WO2023210219A1

WO2023210219A1 - Information processing device, information processing method, and program

Info

Publication number: WO2023210219A1
Application number: PCT/JP2023/011503
Authority: WO
Inventors: 敬長野; 錦涛黄; 良夫齋藤; 康弘村井; 達朗馬場
Original assignee: Tdk株式会社
Priority date: 2022-04-28
Filing date: 2023-03-23
Publication date: 2023-11-02

Abstract

This information processing device comprises a prescription candidate output unit that, on the basis of a first time frequency spectrum image indicating a first time frequency spectrum corresponding to a waveform of a pulse wave of a first subject, outputs output information including Chinese medicine candidate information which indicates a candidate of a Chinese medicine to be prescribed to the first subject.

Description

Information processing device, information processing method, and program

The present disclosure relates to an information processing device, an information processing method, and a program.
This application claims priority based on Japanese Patent Application No. 2022-074671 filed in Japan on April 28, 2022, the contents of which are incorporated herein.

Research and development is being conducted on technology that assists doctors in prescribing medicines to patients.

In this regard, a herbal medicine doctor diagnoses the symptoms of multiple subjects based on a database that includes information that associates the symptoms of the subjects with the history of prescriptions of Chinese herbal medicine to the subjects. An information processing device is known that outputs information indicating a prescription for a Chinese herbal medicine that is associated with the symptoms of a target subject (see Patent Document 1).

Japanese Patent Application Publication No. 2016-139268

Here, the symptoms of the target test subject described in Patent Document 1 include headache, dizziness, menopausal disorder, etc., and are based on the physical or mental state of the test subject as diagnosed by a Chinese herbalist. That's true. Therefore, in the information processing device described in Patent Document 1, a Chinese herbalist diagnoses symptoms based on diagnostic methods such as interview, facial examination, tongue examination, abdominal examination, and pulse examination, and displays the diagnosis results. It is necessary to input information into the information processing device. For this reason, the information processing device may not be able to sufficiently reduce the effort required for a Chinese herbalist to prescribe a Chinese herbal medicine to a target subject.

The present disclosure has been made in consideration of such circumstances, and provides an information processing device, an information processing method, and a program that can reduce the effort required for a Chinese medicine doctor to prescribe a Chinese medicine to a first subject. The challenge is to provide the following.

One aspect of the present disclosure provides a Chinese herbal medicine candidate to be prescribed to the first subject based on a first time-frequency spectrum image showing a first time-frequency spectrum corresponding to a pulse wave waveform of the first subject. The information processing device includes a prescription candidate output unit that outputs output information including Chinese herbal medicine candidate information indicating.

Further, one aspect of the present disclosure provides a Chinese herbal medicine prescribed to the first subject based on a first time-frequency spectrum image showing a first time-frequency spectrum corresponding to a pulse wave waveform of the first subject. This information processing method includes a prescription candidate output step of outputting output information including Chinese herbal medicine candidate information indicating candidates.

Further, one aspect of the present disclosure is to cause a computer to prescribe a prescription to the first subject based on a first time-frequency spectrum image showing a first time-frequency spectrum corresponding to a pulse wave waveform of the first subject. This is a program for executing a prescription candidate output step of outputting output information including Chinese herbal medicine candidate information indicating Chinese herbal medicine candidates.

According to the present disclosure, it is possible to reduce the effort required for a Chinese herbalist to prescribe a Chinese herbal medicine to the first subject.

1 is a diagram illustrating an example of the configuration of an information processing system 1 including an information processing device 20. FIG. 2 is a diagram illustrating an example of a hardware configuration of an information processing device 20. FIG. 2 is a diagram illustrating an example of a functional configuration of an information processing device 20. FIG. 2 is a diagram illustrating an example of the flow of processing in which the information processing device 20 generates waveform information. FIG. FIG. 4 is a diagram illustrating an example of a process flow in which the information processing device 20 causes the first machine learning model to learn first correspondence information and causes the second machine learning model to learn second correspondence information. It is a figure which shows an example of the waveform shown by the waveform information at the time of learning matched with the target subject identification information selected in step S220. FIG. 3 is a diagram showing an example of a time-frequency spectrum image. FIG. 7 is an image diagram visualizing an example of the process of step S270. FIG. 2 is a diagram illustrating an example of the flow of processing in which the information processing device 20 receives diagnosis result information. It is a figure which shows an example of information reception image PCT1. FIG. 6 is a diagram showing an example of how six drop-down menus are displayed. 2 is a diagram illustrating an example of a process flow in which the information processing device 20 outputs Chinese herbal medicine candidate information. FIG. It is a figure which shows an example of the likelihood for each of several options contained in a diagnostic item like floating veins. FIG. 12 is an image diagram visualizing a flow in which the second machine learning model identifies one or more Chinese herbal medicine candidates that are estimated to be plausible as one or more Chinese herbal medicines to be prescribed to the subject S3. It is a figure which shows an example of information reception image PCT2. 3 is a diagram illustrating an example of a flow in which a time-frequency spectrum image is generated by the information processing device 20 based on waveform information indicating the waveform of a pulse wave detected by a pressure variable arterial wave detection method. FIG.

<Embodiment>
Embodiments of the present disclosure will be described below with reference to the drawings.

<Overview of information processing device>
First, an overview of the information processing apparatus according to this embodiment will be explained.

Currently, Western medicine has become mainstream because of its compatibility with science. However, there are many traditional medicines in the world, such as Greek medicine, Arabic medicine (Unina medicine), Indian medicine (Ayurveda), Tibetan medicine, Mongolian medicine, Chinese medicine (traditional Chinese medicine), Korean medicine, and Japanese and Chinese medicine (herbal medicine). ) etc. exist. Furthermore, in Japan, oriental medicine in a narrow sense that uses acupuncture, medicinal meals, and herbal medicine derived from Chinese medicine is called Kampo medicine. These Chinese herbal medicines have achieved unique development in Japan. In Japan, in recent years, not only Western medicine but also complementary and alternative medicine have been reconsidered. For this reason, homeopathy, natural therapy, aromatherapy, etc. are also attracting attention in Japan. In Japan's medical practice, Western medicine and Chinese herbal medicine are increasingly being used together as part of integrated medicine, including herbal medicine, with approximately 90% of doctors prescribing Chinese herbal medicine.

In Japan, several hundred types of herbal medicines, 148 types of medicinal herbal preparations (preparations approved by the Ministry of Health, Labor and Welfare), and some decoctions are each covered by health insurance in Japan. Additionally, in Japan, in addition to these herbal medicines that are covered by health insurance, there are also OTC (over-the-counter) medicines (medicines that can be purchased without a prescription at pharmacies and drugstores) that are not covered by health insurance, and medicines that specialize in herbal medicine. Chinese herbal medicine is also available for free consultation at hospitals.

In addition to medical history, facial examination, tongue examination, abdominal examination, etc., pulse examination is known as a diagnostic method in Chinese medicine. Pulse diagnosis is based on the idea that the characteristics that appear in the pulse wave according to the state of the organ is based on the idea that the characteristics that appear in the pulse wave according to the state of the organ are diagnosed. In the following, for convenience of explanation, the types of characteristics that appear in pulse waves depending on the state of an organ will be simply referred to as pulse types. As a method for classifying pulse types according to the state of an organ, for example, 28 diseased pulses is known, which classifies the characteristics appearing in a pulse wave into 28 types according to the state of an organ. In these 28 diseased veins, each of the 28 types of veins is divided into six veins called floating veins, sinking veins, slow veins, few veins, virtual veins, and real veins. classified as one of the species. In the following, for convenience of explanation, these six vein types will be referred to as six major vein types. The category of floating veins includes six vein types: floating veins, kou veins (the Chinese character with a crown above the hole), hong veins, leather veins, wet veins, and scatter veins. The category of sink veins includes four vein types: sink veins, dip veins, weak veins, and dungeon veins. The category of slow reticular pulse includes five types of pulses: slow pulse, slow pulse, astringent pulse, conjunctival pulse, and venous vein. The category of reticular pulses includes three types of pulses: reticular pulses, arterial pulses, and facilitatory pulses. The category of ischemic veins includes four types of veins: ischemic veins, short veins, veinlets, and microvenules. The category of real veins includes six vein types: real vein, long vein, chordal vein, tense vein, smooth vein, and major vein.

For example, the Chinese medicine doctor diagnosed the characteristics of the subject's pulse wave and determined the main disease by assigning a diagnostic name to the eight-pronged pulse, which is the most frequently seen in the pulse wave among the 28 diseased pulses. Prescribe the type of herbal medicine that corresponds to the main disease. In other words, a Chinese herbalist's pulse diagnosis is directly linked to herbal medicine prescriptions. The effects of Chinese medicine prescribed by Chinese herbalists are based on statistics.

Information related to diagnosis by a Chinese herbalist is explained in detail on, for example, the following reference website, and therefore further detailed explanation will be omitted in this specification.
・Reference website 1: Ministry of Health, Labor and Welfare homepage/Study group on the state of “integrated medicine”, https://www.mhlw.go.jp/stf/shingi/other-isei_127369.html
・Reference website 2: Japan Oriental Medicine Society homepage/Chinese medicine examination, http://www.jsom.or.jp/universally/examination/index.html

Currently, one of the challenges in promoting integrated medical care using Western medicine and Chinese herbal medicine is the current situation in which Chinese herbal medicines are prescribed based on the qualitative diagnostic methods of herbal medicine and statistics that are difficult to standardize.

Unlike scientific Western medicine, which is based on the subdivision of disease types and pathological elucidation, the qualitative diagnostic method of Chinese medicine is a relative diagnosis based on the subjectivity of the examiner, which captures the imbalance from the normal state of each patient. is considered basic. For this reason, it is considered appropriate to use statistical methods to accumulate data on the diagnostic results of Chinese herbalists with advanced experience and create a mathematical model. However, in addition to the fact that the diagnosis itself by Chinese herbalists includes diagnostic bias, there is currently insufficient understanding of the statistical methods involved. For this reason, regarding pulse diagnosis, which plays a large role in Chinese medicine diagnosis, it is necessary to accurately acquire waveform information indicating the waveform of the pulse wave, to remove noise from the acquired waveform information, and to extract feature values from multiple continuous waveform information. There are high expectations for the emergence of automatic judgment based on a database constructed by extracting relevant information and obtaining related information (for example, the results of medical interviews, facial examinations, tongue examinations, abdominal examinations, etc.).

On the other hand, statistics that are difficult to standardize are difficult to distinguish from the placebo effect due to the slow and long-term improvement effect of herbal medicine. Although advanced data accumulation and modeling approaches are considered suitable, they are still poorly understood.

As references for pulse wave diagnosis, the following three documents are listed: References 1 to 3.

Reference 1: JP 2009-011585 Reference 2: JP 2004-195204 Reference 3: JP 2020-108819

Reference 1 describes a wristwatch-type 24-hour wearable pulse wave monitoring device. This 24-hour wearable pulse wave monitoring device is a device that performs health management based on exercise and heart rate.

Reference 2 describes a device that detects indicators representing the characteristics of blood pressure waveforms and determines the prescription of Western medical drugs such as Ca antagonists and β-blockers based on systolic blood pressure and AI (Augmentation Index) values. .

Reference 3 describes a device that estimates blood sugar levels by using the correlation between the AI value of blood pressure waveforms and postprandial blood sugar levels.

As mentioned above, although the devices described in each of References 1 to 3 automatically monitor physical condition, prescribe Western medicine, estimate blood sugar level, etc. by automatic diagnostic analysis of pulse waves, It has not yet been possible to automatically prescribe Chinese medicine from automatic pulse wave analysis.

On the other hand, as a prior art technology for prescribing Chinese herbal medicine based on pulse wave diagnosis results, there is an effort to have a natural language processing AI (Artificial Intelligence) learn disease diagnosis names and prescription drugs from electronic medical records containing Chinese herbal medicine prescriptions. Examples include a quick reference table of diagnosis names and prescription drugs based on a Chinese medicine prescription wheel based on dialectic theory.

In recent years, information processing devices that utilize estimation models based on statistical learning have appeared with regard to automating the prescription of Chinese herbal medicines. Details of this information processing device are described in Patent Document 1 cited as a prior art document. That is, as described above, this information processing device is based on a database that includes information for each of a plurality of subjects, in which the symptoms of the subject are associated with the history of prescriptions of Chinese herbal medicines to the subject. Then, the Chinese herbal medicine doctor outputs information indicating the prescription of the Chinese herbal medicine that is associated with the symptoms of the target subject whose symptoms are to be diagnosed.

Therefore, the information processing apparatus according to the embodiment provides a prescription for a first subject based on a first time-frequency spectrum image showing a first time-frequency spectrum corresponding to the waveform of the pulse wave of the first subject. A prescription candidate output unit is provided that outputs output information including Chinese herbal medicine candidate information indicating Chinese herbal medicine candidates.

Thereby, the information processing device according to the embodiment can automate the process from pulse diagnosis to prescription of Chinese medicine. As a result, the information processing device can reduce the effort required for the Chinese medicine doctor to prescribe the Chinese medicine to the first subject.

Below, the configuration of the information processing device according to the embodiment and the processing performed by the information processing device will be described in detail.

