Method for generating learning model, program, method for generating training data, information processing apparatus, and information processing method

Abstract

To provide a method of generating a learning model capable of appropriately presenting a plurality of possible diseases.SOLUTION: A learning model generation method is provided, involving making a computer execute steps of: acquiring training data associating a plurality of symptoms, a plurality of candidate diseases, and scores of respective diseases, and generating a learning model for outputting a plurality of diseases in candidate order upon receipt of an input of a plurality of symptoms based on the acquired training data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a method for generating a learning model, a program, a method for generating training data, an information processing apparatus, and an information processing method.

Background Art

[0002] Conventionally, a technique for automatically extracting and presenting possible diseases has been disclosed to assist a doctor's diagnosis.

[0003] Patent Document 1 describes an information processing apparatus including: inspection data acquisition means for acquiring one or more pieces of inspection data regarding a patient to be diagnosed; identification means for identifying, based on the acquired inspection data, candidates for disease names that the patient to be diagnosed may have, and subsequent inspections to be performed to determine the disease name of the patient to be diagnosed; and output means for outputting the identified disease name candidates and subsequent inspections.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The technique described in Patent Document 1 has a problem that it does not necessarily preferably present a plurality of possible diseases.

[0006] An object of the present disclosure is to provide a method for generating a learning model and the like that can preferably present a plurality of possible diseases.

Means for Solving the Problems

[0007] A method for generating a learning model according to one aspect of the present disclosure acquires training data in which a plurality of symptoms are associated with a plurality of candidate diseases and scores for each of the diseases, and based on the acquired training data, a computer executes a process of generating a learning model that outputs a plurality of diseases in candidate order when a plurality of symptoms are input.

[0008] A program according to one aspect of the present disclosure causes a computer to execute a process of inputting the acquired plurality of symptoms into a learning model trained to output a plurality of diseases in candidate order when the plurality of symptoms are input, and outputting the plurality of diseases in candidate order.

[0009] A method for generating training data according to one aspect of the present disclosure receives a plurality of symptoms, a plurality of candidate diseases, and scores for each of the diseases, and associates and stores the received plurality of symptoms with the plurality of diseases and the scores for each of the diseases.

Advantages of the Invention

[0010] According to the present disclosure, a plurality of possible diseases can be suitably presented.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Mode for Carrying Out the Invention

[0012] The present disclosure will be specifically described with reference to the drawings showing its embodiments.

[0013] (First Embodiment) FIG. 1 is a block diagram showing a configuration example of a diagnostic support system 100. The diagnostic support system 100 includes an information processing device 1 and a plurality of information terminal devices 2. Each of the information processing device 1 and the information terminal device 2 is communicatively connected to a network N such as the Internet, and can transmit and receive data via the network N.

[0014] The information processing device 1 is, for example, a server computer, a personal computer, etc., and performs various information processing and information transmission and reception. The information processing device 1 outputs diagnostic support information from the patient's symptoms in response to an inquiry from the information terminal device 2. The diagnostic support information includes ranked information of a plurality of possible (candidate) diseases. The information terminal device 2 is an information terminal device used by experts such as doctors, and is, for example, a personal computer, a smartphone, a tablet terminal, etc. The information terminal device 2 can transmit and receive data to and from the information processing device 1 through communication.

[0015] The information processing device 1 uses the learning model described later to generate diagnostic support information for the symptoms received from the information terminal device 2 and outputs it to the information terminal device 2. Experts such as doctors use the information terminal device 2 to check the ranked list of a plurality of diseases. Here, the symptoms include, for example, the patient's subjective symptoms, objective symptoms, physical findings, clinical test results, and imaging test results, etc.

[0016] When making a diagnosis of a patient, the doctor narrows down a plurality of candidate diseases (differential diagnosis diseases) by comprehensively considering the patient's symptoms based on his or her own experience and knowledge, and proceeds to determine the definitive disease. In order to make an appropriate diagnosis of the definitive disease, it is important to widely extract possible differential diagnosis diseases. By using this diagnostic support system 100 as a disease recall tool, the doctor can grasp diseases that he or she could not recall by himself or herself, which can lead to the prevention of overlooking and misdiagnosis.

[0017] As shown in FIG. 1, the information processing device 1 includes a control unit 11, a storage unit 12, and a communication unit 13. Note that the information processing device 1 may be a multi-computer composed of a plurality of computers, or may be a virtual machine virtually constructed by software.

[0018] The control unit 11 includes a processor using one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), etc. The control unit 11 controls each component using a built-in memory such as a ROM (Read Only Memory) or RAM (Random Access Memory), a clock, a counter, etc., and executes processing.

