Training data generation system, training data generation method, and computer program

Abstract

To more easily acquire data (training data) related to conversations in medical interviews. [Solution] A learning data generation system comprising a control unit that generates learning data showing specific examples of conversations that take place in a medical interview by using basic data which includes one or more of the following: disease information which is information about a disease, patient information which is information about a patient, and background information which is information about the background that led to the medical interview, and a language model.

Description

Technical Field

[0001] The present invention relates to a learning data generation system, a learning data generation method, and a computer program.

Background Art

[0002] Medical education is shifting towards education that cultivates not only medical knowledge but also practical skills for problem-solving. As one of the educational means, education involving simulated patients is being carried out. In recent years, research using generative AI technology as simulated patients for medical interviews has been progressing (see Non-Patent Documents 1 to 3).

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

[0004] However, when using generative AI as a simulated patient for medical interviews, it is necessary to provide prompts to the generative AI and train it on medical interview scenarios based on conversational data. In order to train the AI ​​on medical interview scenarios, it was necessary to prepare a large amount of training data (hereinafter referred to as "training data") in advance. Such training data is composed of data related to conversations that take place between medical professionals and interviewees such as patients during medical interviews.

[0005] Actual conversations that could be used as training data were not easily obtained due to concerns about protecting personal information and equipment limitations. Therefore, there was a problem in efficiently acquiring such training data.

[0006] This invention has been made in view of the circumstances described above, and provides a technology that makes it easier to acquire data (training data) related to conversations in medical interviews. [Means for solving the problem]

[0007] One aspect of the present invention is a learning data generation system comprising a control unit that generates learning data showing specific examples of conversations that take place in a medical interview by using basic data including one or more of the following: disease information which is information about a disease; patient information which is information about a patient; and background information which is information about the background that led to the medical interview; and a language model.

[0008] One aspect of the present invention is the above-described learning data generation system, wherein the disease information is information indicating the name of the disease, the symptoms of the disease, or both.

[0009] One aspect of the present invention is the learning data generation system described above, wherein the patient information includes information about the attributes and / or lifestyle of the patient who is the subject of the medical interview.

[0010] One aspect of the present invention is the learning data generation system described above, wherein the background information includes information on the current medical history of the patient who is the subject of the medical interview.

[0011] One aspect of the present invention is a method for generating learning data that includes the step of generating learning data showing specific examples of conversations that take place in a medical interview by using basic data which includes one or more of the following: disease information which is information about a disease; patient information which is information about a patient; and background information which is information about the background that led to the medical interview; and a language model.

[0012] One aspect of the present invention is a computer program for causing a computer to function as a learning data generation system, which includes a control unit that generates learning data showing specific examples of conversations that take place in a medical interview by using basic data that includes one or more of the following: disease information which is information about a disease; patient information which is information about a patient; and background information which is information about the background that led to the medical interview; and a language model. [Effects of the Invention]

[0013] This invention makes it easier to acquire data (training data) related to conversations in medical interviews. [Brief explanation of the drawing]

[0014] [Figure 1] This is a schematic block diagram showing the system configuration of the learning data generation system 100 of the present invention. [Figure 2] This is a schematic block diagram showing a specific example of the functional configuration of the generation request device 10. [Figure 3] This is a schematic block diagram showing a specific example of the functional configuration of the learning data generation device 20. [Figure 4] This is a sequence chart showing a specific example of the operation flow of the learning data generation system 100. [Figure 5] This figure shows a schematic example of the hardware configuration of the information processing device 90 applied to this embodiment. [Figure 6] This figure shows a modified example of the learning data generation device 20. [Modes for carrying out the invention]

[0015] Figure 1 is a schematic block diagram showing the system configuration of the learning data generation system 100 of the present invention. The learning data generation system 100 is used to generate data (learning data) related to conversations in medical interviews. The learning data is data that shows specific examples of conversations that take place in medical interviews.

[0016] The learning data generation system 100 includes a generation request device 10 and a learning data generation device 20. The generation request device 10 and the learning data generation device 20 are communicably connected via a network 70. The network 70 may be a network using wireless communication or a network using wired communication. The network 70 may be configured using, for example, the Internet or a local area network (LAN). The network 70 may be configured by combining a plurality of networks.

[0017] FIG. 2 is a schematic block diagram showing a specific example of the functional configuration of the generation request device 10. The generation request device 10 is configured using an information device such as a personal computer, a server, or the like. The generation request device 10 includes a communication unit 11, an input unit 12, an output unit 13, a storage unit 14, and a control unit 15.