<Configuration of information processing device>
Hereinafter, the configuration of the information processing apparatus according to the embodiment will be described using the information processing apparatus 20 as an example of the information processing apparatus according to the embodiment. Note that, in the embodiment, the subject's symptoms refer to the subject's physical or mental state diagnosed by a Chinese herbalist. Therefore, in the embodiment, the symptoms of a certain subject do not include the subject's pulse wave and the waveform of the pulse wave.

FIG. 1 is a diagram showing an example of the configuration of an information processing system 1 including an information processing device 20. Here, the three-dimensional coordinate system TC is a three-dimensional orthogonal coordinate system that indicates the direction in the drawing in which the three-dimensional coordinate system TC is drawn. In the following, for convenience of explanation, the X-axis in the three-dimensional coordinate system TC will be simply referred to as the X-axis. Furthermore, for convenience of explanation, the Y-axis in the three-dimensional coordinate system TC will be simply referred to as the Y-axis. Further, for convenience of explanation, the Z axis in the three-dimensional coordinate system TC will be simply referred to as the Z axis in the following description. In addition, for convenience of explanation, the positive direction of the Z-axis will be referred to as an upward direction, and the negative direction of the Z-axis will be referred to as a downward direction.

The information processing system 1 includes a pulse wave detection device 10 and an information processing device 20 that is an example of an information processing device according to an embodiment.

The pulse wave detection device 10 detects the pulse wave of the subject. Here, in the embodiment, the subject may be any person whose pulse wave is detected by the pulse wave detection device 10.

The pulse wave detection device 10 may have any configuration as long as it is capable of detecting the pulse wave of the subject. In the example shown in FIG. 1, the pulse wave detection device 10 includes a first member 11 on which one arm of a subject can be placed and fixed, and a first member 11 that can be placed in contact with one arm of the subject fixed by the first member 11. The device includes a pulse wave sensor 12 that detects a pulse wave of a subject, and a second member 13 that supports the pulse wave sensor 12.

The first member 11 is, for example, a table on which one arm of the subject can be placed and fixed. Note that the first member 11 may be configured to be able to move the relative position of one arm of the subject with respect to the pulse wave sensor 12 along a horizontal plane; The configuration may be such that it is impossible to do so.

The pulse wave sensor 12 may be any sensor as long as it is capable of detecting the pulse wave of the subject. For example, the pulse wave sensor 12 may be a sensor using MEMS (Micro Electro Mechanical Systems) that can detect pulse waves as pressure fluctuations. ) Pressure sensors, etc. The pulse wave sensor 12 is communicably connected to the information processing device 20 by wire or wirelessly. Therefore, the pulse wave sensor 12 detects the pressure of the pulse wave and outputs an electrical signal corresponding to the detected pressure to the information processing device 20. As a result, the information processing device 20 can generate waveform information indicating the waveform of the subject's pulse wave within the measurement period based on the electrical signal acquired from the pulse wave sensor 12 within the predetermined measurement period. .

The second member 13 may have any configuration as long as it can support the pulse wave sensor 12. Further, the second member 13 may have a configuration in which the position of the pulse wave sensor 12 in the vertical direction (that is, the height of the pulse wave sensor 12) can be adjusted, and it is not necessary to adjust the position. It may be a possible configuration.

The information processing device 20 acquires an electrical signal from the pulse wave sensor 12 within a measurement period specified by the user in accordance with an operation received from the user. The information processing device 20 generates waveform information indicating the waveform of the subject's pulse wave based on the electrical signal acquired from the pulse wave sensor 12 during the measurement period. The information processing device 20 stores the generated waveform information.

Based on the stored waveform information, the information processing device 20 calculates a time-frequency spectrum according to the waveform indicated by the waveform information. The information processing device 20 identifies candidates for Chinese herbal medicines to be prescribed to the subject based on the time-frequency spectrum image showing the calculated time-frequency spectrum. After specifying the Chinese herbal medicine candidate, the information processing device 20 generates output information including Chinese herbal medicine candidate information indicating the identified Chinese herbal medicine candidate. After generating the output information, the information processing device 20 outputs the generated output information. Thereby, the information processing device 20 can reduce the effort required for a Chinese medicine doctor to prescribe a Chinese medicine to a subject.

Here, the information processing device 20 uses, for example, a first machine learning model that has learned the first correspondence information in advance, a second machine learning model that has learned the second correspondence information in advance, and the generated time-frequency spectrum image. Identify candidates for Chinese herbal medicines to be prescribed to the subject based on the The first correspondence information is information in which a time-frequency spectrum image corresponding to the waveform of the pulse wave of the subject is associated with diagnosis result information indicating the diagnosis result of the subject by the Chinese medicine doctor. The first machine learning model is a machine learning model that has learned the first correspondence information. The first machine learning model generates a diagnosis result that indicates a likely diagnosis result for the subject by a Chinese herbalist when a time-frequency spectrum image corresponding to the pulse wave waveform of a certain subject is input. Output information. On the other hand, the second correspondence information includes, for each subject, diagnosis result information indicating the diagnosis result of the subject by the herbalist doctor, and herbal medicine information indicating each of one or more herbal medicines prescribed to the subject by the herbalist doctor. This is the information that is associated with. The second machine learning model is a machine learning model that has learned the second correspondence information. When inputting the diagnosis result information output by the first machine learning model, the second machine learning model selects one or more Chinese medicines that are likely to be prescribed as one or more Chinese medicines to be prescribed to the subject corresponding to the diagnosis result information. Chinese herbal medicine candidate information indicating each of the Chinese herbal medicine candidates is output. In this way, the information processing device 20 uses the first machine learning model and the second machine learning model to identify candidates for Chinese herbal medicines to be prescribed to the subject. Thereby, the information processing device 20 can output Chinese herbal medicine candidate information indicating one or more Chinese herbal medicine candidates to be prescribed to the subject, excluding at least part of the subjectivity of the Chinese medicine doctor. As a result, the information processing device 20 can reduce the time and effort required for a Chinese herbalist to prescribe Chinese medicine to a subject, and can also prescribe Chinese herbal medicine to a subject without being influenced by the experience of the herbalist. Prescriptions can be made with high precision. Note that details of the time-frequency spectrum image and the diagnosis result information will be described later.

The information processing device 20 is, for example, an information processing device such as a notebook PC (Personal Computer), a desktop PC, a workstation, a tablet PC, a multifunctional mobile phone terminal (smartphone), a mobile phone terminal, a PDA (Personal Digital Assistant), etc. However, it is not limited to these.

<Hardware configuration of information processing device>
The hardware configuration of the information processing device 20 will be described below with reference to FIG. 2. FIG. 2 is a diagram showing an example of the hardware configuration of the information processing device 20. As shown in FIG.

The information processing device 20 includes, for example, a processor 21, a storage section 22, an input reception section 23, a communication section 24, and a display section 25. Further, the information processing device 20 communicates with the pulse wave detection device 10 via the communication unit 24. These components are communicatively connected to each other via a bus.

The processor 21 is, for example, a CPU (Central Processing Unit). Note that the processor 21 may be another processor such as an FPGA (Field Programmable Gate Array) instead of the CPU. The processor 21 executes various programs stored in the storage unit 22. Note that the processor 21 may be configured by a CPU included in one information processing device (in this example, the information processing device 20), or may be configured by CPUs included in a plurality of information processing devices.

The storage unit 22 includes, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), an EEPROM (Electrically Erasable Programmable Read Only Memory), a ROM (Read Only Memory), and a RAM (Random Access Memory). Note that the storage unit 22 may be an external storage device connected to a digital input/output port such as a USB (Universal Serial Bus), instead of being built into the information processing device 20. The storage unit 22 stores various information, various programs, etc. processed by the information processing device 20. For example, the storage unit 22 stores the above-mentioned waveform information, the first machine learning model, the second machine learning model, and the like. Note that the storage unit 22 may be configured by one storage device or may be configured by a plurality of storage devices. Further, the plurality of storage devices may include a storage device provided in an information processing device separate from the information processing device 20.

The input reception unit 23 is an input device such as a keyboard, mouse, touch pad, etc. Note that the input receiving section 23 may be a touch panel configured integrally with the display section 25.

The communication unit 24 includes, for example, a digital input/output port such as a USB, an Ethernet (registered trademark) port, a communication antenna, and the like.

The display unit 25 is, for example, a display panel such as a liquid crystal display panel or an organic EL (Electro Luminescence) display panel.

<Functional configuration of information processing device>
The functional configuration of the information processing device 20 will be described below with reference to FIG. 3. FIG. 3 is a diagram showing an example of the functional configuration of the information processing device 20. As shown in FIG.

The information processing device 20 includes a storage section 22, an input reception section 23, a communication section 24, a display section 25, and a control section 26.

The control unit 26 controls the entire information processing device 20. The control unit 26 includes an acquisition unit 261, a reception unit 262, a calculation unit 263, a prescription candidate output unit 264, a first learning unit 265, a second learning unit 266, a display control unit 267, and a generation unit 268. Equipped with These functional units included in the control unit 26 are realized, for example, by the processor 21 executing various programs stored in the storage unit 22. Further, some or all of the functional units may be a hardware functional unit such as an LSI (Large Scale Integration) or an ASIC (Application Specific Integrated Circuit).

The acquisition unit 261 acquires the electrical signal output from the pulse wave sensor 12.

The reception unit 262 receives various information from the user.

The calculation unit 263 calculates various values calculated by the information processing device 20. For example, the calculation unit 263 calculates, based on certain waveform information, a time-frequency spectrum corresponding to a waveform indicated by the waveform information.

The prescription candidate output unit 264 generates the above-mentioned output information and outputs the generated output information. For example, the prescription candidate output unit 264 outputs the generated output information to the storage unit 22, and causes the storage unit 22 to store the output information.

The first learning unit 265 causes the first machine learning model to learn the first correspondence information.

The second learning unit 266 causes the second machine learning model to learn the second correspondence information.

The display control unit 267 generates various images. The display control unit 267 causes the display unit 25 to display the generated image.

The generation unit 268 generates waveform information indicating the waveform of the subject's pulse wave based on the electrical signal acquired by the acquisition unit 261. Further, the generation unit 268 generates a time-frequency spectrum image showing the time-frequency spectrum calculated by the calculation unit 263.

<Processing in which the information processing device generates waveform information>
Hereinafter, with reference to FIG. 4, a process in which the information processing device 20 generates waveform information will be described. FIG. 4 is a diagram illustrating an example of a process flow in which the information processing device 20 generates waveform information. In the following, as an example, the information processing apparatus 20 receives a first operation that causes the information processing apparatus 20 to start the process of generating waveform information at a timing before the process of step S110 shown in FIG. 4 is performed. I will explain the case where there is. In addition, in the following, as an example, the information processing device 20 receives an operation that specifies a period from the timing at which the first operation is received until the first time specified by the user has elapsed as the measurement period. I will explain the case when The first time is, for example, one minute, but may alternatively be shorter than one minute or longer than one minute. Moreover, below, as an example, a case will be described in which the pulse wave detection device 10 starts detecting the pulse wave of the subject S1 at the timing. Here, when the pulse wave detection device 10 starts detecting the pulse wave of the subject S1, the pulse wave sensor 12 starts outputting the above-mentioned electrical signal to the information processing device 20. Moreover, below, as an example, a case will be described in which the information processing device 20 receives patient identification information for identifying the patient S1 at the timing.

After receiving the first operation, the acquisition unit 261 acquires an electrical signal from the pulse wave sensor 12 at a predetermined sampling period within the measurement period until the first time elapses (step S110). Note that in the embodiment, the process of converting this electrical signal from an analog signal to a digital signal is a known process, and therefore a description thereof will be omitted. Here, the sampling period is, for example, 0.002 seconds, but instead may be a period shorter than 0.002 seconds or a period longer than 0.002 seconds. The acquisition unit 261 causes the storage unit 22 to store information indicating the electrical signal acquired within the measurement period in this manner.

Next, the generation unit 268 generates waveform information indicating the waveform of the pulse wave of the subject S1 within the measurement period based on the information indicating the electrical signal stored in the storage unit 22 in step S110 (step S120 ). Here, the method for generating the waveform information based on the information may be a known method or a method to be developed in the future. Further, the information processing device 20 may be configured to perform the process in step S120 in parallel with the process in step S110.

Next, the generation unit 268 causes the storage unit 22 to store the waveform information generated in step S120 (step S130). At this time, the generation unit 268 causes the storage unit 22 to store the subject identification information received in advance (that is, the subject identification information identifying the subject S1) in association with the waveform information. After the process of step S130 is performed, the generation unit 268 ends the process of the flowchart shown in FIG. 4.

In this way, the information processing device 20 can cause the storage unit 22 to store waveform information for each subject.