[0019] The storage unit 12 includes a non-volatile memory such as a hard disk, a flash memory, an SSD (Solid State Drive), etc. The storage unit 12 may be an external storage device connected to the information processing apparatus 1. The storage unit 12 stores programs and data referred to by the control unit 11. The programs stored in the storage unit 12 include a program 12P for causing a computer to execute processing related to generation of the learning model 121, estimation of diseases, etc. The data stored in the storage unit 12 includes the learning model 121, a master DB (Data Base) 122, and a training data DB 123. The learning model 121 is a machine learning model that has learned training data. The learning model 121 is assumed to be used as a program module constituting artificial intelligence software.

[0020] The programs stored in the storage unit 12 may be in a state of being recorded in a computer-readable manner on a recording medium. The storage unit 12 stores a program read from the recording medium 1A by a reading device (not shown). Also, it may be a program downloaded from an external computer (not shown) connected to a communication network (not shown) and stored in the storage unit 12.

[0021] The communication unit 13 includes a communication module for communicating with an external device via the network N. The control unit 11 transmits and receives data to and from the information terminal device 2 via the communication unit 13.

[0022] The configuration of the information processing apparatus 1 is not limited to the above example. For example, it may include an operation unit for receiving user operations, a display unit for displaying various information, and the like.

[0023] The information terminal device 2 includes a control unit 21, a storage unit 22, a communication unit 23, a display unit 24, and an operation unit 25.

[0024] The control unit 21 includes a processor using one or more CPUs, GPUs, etc. The control unit 21 uses a memory such as a built-in ROM or RAM, a clock, a counter, etc., to control each component and execute processing.

[0025] The storage unit 22 includes a non-volatile memory such as a hard disk, a flash memory, an SSD, etc. The storage unit 22 stores programs and data referred to by the control unit 21. The programs stored in the storage unit 22 include a program 22P for causing a computer to execute processing related to the acquisition of diagnostic support information. The programs stored in the storage unit 22 may be in a form recorded in a computer-readable manner on a recording medium. The storage unit 22 stores the program read from the recording medium 2A by a reading device (not shown). Also, it may be a program downloaded from an external computer (not shown) connected to a communication network (not shown) and stored in the storage unit 22.

[0026] The communication unit 23 is a communication module for performing communication-related processing. The control unit 21 transmits and receives data to and from the information processing apparatus 1 via the communication unit 23.

[0027] The display unit 24 includes a display device such as a liquid crystal display or an organic EL (electroluminescence) display. The display unit 24 displays various information according to instructions from the control unit 21. The operation unit 25 is an interface that receives user operations. The operation unit 25 includes, for example, a keyboard, a touch panel device built into the display, a speaker, and a microphone. The operation unit 25 receives an operation input from the user and sends a control signal corresponding to the operation content to the control unit 21.

[0028] FIG. 2 is a diagram showing an example of the content of information stored in the master DB 122. The master DB 122 includes a symptom information table and a disease information table. The symptom information table includes a symptom ID column and a symptom column for identifying symptoms. In the symptom column, information regarding the patient's subjective symptoms, objective symptoms, physical findings, clinical examination results, image examination results, etc. is stored. The information regarding symptoms may include information indicating symptoms such as information regarding symptoms such as fever and dry cough, information regarding clinical examination results such as influenza virus kit: -, information regarding image examination results such as chest CT examination: ground glass opacity: -, etc.

[0029] The disease information table includes a disease ID column, a disease column, a related information column, a related symptom column, etc. for identifying diseases. In the disease column, information regarding each disease is stored. In the example of FIG. 2, in the disease column, the disease names of each disease are stored. The related information column includes information related to each disease. The information related to each disease may include, for example, examination information, medication information, etc. The related symptom column includes information regarding symptoms highly related to each disease. The symptoms highly related mean symptoms that are likely to appear in the case of the corresponding disease. The information processing device 1 collects the above-mentioned each data from an external server or the like in advance and stores it in the master DB 122.

[0030] FIG. 3 and FIG. 4 are diagrams showing example contents of information stored in the training data DB 123. The information processing apparatus 1 collects training data for training (learning) the learning model 121 and stores it in the training data DB 123. The training data DB 123 stores training data including one or more symptoms among the symptom information stored in the master DB 122, a plurality of diseases among the disease information, and scores for each disease. The training data DB 123 includes, as management items, for example, case ID, sub-case ID, symptoms, diseases, and scores.

[0031] In the case ID column and the sub-case ID column, identification information for identifying a case, such as a case number and a sub-case number, is stored. A case is composed of symptoms and diseases. In the symptom column, information regarding one or more symptoms included in the same case is stored. The information regarding the symptoms is the same as the symptom information table.

[0032] In the disease column, information regarding a plurality of diseases (candidate diseases) for the same case is stored. The information regarding the diseases is the same as the disease information table. In the score column, the score for each disease is stored. The score is represented by, for example, a value from 1 to 100, and the higher the numerical value associated with the disease, the higher the possibility that the disease is the case. α shown in FIGS. 3 and 4 is a predetermined coefficient assigned to the confirmed disease. α may be set to an appropriate value, such as 10, 20, etc. The predetermined coefficient α is automatically added to the confirmed diseases among the candidate diseases. α may be a ratio coefficient for multiplication with respect to the original score. The score is a determining factor when ranking a plurality of diseases by the learning model 121 described later.