[0018] The communication unit 11 is a communication device. The communication unit 11 may be configured as, for example, a network interface. The communication unit 11 performs data communication with other devices via the network 70 according to the control of the control unit 15. The communication unit 11 may be a device that performs wireless communication or a device that performs wired communication.

[0019] The input unit 12 is configured using existing input devices such as a keyboard, a pointing device (mouse, tablet, etc.), buttons, a touch panel, or the like. The input unit 12 is operated by the user when inputting the user's instruction to the generation request device 10. The input unit 12 may be an interface for connecting the input device to the generation request device 10. In this case, the input unit 12 inputs an input signal generated in response to the user's input in the input device to the generation request device 10. The input unit 12 may be configured using a microphone and a voice recognition device. In this case, the input unit 12 acquires an acoustic signal generated by the user's speech, recognizes the words spoken by the user, and inputs the character string information of the recognition result to the generation request device 10. The voice recognition processing may be executed by the control unit 15. The input unit 12 may be configured using a device for inputting electronic data such as a text file or PDF to the generation request device 10. For example, the input unit 12 may be configured using a device that reads electronic data recorded on a recording medium such as a USB memory from the recording medium. The input unit 12 may be configured using a device that receives the above electronic data from another device via communication such as short-range wireless communication or a network. The input unit 12 may be configured in any way as long as it can input the user's instruction to the generation request device 10.

[0020] The output unit 13 outputs information in a format that the user can recognize. The output unit 13 may be an image display device such as a liquid crystal display or an organic EL (Electro-Luminescence) display. The output unit 13 may also be an interface for connecting an image display device to the generation request device 10. In this case, the output unit 13 generates a video signal for displaying image data and outputs the video signal to the image display device connected to it. The output unit 13 may also be a device that outputs sound, such as a speaker. The output unit 13 may also be an interface for connecting an audio output device such as a speaker or headphones to the generation request device 10. In this case, the output unit 13 generates an audio signal for playing audio data and outputs the audio signal to the audio output device connected to it. The output unit 13 may also be configured as a touch panel integrated with the input unit 12.

[0021] The storage unit 14 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 14 stores data used by the control unit 15. The storage unit 14 stores data necessary when the control unit 15 performs processing. The storage unit 14 functions, for example, as a disease information storage unit 141, a patient information storage unit 142, a background information storage unit 143, a patient-related information storage unit 144, and a learning data storage unit 145.

[0022] The disease information storage unit 141 stores information about the disease being discussed in a medical interview in the training data to be generated. For example, training data related to a medical interview where the primary complaint is a disease stored in the disease information storage unit 141 may be generated, or training data related to a medical interview where the secondary complaint is a disease stored in the disease information storage unit 141 may be generated. A list of diseases and symptoms described in the Objective Structured Clinical Examination (OSCE) in the common examination may be used as disease information. The disease information storage unit 141 may store, for example, the name of the disease, or the symptoms caused by the disease, or both. The disease information storage unit 141 may store strings representing disease names such as "pneumonia," "rheumatoid arthritis," "tuberculosis," "cirrhosis," "heatstroke," and "dementia," or it may store strings representing symptoms and the severity of symptoms such as "fever," "cough," "chest pain," "abdominal pain," "back pain," "intellectual disability," and "physical disability." By using this kind of information, it is possible to generate training data for diseases and symptoms that are tailored to the learning objectives.

[0023] The patient information storage unit 142 stores information about the patient (hereinafter referred to as "patient information"). Patient information includes the patient's name, patient attribute information, and information about the patient's lifestyle (hereinafter referred to as "lifestyle information"). In other words, patient information includes information about the patient's attribute information, lifestyle, or both. Attribute information includes, for example, information about gender, age, place of origin, preferences, occupation, personality, family structure, pets owned, overseas travel experience, hobbies, etc. Lifestyle information includes, for example, information about drinking habits, smoking habits, eating habits, exercise habits, etc. By using such information, learning data can be generated that is tailored to the diverse patient characteristics according to the learning objectives. Such patients may include, for example, patients with dementia, intellectual disabilities, and physical disabilities.

[0024] The background information storage unit 143 stores information about the background that led the interviewee to undergo a medical interview (hereinafter referred to as "background information"). Background information includes information about the patient's condition before the medical interview. For example, background information may include the patient's medical history. More specifically, background information may include information such as the onset time of symptoms for a disease, the duration of symptoms, the frequency and severity of symptoms, the circumstances under which symptoms occur, factors that exacerbate or alleviate symptoms, sleep patterns, and appetite. By using such information, diverse learning data with various stories can be generated to suit the learning objectives. Background information may also include information about past medical history.

