Disease risk evaluation device, disease risk evaluation system and disease risk evaluation method

Abstract

To provide a disease risk evaluation device, a disease risk evaluation system, and a disease risk evaluation method that determine whether a user has a high risk of contracting a specific disease and propose disease candidates.SOLUTION: A disease risk evaluation device according to the present invention includes: a disease level prediction unit that predicts a disease level using a first learned model with diagnostic data of a user as input, and outputs a disease level prediction score; a prediction reason analysis unit that analyzes a disease level prediction reason indicated by the disease level prediction score using the first learned model and the diagnostic data input to the first learned model, and outputs data on the disease level prediction reason; a symptom prediction unit that predicts symptoms using a second learned model with the data on the disease level prediction reason output from the prediction reason analysis unit as input and outputs the predicted symptoms; and a disease information output unit that outputs disease information on a disease corresponding to the symptoms on the basis of the symptoms output from the symptom prediction unit.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a disease risk assessment device, a disease risk assessment system, and a disease risk assessment method for determining whether or not there is a high risk of contracting a specific disease at a health stage. [Background technology]

[0002] In recent years, interest in preventive medicine has increased, and there is a demand for determining whether a person is at high risk of contracting a particular disease at a health stage where they are not yet affected by the disease. In addition, there are attempts to use AI machine learning models to determine the risk of contracting a particular disease from patient diagnostic data obtained from health checkups, etc.

[0003] However, when using an AI-based machine learning model to assess risk, the reasons for the results output by the AI-based machine learning model were a black box from the user's perspective, making it difficult for users to understand how to evaluate and use the results obtained.

[0004] For this reason, there is a need to clarify the reasons why results output by AI were obtained, in order to assist users of disease risk assessment systems that use AI-based machine learning models in making decisions when considering introducing the system or when interpreting and using the output results from the system.

[0005] Furthermore, for example, when a person is determined to be at high risk of contracting a particular disease, by clarifying the reason for the result output by the AI, it will be possible to infer the reason for the increased risk of contracting that particular disease, which is expected to be useful in preventing the disease.

[0006] Furthermore, because specialized knowledge is required to interpret the reasons for the results output by AI, it is necessary to present the information in a way that its intended users, medical professionals such as doctors, can use it to assess disease risk and as a reference for actual diagnoses. [Prior art documents] [Non-patent literature]

[0007] [Non-Patent Document 1] Xiao Liu et al., “Development of a Novel Retina-Based Diagnostic Score for Early Detection of Major Depressive Disorder: An Interdisciplinary View”, Frontiers in Psychiatry, May 2022, Volume 13, Article 897759 Summary of the Invention [Problem to be solved by the invention]

[0008] For example, Non-Patent Document 1 attempts to predict the progression of depressive disorders using data from OCTA (Optical Coherence Tomography Angiography). However, the reason for the obtained prediction result is not visible to the user.

[0009] Furthermore, even in cases where the AI ​​was able to output information on the reasons for the obtained prediction results, it was not possible to present the information in a way that would allow an appropriate interpretation of the output and enable it to be used as information to assess disease risk and to refer to actual diagnosis.

[0010] Therefore, according to the present invention, in a disease risk assessment system that determines whether or not there is a high risk of contracting a specific disease at a health stage, a disease risk assessment device, a disease risk assessment system, and a disease risk assessment method are provided that obtain data on the reasons for the prediction as to the basis for the prediction result output from an AI-based machine learning model, determine symptoms from the obtained data on the reasons for the prediction, and suggest candidate diseases. [Means for solving the problem]

[0011] In order to solve the above problems, the present invention provides a disease risk assessment device comprising: a disease level prediction unit that takes a user's diagnostic data as input, predicts a disease level using a first trained model, and outputs a disease level prediction score; a prediction reason analysis unit that uses the first trained model and the diagnostic data input to the first trained model to analyze a reason for disease level prediction indicated by the disease level prediction score, and outputs data on the reason for disease level prediction; a symptom prediction unit that takes the data on the reason for disease level prediction output from the prediction reason analysis unit as input, predicts symptoms using a second trained model, and outputs the predicted symptoms; and a disease information output unit that outputs disease information of a disease corresponding to the symptoms based on the symptoms output from the symptom prediction unit.

[0012] In addition, a disease risk assessment device according to another aspect of the present invention is characterized in that it comprises a disease level prediction unit that takes a user's diagnostic data as input, predicts a disease level using a first trained model, and outputs a disease level prediction score, a symptom prediction unit that takes the user's diagnostic data as input, predicts symptoms using a second trained model, and outputs the predicted symptoms, and a disease information output unit that outputs disease information of the disease corresponding to the symptoms based on the output symptoms.

[0013] In a disease risk assessment device according to one aspect of the present invention, the disease information output unit is characterized in that it uses the symptoms output from the symptom prediction unit as input, searches an external disease database, and outputs the disease name or disease code corresponding to the symptoms.

[0014] In a disease risk assessment device according to one embodiment of the present invention, the first trained model is characterized in that it is generated by inputting a user's diagnostic data as learning diagnostic data and inputting diagnostic data for each disease level related to a specific disease as learning score data and performing machine learning.

[0015] In a disease risk assessment device according to one embodiment of the present invention, the learning diagnostic data is characterized in that it is one or more of the user's fundus image, OCT (Optical Coherence Tomography) measurement values, BMI (Body Mass Index), gender, age, and health check parameters.

[0016] In a disease risk assessment device according to one embodiment of the present invention, the prediction reason analysis unit extracts values ​​related to the propagation coefficients of each layer of the neural network in the first trained model, and analyzes the main factors that led to the disease level prediction score.

[0017] In a disease risk assessment device according to one embodiment of the present invention, the prediction reason analysis unit extracts an integrated value of the propagation coefficients of each layer of the neural network in the first trained model, and quantifies the contribution of each pixel or parameter of the diagnostic data to the derivation of the disease level prediction score.

[0018] In the disease risk assessment device according to one aspect of the present invention, the quantified contribution degree is output as a map or a graph.

[0019] In a disease risk assessment device according to a certain embodiment of the present invention, the second trained model is characterized in that it is generated by inputting image data showing a specific symptom and an ID indicating the specific symptom as learning data and performing machine learning.

[0020] In a disease risk assessment device according to one embodiment of the present invention, the second trained model is characterized in that it is generated by inputting numerical data indicating a specific symptom and an ID indicating the specific symptom as training data and performing machine learning.

[0021] In a disease risk assessment device according to one embodiment of the present invention, the specific disease is depression, and the first trained model is generated by inputting a user's fundus image and OCT measurement values ​​as training health check data, and inputting a disease level related to depression as training score data, and performing machine learning.

[0022] In a disease risk assessment device according to one embodiment of the present invention, the disease level prediction unit is characterized in that it inputs at least a user's fundus image and OCT measurement values ​​as the user's diagnostic data, predicts the disease level using a first trained model, and outputs a disease level prediction score.