[0033] The training data is classified into a plurality of types of cases managed by the case ID, and further subdivided by the sub-case ID. One set of training data is composed of one symptom group including one or more symptoms, one disease group including a plurality of diseases, and scores for each disease.

[0034] Training data is generated for a plurality of disease groups for each case. As shown in FIGS. 3 and 4, in each training data having the same case ID, the diseases included in each disease group are the same, and the symptom groups associated with each disease group differ in the types and numbers of symptoms included in the symptom group. For each disease in each disease group, the score is set to a different value according to the types and numbers of symptoms included in the corresponding symptom group, even for the same disease.

[0035] In the example shown in FIG. 3, the training data with case ID 1 and child case ID 1 corresponds to the first disease group training data composed of the first symptom group including three symptoms such as fever, the first disease group including four diseases such as seasonal influenza, and its score. The training data with child case ID 2 corresponds to the second disease group training data composed of the second symptom group including six symptoms such as fever and the second disease group including the same four diseases as the first disease group such as seasonal influenza and its score. Similarly, the third disease group training data, …, the Kth disease group training data, are stored hereinafter.

[0036] As shown in FIG. 4, similarly for cases with case ID 2 and later, a plurality of training data associating the first to Kth symptom groups with the first to Kth disease groups and their scores are stored. As an example, the training data with case ID 2 and child case ID 1 is composed of a symptom group including one symptom such as acute abdominal pain and a disease group including nine diseases such as appendicitis and its score.

[0037] In this way, the training data DB123 stores, for each case, a plurality of training data from the first disease group training data to the K-th (where K is an integer of 2 or more) disease group training data. In the learning of the learning model 121 described later, for example, when K = 6, the information processing device 1 performs learning for one case using each of the training data from the first disease group training data to the sixth disease group training data. Similarly, for other cases, learning is performed using each of the first disease group training data to the K-th disease group training data. That is, the information processing device 1 learns the learning model 121 using the first disease group training data to the K-th disease group training data for each of the plurality of cases. Note that FIGS. 3 and 4 are examples, and the stored content of the training data DB123 is not limited.

[0038] FIG. 5 is an explanatory diagram showing an overview of the learning model 121. The learning model 121 is a machine learning model that takes a plurality of symptoms as inputs and outputs a plurality of diseases in the order of candidates (probabilities) for each disease.

[0039] The information processing device 1 performs machine learning to learn predetermined training data and pre-generates the learning model 121. Then, the information processing device 1 inputs the symptoms acquired from the information terminal device 2 into the learning model 121 and outputs a plurality of diseases in the order of candidates.

[0040] The learning model 121 is a neural network model generated by deep learning and is constructed, for example, by learning to rank (LTR). Learning to rank is a method of learning the order of output target data for input data. The learning model 121 outputs a list of data in descending order of probability (estimated score) for the input data. The learning model 121 has, for example, an input layer where symptoms are input, an intermediate layer that extracts feature amounts of the symptoms using various parameters, and an output layer that outputs information indicating the probability for each disease.

[0041] The input information input to the input layer of the learning model 121 is a plurality of symptoms. The input layer of the learning model 121 may include a plurality of nodes that receive the input of each symptom stored in, for example, the master DB 122. The output layer of the learning model 121 has a plurality of nodes corresponding to each of the plurality of diseases, and outputs information indicating the probability of each associated disease. The output layer of the learning model 121 may include a plurality of nodes corresponding to each disease stored in, for example, the master DB 122. The learning model 121 outputs a list of a plurality of diseases arranged in the order of probability for each disease based on the probability of each output disease. In this way, the learning model 121 ranks a plurality of diseases in descending order of probability for the input symptom group, and outputs a list of the plurality of diseases arranged in the order of probability.

[0042] In the above, an example where the learning model 121 is a ranking learning model has been described, but the configuration of the learning model 121 is not limited. The learning model 121 only needs to be able to estimate a plurality of candidate diseases and their candidate order for a plurality of symptoms. Note that the learning model 121 may take one symptom as an input.

[0043] In the learning phase, which is a stage prior to the operation phase of performing diagnostic support, the information processing apparatus 1 generates the learning model 121 using the above-described training data, and stores the generated learning model 121. Then, in the operation phase, diagnostic support information including a plurality of candidate diseases is generated using the stored learning model 121.

[0044] FIG. 6 is a flowchart showing an example of the generation processing procedure of the training data. The following processing is executed by the control unit 11 according to the program 12P stored in the storage unit 12 of the information processing apparatus 1.