[0025] The patient-related information storage unit 144 stores information about patient-related individuals. Patient-related individuals are those who can be interviewed in a medical interview on behalf of the patient. Interviewees are those who conduct interviews with medical professionals in a medical interview. Interviewees can be the patient themselves or patient-related individuals. Patient-related individuals include, for example, the patient's relatives, those who provide care for the patient (including other medical professionals), etc., who communicate symptoms and other information about the patient to medical professionals on behalf of the patient during the medical interview. The patient-related information storage unit 144 may also store information indicating the relationship between patient-related individuals and the patient, or attribute information of patient-related individuals. The learning data storage unit 145 stores the generated learning data. Medical professionals may include doctors, dentists, nurses, pharmacists, veterinary students (before clinical training), faculty members of universities and vocational schools, and those in leadership positions responsible for educational work in medical settings. In this case, the system is particularly effective when used as a simulated patient for an OSCE as a competency test. Furthermore, although OSCE is not mandatory, this system is also effective when used for educational training of "patients and those requiring care." It is particularly effective for the following individuals (healthcare professionals, etc.). For this reason, the following individuals may be defined as healthcare professionals: physical therapists, occupational therapists, speech-language pathologists, radiological technologists, clinical psychologists, clinical engineers, clinical laboratory technologists, care workers (caregivers, certified care workers, care staff, care managers, home care workers, etc.), faculty members at universities and vocational schools, and those in leadership positions responsible for educational work in medical settings.

[0026] The control unit 15 is composed of a processor such as a CPU (Central Processing Unit) and memory (main memory). The control unit 15 functions when the processor executes a program. Note that all or part of the functions of the control unit 15 may be implemented using hardware such as an ASIC (Application Specific Integrated Circuit), PLD (Programmable Logic Device), or FPGA (Field Programmable Gate Array). The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include, for example, portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and semiconductor memory devices (e.g., SSDs: Solid State Drives), as well as storage devices such as hard disks and semiconductor memory devices built into computer systems. The above program may be transmitted via a telecommunications line.

[0027] The control unit 15 may execute an application installed on its own device (the generation request device 10). A specific example of such an application is an application provided to the generation request device 10 as a dedicated application for the learning data generation system 100. Another specific example of such an application is a web browser application. Such an application may be pre-installed on the generation request device 10, or it may be downloaded each time the learning data generation process is executed. For example, if it is implemented as a web browser application, the generation request device 10 may download and execute the application from a device specified by the web server (for example, the web server itself or another server) when the generation request device 10 connects to a specific web server. The control unit 15 operates according to the program of the application being executed.

[0028] The control unit 15 controls the generation request device 10 in response to user operations and information received from the learning data generation device 20. For example, the control unit 15 transmits information input by the user through the operation of the input unit 12 to the learning data generation device 20 using the communication unit 11. For example, when the information transmitted from the learning data generation device 20 is received by the communication unit 11 via the network 70, the control unit 15 generates learning data based on the received information and records it in the learning data storage unit 145. The control unit 15 may also output the generated learning data from the output unit 13.

[0029] The information control unit 151 controls the input and output of data in the generation request device 10. For example, when disease information, patient information, background information, and patient-related information are input from another device, the information control unit 151 records the input data in the storage unit 14. The information control unit 151 may also output the learning data recorded in the learning data storage unit 145 to another device or another recording medium.

[0030] The generation control unit 152 performs processing to generate training data using the information stored in the storage unit 14. Specifically, it is as follows: The generation control unit 152 includes at least one of disease information, patient information, and background information. The basic data may be configured to include disease information, patient information, and background information. With the basic data configured in this way, it becomes possible to generate training data for when a medical professional and the patient themselves conduct a medical interview. Furthermore, the basic data may be configured to include disease information, patient information, background information, and patient-related information. With the basic data configured in this way, it becomes possible to generate training data for when a medical professional and patient-related individuals conduct a medical interview.

[0031] The information used by the generation control unit 152 when generating basic data may be selected in any way from among the multiple pieces of information stored in each storage unit (141-144). For example, if the basic data includes disease information, patient information, and background information, the generation control unit 152 may select one piece of disease information from among the multiple pieces of disease information stored in the disease information storage unit 141, one piece of patient information from among the multiple pieces of patient information stored in the patient information storage unit 142, and one piece of background information from among the multiple pieces of background information stored in the background information storage unit 143, and then generate basic data using the selected pieces of information. Alternatively, the generation control unit 152 may use the information stored in each storage unit (141-144) as basic data as is. The selection of each piece of information may be done randomly or in a predetermined order. This process is the same even if the basic data includes patient-related information.