[0023] In a disease risk assessment device according to one embodiment of the present invention, the prediction reason analysis unit uses a first trained model and the user's fundus image and OCT measurement values ​​input into the first trained model as the diagnostic data to analyze the reason for disease level prediction indicated by the disease level prediction score, and outputs a contribution map corresponding to at least the user's fundus image as the reason for disease level prediction.

[0024] In a disease risk assessment device according to one embodiment of the present invention, the specific disease is diabetes, and the first trained model is generated by inputting at least basic bio-values ​​as training health check data and inputting a disease level related to diabetes as training score data and performing machine learning.

[0025] In a disease risk assessment device according to one embodiment of the present invention, the disease level prediction unit is characterized in that it inputs at least basic bio-values ​​as the user's diagnostic data, predicts the disease level using a first trained model, and outputs a disease level prediction score.

[0026] In a disease risk assessment device according to a certain embodiment of the present invention, the prediction reason analysis unit analyzes the reason for disease level prediction indicated by the disease level prediction score using a first trained model and basic bio numerical values ​​input to the first trained model as the diagnostic data, and outputs at least the relevant basic bio numerical values ​​as the reason for disease level prediction. The disease risk assessment device described in claim 13, characterized in that

[0027] In a disease risk assessment device according to a certain embodiment of the present invention, the specific disease is a disease related to HbA1C values, and the first trained model is generated by inputting at least a user's fundus image, OCT measurement values, BMI, age, and gender as training health check data, and inputting the HbA1C value as training score data, and performing machine learning.

[0028] In a disease risk assessment device according to one embodiment of the present invention, the disease level prediction unit is characterized in that it inputs at least the user's fundus image, OCT measurement values, BMI, age, and gender as the user's diagnostic data, predicts the disease level using a first trained model, and outputs the HbA1C value as a disease level prediction score.

[0029] In a disease risk assessment device according to one embodiment of the present invention, the prediction reason analysis unit analyzes the reason for disease level prediction indicated by the disease level prediction score using a first trained model and the user's fundus image, OCT measurement values, BMI, age, and gender inputted into the first trained model as diagnostic data, and outputs at least related parameters as the reason for disease level prediction.

[0030] In a disease risk assessment device according to one embodiment of the present invention, the specific disease is a disease related to heart disease, and the first trained model is generated by inputting at least one of a user's fundus image, electrocardiogram, OCT measurement value, BMI, age, and gender as training health check data and performing machine learning.

[0031] In a disease risk assessment device according to one embodiment of the present invention, the disease level prediction unit is characterized in that it inputs at least one of the user's fundus image, electrocardiogram, OCT measurement value, BMI, age, and gender as the user's diagnostic data, predicts the disease level using a first trained model, and outputs a disease level prediction score.

[0032] In a disease risk assessment device according to one embodiment of the present invention, the prediction reason analysis unit analyzes the disease level prediction reason indicated by the disease level prediction score using a first trained model and at least one of the user's fundus image, electrocardiogram, OCT measurement values, BMI, age, and gender inputted into the first trained model as diagnostic data, and outputs at least related parameters as the disease level prediction reason.

[0033] In a disease risk assessment device according to one embodiment of the present invention, the specific disease is diabetic retinopathy, and the first trained model is generated by inputting at least a fundus image as training health check data and a disease level related to diabetic retinopathy as training score data and performing machine learning.

[0034] In a disease risk assessment device according to one embodiment of the present invention, the disease level prediction unit is characterized in that it inputs at least a fundus image as diagnostic data for the user, predicts the disease level using a first learned model, and outputs a disease level prediction score.

[0035] In a disease risk assessment device according to one embodiment of the present invention, the prediction reason analysis unit uses a first trained model and a fundus image input to the first trained model as the diagnostic data to analyze the reason for disease level prediction indicated by the disease level prediction score, and outputs a contribution image indicating at least the contribution of each pixel of the fundus image as the reason for disease level prediction.

[0036] In a disease risk assessment device according to one embodiment of the present invention, the specific disease is arteriosclerosis, and the first trained model is generated by inputting at least a fundus image as training health check data and a disease level related to arteriosclerosis as training score data and performing machine learning.

[0037] In a disease risk assessment device according to one embodiment of the present invention, the disease level prediction unit is characterized in that it inputs at least a fundus image as diagnostic data for the user, predicts the disease level using a first learned model, and outputs a disease level prediction score.

[0038] In a disease risk assessment device according to one embodiment of the present invention, the prediction reason analysis unit uses a first trained model and a fundus image input to the first trained model as diagnostic data to analyze the reason for disease level prediction indicated by the disease level prediction score, and outputs a contribution image indicating at least the contribution of each pixel of the fundus image as the reason for disease level prediction.

[0039] The present invention also provides a disease risk assessment system comprising the above-mentioned disease risk assessment device, a user terminal, and a diagnostic data database for storing diagnostic data of the user input from the user terminal, wherein a disease level prediction unit acquires the diagnostic data of the user from the diagnostic data database, and a disease information output unit transmits disease information corresponding to the symptoms predicted by the disease level prediction unit to the user terminal.

[0040] The present invention also provides a disease risk assessment method in which a disease risk assessment device assesses disease risk, comprising the steps of: using a user's diagnostic data as input, predicting a disease level using a first trained model, and outputting a disease level prediction score; using the disease level prediction score, the first trained model, and the diagnostic data input to the first trained model, analyzing the reason for the disease level prediction indicated by the disease level prediction score, and outputting the reason for the disease level prediction; using the outputted data of the reason for the disease level prediction as input, predicting symptoms using a second trained model, and outputting the predicted symptoms; and outputting disease information of the disease corresponding to the symptoms based on the outputted symptoms.

[0041] In addition, a disease risk assessment method in which a disease risk assessment device according to another aspect of the present invention assesses disease risk is characterized in that the disease risk assessment device includes a step of taking a user's diagnostic data as input, predicting a disease level using a first trained model, and outputting a disease level prediction score, a step of taking the user's diagnostic data as input, predicting symptoms using a second trained model, and outputting the predicted symptoms, and a step of outputting disease information of the disease corresponding to the symptoms based on the symptoms output from the symptom prediction unit.

[0042] The present invention also provides a program for causing a computer to execute each step of the above-mentioned disease risk assessment method. Effect of the Invention

[0043] According to the present invention, in a disease risk assessment system that determines whether or not there is a high risk of contracting a specific disease at a health stage, it is possible to obtain data on the reasons for the prediction output from an AI-based machine learning model as to the reasons on which the prediction result was determined, and to determine symptoms from the obtained data on the reasons for the prediction and suggest candidate diseases.