<Processing in which the information processing device causes the first machine learning model to learn the first correspondence information and causes the second machine learning model to learn the second correspondence information>
Hereinafter, with reference to FIG. 5, a process in which the information processing device 20 causes the first machine learning model to learn the first correspondence information and causes the second machine learning model to learn the second correspondence information will be described. FIG. 5 is a diagram illustrating an example of a process flow in which the information processing device 20 causes the first machine learning model to learn the first correspondence information and causes the second machine learning model to learn the second correspondence information. In the following, as an example, N pieces of waveform information generated by the process of the flowchart shown in FIG. 4 are stored in the storage unit 22 at a timing before the process of step S210 shown in FIG. 5 is performed. I will explain the case where there is. N may be any integer greater than or equal to 1. These N pieces of waveform information are waveform information for each of the N subjects. Therefore, for convenience of explanation, each of these N subjects will be referred to as a learning subject in the following description. Further, for convenience of explanation, the waveform information for each of the N learning subjects will be referred to as learning waveform information in the following description. That is, below, as an example, a case will be described in which learning waveform information indicating the waveform of the pulse wave of each of N learning subjects is stored in the storage unit 22 at the timing. Further, below, as an example, a case will be described in which N pieces of diagnosis result information are stored in the storage unit 22 at the timing. These N pieces of diagnostic result information are diagnostic result information for each of the N learning subjects. The diagnosis result information regarding a certain study subject is information indicating the diagnosis result of the study subject by a Chinese herbalist. In this case, diagnostic result information regarding a certain learning subject is associated with subject identification information that identifies the learning subject. Further, below, as an example, a case will be described in which N pieces of Chinese herbal medicine information are stored in the storage unit 22 at the timing. These N pieces of Chinese herbal medicine information are Chinese herbal medicine information for each of the N learning subjects. The Chinese herbal medicine information for a certain learning subject is information indicating each of one or more Chinese herbal medicines that a Chinese medicine doctor has prescribed to the learning subject. In this case, Chinese herbal medicine information regarding a certain study subject is associated with subject identification information that identifies the study subject.

The control unit 26 reads each of the N pieces of learning waveform information stored in advance in the storage unit 22 from the storage unit 22 (step S210). In addition, in FIG. 5, the process of step S210 is shown as "waveform information reading".

Next, the control unit 26 selects one piece of subject identification information associated with each of the N pieces of learning waveform information read out in step S210 as target subject identification information, and The processes of steps S230 to S270 are repeated for each examiner identification information (step S220). In addition, in FIG. 5, the process of step S220 is shown by "each subject identification information."

After the target patient identification information is selected in step S220, the first learning unit 265 refers to the N pieces of diagnostic result information stored in the storage unit 22 and identifies the target patient identified in step S220. Diagnosis result information associated with the information is read from the storage unit 22 (step S230).

Here, the diagnosis result information will be explained. As described above, the diagnosis result information regarding a certain study subject is information indicating the diagnosis result of the study subject by a Chinese medicine doctor. More specifically, the diagnosis result information is information indicating the results of the Chinese medicine doctor's diagnosis of the study subject (for example, the results of pulse diagnosis, etc.) for each of the M diagnosis items. M may be any integer as long as it is 1 or more. Therefore, each of the M diagnosis items includes a plurality of options that can be selected as the diagnosis result. Information indicating a diagnosis result for a certain diagnostic item among the M diagnostic items is represented by a vector having as components variables associated with each of a plurality of options included in the diagnostic item. Therefore, the diagnosis result information is represented by the direct sum of vectors indicating the diagnosis results of each of the M diagnosis items. For example, if a certain diagnostic item among M diagnostic items includes three options X1 to X3, and if X1 is selected by the Chinese herbalist as the diagnosis result of the diagnostic item, The information indicating the result has variables associated with each of X1 to X3 as components, 1 is assigned to the variable associated with X1, and 0 is assigned to the variables associated with each of X2 and X3. This is the vector being assigned. Diagnosis result information indicating the diagnosis results of each of the M diagnosis items is represented by the direct sum of these six vectors.

In the embodiment, the M diagnostic items are six major vein types in the 28 diseased veins, namely, floating veins, sinking veins, slow veins, few veins, ischemic veins, and real veins. In this case, M is 6. As a result of the diagnosis, the options that can be selected by the Chinese herbalist include the multiple options included in Floating Vein, Kui Vein (the Chinese character with a grass crown above the hole), Hong Vein, Lea vein, Wet Vein, There are six types of veins. Further, the plurality of options included in the sink vein as options that can be selected by the Chinese herbalist as a diagnosis result are four pulse types: sink vein, dip vein, weak vein, and dead vein. In addition, the plurality of options included in the slow reticular pulse as options that can be selected by the Chinese herbalist as a diagnosis result are five pulse types: slow pulse, bradycardia, astringent pulse, tubercles, and grand veins. Further, the plurality of options included in the number of pulses as options that can be selected by the Chinese herbalist as a diagnosis result are three pulse types: the number of pulses, the arterial, and the accelerated pulse. Furthermore, the plurality of options included in the ischemic pulse as options that can be selected by the Chinese herbalist as a diagnosis result are four pulse types: ischemic pulse, short vein, arteriole, and microvenule. Further, the plurality of options included in the real reticular vein as options that can be selected by the Chinese herbalist as a diagnosis result are six pulse types: real pulse, long pulse, chordal pulse, tense pulse, smooth pulse, and large vein. For example, during a certain learning session, the information showing the diagnosis result of a test subject's floating veins is: floating veins, go veins (kanji with a crown above the hole), hong veins, leather veins, wet veins, and scattered veins. This is a vector having variables associated with each as components. For example, in the variable associated with floating veins among the components of the vector, if floating veins are selected by the Chinese herbalist as the diagnosis result for the floating veins, 1 is assigned. In addition, in this case, among the components included in the vector, the variables associated with each of the components koume (kanji with a grass crown above the hole), ko vein, leather vein, wet vein, and san vein are set to 0. It has been assigned. Here, certain diagnosis result information is represented by the direct sum of vectors indicating the diagnosis results for each of the six major vein types. That is, in the embodiment, the dimension of a vector representing certain diagnostic result information is 28 dimensions. In the following, for convenience of explanation, a combination of pulse types associated with each of the six components to which one of the components of a vector representing certain diagnostic result information is assigned will be referred to as a target pulse type set. do. That is, the diagnosis result information is information indicating the target pulse type set. Note that other information may be associated with this diagnosis result information. In the following, as an example, information about the subject at the time of learning, information indicating the detection position of the pulse wave of the subject at the time of learning, and information indicating the past illness of the subject at the time of learning are associated with the diagnosis result information. We will explain the case where Here, the information regarding the learning subject includes, for example, information indicating the gender of the learning subject, information indicating the age of the learning subject, and height of the learning subject. information, and information indicating the weight of the subject at the time of learning.

Next, the first learning unit 265 refers to the N pieces of Chinese herbal medicine information stored in the storage unit 22 and extracts the Chinese herbal medicine information associated with the target patient identification information selected in step S220 from the storage unit 22. Read out (step S240).

Next, the calculation unit 263 selects the learning waveform information associated with the target subject identification information selected in step S220 from the learning waveform information read out in step S210, and selects the learning waveform information associated with the target subject identification information selected in step S220. A time-frequency spectrum is calculated according to the waveform indicated by the waveform information. Then, the generation unit 268 generates a time-frequency spectrum image indicating the time-frequency spectrum calculated by the calculation unit 263 (step S250). In the following, for convenience of explanation, a time-frequency spectrum image showing a time-frequency spectrum corresponding to a waveform indicated by the learning waveform information will be referred to as a learning time-frequency spectrum image.

Here, the time-frequency spectrum image will be explained. FIG. 6 is a diagram illustrating an example of a waveform indicated by the learning waveform information associated with the target subject identification information selected in step S220. The vertical axis of the graph shown in FIG. 6 indicates the signal amplitude of the pulse wave. Further, the horizontal axis of the graph indicates elapsed time. The curve plotted on the graph indicates the waveform. The calculation unit 263 removes frequency components equal to or higher than a predetermined frequency from the waveform indicated by the learning waveform information. The calculation unit 263 removes the frequency component using a bandpass filter. The information processing device 20 may be configured to accept a predetermined frequency from a user, or may be configured to accept a predetermined frequency by another method. The calculation unit 263 calculates a time-frequency spectrum based on information indicating the waveform after removing the frequency component. The calculation unit 263 calculates a time-frequency spectrum by STFT (Short Time Fourier Transform). The generation unit 268 then generates a time-frequency spectrum image showing the time-frequency spectrum calculated by the calculation unit 263. The time-frequency spectral image is, for example, a contour map in which spectral intensity is plotted as a function of time and frequency. FIG. 7 is a diagram showing an example of a time-frequency spectrum image. The contour map shown in FIG. 7 is an example of a time-frequency spectrum image for 5 seconds. The horizontal axis of the contour map indicates time. The vertical axis of the contour map indicates frequency. Each of the plurality of curves plotted on the contour map is a contour line regarding spectral intensity.

After the process of step S250 is performed, the first learning unit 265 generates first correspondence information (step S260). More specifically, the first learning unit 265 generates, as first correspondence information, information that associates the learning time-frequency spectrum image generated in step S250 with the diagnosis result information read out in step S230.

Next, the second learning unit 266 generates second correspondence information (step S270). More specifically, the second learning unit 266 generates, as second correspondence information, information that associates the diagnosis result information read in step S230 with the herbal medicine information read in step S240. The second learning unit 266 stores the generated second correspondence information in a second database stored in the storage unit 22 in advance. That is, the second database is a database that stores N pieces of second correspondence information generated in the repeated processing of steps S220 to S270 shown in FIG. Further, the second database is a database with a two-dimensional table structure of n1×m1. That is, the second database shows a two-dimensional table of n1×m1. n1 is the number of pulse type combinations that can be selected as the target pulse type set. For n1, the number of vein types included in the category of floating veins is 6, the number of vein types included in the category of sinking veins is 4, the number of vein types included in the category of slow veins is 5, Since the number of pulse types included in the category of phantom veins is 3, the number of vein types included in the category of virtual veins is 4, and the number of pulse types included in the category of real veins is 6, 8640 be. Furthermore, m1 is the number of types of Chinese medicine that the information processing device 20 can handle. The second learning unit 266 adds 1 to the value assigned to the field where the target vein type set and each of the one or more herbal medicines indicated by the herbal medicine information cross on the two-dimensional table. The second correspondence information obtained is stored in the second database. Note that 0 is assigned as an initial value to each field on the two-dimensional table. Moreover, FIG. 8 is an image diagram visualizing an example of the process of step S270.

After the process of step S270 is performed, the control unit 26 moves to step S220 and selects the next target subject identification information. Note that if there is no unselected subject identification information in step S220, the control unit 26 ends the repetitive processing of steps S220 to S270.

After the iterative process from step S220 to step S270 is completed, the first learning unit 265 causes the first machine learning model to learn the N pieces of first correspondence information stored in the first database in the iterative process (step S280 ). Here, the process of step S280 will be explained.

In step S280, the first learning unit 265 trains the first machine learning model for each of the M diagnostic items. Thereby, the first learning unit 265 can acquire the coefficient sequence of the learned first machine learning model from the learned first machine learning model for each of the M diagnostic items. In the embodiment, the M diagnostic items were each of the six major vein types, as described above. In this case, the first learning unit 265 learns the first machine learning model for each of the floating network, sinking network, slow network, few network, imaginary network, and real network. For example, when learning the first machine learning model for floating net veins, the first learning unit 265 calculates vectors indicating the diagnosis results of floating net veins and correspondences to these vectors from each of the N pieces of first correspondence information. The attached time-frequency spectrum image is extracted. In this case, for each extracted vector, the first learning unit 265 uses the time-frequency spectrum image associated with the vector as an input, and causes the first machine learning model to learn using the output as a vector. The first machine learning model that has learned such inputs and outputs is able to identify pulse types included in the category of floating veins when a time-frequency spectrum image corresponding to the waveform of a pulse wave of a certain subject is input. A vector indicating the type of pulse that is estimated to appear most strongly in the pulse wave of the subject is output as information indicating the diagnosis result regarding floating veins. Here, the first machine learning model is a CNN (Convolutional Neural Network) for deep learning. In training the first machine learning model with such inputs and outputs, the input time-frequency spectrum image is downsized by a kernel filter and nonlinear processing with one output, and finally becomes one-dimensional data. Compressed. In this learning, the weights and biases of the first machine learning model are determined based on the distribution state of this one-dimensional data based on the number of multiple options included in the diagnostic item (in this case, the number of vein types included in the category of floating veins). It is optimized so that it can be distinguished accurately to the number of 6). By performing such learning on the first machine learning model for each of the six major vein types, the first learning unit 265 calculates the coefficient string of the first machine learning model after learning for each of the M diagnostic items. can be obtained from the first machine learning model after learning. Here, the coefficient sequence of a certain first machine learning model is a combination of weights and biases of the first machine learning model. In the following, for convenience of explanation, the coefficient sequence of the first machine learning model after learning about floating veins will be referred to as a first coefficient sequence. Furthermore, for convenience of explanation, the coefficient sequence of the first machine learning model after learning about the sinking vein will be referred to as a second coefficient sequence in the following description. Further, in the following description, for convenience of explanation, the coefficient sequence of the first machine learning model after learning about the slow network pulse will be referred to as a third coefficient sequence. Furthermore, for convenience of explanation, the coefficient sequence of the first machine learning model after learning about the number network will be referred to as a fourth coefficient sequence. Furthermore, for convenience of explanation, the coefficient sequence of the first machine learning model after learning about the imaginary network will be referred to as the fifth coefficient sequence. Furthermore, for convenience of explanation, the coefficient sequence of the first machine learning model after learning about the real network will be referred to as the sixth coefficient sequence. Note that when acquiring each of the first to sixth coefficient sequences, the first learning unit 265 prepares a first machine learning model for each diagnostic item, and performs learning of the first machine learning model for each diagnostic item. The learning may be performed in parallel, or a single first machine learning model may be prepared and the learning of the first machine learning model may be performed sequentially for each diagnostic item. After acquiring each of the first coefficient sequence to the sixth coefficient sequence in step S280, the first learning unit 265 causes the storage unit 22 to store coefficient sequence information indicating each of the first coefficient sequence to the sixth coefficient sequence. Thereby, the information processing device 20 can quickly reproduce the first machine learning model after learning about the floating network based on the first coefficient sequence and the first machine learning model, for example.