[0045] The control unit 11 of the information processing apparatus 1 acquires case data including a plurality of symptoms, a plurality of candidate diseases, and scores for each disease (step S11). In step S11, the control unit 11 may acquire the case data, for example, by receiving input of information by an expert such as a doctor operating an operation unit or the like using a screen for registering case data, or may acquire the case data by communication through an external device such as the information terminal device 2. The plurality of candidate diseases include not only confirmed diseases but also differential diagnosis diseases. The score for each disease may be a score determined by an expert or the like. The expert or the like associates a plurality of candidate diseases with a plurality of symptoms based on past diagnosis histories and the like, and further assigns a score according to the possibility of each disease. The score for each disease changes according to the types and numbers of symptoms in the case data and the types and numbers of diseases.

[0046] The control unit 11 generates training data, which is a data set in which a plurality of candidate diseases and correct scores for each of the diseases are labeled for the acquired plurality of symptoms (step S12). The control unit 11 stores the acquired symptoms and diseases in the master DB 122, stores the generated training data in the training data DB 123 (step S13), and ends a series of processes. The control unit 11 collects a large amount of case data, generates a plurality of information groups based on the collected case data, and accumulates the generated information groups in the DB 122 as training data.

[0047] The case data may be generated by text mining or the like from, for example, past papers, professional books, case reports, medical record data, and the like. The case data is preferably widely extracted to include information on differential diagnosis diseases included in the text in addition to information on confirmed diseases extracted from the abstracts of papers or the like.

[0048] FIG. 7 is a flowchart showing an example of the generation processing procedure of the learning model 121. The following processing is executed by the control unit 11 according to a program 12P stored in the storage unit 12 of the information processing apparatus 1, for example, after the processing of FIG. 6 is completed.

[0049] The control unit 11 of the information processing apparatus 1 acquires a set of training data extracted from the information group based on the information stored in the training data DB 123 (step S21). The control unit 11 uses the acquired training data to generate a learning model 121 that outputs information indicating a plurality of diseases in candidate order for the corresponding symptoms when a plurality of symptoms are input (step S22).

[0050] The control unit 11 may perform learning using, for example, a known listwise method. The control unit 11 inputs a plurality of symptoms included in the training data as input data into the learning model 121 and acquires a list of a plurality of diseases in candidate order output from the learning model 121. The control unit 11 adjusts various parameters using, for example, the gradient descent method so as to optimize the error between the output disease list and the disease list of the correct value. As a loss function indicating the error, for example, Approximate NDCG obtained by converting NDCG (Normalized Discounted Cumulative Gain), which is an example of an evaluation function, may be used. The listwise method is a method of directly optimizing the output ranking itself with an evaluation index based on NDCG. The control unit 11 learns using the k-th disease group training data from the first disease group training data. The control unit 11 completes the learning when the loss function satisfies a predetermined criterion.

[0051] The control unit 11 stores, in the storage unit 12, definition information regarding the learned learning model 121 as the learned learning model 121 (step S23) and ends a series of processes. Through the above-described processes, the control unit 11 can construct a learning model 121 that is learned to appropriately output a plurality of diseases in candidate order for a plurality of symptoms.

[0052] In the above, an example in which the information processing apparatus 1 executes processes related to the generation of training data and the generation of the learning model 121 has been described, but the processing entity is not limited. For example, the learning model 121 may be generated by the information terminal device 2 or may be generated by an external computer (not shown) communicably connected to the information processing apparatus 1.

[0053] Using the learning model 121 generated as described above, the diagnostic support system 100 executes diagnostic support. Hereinafter, the processing procedure executed by the diagnostic support system 100 in the operation phase will be described.

[0054] FIG. 8 is a flowchart showing an example of the processing procedure executed by the diagnostic support system 100. The following processing is executed by the control unit 11 according to the program 12P stored in the storage unit 12 of the information processing apparatus 1 and is executed by the control unit 21 according to the program 22P stored in the storage unit 22 of the information terminal apparatus 2.

[0055] The control unit 11 of the information processing apparatus 1 generates a screen for receiving a selection of symptoms and transmits the generated screen for receiving a selection of symptoms to the information terminal apparatus 2 (step S31).

[0056] The control unit 21 of the information terminal apparatus 2 receives a screen for receiving a selection of symptoms (step S32), displays the received screen for receiving a selection of symptoms on the display unit 24 (step S33). The control unit 21 receives a selection of a plurality of symptoms by the doctor operating the operation unit 25 (step S34). The control unit 21 transmits the received symptoms to the information processing apparatus 1 (step S35).

[0057] The control unit 11 of the information processing apparatus 1 receives a plurality of symptoms (step S36). The control unit 11 inputs the acquired plurality of symptoms as input data to the learning model 121 (step S37). The control unit 11 acquires a list of a plurality of diseases arranged in the order of candidates output from the learning model 121 (step S38).

[0058] The control unit 11 generates a screen including the acquired list of diseases (step S39) and transmits the generated screen including the list of diseases to the information terminal apparatus 2 (step S40).

[0059] The control unit 21 of the information terminal device 2 receives a screen including a list of diseases (step S41). The control unit 21 displays the received screen including the list of diseases on the display unit 24 (step S42) and ends a series of processes. The diagnostic support system 100 may execute a loop process of returning the process to step S34, receiving new symptoms, and repeatedly extracting the received new symptoms by the learning model 121.