[0044] According to another aspect of the disease risk assessment system of the present invention, it is possible to determine whether or not a user is at high risk of contracting a specific disease at a health stage, as well as to determine symptoms from the user's diagnostic data and suggest candidate diseases. Other objects, features and advantages of the present invention will become apparent from the following detailed description of the preferred embodiments of the present invention taken in conjunction with the accompanying drawings. [Brief description of the drawings]

[0045] [Figure 1] FIG. 1 is a schematic diagram showing an entire disease risk assessment system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram illustrating generation of a first trained model according to one embodiment of the present invention. [Diagram 3] FIG. 3 is a diagram illustrating a disease level prediction unit according to one embodiment of the present invention. [Figure 4] FIG. 4 is a diagram showing a prediction reason analysis unit according to one embodiment of the present invention. [Diagram 5] FIG. 5 is a diagram illustrating a symptom prediction unit according to an embodiment of the present invention. [Figure 6A] FIG. 6A is a diagram showing a disease information output unit according to one embodiment of the present invention. [Figure 6B] FIG. 6B is a diagram showing a disease information output unit according to one embodiment of the present invention. [Figure 6C] FIG. 6C is a diagram illustrating a disease information output unit according to one embodiment of the present invention. [Figure 7] FIG. 7 is a diagram showing a flow of a disease risk assessment process according to one embodiment of the present invention. [Figure 8] FIG. 8 is a schematic diagram showing an entire disease risk assessment system according to another embodiment of the present invention. [Figure 9] FIG. 9 is a diagram showing a flow of a disease risk assessment process according to another embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS EXAMPLES

[0046] FIG. 1 is a schematic diagram showing an entire disease risk assessment system 1 according to the present invention. The disease risk assessment system 1 comprises a disease risk assessment device 10, a user terminal 60, and a diagnostic data database 70 that stores the user's diagnostic data input from the user terminal 60. In the disease risk assessment device 10, a disease level prediction unit 20 acquires the user's diagnostic data from the diagnostic data database 70 and predicts the disease level, a prediction reason analysis unit 30 analyzes the reason for predicting the disease level, a symptom prediction unit 40 predicts symptoms, and a disease information output unit 50 transmits disease information corresponding to the symptoms predicted by the disease level prediction unit to the user terminal 60.

[0047] The disease risk assessment device 10 comprises a disease level prediction unit 20 that receives a user's diagnostic data as input, predicts a disease level using a first trained model M1, and outputs a disease level prediction score; a prediction reason analysis unit 30 that uses the first trained model M1 and the diagnostic data input to the first trained model to analyze the reason for disease level prediction indicated by the disease level prediction score and output data of the reason for disease level prediction; a symptom prediction unit 40 that receives input data of the reason for disease level prediction output from the prediction reason analysis unit 30, predicts symptoms using a second trained model M2, and outputs the predicted symptoms; and a disease information output unit 50 that outputs disease information of the disease corresponding to the symptoms based on the symptoms output from the symptom prediction unit 40.

[0048] The disease information output unit 50 receives the symptoms output from the symptom prediction unit 40, searches an external disease database, and outputs the disease name or disease code corresponding to the symptoms. The search key for searching the external disease database is a keyword such as a symptom name or an ID indicating a specific symptom. The ID indicating a specific symptom may be based on, for example, the ICD10 International Classification of Diseases or the DPC Disease Name Classification.

[0049] FIG. 2 is a diagram illustrating generation of a first trained model according to one embodiment of the present invention. The first trained model M1 is generated in advance by a machine learning model generation unit 90 outside the disease risk assessment device 10. The machine learning model generation unit 90 may be included in the disease risk assessment system 1 or may be provided outside the disease risk assessment system 1.

[0050] The first trained model M1 is generated by inputting a user's diagnostic data as training diagnostic data, inputting diagnostic data for each disease level relating to a specific disease as training score data, and performing machine learning.

[0051] The learning diagnostic data may be, for example, one or more of a user's fundus image, OCT (Optical Coherence Tomography) measurement value, BMI (Body Mass Index), gender, age, and health check parameters. Here, the "health check parameters" may include any arbitrary health check parameters that may be examination items in general health checks, medical checkups, or examinations related to diseases. The "diagnostic data" refers to the user's biological data obtained in a health check, medical checkup, or examinations related to diseases. The diagnostic data may include basic physical data such as the user's gender, age, height, weight, blood type, BMI, blood pressure, and any arbitrary health check parameters that may be examination items in general health checks, medical checkups, or examinations related to diseases. In addition, eye-related data such as a fundus image obtained by an ophthalmological examination may also be included in the diagnostic data. For example, the diagnostic data may include items measured by blood tests, such as white blood cell count, red blood cell count, platelet count, cholesterol level (LDL cholesterol level, HDL cholesterol level, total cholesterol level), vitamin B12 level, 25-hydroxyvitamin D2 level, segmented neutrophil percentage, red blood cell distribution width, red blood cell folate level, monocyte percentage, mean platelet volume, mean corpuscular volume, lymphocyte percentage, hemoglobin level, HbA1c level, epi-25-hydroxyvitamin D3 level, 25-hydroxyvitamin D3 level, basophil percentage, eosinophil percentage, etc. In addition, the diagnostic data may include items measured by urine tests, such as creatinine level and albumin level. In addition, the diagnostic data may include results of interviews regarding medical conditions, living conditions, sleep conditions, etc., i.e., answers to questions regarding health conditions, lifestyle habits, etc. The interview results may include, for example, answers of "yes" or "no" to questions such as "Did any of your relatives have diabetes?" and "Have you been told you have asthma?".

[0052] The learning score data is, for example, disease levels grouped into a plurality of stages, such as disease levels 1, 2, ... n. When there is an index that is generally used as a disease level for a specific disease, that index may be used as the disease level. For example, in the case of depression, the value of the Patient Health Questionnaire-9 (PHQ-9), which is an index generally used as a disease level, may be used as the disease level. For example, the value of the PHQ-9 is expressed as a number from 0 to 27.

[0053] FIG. 3 is a diagram illustrating a disease level prediction unit according to one embodiment of the present invention. The disease level prediction unit 20 receives the diagnostic data of the user as input, predicts the disease level using the first trained model M1, and outputs a disease level prediction score. The diagnostic data of the user is, for example, the user's fundus image, OCT measurement value, BMI, gender, age, and other health check parameters. The diagnostic data of the user is data obtained from an actual health check, a medical checkup, or an examination related to a disease. The prediction score is obtained as an output result of the disease level prediction unit 20, and is, for example, a grouping of disease levels into a plurality of stages such as disease levels 1, 2, ... n. If there is an index that is generally used as a disease level for a specific disease, the index may be used as the disease level. For example, in the case of depression, the value of PHQ-9 (Patient Health Questionnaire-9), which is an index generally used as a disease level, may be used as the disease level. For example, the value of PHQ-9 is expressed as a numerical value from 0 to 27.