Next, the second learning unit 266 causes the second machine learning model to learn the second database (that is, the second correspondence information) generated in the repeated processing of steps S220 to S270 (step S290). More specifically, when diagnosis result information indicating a target pulse type set for a certain subject is input, the second learning unit 266 selects one or more Chinese herbal medicines that are likely to be prescribed to the subject. The second database is trained by the second machine learning model so as to output Chinese herbal medicine candidate information indicating each of the Chinese herbal medicine candidates to be prescribed to the subject. The machine learning model used as the second machine learning model may be any type of machine learning model as long as it can realize such an input/output relationship. After the process of step S290 is performed, the control unit 26 ends the process of the flowchart shown in FIG.

As described above, the information processing device 20 can make the first machine learning model learn the first correspondence information, and can make the second machine learning model learn the second correspondence information.

<Processing in which the information processing device receives diagnosis result information>
Hereinafter, with reference to FIG. 9, a process in which the information processing device 20 receives diagnosis result information will be described. FIG. 9 is a diagram illustrating an example of the flow of processing in which the information processing device 20 receives diagnosis result information. Below, as an example, a case will be described in which the information processing device 20 receives diagnosis result information regarding the subject S2, who is one of the N learning subjects described above. Furthermore, as an example, a case will be described below in which the Chinese medicine doctor diagnoses the subject S2 at a timing before the process of step S310 shown in FIG. 9 is performed. Further, below, as an example, a case will be described in which the information processing device 20 receives a second operation that causes the information processing device 20 to start accepting diagnosis result information at the timing.

After the information processing device 20 receives the second operation, the display control unit 267 generates the information reception image PCT1 (step S310).

Here, the information reception image PCT1 is an image in which the information processing device 20 receives diagnosis result information. FIG. 10 is a diagram showing an example of the information reception image PCT1. The information reception image PCT1 includes, for example, eight images, the first reception image G1 to the eighth reception image G8. Note that the information reception image PCT1 may include other images in addition to these eight images.

The first reception image G1 is a GUI that receives patient identification information. The first reception image G1 includes, for example, an input field into which patient identification information is input.

The second reception image G2 is a GUI that receives information indicating the gender of the subject. The second reception image G2 includes two radio buttons, for example, one that accepts information indicating that the gender of the examinee is male, and the other that accepts information that the gender of the examinee is female. Contains radio buttons.

The third reception image G3 is a GUI that accepts information indicating the age of the subject. The third reception image G3 includes, for example, an input field into which information indicating the age of the subject is input.

The fourth reception image G4 is a GUI that receives information indicating the height of the subject. The fourth reception image G4 includes, for example, an input field into which information indicating the height of the subject is input.

The fifth reception image G5 is a GUI that receives information indicating the subject's weight. The fifth reception image G5 includes, for example, an input field into which information indicating the subject's weight is input.

The sixth reception image G6 is a GUI that receives information indicating the detection position of the subject's pulse wave. The sixth reception image G6 includes, for example, a radio button that accepts information indicating that the arm in which the pulse wave was detected is the left arm among both arms of the subject, A radio button that accepts information indicating that the arm where the test subject's pulse wave was detected is the right arm, a radio button that accepts information that the position where the test subject's pulse wave was detected is the right arm, and a radio button that accepts information that the test subject's pulse wave was detected at the position where the test subject's pulse wave was detected. This includes a radio button that accepts information indicating that the detected position is Seki, and a radio button that accepts information that indicates that the position where the pulse wave of the subject was detected is Shaku.

The seventh reception image G7 is a GUI that receives information indicating the subject's past illnesses. The seventh reception image G7 includes, for example, an input field into which information indicating a medical history of the subject is input.

The eighth reception image G8 is a GUI that receives diagnosis result information. The eighth reception image G8 includes, for example, six images, reception image G81 to reception image G86.

The reception image G81 is a GUI that accepts one of the six vein types included in the category of floating veins. When the reception image G81 is selected, a drop-down menu L81 in which information indicating each of the six pulse types is listed is displayed. When the drop-down menu L81 is displayed, the user of the information processing device 20 can perform an operation to select any one of the information indicating each of the six pulse types listed in the drop-down menu L81. In the information reception image PCT1, when an operation is performed to select any of the information indicating each of the six pulse types listed in the drop-down menu L81, the display of the drop-down menu L81 disappears and the reception image G81 The information selected in the drop-down menu L81 is displayed in the display field.

The reception image G82 is a GUI that accepts one of the four vein types included in the category of sedimentary veins. When the reception image G82 is selected, a drop-down menu L82 in which information indicating each of the four pulse types is listed is displayed. When the drop-down menu L82 is displayed, the user of the information processing device 20 can perform an operation to select any one of the information indicating each of the four pulse types listed in the drop-down menu L82. In the information reception image PCT1, when an operation is performed to select any of the information indicating each of the four pulse types listed in the drop-down menu L82, the display of the drop-down menu L82 disappears and the reception image G82 The information selected in the drop-down menu L82 is displayed in the display field.

The reception image G83 is a GUI that accepts one of the five pulse types included in the category of slow retinal pulse. When the reception image G83 is selected, a drop-down menu L83 in which information indicating each of the five pulse types is listed is displayed. When the drop-down menu L83 is displayed, the user of the information processing device 20 can perform an operation to select any one of the information indicating each of the five pulse types listed in the drop-down menu L83. In information reception image PCT1, when an operation is performed to select any of the information indicating each of the five pulse types listed in drop-down menu L83, the display of drop-down menu L83 disappears and reception image G83 The information selected in the drop-down menu L83 is displayed in the display field.

The reception image G84 is a GUI that accepts one of the three types of pulses included in the category of multiple veins. When the reception image G84 is selected, a drop-down menu L84 in which information indicating each of the three pulse types is listed is displayed. When the drop-down menu L84 is displayed, the user of the information processing device 20 can perform an operation to select any one of the information indicating each of the three pulse types listed in the drop-down menu L84. In the information reception image PCT1, when an operation is performed to select any of the information indicating each of the three pulse types listed in the drop-down menu L84, the display of the drop-down menu L84 disappears and the reception image G84 The information selected in the drop-down menu L84 is displayed in the display field.

The reception image G85 is a GUI that accepts one of the four pulse types included in the category of ischemic pulse. When the reception image G85 is selected, a drop-down menu L85 in which information indicating each of the four pulse types is listed is displayed. When the drop-down menu L85 is displayed, the user of the information processing device 20 can perform an operation to select any one of the information indicating each of the four pulse types listed in the drop-down menu L85. In the information reception image PCT1, when an operation is performed to select any one of the information indicating each of the four pulse types listed in the drop-down menu L85, the display of the drop-down menu L85 disappears and the reception image G85 The information selected in the drop-down menu L85 is displayed in the display field.

The reception image G86 is a GUI that accepts one of the six vein types included in the category of real veins. When the reception image G86 is selected, a drop-down menu L86 is displayed in which information indicating each of the six pulse types is listed. When the drop-down menu L86 is displayed, the user of the information processing device 20 can perform an operation to select any one of the information indicating each of the six pulse types listed in the drop-down menu L86. In the information reception image PCT1, when an operation is performed to select any of the information indicating each of the six pulse types listed in the drop-down menu L86, the display of the drop-down menu L86 disappears and the reception image G86 The information selected in the drop-down menu L86 is displayed in the display field.

Note that FIG. 11 is a diagram showing an example of how each of the six drop-down menus described above is displayed. Here, the information reception image PCT1 may have a configuration that does not include some or all of the second reception image G2 to the seventh reception image G7.

After the process of step S310 is performed, the display control unit 267 displays the information reception image PCT1 generated in step S310 on the display unit 25 (step S320).

Next, the receiving unit 262 waits until the information processing device 20 receives an operation via the information receiving image PCT1 displayed on the display unit 25 in step S320 (step S330).

When the information processing device 20 receives an operation via the information acceptance image PCT1 displayed on the display unit 25 in step S320 (step S330-YES), the reception unit 262 receives the diagnosis result information via the information acceptance image PCT1. It is determined whether the operation to end the reception was accepted in step S330 (step S340).

If the reception unit 262 determines in step S330 that the operation to end the reception of diagnosis result information via the information reception image PCT1 has not been accepted (step S340-NO), the reception unit 262 performs processing according to the operation received in step S330. (Step S370). Here, the processing is, for example, processing in which the information processing device 20 receives diagnosis result information via the eighth reception image G8. That is, through the process of step S370, the information processing device 20 receives various types of information included in the diagnosis result information regarding the subject S2. Note that the process corresponding to the operation received in step S330 may be any process as long as it can be performed according to the operation received via the information reception image PCT1. After the process of step S370 is performed, the reception unit 262 moves to step S330 and waits again until the information processing device 20 accepts the operation via the information reception image PCT1 displayed on the display unit 25 in step S320. .

On the other hand, if the receiving unit 262 determines that the operation to end the reception of diagnosis result information via the information receiving image PCT1 has been received in step S330 (step S340-YES), the receiving unit 262 receives the receiving images G81 to G81 of the information receiving image PCT1. The target vein type set is specified based on the information received through each G86. Then, the reception unit 262 generates diagnostic result information indicating the specified target pulse type group (step S350). At this time, the reception unit 262 associates the generated diagnosis result information with the information received via each of the first reception image G1 to seventh reception image G7 of the information reception image PCT1.

Next, the reception unit 262 stores the diagnosis result information generated in step S350 in the storage unit 22 (step S360), and ends the process of the flowchart shown in FIG.

As described above, the information processing device 20 can receive diagnosis result information.

<Processing in which the information processing device outputs Chinese medicine candidate information>
Hereinafter, with reference to FIG. 12, a process in which the information processing device 20 outputs Chinese herbal medicine candidate information will be described. FIG. 12 is a diagram showing an example of a process flow in which the information processing device 20 outputs Chinese herbal medicine candidate information. In the following, as an example, the first waveform information generated by the process of the flowchart shown in FIG. 4 is stored in the storage unit 22 at a timing before the process of step S410 shown in FIG. 12 is performed. Let me explain the case. The first waveform information is waveform information indicating the waveform of the pulse wave of the subject S3. In addition, as an example, in the following, at the timing, the third operation for causing the information processing device 20 to start the process of outputting the Chinese herbal medicine candidate information, and the test subject identifying the subject to whom the Chinese herbal medicine candidate information is to be provided. A case will be described in which the information processing device 20 receives patient identification information that identifies the patient S3 as the identification information. Further, below, as an example, a case will be described in which coefficient string information is stored in the storage unit 22 by the process of the flowchart shown in FIG. 5 at the timing.

After the information processing device 20 receives the third operation and the subject identification information, the calculation unit 263 reads the first waveform information from the storage unit 22 based on the received subject identification information (step S410). In FIG. 12, the process of step S410 is shown as "waveform information reading".

Next, the calculation unit 263 calculates a time-frequency spectrum according to the waveform indicated by the first waveform information read in step S410. Then, the generation unit 268 generates a time-frequency spectrum image indicating the time-frequency spectrum calculated by the calculation unit 263 (step S420).

Next, the prescription candidate output unit 264 reads out coefficient string information stored in advance in the storage unit 22 from the storage unit 22. Then, the prescription candidate output unit 264 determines whether or not the patient is to be examined based on each of the first to sixth coefficient sequences indicated by the read coefficient sequence information and the time-frequency spectrum image generated by the generation unit 268 in step S420. Candidates for Chinese herbal medicine to be prescribed in S3 are identified (step S430). In FIG. 12, the process of step S430 is shown as "Identification of Chinese medicine candidate." Here, the process of step S430 will be explained.