[0032] The generation control unit 152 may operate to automatically generate multiple training data sets in response to a single instruction from the user. For example, if the user instructs the generation control unit 152 to generate training data for 100 simulated patients, the generation control unit 152 may generate basic data for 100 simulated patients, transmit the basic data to the training data generation device 20, and obtain training data for 100 simulated patients. In this case, it is desirable that the generated basic data sets (e.g., for 100 simulated patients) are all configured to have different combinations of disease information, patient information, and background information. This processing is also applicable when the basic data includes patient-related information.

[0033] The generation control unit 152 obtains training data generated based on the basic data by transmitting the generated basic data to the training data generation device 20. The generation control unit 152 records the generated training data in the training data storage unit 145.

[0034] Figure 3 is a schematic block diagram showing a specific example of the functional configuration of the learning data generation device 20. The learning data generation device 20 is configured using information processing equipment such as a personal computer or a server device. The learning data generation device 20 includes a communication unit 21, a storage unit 22, and a control unit 23.

[0035] The communication unit 21 is a communication device. The communication unit 21 may be configured, for example, as a network interface. The communication unit 21 communicates data with other devices via the network 70 in accordance with the control of the control unit 23. The communication unit 21 may be a wireless communication device or a wired communication device.

[0036] The storage unit 22 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 22 stores data used by the control unit 23. The storage unit 22 may also function, for example, as a trained language model storage unit 221.

[0037] The pre-trained language model storage unit 221 stores language models obtained by performing a training process in advance. The language models may be configured as, for example, large language models.

[0038] The control unit 23 is composed of a processor such as a CPU and memory. The control unit 23 functions as a learning data generation unit 231 when the processor executes a program. Note that all or part of the functions of the control unit 23 may be implemented using hardware such as an ASIC, PLD, or FPGA. The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include, for example, portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and semiconductor storage devices (e.g., SSDs), as well as storage devices such as hard disks and semiconductor storage devices built into computer systems. The above program may be transmitted via a telecommunications line.

[0039] The learning data generation unit 231 generates learning data by using the basic data received from the generation request device 10 and the language model stored in the trained language model storage unit 221. For example, the learning data generation unit 231 may function as a so-called generative AI by using the language model to generate strings of characters that represent specific examples of medical interview conversations based on the basic data as learning data. The learning data generation unit 231 transmits the generated learning data to the generation request device 10.

[0040] The processing of the learning data generation unit 231 can be implemented in any way. For example, the learning data generation unit 231 may generate learning data based on template data of conversations in medical interviews that are pre-recorded in the storage unit 22. In this case, a template of a conversation that is commonly used in medical interviews may be defined as template data, and learning data may be generated by filling in a part of the template with a string based on basic data.

[0041] For example, the learning data generation unit 231 may generate learning data corresponding to the basic data by using a pre-trained language model storage unit 221 that has been fine-tuned in advance using multiple specific examples of conversations in medical interviews. In this case, the pre-trained language model storage unit 221 of the storage unit 22 stores a language model that has been fine-tuned in advance to generate learning data. With this configuration, the pre-trained language model storage unit 221 stores data that has been learned in advance of the flow of conversations during medical interviews.

[0042] The learning data generation unit 231 may be implemented using so-called RAG (Retrieval-Augmented Generation). In this case, the memory unit 22 further stores data necessary for implementing RAG. Specific examples of such data include data showing specific examples or templates of conversations in medical interviews. For example, the learning data generation unit 231 may generate learning data showing conversation content in a format consistent with specific examples or templates by inputting basic data along with data showing specific examples or templates of conversations to the trained language model memory unit 221. In this case, the flow of the conversation itself will follow the specific examples or templates, but the information regarding diseases, symptoms, and backgrounds, which constitute the content of the conversation, will be based on the basic data.

[0043] Figure 4 is a sequence chart showing a specific example of the operation flow of the learning data generation system 100. The generation request device 10 generates basic data (step S101). The generation request device 10 transmits the basic data to the learning data generation device 20 (step S102). When the learning data generation device 20 receives the basic data, it generates learning data based on the received basic data (step S103). The learning data generation device 20 transmits the generated learning data to the generation request device 10 (step S104). When the generation request device 10 receives the learning data, it records the received learning data in the learning data storage unit 145 (step S105).

[0044] The learning data generation system 100 configured in this way makes it easier to acquire data (learning data) related to conversations in medical interviews. Specifically, it works as follows: The generation request device 10 generates basic data that includes at least one of disease information, patient information, and background information. The learning data generation device 20 generates learning data using a language model based on the basic data. Therefore, it becomes possible to acquire learning data without having to perform cumbersome tasks such as recording actual medical interview conversations.