[0054] FIG. 4 is a diagram showing a prediction reason analysis unit according to one embodiment of the present invention. The prediction reason analysis unit 30 uses the first trained model M1 and the diagnostic data input to the first trained model M1 to analyze the disease level prediction reason indicated by the disease level prediction score, and outputs the disease level prediction reason. Specifically, the prediction reason analysis unit 30 extracts values ​​related to the propagation coefficients of each layer of the neural network in the first trained model M1, and analyzes the main factors that led to the disease level prediction score.

[0055] The prediction reason analysis unit 30 extracts the integrated value of the propagation coefficients of each layer of the neural network in the first trained model M1, and quantifies the contribution of each pixel or parameter of the diagnostic data to the derivation of the disease level prediction score.

[0056] The prediction reason analysis unit 30 may output the quantified contribution as a map or a graph. The prediction reason analysis unit 30 may output, for example, a contribution image or a health checkup parameter showing the contribution of each pixel of the user's fundus image as the disease level prediction reason. The quantified contribution may be output as a map or a graph. For example, when predicting the disease level of depression, a so-called heat map showing the contribution of each pixel of the user's fundus image by a change in color may be created to show which part of the inputted fundus image of the user contributes to the prediction of the disease level. In addition, the part of the user's fundus image that contributes highly to the prediction of the disease level may be colored or surrounded by a circle or a rectangle. In addition, for example, when predicting the disease level of diabetes, if BMI and age contribute to the prediction of the disease level among the inputted health checkup parameters of the user, a graph showing the high contribution of BMI and age may be output. The graph may be in any form, such as a bar graph showing rankings or a radar chart.

[0057] FIG. 5 is a diagram illustrating a symptom prediction unit according to an embodiment of the present invention. The symptom prediction unit 40 receives as input the disease level prediction reason data output from the prediction reason analysis unit 30, predicts symptoms using the second trained model M2, and outputs the predicted symptoms. The symptom prediction unit 40 may receive as input at least a contribution image indicating the contribution of each pixel of the user's fundus image, predicts symptoms using the second trained model M2, and outputs the predicted symptoms. The symptom prediction unit 40 may receive as input a contribution image indicating the contribution of each pixel of the user's fundus image and health examination parameters, predicts symptoms using the second trained model M2, and outputs the predicted symptoms.

[0058] The second trained model M2 is generated by inputting image data showing a specific symptom and an ID indicating the specific symptom as training data and performing machine learning. The image data showing a specific symptom is, for example, a fundus image with bleeding from blood vessels in the fundus or a fundus image with damage to the nerve part of the fundus. The ID indicating the specific symptom is, for example, an ID indicating a symptom of bleeding from blood vessels in the fundus or an ID indicating a symptom of damage to the nerve part of the fundus. The ID indicating the specific symptom may be a disease name in the ICD10 International Classification of Diseases, an ICD10 code (ICD-10-Code), and a disease name exchange code. It may also be a disease name in the DPC disease classification showing a specific symptom or a six-digit number of DPC. For example, according to the "ICD10 International Classification of Diseases, 10th Edition (2013 Edition)", in the case of "diabetic cataract", the disease name in the ICD10 International Classification of Diseases is "diabetic cataract", the ICD10 code is "E143 H280", and the disease name interchange code is "M5VP". In addition, for example, in the case of "central chorioretinopathy", which is one of the localized inflammations of the retina and choroid, the disease name in the ICD10 International Classification of Diseases is "central chorioretinopathy", the ICD10 code is "H300", and the disease name interchange code is "SU0Q". The ID indicating a specific symptom is not limited to this, and may be any text data such as a keyword indicating the symptom name, a number, an alphanumeric character, or a symbol.

[0059] For example, a fundus image with bleeding from a blood vessel in the fundus and an ID indicating the symptom of bleeding from a blood vessel in the fundus may be input as learning data, and machine learning may be performed to generate the second trained model M2. In another example, a fundus image with damage to a nerve part of the fundus image and an ID indicating the symptom of damage to the nerve part of the fundus image may be input as learning data, and machine learning may be performed to generate the second trained model M2.

[0060] The second trained model M2 is generated by inputting numerical data indicating a specific symptom and an ID indicating a specific symptom as training data and performing machine learning.

[0061] The predicted symptom output by the symptom prediction unit 40 may be a symptom name or an ID indicating a specific symptom. The symptom name may be fundus hemorrhage, nerve damage, localized inflammation of the retina and central chorioretinopathy. For example, the ID indicating a specific symptom may be, for example, an ID indicating a symptom of bleeding from blood vessels in the fundus or an ID indicating a symptom of damage to the nerve part of the fundus. The ID indicating a specific symptom may be a disease name in the ICD10 International Classification of Diseases, an ICD10 code, and a disease name exchange code. It may also be a disease name in the DPC disease classification indicating a specific symptom or a six-digit number of the DPC. For example, according to the "ICD10 International Classification of Diseases, 10th Edition (2013 Edition)", in the case of "diabetic cataract", the disease name in the ICD10 International Classification of Diseases is "diabetic cataract", the ICD10 code is "E143 H280", and the disease name exchange code is "M5VP". For example, in the case of "central chorioretinopathy," which is one of the localized inflammations of the retina and choroid, the disease name in the International Classification of Diseases (CD10) is "central chorioretinopathy," the ICD10 code is "H300," and the disease name interchange code is "SU0Q." The ID indicating a specific symptom is not limited to this, and may be any text data such as a keyword, number, alphanumeric character, or symbol indicating the symptom name.

[0062] For example, the symptom prediction unit 40 may output the name of the symptom, such as "fundus bleeding," and an ID indicating the symptom.

[0063] 6A to 6C are diagrams showing a disease information output unit according to one embodiment of the present invention. The disease information output unit 50 outputs disease information of a disease corresponding to a symptom based on the symptom output from the symptom prediction unit 40. The disease information is, for example, a disease name or a disease ID. The disease name may be based on, for example, the disease name of the ICD10 International Classification of Diseases or the disease name of the DPC Classification of Diseases.

[0064] The disease information output unit 50 may extract disease names or disease IDs from an external disease data database 80 as disease information, associate the disease level predicted by the disease level prediction unit 20 with the extracted disease names or disease IDs, and store the associated information in advance in the external disease data database 80 or in a memory in the disease information output unit 50. The disease information output unit 50 may extract disease names or disease IDs from an external disease data database 80, and output information associating the disease level predicted by the disease level prediction unit 20 with the extracted disease names or disease IDs to the user terminal 60 as disease information.

[0065] Furthermore, the disease information output unit 50 may extract at least one or more symptoms corresponding to the disease name or disease ID from an external disease data database 80, associate the disease level predicted by the disease level prediction unit 20 with the extracted at least one or more symptoms, and store the association in advance in the external disease data database 80 or in a memory in the disease information output unit 50. The disease information output unit 50 may extract at least one or more symptoms corresponding to the disease name or disease ID from an external disease data database 80, and output information associating the disease level predicted by the disease level prediction unit 20 with the extracted at least one or more symptoms to the user terminal 60 as disease information.

