned example. For example, as the information of the diagnosis result, only the information indicating which level may be output may be output without outputting the probability corresponding to each level. Further, in the above embodiment, the positive level is shown in multiple stages, but for example, information indicating only whether the positive level is positive or negative may be output.

1087 fluorescent antibody images obtained from 362 patients with suspected autoimmune vesicular disease, and the diagnostic results of autoimmune vesicular disease diagnosed by two skilled specialists for each fluorescent antibody image ( Correct answer data) and 2174 sets of data sets consisting of these were prepared. Then, using the data set, a machine learning model was generated by the same method as in the above embodiment. After that, 128 different sets of similar data sets are prepared, the fluorescent antibody images contained in the data set are input to the generated machine learning model, and the level corresponding to the highest probability among the probabilities output at this time. Was the diagnosis result. Then, the degree of agreement between the level of the diagnosis result by the machine learning model and the level of the diagnosis result (correct answer data) included in the same data set was evaluated. The results are shown in FIG. From FIG. 10, the level of agreement between the diagnosis result by the machine learning model and the correct answer data was 78/128 = 0.609 ..., And the agreement rate of 60% or more was obtained. In addition, the matching rate when it was considered that the ones whose levels were shifted by one step were matched was 125/128 = 0.976 ..., Which was close to 100%. As a result, the validity of the diagnosis based on the machine learning model according to the above embodiment was confirmed.

100 Diagnostic system 1 Computer (autoimmune vesicular disease diagnostic device, machine learning model generator)
2 Fluorescence microscope 14 Control unit 14a Acquisition unit 14b Learning unit 14c Diagnosis unit 14d Display control unit 6 Program 8 Machine learning model

Claims (11)

A diagnostic device for diagnosing autoimmune vesicular disease
An acquisition unit that acquires a fluorescent antibody image obtained by the fluorescent antibody method,
A diagnostic device comprising a diagnostic unit that inputs the fluorescent antibody image into a predetermined machine learning model and generates information on the diagnosis result of the autoimmune vesicular disease as an output of the machine learning model. The information on the diagnosis result includes information indicating whether the autoimmune vesicular disease is positive or negative.
The diagnostic device according to claim 1. The diagnostic result information includes information indicating the level of positive defined in multiple stages.
The diagnostic device according to claim 2. The information on the diagnosis result includes information indicating the probability corresponding to each of the positive levels defined in the multistage.
The diagnostic device according to claim 3. The information on the diagnosis result includes information indicating the type of the autoimmune vesicular disease.
The diagnostic device according to any one of claims 1 to 4. A diagnostic program for diagnosing autoimmune vesicular disease
Obtaining a fluorescent antibody image obtained by the fluorescent antibody method and
A diagnostic program that inputs a fluorescent antibody image into a predetermined machine learning model and causes a computer to generate information on a diagnosis result of the autoimmune vesicular disease as an output of the machine learning model. It is a diagnostic method for diagnosing autoimmune vesicular disease.
Obtaining a fluorescent antibody image obtained by the fluorescent antibody method and
A diagnostic method comprising inputting the fluorescent antibody image into a predetermined machine learning model and generating information on the diagnosis result of the autoimmune vesicular disease as an output of the machine learning model. A method for generating a machine learning model for diagnosing autoimmune vesicular disease.
To acquire a large number of data sets including fluorescent antibody images obtained by the fluorescent antibody method and information on the diagnosis results of the autoimmune vesicular disease diagnosed by an expert.
A generation method comprising training a machine learning model in which the fluorescent antibody image is input and the information of the diagnosis result is output, using the large number of data sets as teacher data. The large number of data sets include a data set containing information on the diagnosis result diagnosed by the first expert and information on the diagnosis result diagnosed by the second expert for the same fluorescent antibody image. Contains another dataset,
The generation method according to claim 8. A machine learning model generation program for diagnosing autoimmune vesicular disease.
Acquiring a large number of data sets including fluorescent antibody images obtained by the fluorescent antibody method and information on the diagnosis results of the autoimmune vesicular disease diagnosed by an expert, and
A generation program that causes a computer to train a machine learning model that uses the large number of data sets as teacher data, inputs the fluorescent antibody image, and outputs the information of the diagnosis result. A machine learning model generator for diagnosing autoimmune vesicular disease.
An acquisition unit that acquires a large number of data sets including fluorescent antibody images obtained by the fluorescent antibody method and information on the diagnosis results of the autoimmune vesicular disease diagnosed by an expert.
A generation device including a learning unit for learning a machine learning model in which a large number of data sets are used as teacher data, the fluorescent antibody image is input, and information on the diagnosis result is output.

Images