[0045] In particular, if multiple specific pieces of information are recorded in the generation request device 10 as disease information, patient information, and background information, it becomes possible to easily obtain multiple training data sets according to their combinations. For example, by pre-recording multiple pieces of disease information, it becomes possible to generate training data for various diseases and symptoms that match the learning objective. For example, by pre-recording multiple pieces of patient information, it becomes possible to generate training data for patients with diverse characters that match the learning objective. For example, by pre-recording multiple pieces of background information, it becomes possible to easily generate training data for various stories.

[0046] The one or more training data generated by the training data generation system 100 may be used, for example, in the training process of a language model that simulates (virtually generates) patient statements in a medical interview conversation. By performing such training, it becomes possible to generate a virtual personality (avatar) that speaks in response to the statements of the medical professional in the medical interview. By generating and using such an avatar, it becomes possible to conduct medical interview training without using actual people as interviewees (patients or people related to the patient).

[0047] Figure 5 is a schematic diagram of an example hardware configuration of an information processing device 90 applied to this embodiment. The information processing device 90 comprises a processor 91, main memory 92, communication interface 93, auxiliary storage device 94, input / output interface 95, and internal bus 96. The processor 91, main memory 92, communication interface 93, auxiliary storage device 94, and input / output interface 95 are connected to each other via the internal bus 96 so as to be able to communicate with each other. The information processing device 90 may be applied to, for example, a generation request device 10 and a learning data generation device 20. In this case, for example, the communication unit 11 and the communication unit 21 may be configured using the communication interface 93. For example, the storage unit 14 and the storage unit 22 may be configured using the auxiliary storage device 94. Also, the control unit 15 and the control unit 23 may be configured using the processor 91 and the main memory 92.

[0048] (modified version) In this embodiment, the generation request device 10 and the learning data generation device 20 are configured as separate devices, but they may be configured as a single device. Figure 6 shows a modified example of the learning data generation device 20 configured in this way. The learning data generation device 20 shown in Figure 6 includes an input unit 24 and an output unit 25. The input unit 24 and output unit 25 of the learning data generation device 20 shown in Figure 6 function similarly to the input unit 12 and output unit 13 of the generation request device 10, respectively. The control unit 23 operates in response to operations on the input unit 24 and generates a plurality of basic data based on the information stored in the disease information storage unit 222 to the patient-related information storage unit 225. The control unit 23 then records the generated learning data in the learning data storage unit 226.

[0049] The generation request device 10 may be implemented using multiple information processing devices. For example, the generation request device 10 may be implemented using a cloud or other similar device. For example, in the generation request device 10, the storage unit 14 and the control unit 15 may be implemented in different information processing devices. For example, the storage unit 14 of the generation request device 10 may be distributed and implemented across multiple information processing devices.

[0050] The learning data generation device 20 may be implemented using multiple information processing devices. For example, the learning data generation device 20 may be implemented using a device such as a cloud. For example, in the learning data generation device 20, the storage unit 22 and the control unit 23 may be implemented in different information processing devices. For example, the storage unit 22 of the learning data generation device 20 may be distributed and implemented in multiple information processing devices.

[0051] While embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to these embodiments and includes designs and the like that do not depart from the spirit of this invention. [Explanation of Symbols]

[0052] 100...Learning data generation system, 10...Generation request device, 11...Communication unit, 12...Input unit, 13...Output unit, 14...Storage unit, 15...Control unit, 20...Learning data generation device, 21...Communication unit, 22...Storage unit, 23...Control unit

Claims

1. A learning data generation system comprising a control unit that generates learning data showing specific examples of conversations that take place in a medical interview by using basic data including one or more of the following: disease information, which is information about a disease; patient information, which is information about a patient; and background information, which is information about the background that led to the medical interview; and a language model.

2. The learning data generation system according to claim 1, wherein the disease information is information indicating the name of the disease, the symptoms of the disease, or both.

3. The learning data generation system according to claim 1, wherein the patient information includes information about the attributes and / or lifestyle of the patient who is the subject of the medical interview.

4. The learning data generation system according to claim 1, wherein the background information includes information on the current medical history of the patient who is the subject of the medical interview.

5. A method for generating learning data, comprising the step of generating learning data that shows specific examples of conversations that take place in a medical interview by using basic data that includes one or more of the following: disease information, which is information about a disease; patient information, which is information about a patient; and background information, which is information about the background that led to the medical interview; and a language model.

6. A computer program for causing a computer to function as a learning data generation system, comprising a control unit that generates learning data showing specific examples of conversations that take place in a medical interview by using basic data including one or more of the following: disease information, patient information, and background information, which is information about the circumstances leading to the medical interview; and a language model.

Images