[0066] For example, if the disease name is glaucoma or the disease ID is an ID indicating glaucoma, the disease information output unit 50 may extract symptoms of glaucoma from an external disease data database 80, and output the extracted symptoms of glaucoma, such as "enlarged optic disc cupping," "retinal nerve fiber layer defect," "optic disc hemorrhage," and "suspected glaucomatous disc changes," as disease information to the user terminal 60.

[0067] Similarly, for example, when the disease name is age-related macular degeneration or the disease ID is an ID indicating age-related macular degeneration, the disease information output unit 50 may extract age-related macular degeneration from an external disease data database 80, and output to the user terminal 60 as disease information the symptoms of age-related macular degeneration such as "precursor lesions of age-related macular degeneration (drusen)," "precursor lesions of age-related macular degeneration (abnormalities of the retinal pigment epithelium)," "age-related macular degeneration (exudative type)," "age-related macular degeneration (atrophic type)," etc.

[0068] Similarly, for example, if the disease name is diabetic retinopathy or the disease ID is an ID indicating diabetic retinopathy, the disease information output unit 50 may extract diabetic retinopathy from an external disease data database 80, and output to the user terminal 60 as disease information the symptoms of diabetic retinopathy such as "soft exudates," "hard exudates," "retinal hemorrhage (dots)," "retinal hemorrhage (stain-like)," and "dots," etc.

[0069] Similarly, for example, when the disease name is arteriosclerosis or the disease ID is an ID indicating arteriosclerosis, the disease information output unit 50 may extract arteriosclerosis from the external disease data database 80, and output to the user terminal 60 as disease information the "sclerotic changes" such as "enhanced arterial column reflex," "arteriovenous crossing phenomenon," "copper wire artery," "silver wire artery," and "white wire artery" extracted as symptoms of arteriosclerosis, or the "hypertensive changes" such as "diffuse narrowing," "diffuse stenosis," "focal narrowing," "arterial narrowing," "bleeding on the retinal surface," "white spots," and "papilledema" extracted as symptoms of arteriosclerosis.

[0070] In addition, the disease information output unit 50 may be configured to output to the user terminal 60 as disease information the result calculated using the second trained model M2 of the probability that a finding appearing in the fundus image to be judged corresponds to each of the extracted symptoms.

[0071] In addition, the disease information output unit 50 may be configured to output the pixels and / or parameters focused on by the first trained model M1 to the user terminal 60 as disease information when making a prediction using the first trained model M1.

[0072] Table 1 shows examples of disease types and symptoms that appear in fundus images for those diseases. The disease level prediction unit 20 uses the first trained model M1 to identify the disease name corresponding to the disease level prediction score predicted, and the symptom prediction unit 40 uses the second trained model M2 to calculate the probability that the findings appearing in the fundus image to be judged correspond to each of the extracted symptoms. The disease information output unit 50 may output a table such as Table 1 to the user terminal 60. The disease risk evaluation device 10 performs the process up to outputting such disease information to the user terminal 60, and it is the doctor or the like who performs the actual diagnosis by referring to the output disease information. According to the disease risk evaluation device 10 of the present invention, the prediction result by AI can be displayed in a form that is easy for doctors and the like to understand. This can be expected to have the effect of preventing overlooking findings appearing in the fundus image. [Table 1]

[0073] Furthermore, the disease information output unit 50 may extract at least one or more symptoms corresponding to the disease name or disease ID from an external disease data database 80, associate the disease level prediction reason predicted by the prediction reason analysis unit 30 with the extracted at least one or more symptoms, and store the association in the external disease data database 80 or in a memory in the disease information output unit 50. The disease information output unit 50 may extract at least one or more symptoms corresponding to the disease name or disease ID from an external disease data database 80, and output information associating the disease level prediction reason predicted by the prediction reason analysis unit 30 with the extracted at least one or more symptoms to the user terminal 60 as disease information.

[0074] In the example of FIG. 6A, the disease information output section 50 may output disease information including at least the disease name or disease ID corresponding to the disease level prediction reason predicted by the prediction reason analysis section 30.

[0075] As shown in the example of FIG. 6B, the disease information output section 50 may output disease information including at least the disease name or disease ID corresponding to the symptom predicted by the symptom prediction section 40.

[0076] As shown in the example of FIG. 6C, the disease information output section 50 may output disease information including at least the disease name or disease ID corresponding to the disease level prediction score predicted by the disease level prediction section 20.

[0077] 6A to 6C, the disease information output by the disease information output unit 50 to the user terminal 60 is not limited to that described above, and information related to the disease information extracted by the disease information output unit 50 may be obtained from an external database, search site, or the like and output to the user terminal 60. For example, at least one of papers, case images, ICD10 code, treatment method, area of ​​occurrence, or number of patients related to the disease corresponding to the disease name or disease ID extracted by the disease information output unit 50, or links to these information may be output to the user terminal 60.

[0078] The disease data database 80 may be constructed in an external system or external server of the disease risk assessment device 10. Furthermore, information inside the disease data database 80 may be searchable via a search site that uses an Internet (registered trademark) search engine. Furthermore, the disease information output unit 50 may be connected to an Internet search engine to search for information inside the disease data database 80 or information available from a Web site on the Internet, and output the information to the user terminal 60.

[0079] The specific disease is depression, and the first trained model M1 is generated by inputting the user's fundus image and OCT measurement values ​​as training health check data, and inputting the disease level related to depression as training score data, and performing machine learning.

[0080] The disease level prediction unit 20 receives at least the user's fundus image and OCT measurement values ​​as the user's diagnostic data, predicts the disease level using the first trained model, and outputs a disease level prediction score.

[0081] The prediction reason analysis unit 30 analyzes the reason for disease level prediction indicated by the disease level prediction score using the first trained model M1 and the user's fundus image and OCT measurement values ​​input into the first trained model as the diagnostic data, and outputs a contribution map corresponding to at least the user's fundus image as the reason for disease level prediction.

[0082] When the specific disease is diabetes, the first trained model M1 is generated by inputting at least basic bio values ​​as training health check data and inputting the disease level related to diabetes as training score data and performing machine learning.

[0083] The disease level prediction unit 20 receives at least basic bio-values ​​as the user's diagnostic data as input, predicts the disease level using the first trained model, and outputs a disease level prediction score.

[0084] The prediction reason analysis unit 30 uses the first trained model M1 and the basic bio-values ​​input to the first trained model M1 as diagnostic data to analyze the disease level prediction reason indicated by the disease level prediction score, and outputs at least the relevant basic bio-values ​​as the disease level prediction reason.