For example, the prescription candidate output unit 264 performs learning about floating veins based on the first coefficient sequence among the first to sixth coefficient sequences indicated by the read coefficient sequence information and the first machine learning model. The following first machine learning model is reproduced. The prescription candidate output unit 264 inputs the time-frequency spectrum image generated by the generation unit 268 in step S420 as an input to the reproduced first machine learning model. The first machine learning model into which the time-frequency spectrum image is input calculates the likelihood indicating the likelihood of each of the six options included in the floating mesh vein as a diagnosis result of the floating mesh vein, and A vector indicating the option with the highest calculated likelihood is output as a vector indicating the diagnosis result of floating net veins for person S3. FIG. 13 is a diagram showing an example of the likelihood for each of a plurality of options included in a diagnostic item such as floating veins. The prescription candidate output unit 264 acquires the vector output from the first machine learning model. The prescription candidate output unit 264 performs such processing for each of the first to sixth coefficient sequences. Thereby, the prescription candidate output unit 264 can acquire vectors indicating the diagnosis results of each of the six major vein types for the subject S3 from the first machine learning model. Then, the prescription candidate output unit 264 generates the direct sum of the six vectors outputted by the first machine learning model as diagnosis result information regarding the subject S3. In the following, for convenience of explanation, the vector indicating the diagnosis result information regarding the subject S3 will be referred to as a vector Y. Further, in the following, for convenience of explanation, each of the 28 components of the vector Y is indicated by y1 to y28. In this case, the components included in the vector indicating the diagnosis result of floating net vein are y1 to y6, the components included in the vector indicating the diagnosis result of sinking net vein are y7 to y10, and the components included in the vector indicating the diagnosis result of slow net vein are y1 to y6. The components included in the vector shown are y11 to y15, the components included in the vector showing the diagnosis result of imaginary network are y16 to y18, and the components included in the vector showing the diagnosis result of imaginary network are y19 to y22. , and the components included in the vector indicating the diagnosis result of the real network are y23 to y28.

After the first machine learning model outputs the vector Y, the prescription candidate output unit 264 inputs the vector Y output from the first machine learning model to the second machine learning model as an input. As mentioned above, when the second machine learning model receives the diagnosis result information output by the first machine learning model, it estimates that one or more Chinese herbal medicines are likely to be prescribed to the subject corresponding to the diagnosis result information. Chinese herbal medicine candidate information indicating each of one or more Chinese herbal medicine candidates is output. Therefore, when the second machine learning model receives the vector Y output from the first machine learning model, it selects one or more Chinese herbal medicine candidates that are estimated to be plausible as one or more Chinese herbal medicines to be prescribed to the subject S3. Outputs Chinese herbal medicine candidate information indicating each of the following. Specifically, when the vector Y is input, the second machine learning model calculates the second Among the fields associated with the target pulse type set indicated by the vector Y in the two-dimensional table indicated by the second database, one or more fields to which a value equal to or greater than a predetermined first threshold is assigned are each identified. Then, the second machine learning model selects the Chinese herbal medicines associated with each of the identified one or more fields as one or more Chinese herbal medicine candidates that are estimated to be plausible as the one or more Chinese medicines to be prescribed to the subject S3. Identify. Here, the first threshold value may be any value as long as it is greater than zero. The first threshold value is determined, for example, based on prior experimental results or the like, so as to increase the accuracy of identifying one or more Chinese herbal medicine candidates that are estimated to be plausible as one or more Chinese herbal medicines to be prescribed to the subject S3. After identifying one or more Chinese herbal medicine candidates that are estimated to be plausible as one or more Chinese herbal medicines to be prescribed to the subject S3, the second machine learning model generates Chinese herbal medicine candidate information indicating each of the identified one or more Chinese medicine candidates. Output. Here, FIG. 14 is an image diagram visualizing the flow in which the second machine learning model identifies one or more Chinese herbal medicine candidates that are estimated to be plausible as one or more Chinese herbal medicines to be prescribed to the subject S3.

After the second machine learning model outputs one or more Chinese herbal medicine candidate information, the prescription candidate output unit 264 outputs the Chinese herbal medicine candidate indicated by each of the one or more Chinese medicine candidate information outputted by the second machine learning model to the patient. One or more Chinese herbal medicines that are estimated to be plausible as one or more Chinese herbal medicines to be prescribed to person S3 are identified as candidates. As described above, the prescription candidate output unit 264 identifies the one or more Chinese herbal medicine candidates in step S430.

After the process of step S430 is performed, the prescription candidate output unit 264 generates output information including herbal medicine candidate information indicating each of the one or more herbal medicine candidates identified in step S430. The output information may include any information in addition to the herbal medicine candidate information indicating each of the one or more herbal medicine candidates. The prescription candidate output unit 264 outputs the generated output information to the display control unit 267. Thereby, the display control unit 267 generates, for example, an image including the output information output from the prescription candidate output unit 264. Then, the display control unit 267 displays the generated image on the display unit 25 (step S440), and ends the process of the flowchart shown in FIG. 12. In addition, in FIG. 12, the process of step S440 is shown by "Chinese medicine candidate information display".

As described above, the information processing device 20 generates output information including Chinese herbal medicine candidate information indicating candidates for Chinese herbal medicine to be prescribed to the subject S3 based on the first waveform information indicating the waveform of the pulse wave of the subject S3. Output. Thereby, the information processing device 20 can identify candidates for Chinese herbal medicines to be prescribed to the subject S3 without having the Chinese herbalist perform a pulse diagnosis of the subject S3. As a result, the information processing device 20 can reduce the effort required for the Chinese medicine doctor to prescribe Chinese medicine to the subject S3. Further, thereby, the information processing device 20 can efficiently suppress variations in treatment results depending on the skill level of the Chinese medicine doctor.

<Modification 1 of the embodiment>
Modification 1 of the embodiment will be described below. In Modification 1 of the embodiment, the diagnosis result information for a certain subject includes information indicating the diagnosis result of at least one of an interview, a facial examination, a tongue examination, and an abdominal examination of the subject by a Chinese medicine doctor. include. Below, as an example, a case will be described in which the diagnosis result information for a certain subject includes information indicating the diagnosis result of an interview of the subject by a Chinese medicine doctor. In this case, the diagnostic result information for a certain subject includes, for example, information about the subject, information indicating the detection position of the subject's pulse wave, and information indicating the subject's past diseases. ing. In other words, each of the information regarding the subject, the information indicating the detection position of the subject's pulse wave, and the information indicating the subject's past illness indicates the diagnosis result of the interview of the subject by the Chinese medicine doctor. This is an example of information. In this case, the vector Y described above includes 10 components y29 to y38 in addition to the 28 components y1 to y28 described above. Here, y29 is a variable to which 1 is assigned when the gender of the subject is male, and 0 is assigned when the gender of the subject is female. y30 is a variable to which 1 is assigned when the gender of the subject is female, and 0 is assigned when the gender of the subject is male. y31 is a variable to which the age of the subject is substituted. y32 is a variable to which the height of the subject is substituted. y33 is a variable to which the subject's weight is substituted. y34 is 1 if the arm from which the pulse wave was detected is the left arm, and y34 is 1 if the arm from which the pulse wave was detected is the left arm; This is a variable to which 0 is assigned if it is a right arm. y34 is a variable to which 0 is substituted when the right arm is the one in which the pulse wave has been detected out of both arms of the subject. y35 is a variable to which 1 is substituted when the position where the pulse wave of the subject is detected is Sun, and 0 is substituted when the position where the pulse wave of the subject is detected is Seki or Shaku. y36 is a variable to which 1 is substituted when the position where the pulse wave of the subject is detected is Seki, and 0 is substituted when the position where the pulse wave of the subject is detected is Sun or Shaku. y37 is a variable to which 1 is assigned when the position where the pulse wave of the subject is detected is shaku, and 0 is assigned when the position where the pulse wave of the subject is detected is sun or seki. y38 is a variable to which 1 is assigned when the subject has a pre-existing disease, and 0 is substituted when the subject does not have a pre-existing disease. Note that the vector Y according to the first modification of the embodiment may include a portion of y29 to y38.

In this case, in step S350 shown in FIG. 10, the reception unit 262 identifies the target pulse type group based on the information received through each of the reception images G81 to G86 of the information reception image PCT1. Then, the reception unit 262 generates, as diagnosis result information, information including diagnosis result information indicating the specified target pulse type set and information received via each of the second reception image G2 to seventh reception image G7. do. At this time, the reception unit 262 associates the generated diagnosis result information with the patient identification information received via the first reception image G1 of the information reception image PCT1.

Furthermore, in this case, the aforementioned second database is a database with a two-dimensional table structure of (n1×n2)×m1. Here, n2 is the number of combinations of 10 values assigned to each of y29 to y38. Thereby, the information processing device 20 can obtain information indicating the target pulse type set of a certain subject, information regarding the subject, information indicating the detection position of the subject's pulse wave, and past medical conditions of the subject. Based on the combination of information shown, output information including Chinese herbal medicine candidate information indicating candidates for Chinese medicine prescribed to the subject can be output. As a result, the information processing device 20 can increase the accuracy of identifying candidates for Chinese herbal medicines to be prescribed to the subject, and reduce the effort required for the Chinese herbalist to prescribe Chinese medicines to the subject S3. It can definitely be reduced.

<Modification 2 of embodiment>
Modification 2 of the embodiment will be described below. In the second modification of the embodiment, the first machine learning model performs error correction processing in step S430 shown in FIG. 12. For example, as described above, the prescription candidate output unit 264 generates a formula based on the first coefficient sequence of the first to sixth coefficient sequences indicated by the coefficient sequence information read in step S430 and the first machine learning model. , reproduce the first machine learning model after learning about floating veins. Then, the prescription candidate output unit 264 inputs the time-frequency spectrum image generated by the generation unit 268 in step S420 as an input to the reproduced first machine learning model. The first machine learning model to which the time-frequency spectrum image is input calculates a likelihood indicating the likelihood of each of the six options included in the floating mesh vein as a diagnosis result of the floating mesh vein. At this time, the first machine learning model performs error processing if all six calculated likelihoods are less than a predetermined threshold. The error processing is, for example, a process of replacing these six likelihoods with the average value of these six likelihoods. When the first machine learning model performs error processing in this case, it outputs a vector in which 1 is substituted for all components as a vector indicating the diagnosis result of floating web veins for the subject S3. On the other hand, when at least one of the six calculated likelihoods is equal to or higher than a predetermined threshold, the first machine learning model uses the vector as a vector indicating the diagnosis result of floating veins for the subject S3. A vector indicating the option with the highest calculated likelihood is output. The prescription candidate output unit 264 acquires the vector output from the first machine learning model. The prescription candidate output unit 264 performs such processing for each of the first to sixth coefficient sequences. Here, for example, if at least one of the six vectors acquired from the first machine learning model is a vector in which 1 is assigned to all components, the prescription candidate output unit 264 outputs the formula shown in FIG. End the flowchart processing. Then, the display control unit 267 causes the display unit 25 to display information indicating that an error has occurred, as well as information prompting to reacquire the waveform information indicating the waveform of the pulse wave of the subject S3. Note that the above error processing may be replaced with other processing such as setting an error flag and not storing the error flag. Further, even if at least one of the six vectors acquired from the first machine learning model is a vector in which 1 is assigned to all components, the prescription candidate output unit 264 outputs the formula as shown in FIG. The processing in the flowchart may be continued without ending. In this case, n1 of the two-dimensional table indicated by the second database is not the number of pulse type combinations that can be selected as the target pulse type set, but the number of pulse type combinations that can be selected as a combination of six or more pulse types from among the 28 diseased vein types. Replaced by the number of vein type combinations. Further, in this case, the diagnosis result information is replaced with information indicating a combination of six or more pulse types, that is, a vector indicating a combination of six or more pulse types. By performing these replacements, even in this case, the information processing device 20 can continue the processing of the flowchart shown in FIG. 12 without ending it.

<Variation 3 of the embodiment>
Modification 3 of the embodiment will be described below. In the third modification of the embodiment, the first machine learning model outputs the vector Z indicating the likelihood calculated for each of the 28 diseased pulses together with the vector Y in step S430 shown in FIG. 12 as diagnosis result information. It may be a configuration. In this case, in step S430, the prescription candidate output unit 264 inputs the vector Y output from the first machine learning model to the second machine learning model, thereby outputting one or more Chinese herbal medicines output from the second machine learning model. Based on the combination of candidates and the vector Z output from the first machine learning model, the values of each field of the two-dimensional table indicated by the second correspondence information are updated. Specifically, the prescription candidate output unit 264 outputs the likelihood indicated by the vector Z to the value of the field where each of the 28 disease veins and the one or more herbal medicine candidates cross each other on the two-dimensional table. Add degrees. For example, the prescription candidate output unit 264 adds the likelihood of floatation among the likelihoods indicated by the vector Z to the value of the field where floatation and the one or more Chinese medicine candidates cross each other. As a result, the information processing device 20 decreases the accuracy of identifying one or more Chinese herbal medicine candidates that are estimated to be plausible as one or more Chinese herbal medicines to be prescribed to the subject S3 due to variations in the diagnostic accuracy of the Chinese medicine doctor's vein type. You can prevent it from happening. In this case, the first machine learning model is configured to estimate at least one of the factor ranking and the degree of contribution of the Chinese medicine prescription determining factors for each of the six major vein types obtained by analyzing the second correspondence information. Good too. In this case, the first machine learning model can, for example, multiply the likelihood calculated for each of the six major vein types by a weight based on at least one of the estimated factor rank and the contribution. Here, this weight is a weight normalized so that the weights multiplied by the likelihoods calculated for each of the six major vein types add up to 1. For example, the first machine learning model determines the weight to be multiplied by the likelihood of each of the six options included in the floating net based on at least one of the factor ranking and the contribution degree estimated for the floating net. The weight becomes larger as each of the factor ranking and the degree of contribution become larger. After determining the weight, the first machine learning model multiplies the likelihood of each of the six options included in the floating network by the determined weight. As a result, the information processing device 20 decreases the accuracy of identifying one or more Chinese herbal medicine candidates that are estimated to be plausible as one or more Chinese herbal medicines to be prescribed to the subject S3 due to variations in the diagnostic accuracy of the Chinese medicine doctor's vein type. It is possible to more reliably prevent this from happening.