[0060] In the above-described process, the processing entity of each process is not limited. For example, part or all of the processes executed by the control unit 11 of the information processing device 1 may be executed by the control unit 21 of the information terminal device 2. The control unit 21 of the information terminal device 2 may store the learning model 121 acquired from the information processing device 1 in the storage unit 22 and execute estimation processing based on the learning model 121.

[0061] FIG. 9 and FIG. 10 are schematic diagrams showing an example of a screen 240 displayed on the display unit 24. In step S33, the control unit 21 of the information terminal device 2 displays a screen 240 including reception fields 241a and 241b for receiving selection of symptoms as shown in FIG. 9, and receives selection of symptoms from a doctor using the screen 240. The reception fields 241a and 241b include a plurality of pre-registered symptoms and check boxes associated with each symptom. The control unit 21 receives selection of symptoms by receiving activation of the check box of the symptom to be selected via the operation unit 25. The reception field 241a includes all the symptoms recorded in the master DB 122. The reception field 241b includes the selected symptoms among all the symptoms.

[0062] When the control unit 21 receives activation of the check box of the symptom to be selected in the reception field 241a via the operation unit 25, the control unit 21 causes the check box of the selected symptom to be displayed in the reception field 241b in an activated state. The control unit 21 transmits the symptom selection result to the information processing device 1.

[0063] When the information processing apparatus 1 receives the selection result of symptoms, as shown in FIG. 10, it generates a screen 240 including a display column 242 that displays a plurality of candidate diseases, and outputs the screen 240 to the information terminal apparatus 2. The display column 242 displays the names of a plurality of candidate diseases in the order of candidates. For each disease name, a probability indicating the likelihood of being a candidate is displayed in association with the candidate rank. The information processing apparatus 1 acquires a list of a plurality of diseases arranged in the candidate order (probability order) output from the learning model 121 for the selected symptoms, and displays the acquired plurality of diseases in the display column 242 in the order of probability. In this case, the information processing apparatus 1 displays the probability in the learning model 121 in association with each disease. Note that the information processing apparatus 1 may extract a predetermined number (for example, 50) of diseases in descending order of probability from the diseases output from the learning model 121 and display them in the display column 242.

[0064] The selection result of the symptoms is held until a predetermined end operation is performed, and is displayed in, for example, the reception column 241b. When the doctor repeatedly selects new symptoms, the doctor can perform corrected input of symptoms using the reception column 241b. Specifically, the information processing apparatus 1 stores the acquisition history of symptoms in the storage unit 12, and outputs a screen 240 including the reception column 241b that displays the acquisition history to the information terminal apparatus 2. In the reception column 241b, the information processing apparatus 1 displays the symptoms in a state where the selection of the check box for the symptoms acquired in the past is canceled. The information processing apparatus 1 receives a correction to the acquisition history using the reception column 241b via the information terminal apparatus 2, and acquires a selection of new symptoms. The doctor can efficiently perform the selection operation by correcting the check presence / absence for each symptom displayed in the reception column 241b and adding new symptoms as necessary.

[0065] According to the present embodiment, using the learning model 121, a plurality of disease candidates corresponding to symptoms can be accurately estimated together with their candidate order, and can be efficiently and effectively presented in a ranking format. The learning model 121 can output a list of a plurality of diseases suitably for various combinations of symptoms by learning the order of a plurality of diseases for the symptoms.

[0066] The learning model 121 is trained using training data in which a disease group including a plurality of diseases is labeled for a symptom group. The disease group of the training data is generated based on widely extracted information including not only definitive diseases but also differential diseases. The learning model 121 can effectively reflect the information of differential diseases as compared with the case of learning using training data in which only one definitive disease is labeled for a symptom. Also, the learning model 121 is efficiently trained using training data classified for each of a plurality of disease groups. Even when the input data includes symptoms with a low appearance rate, the learning model 121 can display highly relevant diseases at the top, so that even when the correlation between the input symptoms and the symptoms of the training data is low, diseases with high scores can be accurately extracted.

[0067] A plurality of diseases that are the estimation results of the learning model 121 are displayed including the accuracy of the learning model in the order of candidates. A doctor or the like can easily grasp the estimation result of the learning model 121 and the quantified basis.

[0068] (Second Embodiment) In the second embodiment, a configuration for outputting additional information corresponding to a disease will be described. In the following embodiments, the differences from the first embodiment will be mainly described, and the same reference numerals will be given to the configurations common to the first embodiment, and the detailed description thereof will be omitted.

[0069] The information processing apparatus 1 of the second embodiment presents further additional information for a plurality of diseases in the order of candidates output from the learning model 121.

[0070] FIG. 11 is a flowchart showing an example of a processing procedure executed by the diagnostic support system 100 of the second embodiment.