[0085] When the specific disease is a disease related to HbA1C values, the first trained model M1 is generated by inputting at least the user's fundus image, OCT measurement values, BMI, age, and gender as training health check data, and inputting the HbA1C value as training score data and performing machine learning.

[0086] The disease level prediction unit 20 inputs at least the user's fundus image, OCT measurement values, BMI, age, and gender as the user's diagnostic data, predicts the disease level using the first trained model M1, and outputs the HbA1C value as a disease level prediction score.

[0087] The prediction reason analysis unit 30 analyzes the reason for disease level prediction indicated by the disease level prediction score using the first trained model M1 and the user's fundus image, OCT measurement values, BMI, age, and gender input into the first trained model M1 as diagnostic data, and outputs at least related parameters as the reason for disease level prediction.

[0088] When the specific disease is a disease related to heart disease, the first trained model M1 is generated by inputting at least one of the user's fundus image, electrocardiogram, OCT measurement value, BMI, age, and gender as training health check data and performing machine learning.

[0089] The disease level prediction unit 20 inputs at least one of the user's fundus image, electrocardiogram, OCT measurement values, BMI, age, and gender as the user's diagnostic data, predicts the disease level using a first learned model, and outputs a disease level prediction score.

[0090] The prediction reason analysis unit 30 analyzes the reason for disease level prediction indicated by the disease level prediction score using the first trained model M1 and at least one of the user's fundus image, electrocardiogram, OCT measurement values, BMI, age, and gender input into the first trained model as diagnostic data, and outputs at least related parameters as the reason for disease level prediction.

[0091] When the specific disease is diabetic retinopathy, the first trained model M1 is generated by inputting at least a fundus image as training health check data and a disease level related to diabetic retinopathy as training score data and performing machine learning.

[0092] The disease level prediction unit 20 receives at least a fundus image as input as diagnostic data for the user, predicts the disease level using the first trained model M1, and outputs a disease level prediction score.

[0093] The prediction reason analysis unit 30 analyzes the reason for disease level prediction indicated by the disease level prediction score using the first trained model M1 and the fundus image input to the first trained model M1 as diagnostic data, and outputs a contribution image indicating at least the contribution of each pixel of the fundus image as the reason for disease level prediction.

[0094] When the specific disease is arteriosclerosis, the first trained model M1 is generated by inputting at least a fundus image as training health check data and a disease level related to arteriosclerosis as training score data and performing machine learning.

[0095] The disease level prediction unit 20 receives at least a fundus image as input as diagnostic data for the user, predicts the disease level using the first trained model M1, and outputs a disease level prediction score.

[0096] The prediction reason analysis unit 30 analyzes the reason for disease level prediction indicated by the disease level prediction score using the first trained model M1 and the fundus image input to the first trained model M1 as diagnostic data, and outputs a contribution image indicating at least the contribution of each pixel of the fundus image as the reason for disease level prediction.

[0097] The health service introduction method according to the present invention will now be described. FIG. 7 is a diagram showing a flow of a disease risk assessment process according to one embodiment of the present invention. The disease risk assessment method according to the present invention includes the steps of: using a user's diagnostic data as input, predicting a disease level using a first trained model M1, and outputting a disease level prediction score (step S701); using the disease level prediction score, the first trained model M1, and the diagnostic data input to the first trained model, analyzing the reason for the disease level prediction indicated by the disease level prediction score, and outputting the reason for the disease level prediction (step S702); using the output data of the reason for the disease level prediction as input, predicting symptoms using a second trained model, and outputting the predicted symptoms (step S703); and outputting disease information for the disease corresponding to the symptoms based on the output symptoms (step S704).

[0098] The present invention also provides a program for causing a computer to execute each step of the above-mentioned disease risk assessment method.

[0099] The present invention also provides a program for causing a computer to execute each step of the disease risk assessment method of the present invention. The program may be recorded in a computer-readable recording medium. The program may also be stored in a server, executed on the server, and / or provide its functions via a network. EXAMPLES

[0100] FIG. 8 is a schematic diagram showing an entire disease risk assessment system 1' according to another embodiment of the present invention. The disease risk assessment system 1' includes a disease risk assessment device 10', a user terminal 60, and a diagnostic data database 70 that stores diagnostic data of the user input from the user terminal 60. In the disease risk assessment device 10', the disease level prediction unit 20 acquires diagnostic data of the user from the diagnostic data database 70, predicts the disease level, the prediction reason analysis unit 30 analyzes the prediction reason of the disease level, the symptom prediction unit 40 predicts the symptoms, and the disease information output unit 50 transmits disease information corresponding to the predicted symptoms to the user terminal 60. The difference from the disease risk assessment device 10 of the first embodiment is that the prediction reason analysis unit 30 is not included. Hereinafter, unless otherwise specified, the disease risk assessment device 10' of the second embodiment has the same configuration as the above-described configuration of the disease risk assessment device 10 of the first embodiment.

[0101] The disease risk assessment device 10' comprises a disease level prediction unit 20 that takes a user's diagnostic data as input, predicts a disease level using a first trained model M1, and outputs a disease level prediction score, a symptom prediction unit 40 that takes the user's diagnostic data as input, predicts symptoms using a second trained model, and outputs the predicted symptoms, and a disease information output unit 50 that outputs disease information of the disease corresponding to the symptoms based on the symptoms output from the symptom prediction unit.

[0102] In the second embodiment, the symptom prediction unit 40 receives at least a fundus image of the user as an input, predicts symptoms using the second trained model M2, and outputs the predicted symptoms. In the second embodiment, the symptom prediction unit 40 may receive a fundus image of the user and medical examination parameters as input, predict symptoms using the second trained model M2, and output the predicted symptoms.

[0103] FIG. 9 is a diagram showing a flow of a disease risk assessment process according to another embodiment of the present invention. In addition, a disease risk assessment method in which the disease risk assessment device 10' assesses disease risk according to another aspect of the present invention includes a step (step S901) in which the disease risk assessment device 10' takes a user's diagnostic data as input, predicts a disease level using a first trained model M1, and outputs a disease level prediction score, a step (step S902) in which the disease risk assessment device 10' takes the user's diagnostic data as input, predicts symptoms using a second trained model M2, and outputs the predicted symptoms, and a step (step S903) in which the disease information of the disease corresponding to the symptoms based on the symptoms output from the symptom prediction unit is output.

[0104] According to the disease risk assessment system 1 of the present invention described above, in a disease risk assessment system that determines whether or not there is a high risk of contracting a specific disease at a health stage, data on the reasons for the prediction output from the AI-based machine learning model is obtained as to the reasons on which the prediction result was determined, and symptoms can be determined from the obtained data on the reasons for the prediction, and candidate diseases can be suggested.