<Modification 4 of embodiment>
Modification 4 of the embodiment will be described below. In the fourth modification of the embodiment, the M diagnostic items are each of the six diagnostic items in the Japanese pulse diagnosis method, instead of the 28 diseased pulses. In this case as well, M is 6. Here, these six diagnostic items are the strength between floating pulses and sinking pulses, the strength between several pulses and slow pulses, the strength between large and small veins, and the strength between ischemic pulses and real pulses. The strength is the strength between the pulses, the strength between the tense and slow pulses, and the strength between the smooth and astringent pulses. In the Japanese pulse diagnosis method, floating veins and sinking veins have an opposing relationship. For this reason, Chinese herbalists who use the Japanese pulse diagnosis method diagnose whether floating pulses or sinking pulses appear more strongly in the subject's pulse wave. This is a diagnosis of the strength between floating veins and sinking veins. Similarly, in the Japanese pulse diagnosis method, slow pulse and slow pulse, large pulse and small pulse, ischemic pulse and real pulse, tense pulse and brady pulse, and smooth pulse and astringent pulse are opposed to each other. For this reason, in the Japanese pulse diagnosis method, there are differences in strength between floating pulse and sinking pulse, strength between several pulses and slow pulse, strength and weakness between large and small pulses, and weak and weak pulses between ischemic and real pulses. The strength between the two, the strength between the tense and slow pulses, and the strength between the smooth and astringent pulses are each diagnosed as six diagnostic items. Below, as an example, a case will be described in which each of these six diagnostic items is diagnosed by the Chinese herbalist as one of five levels from level 1 to level 5. In this case, a plurality of options included in a certain diagnostic item among these six diagnostic items are levels 1 to 5, respectively. In this case, the strength between the floating veins and the sinking veins indicates that the lower the level value, the stronger the floating veins, and the higher the level value, the stronger the sinking veins. Further, in this case, the strength between the several pulses and the slow pulse indicates that the lower the level value is, the stronger the several pulses are, and the higher the level value is, the stronger the slow pulse is. Further, in this case, regarding the strength between the large vein and the small vein, the lower the level value, the stronger the large vein, and the higher the level value, the stronger the small vein. Further, in this case, the strength between the ischemic pulse and the real pulse indicates that the lower the level value is, the stronger the isty pulse is, and the higher the level value is, the stronger the real pulse is. Further, in this case, regarding the intensity between the tense pulse and the bradycardia, the lower the level value, the stronger the bradycardia, and the higher the level value, the stronger the bradycardia. Further, in this case, regarding the strength between the smooth vein and the astringent vein, the lower the level value, the stronger the smooth vein, and the higher the level value, the stronger the astringent vein. For example, information indicating a diagnosis result regarding the strength of a floating pulse and a sinking pulse of a certain learning subject is a vector having variables associated with each of levels 1 to 5 as components. For example, if level 1 is selected by the Chinese herbalist as a diagnostic result of the strength between floating veins and sinking veins, the variable associated with level 1 among the components included in the vector is 1. It has been assigned. Furthermore, in this case, 0 is assigned to variables associated with each of levels 2 to 5 among the components of the vector. Here, the diagnosis result information according to the fourth modification of the embodiment is represented by the direct sum of vectors indicating the diagnosis results for each of these six diagnosis items. That is, in the fourth modification of the embodiment, the dimension of a vector representing certain diagnostic result information is 30 dimensions. In a fourth modification of the embodiment, the diagnosis result information indicates a target level group instead of a target pulse type group.

The information processing device 20 receives diagnosis result information indicating such a target level group via the information reception image PCT2 instead of the information reception image PCT1 shown in FIG. Therefore, the display control unit 267 generates the information reception image PCT2 in step S310 shown in FIG. Here, FIG. 15 is a diagram showing an example of the information reception image PCT2.

For example, instead of the eighth reception image G8, the information reception image PCT2 includes a ninth reception image G9 together with the first reception image G1 to the seventh reception image G7. Note that the information reception image PCT2 may include other images in addition to these images. Further, since the first reception image G1 to the seventh reception image G7 have already been explained, their explanation will be omitted.

The ninth reception image G9 is a GUI that receives diagnosis result information. The ninth reception image G9 includes, for example, six images, reception image G91 to reception image G96.

The reception image G91 is a GUI that accepts the value of the strength level between the floating vein and the sinking vein. In the reception image G91, five radio buttons are arranged in a line between information indicating a floating vein and information indicating a sinking vein. These five radio buttons are arranged in order from the radio button closest to the information indicating floating veins to the information indicating sinking veins: the radio button associated with level 1, the radio button associated with level 2, and the radio button associated with level 3. , a radio button associated with level 4, and a radio button associated with level 5. When an operation is performed to select any one of these five radio buttons, information indicating that the radio button has been selected is superimposed on the radio button selected by the operation. Thereby, the information processing device 20 receives the value of the level associated with the radio button selected by the operation as the value of the level of strength between the floating vein and the sinking vein. For example, when receiving an operation to select a radio button associated with level 1 in the reception image G91, the information processing device 20 accepts level 1 as the value of the level of strength between floating veins and sinking veins. .

The reception image G92 is a GUI that accepts the value of the strength level between the slow pulse and the slow pulse. In the reception image G92, five radio buttons are arranged in a line between information indicating a slow pulse and information indicating a slow pulse. These five radio buttons are arranged in order from the radio button closest to the information indicating slow pulse to the information indicating slow pulse: the radio button associated with level 1, the radio button associated with level 2, and the radio button associated with level 3. , a radio button associated with level 4, and a radio button associated with level 5. When an operation is performed to select any one of these five radio buttons, information indicating that the radio button has been selected is superimposed on the radio button selected by the operation. Thereby, the information processing device 20 receives the value of the level associated with the radio button selected by the operation as the value of the level of intensity between the slow pulse and the slow pulse. For example, when receiving an operation to select a radio button associated with level 1 in the reception image G92, the information processing device 20 accepts level 1 as the strength level value between multiple pulses and slow pulses. .

The reception image G93 is a GUI that accepts the value of the strength level between the large vein and the small vein. In the reception image G93, five radio buttons are arranged in a line between information indicating large veins and information indicating small veins. These five radio buttons are arranged in order from the radio button closest to the information indicating the large vein to the information indicating the small vein: the radio button associated with level 1, the radio button associated with level 2, and the radio button associated with level 3. , a radio button associated with level 4, and a radio button associated with level 5. When an operation is performed to select any one of these five radio buttons, information indicating that the radio button has been selected is superimposed on the radio button selected by the operation. Then, the information processing device 20 receives the value of the level associated with the radio button selected by the operation as the value of the level of strength between the large and small veins. For example, when receiving an operation to select a radio button associated with level 1 in the reception image G93, the information processing device 20 accepts level 1 as the value of the level of strength between the large and small veins. .

The reception image G94 is a GUI that accepts the value of the strength level between the ischemic pulse and the real pulse. In the reception image G94, five radio buttons are arranged in a line between information indicating an ischemic pulse and information indicating a real pulse. These five radio buttons are arranged in order from the radio button closest to the information indicating the ischemic pulse to the information indicating the actual pulse: the radio button associated with level 1, the radio button associated with level 2, and the radio button associated with level 3. , a radio button associated with level 4, and a radio button associated with level 5. When an operation is performed to select any one of these five radio buttons, information indicating that the radio button has been selected is superimposed on the radio button selected by the operation. Then, the information processing device 20 receives the value of the level associated with the radio button selected by the operation as the value of the level of strength between the ischemic pulse and the real pulse. For example, when receiving an operation to select a radio button associated with level 1 in the reception image G94, the information processing device 20 accepts level 1 as the value of the level of strength between the ischemic pulse and the real pulse. .

The reception image G95 is a GUI that accepts the value of the level of strength between systolic and bradycardia. In the reception image G95, five radio buttons are arranged in a line between information indicating a tense heartbeat and information indicating a slow heartbeat. These five radio buttons are arranged in order from the radio button closest to the information indicating rapid pulse to the information indicating brady pulse: the radio button associated with level 1, the radio button associated with level 2, and the radio button associated with level 3. , a radio button associated with level 4, and a radio button associated with level 5. When an operation is performed to select any one of these five radio buttons, information indicating that the radio button has been selected is superimposed on the radio button selected by the operation. Thereby, the information processing device 20 receives the value of the level associated with the radio button selected by the operation as the value of the level of strength between the strong and weak pulses. For example, when the information processing device 20 receives an operation to select a radio button associated with level 1 in the reception image G95, it accepts level 1 as the value of the level of strength between tense and slow pulses. .

The reception image G96 is a GUI that accepts the value of the strength level between the smooth vein and the astringent vein. In the reception image G96, five radio buttons are arranged in a line between information indicating smooth pulse and information indicating slow pulse. These five radio buttons are arranged in order from the radio button closest to the information indicating smooth pulse to the information indicating difficult pulse: the radio button associated with level 1, the radio button associated with level 2, and the radio button associated with level 3. , a radio button associated with level 4, and a radio button associated with level 5. When an operation is performed to select any one of these five radio buttons, information indicating that the radio button has been selected is superimposed on the radio button selected by the operation. Thereby, the information processing device 20 receives the value of the level associated with the radio button selected by the operation as the value of the level of strength between the smooth pulse and the astringent pulse. For example, when receiving an operation to select a radio button associated with level 1 in reception image G96, the information processing device 20 accepts level 1 as the value of the level of strength between smooth pulse and astringent pulse. .

The information processing device 20 can receive the diagnosis result information according to the fourth modification of the embodiment via the information reception image PCT2 having such a configuration. Then, the information processing device 20 performs the process of the flowchart shown in FIG. 5 using the diagnosis result information received via the information reception image PCT2 having such a configuration, thereby obtaining the first result as in the embodiment. First correspondence information and second correspondence information are generated, and a first machine learning model is trained based on the generated first correspondence information, and a second machine learning model is trained based on the generated second correspondence information, respectively. As a result, the information processing device 20 can perform the process shown in the flowchart shown in FIG. 12 similarly to the embodiment, reducing the effort required for the Chinese herbalist to prescribe the Chinese herbal medicine to the first subject. I can do it. Note that n1 in the second database in which the second correspondence information according to the fourth modification of the embodiment is stored is the number of combinations of level values that can be selected as the target level set. And n1 is the strength between floating veins and sinking veins, the strength between several pulses and slow pulses, the strength between large veins and small veins, the strength and weakness between weak pulses and real pulses, and tension. Since the number of levels included in each of the strength between the pulse and slow pulse and the strength between smooth pulse and astringent pulse is 5, the number is 15,625.

<Variation 5 of the embodiment>
Modification 5 of the embodiment will be described below. Modification 5 of the embodiment is a modification of Modification 4 of the embodiment. In modification 5 of the embodiment, the first learning unit 265 analyzes each of the N pieces of diagnosis result information read out in step S230 shown in FIG. Correct by weight vector.

Hereinafter, this correction will be specifically explained. In modification 5 of the embodiment, the storage unit 22 stores correction filters for each Chinese medicine doctor. In this case, each of the N pieces of diagnosis result information stored in the storage unit 22 is associated with Chinese herbalist identification information that identifies the Chinese herbalist who diagnosed the diagnosis result indicated by each piece of diagnostic result information.

A Chinese medicine doctor's correction filter is based on the strength between floating pulses and sinking pulses, the strength between several pulses and slow pulses, the strength between large and small pulses, and the strength between imaginary pulses and real pulses. It has weight vectors associated with each of the strength and weakness, the strength between tense and slow pulses, and the strength between smooth and astringent pulses. For example, the weight vector associated with the strength between floating veins and sinking veins is a vector that corrects a vector indicating the diagnosis result by the Chinese medicine doctor regarding the strength between floating veins and sinking veins. More specifically, the weight vector has different magnitudes, which are multiplied by each of the five variables that the vector indicating the diagnosis result by the herbalist doctor regarding the strength between floating veins and sinking veins has as a component. This is a vector having five weights as components. Then, the vector indicating the diagnosis result by the Chinese herbalist about the strength between the floating vein and the sinking vein is the vector indicating the diagnosis result by the Chinese herbalist about the strength between the floating vein and the sinking vein, It is corrected by the Hadamard product of the weight vectors associated with the strengths and weaknesses between and the sedimentation veins. Here, the Hadamard product is a product of each component of a matrix, and is sometimes called a Schur product or the like. That is, the Hadamard product of the vector and the weight vector is the Hadamard product of matrices with 5 rows and 1 column. By such correction, the value of the variable to which 1 is substituted is corrected in the vector indicating the diagnosis result by the Chinese medicine doctor regarding the strength between the floating vein and the sinking vein. These circumstances include the strength between slow and slow pulses, the strength between large and small pulses, the strength between ischemic and real pulses, the strength between tense and slow pulses, The same applies to the strength of the smooth vein and the astringent vein.