[0071] The control unit 21 of the information terminal device 2 accepts the selection of any one of a plurality of diseases in the order of candidates output from the learning model 121 (step S51). In step S51, the control unit 21 may accept the selection of any one of the diseases displayed in the display column 242 by the doctor operating the operation unit 25 in a state where the screen 240 shown in FIG. 10 is displayed, for example. The control unit 21 transmits the accepted disease to the information processing device 1 (step S52).

[0072] The control unit 11 of the information processing device 1 receives the disease (step S53). The control unit 11 specifies additional information corresponding to the received disease based on the information stored in the master DB 122 (step S54). The additional information may include, for example, related information related to the disease or related symptoms corresponding to the disease. The control unit 11 generates a screen including the specified additional information and transmits the generated screen including the additional information to the information terminal device 2 (step S55).

[0073] The control unit 21 of the information terminal device 2 receives the screen including the additional information (step S56). The control unit 21 displays the screen including the received additional information on the display unit 24 (step S57) and ends the series of processes.

[0074] FIG. 12 is a schematic diagram showing a first example of the screen 240 in the second embodiment. FIG. 12 shows an example of a screen 240 including a related information column 243 that displays related information related to the disease as additional information. The related information column 243 displays related information related to the candidate disease, such as information on the examination content and medication content corresponding to the candidate disease. The control unit 11 of the information processing device 1 specifies the related information corresponding to the received disease based on the information stored in the master DB 122 and causes the related information column 243 including the specified related information to be displayed near the selected disease.

[0075] FIG. 13 is a schematic diagram showing a second example of the screen 240 in the second embodiment. FIG. 13 shows an example of the screen 240 including a related symptom column 244 that displays related symptoms corresponding to a disease as additional information. The related symptom column 244 displays, for example, symptoms highly relevant to a candidate disease. The control unit 11 of the information processing apparatus 1 specifies related symptoms corresponding to the received disease based on the information stored in the master DB 122, and causes the related symptom column 244 including the specified related symptoms to be displayed near the selected disease.

[0076] The related symptom column 244 may be configured to be able to receive the presence or absence of the related symptom in association with the related symptom. The control unit 11 of the information processing apparatus 1 uses the related symptom column 244 to receive the presence or absence of the related symptom specified by a doctor or the like. When receiving information that the specified related symptom is present, the control unit 11 acquires a series of symptoms with the related symptom added as new input information for the learning model 121. The control unit 11 inputs a series of symptoms with the related symptom added to the learning model 121, and acquires a list of a plurality of diseases in the order of candidates for the series of symptoms with the related symptom added. Thereby, the presence or absence of the related symptom can be reflected in the inference of the disease.

[0077] In the above, an example of specifying related symptoms using the master DB 122 that stores the association between diseases and related symptoms has been described, but the present embodiment is not limited thereto. The control unit 11 may specify related symptoms using a learned estimation model that estimates symptoms highly relevant to a disease, for example.

[0078] According to the present embodiment, more information can be presented for the disease output from the learning model 121, and the utilization rate of the diagnostic support system 100 is increased. Further, using the output result of the learning model 121, the symptoms to be confirmed can be proposed in reverse. By repeating the inference using the learning model 121 and the answer to the reverse proposal, the reliability of the extraction result by the learning model 121 can be improved.

[0079] (Third Embodiment) In the third embodiment, a configuration for registering new case data using the output result of the learning model 121 will be described.

[0080] FIG. 14 is a schematic diagram showing an example of a screen 240 displayed on the display unit 24 in the third embodiment. A registrant of case data registers new case data by pressing a case data registration button or the like, for example, in a state where the screen 240 shown in FIG. 10 is displayed.

[0081] When the control unit 11 of the information processing apparatus 1 receives an instruction to register new case data, it generates a screen 240 that displays a plurality of symptoms and a plurality of diseases in the order of candidates by the learning model 121 in association with each other as shown in FIG. 14. The screen 240 includes a symptom column 245 including a plurality of symptoms and a disease registration column 246 for registering diseases for the symptoms.

[0082] The symptom column 245 includes, for example, the symptoms selected in the reception column 241b shown in FIG. 10. The disease registration column 246 includes the ranking results of the learning model 121 for the symptoms in the symptom column 245. The disease registration column 246 displays a plurality of diseases in the order of candidates by the learning model 121 in association with the probability of each disease. Further, the disease registration column 246 includes, as registration information for each disease, a check box for inputting either a confirmed disease or a differential disease and a score input column for registering a score. The registrant reflects the confirmed disease and the differential disease based on information such as the confirmed diagnosis information for the plurality of diseases output from the learning model 121, and inputs a score for each disease corresponding to the confirmed disease and the differential disease. The control unit 11 uses the screen 240 for registration reception to obtain new case data in which the disease type, the distinction between the confirmed disease and the differential disease, the registration score, etc. of each of the plurality of candidate diseases are associated with a plurality of symptoms.

[0083] FIG. 15 is a flowchart showing an example of a processing procedure executed by the diagnostic support system 100 of the third embodiment.