[0105] In addition, according to the disease risk assessment system 1' of the present invention in Example 2, it is possible to determine whether there is a high risk of contracting a specific disease at a health stage, as well as to determine symptoms from the user's diagnostic data and suggest candidate diseases. Although the above description has been given with reference to the embodiment, it will be apparent to those skilled in the art that the present invention is not limited thereto, and that various changes and modifications can be made within the scope of the principles of the present invention and the appended claims. [Explanation of symbols]

[0106] 1. Disease risk assessment system 10. Disease risk assessment device 20 Disease Level Prediction Department 30 Prediction Reason Analysis Department 40 Symptom Prediction Section 50 Disease information output section 60 User terminals 70 Diagnostic Data Database 80 Disease Data Database 90 Machine learning model generation unit

Claims

1. A disease level prediction unit that uses a user's diagnostic data as input, predicts a disease level using a first trained model, and outputs a disease level prediction score; A prediction reason analysis unit that analyzes a disease level prediction reason indicated by the disease level prediction score using the first trained model and the diagnostic data input to the first trained model, and outputs data on the disease level prediction reason; A symptom prediction unit that predicts a symptom using a second trained model with the data of the disease level prediction reason output from the prediction reason analysis unit as an input, and outputs the predicted symptom; a disease information output unit that outputs disease information of a disease corresponding to a symptom based on the symptom output from the symptom prediction unit; A disease risk assessment device comprising:

2. The disease risk assessment device of claim 1, characterized in that the disease information output unit uses the symptoms output from the symptom prediction unit as input, searches an external disease database, and outputs the disease name or disease code corresponding to the symptoms.

3. The disease risk assessment device of claim 1, wherein the first trained model is generated by inputting a user's diagnostic data as training diagnostic data and inputting diagnostic data for each disease level related to a specific disease as training score data and performing machine learning.

4. 4. The disease risk assessment device according to claim 3, wherein the learning diagnostic data is one or more of a user's fundus image, an OCT (Optical Coherence Tomography) measurement value, a BMI (Body Mass Index), a gender, an age, and a health checkup parameter.

5. 2. The disease risk assessment device according to claim 1, wherein the prediction reason analysis unit extracts values ​​related to a propagation coefficient of each layer of the neural network in the first trained model and analyzes main factors for deriving the disease level prediction score.

6. The disease risk assessment device according to claim 1, wherein the prediction reason analysis unit extracts an integrated value of the propagation coefficients of each layer of the neural network in the first trained model, and quantifies the contribution of each pixel or parameter of the diagnostic data to the derivation of the disease level prediction score.

7. 7. The disease risk assessment device according to claim 6, wherein the quantified contribution degree is output as a map or a graph.

8. The disease risk assessment device of claim 1, wherein the second trained model is generated by inputting image data in which a specific symptom appears and an ID indicating the specific symptom as learning data and performing machine learning.

9. The disease risk assessment device of claim 1, wherein the second trained model is generated by inputting numerical data indicating a specific symptom and an ID indicating the specific symptom as training data and performing machine learning.

10. The specific disease is depression, The disease risk assessment device of claim 3, wherein the first trained model is generated by inputting a user's fundus image and OCT measurement values ​​as training health check data and inputting a disease level related to the depression as training score data and performing machine learning.

11. The disease risk assessment device described in claim 10, characterized in that the disease level prediction unit inputs at least a user's fundus image and OCT measurement values ​​as the user's diagnostic data, predicts a disease level using a first learned model, and outputs a disease level prediction score.

12. The disease risk assessment device of claim 10, wherein the prediction reason analysis unit analyzes the reason for disease level prediction indicated by the disease level prediction score using the first trained model and the user's fundus image and the OCT measurement value input to the first trained model as the diagnostic data, and outputs a contribution map corresponding to at least the user's fundus image as the reason for the disease level prediction.

13. the specific disease is diabetes, The disease risk assessment device of claim 3, wherein the first trained model is generated by inputting at least basic bio-values ​​as training health check data and inputting a disease level related to diabetes as training score data and performing machine learning.

14. The disease risk assessment device described in claim 13, characterized in that the disease level prediction unit inputs at least basic bio-values ​​as the user's diagnostic data, predicts the disease level using the first learned model, and outputs a disease level prediction score.

15. The disease risk assessment device according to claim 13, characterized in that the prediction reason analysis unit analyzes a disease level prediction reason indicated by the disease level prediction score using the first trained model and the basic bio-numerals input to the first trained model as the diagnostic data, and outputs at least the related basic bio-numerals as the disease level prediction reason.

16. The specific disease is a disease related to HbA1C value, The disease risk assessment device of claim 3, wherein the first trained model is generated by inputting at least a user's fundus image, OCT measurement value, BMI, age, and gender as training health check data, and inputting an HbA1C value as training score data and performing machine learning.

17. The disease risk assessment device described in claim 16, characterized in that the disease level prediction unit inputs at least the user's fundus image, OCT measurement value, BMI, age, and gender as the user's diagnostic data, predicts the disease level using a first learned model, and outputs the HbA1C value as a disease level prediction score.

18. The disease risk assessment device according to claim 16, wherein the prediction reason analysis unit analyzes a disease level prediction reason indicated by the disease level prediction score using the first trained model and the user's fundus image, OCT measurement value, BMI, age, and gender inputted as the diagnostic data to the first trained model, and outputs at least related parameters as the disease level prediction reason.

19. The specific disease is a heart disease, The disease risk assessment device of claim 3, wherein the first trained model is generated by inputting at least one of a user's fundus image, electrocardiogram, OCT measurement value, BMI, age, and gender as training health check data and performing machine learning.

20. The disease risk assessment device described in claim 19, characterized in that the disease level prediction unit inputs at least one of the user's fundus image, electrocardiogram, OCT measurement value, BMI, age, and gender as the user's diagnostic data, predicts the disease level using a first learned model, and outputs a disease level prediction score.

21. The disease risk assessment device according to claim 19, characterized in that the prediction reason analysis unit analyzes a disease level prediction reason indicated by the disease level prediction score using the first trained model and at least one of a fundus image, an electrocardiogram, an OCT measurement value, a BMI, an age, and a gender of the user inputted as the diagnostic data to the first trained model, and outputs at least a related parameter as the disease level prediction reason.

22. The disease risk assessment device of claim 1, wherein the symptom prediction unit inputs a contribution image indicating the contribution of each pixel of the user's fundus image, predicts symptoms using a second trained model M2, and outputs the predicted symptoms.

23. A disease level prediction unit that uses a user's diagnostic data as input, predicts a disease level using a first trained model, and outputs a disease level prediction score; A symptom prediction unit that uses the diagnostic data of the user as input, predicts a symptom using a second trained model, and outputs the predicted symptom; a disease information output unit that outputs disease information of a disease corresponding to a symptom based on the symptom output from the symptom prediction unit; A disease risk assessment device comprising:

24. The disease risk assessment device described in claim 23, characterized in that the disease information output unit uses the symptoms output from the symptom prediction unit as input, searches an external disease database, and outputs the disease name or disease code corresponding to the symptoms.