Here, in the process of the flowchart shown in FIG. Identify the identified herbalist's correction filter. The first learning unit 265 corrects the vector indicated by the diagnosis result information based on the specified correction filter. The first learning unit 265 then generates, as first correspondence information, information in which the corrected vector is associated with the learning time-frequency spectrum image generated in step S250. Further, the second learning unit 266 generates second correspondence information that associates the corrected vector with each of the one or more Chinese medicines indicated by the Chinese medicine information read in step S240, and uses the generated second correspondence information. Store in the second database. At this time, the second learning unit 266 identifies the aforementioned target level set based on the corrected vector, and each of the identified target level set and the one or more Chinese medicine candidates crosses in the second database. The value of the non-zero variable among the variables included in the vector is added to the value of the field. Thereby, the information processing device 20 can reduce variations in diagnosis due to the subjectivity of the Chinese medicine doctor regarding the second database. As a result, the information processing device 20 can more reliably reduce the effort required for the Chinese medicine doctor to prescribe the Chinese medicine to the first subject.

<Variation 6 of the embodiment>
Modification 6 of the embodiment will be described below. Modification 6 of the embodiment is a modification of Modification 4 of the embodiment. In modification 6 of the embodiment, the second machine learning model has a herbal medicine contraindication filter. As mentioned above, when the second machine learning model receives the diagnosis result information output by the first machine learning model, it estimates that one or more Chinese herbal medicines are likely to be prescribed to the subject corresponding to the diagnosis result information. Chinese herbal medicine candidate information indicating each of one or more Chinese herbal medicine candidates is output. The herbal medicine contraindication filter is a filter that prevents one or more herbal medicine candidates indicated by the herbal medicine candidate information outputted from the second machine learning model from including a combination of contraindicated herbal medicines.

Hereinafter, a process in which the second machine learning model uses the herbal medicine contraindication filter will be described. For example, when a vector indicating diagnostic result information for a certain subject is input, the second machine learning model combines the input vector and a second database learned in advance (i.e., second correspondence information). Based on this, among the fields associated with the target level set indicated by the vector in the two-dimensional table indicated by the second database, each of one or more fields to which a value equal to or greater than a predetermined first threshold is assigned is identified. . The value of a certain field among these one or more fields is treated as a likelihood indicating the likelihood that the herbal medicine associated with the field is the herbal medicine to be prescribed to the subject. Therefore, the Chinese herbal medicine contraindication filter selects these one or more fields as target fields one by one in descending order of assigned likelihood. Then, the herbal medicine contraindication filter performs the following processing for each selected target field. The herbal medicine contraindication filter specifies fields associated with each of one or more herbal medicines that are contraindicated in combination with the herbal medicine associated with the selected target field as contraindication fields, and filters one or more fields other than the target field and the contraindication field. Specify as another field. Then, the Chinese herbal medicine contraindication filter multiplies the value of the target field by 1, the value of the contraindication field by -1, and the values of other fields by 0. The herbal medicine contraindication filter performs such processing for each selected target field, and finally displays the combination of herbal medicines associated with each of the one or more fields to which a positive value has been assigned to the subject. The second machine learning model identifies a combination of one or more Chinese herbal medicine candidates that is estimated to be plausible as one or more Chinese herbal medicines to be prescribed, and does not include any contraindicated combination of Chinese herbal medicines. be able to. In this way, the Chinese herbal medicine contraindication filter is a filter that excludes combinations of Chinese herbal medicines that are contraindicated from among one or more Chinese medicine candidates indicated by the Chinese medicine candidate information.

<Modification 7 of embodiment>
Modification 7 of the embodiment will be described below. In the seventh modification of the embodiment, the pulse wave waveform of the subject S3 is measured by the pulse wave sensor 12 that detects the pulse wave waveform of the subject S3 within a predetermined measurement time. This is a waveform detected by the pulse wave sensor 12 under a situation where the pressure applied to S3 is not constant. Even in this case, the information processing device 20 can accurately identify Chinese herbal medicine candidates to be prescribed to the subject. This means that even if the way the pulse wave sensor 12 is pressed against the subject changes each time a pulse diagnosis is performed, an appropriate Chinese herbal medicine can be prescribed to the subject.

To achieve this, the storage unit 22 of the information processing device 20 stores waveform information indicating a waveform detected by the following pressure-variable arterial wave detection method through the process shown in the flowchart shown in FIG. . In the pressure variable arterial wave detection method, the waveform of the pulse wave of the subject is determined by applying pressure to the subject against the pulse wave sensor 12 that detects the waveform of the subject's pulse wave within a predetermined measurement time. is detected by the pulse wave sensor 12 while changing the pulse wave. In this case, the pulse wave detection device 10 includes, for example, a first member 11 configured to be able to move one arm of the subject in the vertical direction using an actuator or the like, and a pulse wave sensor 12 that can move the pulse wave sensor 12 in the vertical direction using the actuator or the like. This configuration includes at least one of the second member 13 and the second member 13 that can be configured to Thereby, the pulse wave detection device 10 can, for example, cause the pulse wave sensor 12 to detect the pressure of the pulse wave while intermittently or continuously increasing the pressure with which the pulse wave sensor 12 is pressed against the subject. . In the following, as an example, the pulse wave detection device 10 applies pulse wave pressure to the pulse wave sensor 12 while continuously increasing the pressure with which the pulse wave sensor 12 is pressed against the subject in the range of 40 gf to 300 gf. The case of detection will be explained. Note that the pulse wave detection device 10 may have a configuration in which the pulse wave sensor 12 detects the pressure of the pulse wave while intermittently or continuously lowering the pressure with which the pulse wave sensor 12 is pressed against the subject. . Here, FIG. 16 is a diagram showing an example of a flow in which a time-frequency spectrum image is generated by the information processing device 20 based on waveform information indicating the waveform of a pulse wave detected by the pressure variable arterial wave detection method. . The waveform WP1 shown in FIG. 16 shows an example of the waveform of a pulse wave detected by the pressure variable arterial wave detection method. In step S110 shown in FIG. 4, the information processing device 20 acquires, from the pulse wave sensor 12, an electrical signal corresponding to the pressure of the pulse wave detected by the pressure variable arterial wave detection method. Then, the information processing device 20 generates waveform information indicating the waveform of the subject's pulse wave within the measurement period, based on the electrical signal thus acquired during the measurement period. Thereafter, in step S250 shown in FIG. 5, the information processing device 20 generates a time-frequency spectrum image based on the waveform information generated in this way. At this time, the information processing device 20 removes frequency components equal to or higher than a predetermined frequency from the waveform indicated by the waveform information using a band-pass filter, and provides information indicating the waveform after removing the frequency components. The time-frequency spectrum is calculated by STFT based on . As a result, the information processing device 20 can generate a time-frequency spectrum image showing the calculated time-frequency spectrum. Image WP2 shown in FIG. 16 shows an example of the time-frequency spectrum image generated in this way.

The information processing device 20 generates, as first correspondence information, information in which the time-frequency spectrum image generated in this manner is associated with the diagnosis result information received through the processing of the flowchart shown in FIG. The first machine learning model is caused to learn the first correspondence information. Therefore, the first machine learning model is trained to output appropriate diagnosis result information according to the time-frequency spectrum image according to the seventh modification of the embodiment. As a result, the information processing device 20 detects that the pulse wave sensor 12 detects the pulse wave of the subject S3 within the predetermined measurement time. Even if the waveform is detected by the pulse wave sensor 12 in a situation where the pressure applied to the test subject S3 is not constant, it is possible to accurately identify a candidate for a Chinese herbal medicine that is suitable as a Chinese medicine to be prescribed to the subject S3. In other words, the information processing device 20 can eliminate ambiguity caused by the amount of pressure applied during a pulse diagnosis by a Chinese herbalist doctor, and can accurately identify candidates for Chinese herbal medicines that are appropriate as the Chinese herbal medicines to be prescribed to the subject S3. .

Note that the matters described above may be combined in any manner.
Further, the method of classifying pulse types according to the state of the organ explained above may be other classification methods such as 38 diseased pulses instead of 28 diseased pulses. In this case, the aforementioned target vein type set may be a combination of seven or more vein types. However, even in this case, the overall flow of the processing performed by the information processing device 20 is the same as the flow described above.

<Additional notes>
[1] Based on the first time-frequency spectrum image showing the first time-frequency spectrum corresponding to the waveform of the pulse wave of the first subject (in the example explained above, subject S3), the first subject is An information processing device (in the example described above, the information processing device 20) includes a prescription candidate output unit that outputs output information including Chinese herbal medicine candidate information indicating candidates for Chinese herbal medicines to be prescribed to an examiner.

[2] A calculation unit (calculation unit 263 in the example described above) that calculates the first time-frequency spectrum based on first waveform information indicating the pulse wave waveform of the first subject as a first waveform. The information processing device according to [1], further comprising: a generation unit (generation unit 268 in the example described above) that generates the first time-frequency spectrum image based on the first time-frequency spectrum.

[3] The calculation unit removes a frequency component equal to or higher than a predetermined frequency from the first waveform, and generates information indicating the first waveform after removing the frequency component as the first waveform information. , the information processing device according to [2], wherein the first time-frequency spectrum is calculated based on the generated first waveform information.

[4] The prescription candidate output unit generates candidates for Chinese herbal medicine to be prescribed to the first subject based on the first machine learning model, the second machine learning model, and the first time-frequency spectrum image. and the first machine learning model has a second time-frequency spectrum that indicates a second time-frequency spectrum corresponding to a pulse wave waveform of a second subject (in the example described above, the subject during learning). A machine learning model that has learned first correspondence information in which an image is associated with second diagnosis result information indicating a diagnosis result of the second subject by a Chinese medicine doctor, and the second machine learning model is , is a machine learning model that has learned second correspondence information in which the second diagnosis result information and Chinese herbal medicine information are associated with each other, and the Chinese herbal medicine information is the one that the Chinese medicine doctor prescribed to the second subject. The information processing device according to [2] or [3], wherein the information indicates each of one or more Chinese herbal medicines.

[5] The information processing device according to [4], wherein the first machine learning model is a convolutional neural network for deep learning.

[6] The calculation unit calculates the second time-frequency spectrum based on second waveform information indicating the pulse wave waveform of the second subject as a second waveform, and the generation unit The information processing device generates the second time-frequency spectrum image based on the time-frequency spectrum, and includes a reception unit (in the example described above, which receives the second diagnosis result information and the one or more Chinese herbal medicine information). a reception unit 262), generates, as the first correspondence information, information in which the second time-frequency spectrum image generated by the generation unit and the second diagnosis result information received by the reception unit are associated; a first learning unit (in the example described above, the first learning unit 265) that causes the first machine learning model to learn the first correspondence information, and the second diagnosis result information received by the reception unit; a second learning unit that generates information that is associated with the one or more Chinese herbal medicine information received by the reception unit as the second correspondence information, and causes the second machine learning model to learn the generated second correspondence information; In the example described above, the information processing device according to [4], further comprising a second learning section 266).

[7] The second diagnosis result information includes a herbalist's information regarding the predetermined first diagnosis item (in the example explained above, for example, floating veins, strength and weakness between floating veins and sinking veins, etc.) information indicating the diagnosis result of the second subject (in the example explained above, for example, floating pulse, level 1, etc.) and a predetermined second diagnosis item (in the example explained above, for example, information indicating the diagnosis result of the second subject by the Chinese herbalist (in the example explained above, for example, sinking pulse, level 2, etc.) Based on the first correspondence information, the first learning section includes information indicating the diagnosis result of the second subject by the Chinese medicine doctor regarding the first diagnosis item, and the second learning section. the first machine learning model learns the time-frequency spectrum image, and obtains a coefficient sequence of the first machine learning model for the first diagnosis item as a first coefficient sequence, and based on the first correspondence information, , causing the first machine learning model to learn information indicating the diagnosis result of the second subject by the Chinese medicine doctor regarding the second diagnosis item and the second time-frequency spectrum image, and determining the second diagnosis item. The prescription candidate output unit obtains the coefficient sequence of the first machine learning model for information indicating a plausible diagnosis result of the first subject by the Chinese herbalist regarding the first diagnostic item based on the first time-frequency spectrum image; and causing the first machine learning model to output first diagnosis result information including information indicating a plausible diagnosis result as a diagnosis result of the first subject, and the first diagnosis caused to be output by the first machine learning model. The information processing device according to [6], which identifies candidates for Chinese herbal medicine to be prescribed to the first subject based on the result information and the second machine learning model.

[8] Each of the first diagnostic item and the second diagnostic item is one of floating network, sinking network, slow network, few network, virtual network, and real network in the 28 diseased veins. The information processing device according to [7], wherein the diagnostic items are different from each other.