[0084] The control unit 11 of the information processing apparatus 1 generates a screen for receiving the registration of case data, and transmits the generated reception screen to the information terminal device 2 (step S61). As shown in FIG. 14, the reception screen includes a plurality of symptoms and a plurality of diseases in the order of candidates output by the learning model 121 when a plurality of symptoms are input.

[0085] The control unit 21 of the information terminal device 2 receives the reception screen (step S62), and displays the received reception screen on the display unit 24 (step S63). The control unit 21 receives the registration of case data when the registrant operates the operation unit 25 (step S64). Specifically, the control unit 21 uses the reception screen to receive corrections such as the type of candidate disease, the distinction between the confirmed disease or differential diagnosis disease, and the registration score for a plurality of symptoms, thereby obtaining a plurality of symptoms, a candidate disease, and a score for each disease. The control unit 21 receives new case data in which the above are associated, and transmits the received case data to the information processing apparatus 1 (step S65).

[0086] The control unit 11 of the information processing apparatus 1 receives the case data (step S66). The control unit 11 generates training data, which is a data set in which a plurality of candidate diseases and the correct score (registration score) for each of the diseases are labeled for a plurality of symptoms included in the received case data (step S67). The control unit 11 stores the generated training data in the training data DB 123 (step S68), and ends a series of processes. The control unit 11 may perform re-learning of the learning model 121 using the generated training data.

[0087] In step S67 described above, the control unit 11 may add a predetermined coefficient α to the confirmed disease as the score for each disease. Thereby, even when there are few symptoms, the confirmed disease is listed at the top. Further, the control unit 11 acquires information regarding the progress of the differential diagnosis, and increases the scores of the differential disease and the confirmed disease corresponding to the symptoms related to the differential diagnosis as the differential diagnosis progresses according to the acquired progress of the differential diagnosis. Thereby, the inference accuracy of the learning model 121 can be improved.

[0088] According to this embodiment, when registering case data, by using the output information of the learning model 121, it is possible to omit the trouble of generating the combination of symptoms and diseases from the beginning, and the efficiency of the registration operation can be improved.

[0089] (Fourth Embodiment) In the fourth embodiment, a configuration for receiving correction information for the output result of the learning model 121 will be described. The diagnostic support system 100 receives the input of correction information including, for example, a doctor's comment (for example, the doctor's diagnosis result is different from the inference result of the diagnostic support system), the input person of the comment, the input date and time, the patient name, etc., for the output result by the learning model 121. The diagnostic support system 100 uses the received correction information to correct the case data and perform re-learning of the learning model 121, thereby improving the inference accuracy of the learning model 121.

[0090] FIG. 16 is a flowchart showing an example of a processing procedure executed by the diagnostic support system 100 of the fourth embodiment.

[0091] The control unit 21 of the information terminal device 2 receives the correction information (step S71). In step S71, the control unit 21 may receive the input of the correction information by, for example, a doctor or the like operating the operation unit 25 in a state where the screen 240 shown in FIG. 10 is displayed. The control unit 21 transmits the received correction information to the information processing device 1 (step S72).

[0092] The control unit 11 of the information processing apparatus 1 receives correction information (step S73). The control unit 11 performs relearning using the correction information and updates the learning model 121 (step S74). Specifically, the control unit 11 corrects the training data stored in the training data DB 123 based on the received correction information. For example, when correction information indicating different certified diseases for the predicted diseases by the learning model 121 is obtained for a plurality of symptoms targeted by the correction information, the control unit 11 acquires training data including the plurality of symptoms targeted by the correction information and the plurality of diseases including the certified disease as the corrected training data. It is preferable that the highest score is given to the certified disease in the corrected training data. The control unit 11 optimizes various parameters so that the plurality of diseases in the candidate order output from the learning model 121 approximate the plurality of diseases in the corrected candidate order, and regenerates the learning model 121.

[0093] Note that the correction information is not limited to being acquired via the information terminal device 2, and may be acquired, for example, by cooperating with an electronic medical record system. Further, the control unit 11 of the information processing apparatus 1 may store the correction information, the search history of diseases, etc. in association with the account information of a doctor or the like, and display them as history data when used after the next time.

[0094] According to the present embodiment, the training data DB 123 and the learning model 121 can be further optimized through the operation of the present diagnosis support system 100.

[0095] The embodiments disclosed this time should be considered as illustrative in all respects and not restrictive. The technical features described in each example can be combined with each other, and the scope of the present invention is intended to include all changes within the scope of the claims and the scope equivalent to the claims.