25. The disease risk assessment device described in claim 23, characterized in that the first learned model is generated by inputting a user's diagnostic data as learning diagnostic data and inputting diagnostic data for each disease level related to a specific disease as learning score data and performing machine learning.

26. 26. The disease risk assessment device according to claim 25, wherein the learning diagnostic data is one or more of a user's fundus image, an OCT (Optical Coherence Tomography) measurement value, a BMI (Body Mass Index), a gender, an age, and a health checkup parameter.

27. The disease risk assessment device described in claim 23, characterized in that the second trained model is generated by inputting image data in which a specific symptom is present and an ID indicating the specific symptom as training data and performing machine learning.

28. The disease risk assessment device described in claim 23, characterized in that the second trained model is generated by inputting numerical data indicating a specific symptom and an ID indicating the specific symptom as training data and performing machine learning.

29. The specific disease is depression, The disease risk assessment device described in claim 25, characterized in that the first trained model is generated by inputting a user's fundus image and OCT measurement values ​​as training health check data and inputting a disease level related to the depression as training score data and performing machine learning.

30. The disease risk assessment device described in claim 29, characterized in that the disease level prediction unit inputs at least a user's fundus image and OCT measurement values ​​as the user's diagnostic data, predicts a disease level using a first learned model, and outputs a disease level prediction score.

31. the specific disease is diabetes, The disease risk assessment device described in claim 25, characterized in that the first trained model is generated by inputting at least basic bio values ​​as training health check data and inputting a disease level related to diabetes as training score data and performing machine learning.

32. The disease risk assessment device described in claim 31, characterized in that the disease level prediction unit inputs at least basic bio-values ​​as the user's diagnostic data, predicts the disease level using the first learned model, and outputs a disease level prediction score.

33. The specific disease is a disease related to HbA1C value, The disease risk assessment device described in claim 25, characterized in that the first trained model is generated by inputting at least a user's fundus image, OCT measurement value, BMI, age, and gender as training health check data, and inputting an HbA1C value as training score data and performing machine learning.

34. The disease risk assessment device described in claim 33, characterized in that the disease level prediction unit inputs at least the user's fundus image, OCT measurement value, BMI, age, and gender as the user's diagnostic data, predicts the disease level using a first learned model, and outputs the HbA1C value as a disease level prediction score.

35. The specific disease is a heart disease, The disease risk assessment device of claim 25, wherein the first trained model is generated by inputting at least one of a user's fundus image, electrocardiogram, OCT measurement value, BMI, age, and gender as training health check data and performing machine learning.

36. The disease risk assessment device described in claim 35, characterized in that the disease level prediction unit inputs at least one of a user's fundus image, electrocardiogram, OCT measurement value, BMI, age, and gender as the user's diagnostic data, predicts the disease level using a first learned model, and outputs a disease level prediction score.

37. The disease risk assessment device according to claim 23 , wherein the symptom prediction unit inputs at least a fundus image of the user, predicts symptoms using a second trained model M2, and outputs the predicted symptoms.

38. The specific disease is diabetic retinopathy, The disease risk assessment device of claim 3, wherein the first trained model is generated by inputting at least a fundus image as training health check data and a disease level related to diabetic retinopathy as training score data and performing machine learning.

39. The disease risk assessment device described in claim 38, characterized in that the disease level prediction unit inputs at least a fundus image as diagnostic data for the user, predicts the disease level using the first learned model, and outputs a disease level prediction score.

40. The disease risk assessment device described in claim 38, characterized in that the prediction reason analysis unit analyzes the disease level prediction reason indicated by the disease level prediction score using the first trained model and the fundus image input to the first trained model as the diagnostic data, and outputs a contribution image indicating at least the contribution of each pixel of the fundus image as the disease level prediction reason.

41. the specific disease is arteriosclerosis, The disease risk assessment device of claim 3, wherein the first trained model is generated by inputting at least a fundus image as training health check data and a disease level related to arteriosclerosis as training score data and performing machine learning.

42. The disease risk assessment device described in claim 41, characterized in that the disease level prediction unit inputs at least a fundus image as diagnostic data for the user, predicts the disease level using the first learned model, and outputs a disease level prediction score.

43. The disease risk assessment device described in claim 41, characterized in that the prediction reason analysis unit analyzes the disease level prediction reason indicated by the disease level prediction score using the first trained model and the fundus image input to the first trained model as the diagnostic data, and outputs a contribution image indicating at least the contribution of each pixel of the fundus image as the disease level prediction reason.

44. The specific disease is diabetic retinopathy, The disease risk assessment device described in claim 25, characterized in that the first trained model is generated by inputting at least a fundus image as training health check data and a disease level related to diabetic retinopathy as training score data and performing machine learning.

45. The disease risk assessment device described in claim 44, characterized in that the disease level prediction unit inputs at least a fundus image as diagnostic data for the user, predicts the disease level using the first learned model, and outputs a disease level prediction score.

46. the specific disease is arteriosclerosis, The disease risk assessment device described in claim 25, characterized in that the first trained model is generated by inputting at least a fundus image as training health check data and a disease level related to arteriosclerosis as training score data and performing machine learning.

47. The disease risk assessment device described in claim 46, characterized in that the disease level prediction unit inputs at least a fundus image as diagnostic data for the user, predicts the disease level using the first learned model, and outputs a disease level prediction score.

48. A disease risk assessment device according to any one of claims 1 to 47; A user terminal; a diagnostic data database for storing diagnostic data of the user inputted from the user terminal; Equipped with The disease level prediction unit obtains diagnostic data of the user from the diagnostic data database, A disease risk assessment system, characterized in that the disease information output unit transmits disease information corresponding to the symptoms predicted by the disease level prediction unit to the user terminal.

49. A disease risk assessment method in which a disease risk assessment device assesses a disease risk, the disease risk assessment device comprising: A step of predicting a disease level using a first trained model based on the user's diagnostic data as input, and outputting a disease level prediction score; A step of analyzing a disease level prediction reason indicated by the disease level prediction score using the disease level prediction score, the first trained model, and the diagnostic data input to the first trained model, and outputting the disease level prediction reason; A step of predicting symptoms using the output data of the disease level prediction reason by a second trained model and outputting the predicted symptoms; outputting disease information of a disease corresponding to the symptom based on the outputted symptom; A method for assessing a disease risk, comprising:

50. A disease risk assessment method in which a disease risk assessment device assesses a disease risk, the disease risk assessment device comprising: A step of predicting a disease level using a first trained model based on the user's diagnostic data as input, and outputting a disease level prediction score; A step of predicting a symptom using a second trained model by using the diagnostic data of the user as an input, and outputting the predicted symptom; outputting disease information of a disease corresponding to the symptom based on the output symptom; A method for assessing a disease risk, comprising:

51. A program for causing a computer to execute each step of the method according to claim 49 or 50.

Images