Explanation of Signs

[0096] 1 Information processing apparatus 11 Control unit 12 Storage unit 13 Communication unit 12P Program 121 Learning model 122 Master DB 123 Training data DB 1A Recording medium 2 Information processing terminal 21 Control unit 22 Memory unit 23 Communication unit 24 Display unit 25 Operation unit 22P Program 2A Recording medium

Claims

1. First disease group training data associating a first symptom group including a plurality of symptoms with a first disease group including a plurality of candidate diseases and scores for each of the diseases, a Kth (K is an integer of 2 or more) symptom group including a plurality of symptoms different in combination from the first symptom group, and Kth disease group training data including a plurality of diseases common in combination with the first disease group, obtain training data including, Based on the obtained training data, generate a learning model that outputs a plurality of diseases in candidate order when a plurality of symptoms are input A method for generating a learning model in which a computer executes a process.

2. The scores related to the same disease in the plurality of pieces of training data including the same disease differ according to the types or the number of the plurality of symptoms included in each piece of training data The method for generating a learning model according to claim 1.

3. A predetermined coefficient is added to the score related to the confirmed disease among the plurality of diseases in the training data The method for generating a learning model according to claim 1 or claim 2.

4. The learning model is learned by ranking learning using the first disease group training data to the Kth disease group training data The method for generating a learning model according to any one of claims 1 to 3.

5. Obtain a plurality of symptoms, A learning model learned to output a plurality of diseases in candidate order when a plurality of symptoms are input, a first symptom group including a plurality of symptoms, first disease group training data associating a first disease group including a plurality of candidate diseases and scores for each of the diseases, a Kth (K is an integer of 2 or more) symptom group including a plurality of symptoms different in combination from the first symptom group, and Kth disease group training data including a plurality of diseases common in combination with the first disease group, input the obtained plurality of symptoms into the learning model learned using the training data including, and output a plurality of diseases in candidate order A program for causing a computer to execute a process.

6. The scores related to the same disease in the plurality of pieces of training data including the same disease differ according to the types or the number of the plurality of symptoms included in each piece of training data The program according to claim 5.

7. The learning model is learned using the first disease group training data to the Kth disease group training data The program according to claim 5 or claim 6.

8. Output by associating the disease output by the learning model with inspection information or medication information corresponding to the disease The program according to any one of claims 5 to 7.

9. Display the acquisition history of the plurality of symptoms, Obtain a new plurality of symptoms by accepting a correction to the displayed acquisition history The program according to any one of claims 5 to 8.

10. Accept correction information for the disease output by the learning model The program according to any one of claims 5 to 9.

11. Identify symptoms highly relevant to the disease output by the learning model The program according to any one of claims 5 to 10.

12. Output a plurality of the diseases in candidate order by the learning model for the plurality of symptoms, Using the plurality of diseases in candidate order by the output learning model, accept registration of the plurality of symptoms, a plurality of candidate diseases, and scores for each of the diseases The program according to any one of claims 5 to 11.

13. An acquisition unit that acquires training data including first disease group training data in which a first symptom group including a plurality of symptoms is associated with a first disease group including a plurality of candidate diseases and scores for each of the diseases, and a Kth (K is an integer of 2 or more) symptom group including a plurality of symptoms different in combination from the first symptom group, and Kth disease group training data including a plurality of diseases common in combination with the first disease group; A generation unit that generates a learning model that outputs a plurality of diseases in candidate order when a plurality of symptoms are input based on the acquired training data; An information processing apparatus comprising:

14. Acquire training data including first disease group training data in which a first symptom group including a plurality of symptoms is associated with a first disease group including a plurality of candidate diseases and scores for each of the diseases, and a Kth (K is an integer of 2 or more) symptom group including a plurality of symptoms different in combination from the first symptom group, and Kth disease group training data including a plurality of diseases common in combination with the first disease group; Generate a learning model that outputs a plurality of diseases in candidate order when a plurality of symptoms are input based on the acquired training data A program for causing a computer to execute the process.

15. An acquisition unit that acquires a plurality of symptoms,

15. An acquisition unit that acquires a plurality of symptoms, A learning model trained to output a plurality of diseases in the order of candidates when a plurality of symptoms are input, a first symptom group including a plurality of symptoms, a first disease group including a plurality of candidate diseases, and a score for each disease, and a first disease group training data associating them; a Kth (K is an integer of 2 or more) symptom group including a plurality of symptoms different from the combination of the first symptom group; and a Kth disease group training data including a plurality of diseases common to the combination with the first disease group. An output unit that inputs the acquired plurality of symptoms to the learning model learned using the training data including the above and outputs a plurality of diseases in the order of candidates An information processing apparatus comprising the same. Acquire a plurality of symptoms, A learning model trained to output a plurality of diseases in the order of candidates when a plurality of symptoms are input, a first symptom group including a plurality of symptoms, a first disease group including a plurality of candidate diseases, and a score for each disease, and a first disease group training data associating them; a Kth (K is an integer of 2 or more) symptom group including a plurality of symptoms different from the combination of the first symptom group; and a Kth disease group training data including a plurality of diseases common to the combination with the first disease group. Input the acquired plurality of symptoms to the learning model learned using the training data including the above and output a plurality of diseases in the order of candidates An information processing method in which a computer executes the process.

Images