[9] The first diagnosis item includes a plurality of first options selectable as a diagnosis result, and the second diagnosis item includes a plurality of second options selectable as a diagnosis result. The first machine learning model is configured to calculate the likelihood of each of the plurality of first options as a diagnosis result by a Chinese herbalist based on the first coefficient sequence and the first time-frequency spectrum image. Based on the calculated likelihood, information indicating a likely diagnosis result of the first subject by the Chinese medicine doctor regarding the first diagnosis item is estimated, and the second coefficient sequence and , a likelihood indicating the likelihood of each of the plurality of second options as a diagnosis result by a Chinese herbalist is calculated based on the first time-frequency spectrum image, and based on the calculated likelihood, the second diagnosis is made. The information processing device according to [7], which estimates information indicating a plausible diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor regarding the item.

[10] The first diagnostic item and the second diagnostic item each include the strength between floating pulse and sinking pulse, the strength between few pulses and slow pulse, and the strength between large vein and slow pulse in Japanese pulse diagnosis method. It is either the strength between small veins, the strength between ischemic pulse and real pulse, the strength between tense and slow pulse, or the strength between smooth pulse and astringent pulse, and is different from each other. The information processing device according to [7], which is a diagnostic item.

[11] The first diagnosis item includes a plurality of first options selectable as a diagnosis result, and the second diagnosis item includes a plurality of second options selectable as a diagnosis result. The first machine learning model is configured to perform a diagnosis by a Chinese herbalist of each of the plurality of first options based on the first coefficient sequence, the first machine learning model, and the first time-frequency spectrum image. A likelihood indicating the likelihood of the result is calculated, and based on the calculated likelihood, information indicating a likely diagnosis result as a diagnosis result of the first subject by the Chinese herbalist regarding the first diagnosis item is estimated. and a likelihood indicating the likelihood of each of the plurality of second options as a diagnosis result by a Chinese herbalist based on the second coefficient sequence, the first machine learning model, and the first time-frequency spectrum image. and, based on the calculated likelihood, estimate information indicating a likely diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor regarding the second diagnosis item, the information according to [9] Processing equipment.

[12] The first machine learning model generates a first vector as information indicating a plausible diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor regarding the first diagnosis item, and 1 vector is corrected by a Hadamard product with a weight vector associated with the first diagnostic item, and a plausible diagnosis result is indicated as a diagnosis result of the first subject by the Chinese herbalist regarding the second diagnostic item. A second vector is generated as information, the generated second vector is corrected by a Hadamard product with a weight vector associated with the second diagnostic item, and the first vector after the correction and the first vector after the correction are obtained. The information processing device according to [10], which outputs the first diagnosis result information including the second vector.

[13] The first machine learning model is configured such that if the likelihoods for each of the plurality of first options are all less than a predetermined threshold, or the likelihood for each of the plurality of second options is The information processing device according to [11], which performs error processing when all of the values are less than the threshold values.

[14] The first machine learning model determines at least one of the factor ranking and the degree of contribution of the Chinese herbal medicine prescription determining factors for each of the first diagnostic item and the second diagnostic item, based on the second correspondence information. the likelihood for each of the plurality of first options is multiplied by the first weight based on the estimated at least one, and the second weight based on the estimated at least one is multiplied by the likelihood for each of the plurality of second options. The prescription candidate output unit multiplies the likelihood for each of the plurality of first options, the first weight, the likelihood for each of the plurality of second options, and the second weight, The information processing device according to [11], wherein the second correspondence information is updated based on the Chinese herbal medicine candidate information output from the second machine learning model.

[15] The information processing device according to [10], wherein the second machine learning model has a herbal medicine contraindication filter that excludes contraindicated combinations of herbal medicines from among the one or more Chinese medicine candidates indicated by the herbal medicine candidate information. .

[16] The waveform of the pulse wave of the first subject is determined by pressing a sensor that detects the waveform of the pulse wave of the first subject against the first subject within a predetermined measurement time. The information processing device according to any one of [1] to [15], wherein the waveform is detected by the sensor while changing pressure.

[17] A Chinese herbal medicine candidate indicating a Chinese herbal medicine candidate to be prescribed to the first subject based on a first time-frequency spectrum image showing a first time-frequency spectrum corresponding to the waveform of the pulse wave of the first subject. An information processing method comprising a prescription candidate output step of outputting output information including information.

[18] A computer is configured to select candidates for Chinese herbal medicines to be prescribed to the first subject based on a first time-frequency spectrum image showing a first time-frequency spectrum corresponding to the waveform of the pulse wave of the first subject. A program for executing a prescription candidate output step of outputting output information including Chinese herbal medicine candidate information shown in FIG.

Although the embodiments of this disclosure have been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and modifications, substitutions, deletions, etc. may be made without departing from the gist of this disclosure. may be done.

Furthermore, a program for realizing the functions of arbitrary components in the apparatus described above may be recorded on a computer-readable recording medium, and the program may be read and executed by a computer system. Here, the device is, for example, the pulse wave detection device 10, the information processing device 20, etc. Note that the "computer system" herein includes hardware such as an OS (Operating System) and peripheral devices. Furthermore, "computer-readable recording media" refers to portable media such as flexible disks, magneto-optical disks, ROMs, and CD (Compact Disk)-ROMs, and storage devices such as hard disks built into computer systems. . Furthermore, "computer-readable recording media" refers to volatile memory inside a computer system that serves as a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. This also includes those that hold time programs.

Further, the above program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has a function of transmitting information, such as a network such as the Internet or a communication line such as a telephone line.
Moreover, the above-mentioned program may be for realizing a part of the above-mentioned functions. Furthermore, the above-mentioned program may be a so-called difference file or difference program that can realize the above-described functions in combination with a program already recorded in the computer system.

DESCRIPTION OF SYMBOLS 1... Information processing system, 10... Pulse wave detection device, 11... First member, 12... Pulse wave sensor, 13... Second member, 20... Information processing device, 21... Processor, 22... Storage unit, 23... Input reception Department, 24...Communication section, 25...Display section, 26...Control section, 261...Acquisition section, 262...Reception section, 263...Calculation section, 264...Prescription candidate output section, 265...First learning section, 266...Second Learning section, 267... Display control section, 268... Generation section, TC... Three-dimensional coordinate system

Claims

Contains Chinese herbal medicine candidate information indicating a Chinese herbal medicine candidate to be prescribed to the first subject based on a first time frequency spectrum image indicating a first time frequency spectrum corresponding to the pulse wave waveform of the first subject. comprising a prescription candidate output unit that outputs output information;
Information processing device.
a calculation unit that calculates the first time-frequency spectrum based on first waveform information indicating a pulse wave waveform of the first subject as a first waveform;
a generation unit that generates the first time-frequency spectrum image based on the first time-frequency spectrum;
The information processing device according to claim 1, further comprising:.
The calculation unit removes a frequency component equal to or higher than a predetermined frequency from the first waveform, and generates information indicating the first waveform after removing the frequency component as the first waveform information. calculating the first time-frequency spectrum based on the first waveform information;
The information processing device according to claim 2.
The prescription candidate output unit identifies candidates for Chinese herbal medicine to be prescribed to the first subject based on the first machine learning model, the second machine learning model, and the first time frequency spectrum image,
The first machine learning model shows a second time-frequency spectrum image showing a second time-frequency spectrum according to a pulse waveform of a second subject, and a diagnosis result of the second subject by a Chinese herbalist. A machine learning model that has learned first correspondence information associated with second diagnosis result information,
The second machine learning model is a machine learning model that has learned second correspondence information in which the second diagnosis result information and Chinese herbal medicine information are associated with each other,
The Chinese herbal medicine information is information indicating each of one or more Chinese medicines prescribed by the Chinese herbalist to the second subject.
The information processing device according to claim 2 or 3.
The first machine learning model is a convolutional neural network for deep learning,
The information processing device according to claim 4.
The calculation unit calculates the second time-frequency spectrum based on second waveform information indicating a pulse wave waveform of the second subject as a second waveform,
The generation unit generates the second time-frequency spectrum image based on the second time-frequency spectrum,
The information processing device includes:
a reception unit that receives the second diagnosis result information and the one or more Chinese herbal medicine information;
generating, as the first correspondence information, information in which the second time-frequency spectrum image generated by the generation unit and the second diagnosis result information received by the reception unit are associated; and the generated first correspondence information a first learning unit that causes the first machine learning model to learn;
Generate information in which the second diagnosis result information received by the reception unit and the one or more Chinese herbal medicine information received by the reception unit are associated as the second correspondence information, and use the generated second correspondence information. a second learning unit that causes the second machine learning model to learn;
further comprising;
The information processing device according to claim 4.
The second diagnosis result information includes information indicating the diagnosis result of the second subject by the Chinese medicine doctor regarding the predetermined first diagnosis item, and information indicating the diagnosis result of the second subject by the Chinese medicine doctor regarding the predetermined second diagnosis item. Information indicating the diagnosis result of the second subject is included,
The first learning unit is configured to acquire information indicating a diagnosis result of the second subject by the Chinese medicine doctor regarding the first diagnosis item and the second time-frequency spectrum image based on the first correspondence information. A first machine learning model is trained to obtain a coefficient sequence of the first machine learning model for the first diagnostic item as a first coefficient sequence, and based on the first correspondence information, for the second diagnostic item. The first machine learning model is made to learn information indicating the diagnosis result of the second subject by the Chinese medicine doctor and the second time frequency spectrum image, and the first machine learning about the second diagnosis item is performed. Obtain the coefficient sequence of the model as the second coefficient sequence,
The prescription candidate output unit is configured to select a Chinese herbal medicine doctor for the first diagnostic item based on the first coefficient sequence, the second coefficient sequence, the first machine learning model, and the first time-frequency spectrum image. information indicating a plausible diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor regarding the second diagnosis item, and information indicating a plausible diagnosis result as the diagnosis result of the first subject by the Chinese medicine doctor regarding the second diagnosis item. The first test result information is output to the first machine learning model, and the first test result information is output to the first test subject based on the first diagnosis result information output by the first machine learning model and the second machine learning model. identify candidates for herbal medicines to be prescribed to patients;
The information processing device according to claim 6.
Each of the first diagnostic item and the second diagnostic item is any one of floating reticular vein, sinking reticular pulse, slow reticular pulse, few reticular pulse, ischemic reticular pulse, and real reticular pulse in the 28 diseased pulses, Diagnostic items that are different from each other,
The information processing device according to claim 7.
The first diagnosis item includes a plurality of first options that can be selected as a diagnosis result,
The second diagnosis item includes a plurality of second options that can be selected as a diagnosis result,
The first machine learning model calculates a likelihood indicating the likelihood of each of the plurality of first options as a diagnosis result by a Chinese herbalist based on the first coefficient sequence and the first time-frequency spectrum image. Based on the calculated likelihood, information indicating a likely diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor regarding the first diagnosis item is estimated, and information is estimated based on the second coefficient sequence and the first subject. Based on the one-time frequency spectrum image, a likelihood indicating the likelihood of each of the plurality of second options as a diagnosis result by a Chinese herbalist is calculated, and based on the calculated likelihood, the second diagnosis item is determined based on the calculated likelihood. estimating information indicating a plausible diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor;
The information processing device according to claim 7.
The first diagnostic item and the second diagnostic item each include the strength between floating pulse and sinking pulse, the strength between several pulses and slow pulse, and the strength between large and small veins in the Japanese pulse diagnosis method. The strength is between the strength between ischemic and real pulses, the strength between tense and slow pulses, the strength between smooth and astringent pulses, and these are different diagnostic items. be,
The information processing device according to claim 7.
The first diagnosis item includes a plurality of first options that can be selected as a diagnosis result,
The second diagnosis item includes a plurality of second options that can be selected as a diagnosis result,
The first machine learning model is configured to determine a diagnosis result by a Chinese herbalist for each of the plurality of first options based on the first coefficient sequence, the first machine learning model, and the first time-frequency spectrum image. A likelihood indicating likelihood is calculated, and based on the calculated likelihood, information indicating a likely diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor regarding the first diagnosis item is estimated, and the Based on the second coefficient sequence, the first machine learning model, and the first time-frequency spectrum image, a likelihood indicating the likelihood of each of the plurality of second options as a diagnosis result by a Chinese herbalist is calculated. , based on the calculated likelihood, estimate information indicating a likely diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor regarding the second diagnosis item;
The information processing device according to claim 9.
The first machine learning model generates a first vector as information indicating a plausible diagnosis result as a diagnosis result of the first subject by the Chinese medicine doctor regarding the first diagnosis item, and , corrected by the Hadamard product with the weight vector associated with the first diagnostic item, and as information indicating a likely diagnosis result of the first subject by the Chinese herbalist regarding the second diagnostic item. 2 vectors are generated, the generated second vector is corrected by a Hadamard product with a weight vector associated with the second diagnostic item, and the corrected first vector and the corrected second vector are outputting the first diagnosis result information including two vectors;
The information processing device according to claim 10.
The first machine learning model is configured such that if all the likelihoods for each of the plurality of first options are less than a predetermined threshold, or all the likelihoods for each of the plurality of second options are If it is less than the threshold, perform error processing,
The information processing device according to claim 11.
The first machine learning model estimates at least one of a factor ranking and a degree of contribution of Chinese herbal medicine prescription determining factors for each of the first diagnostic item and the second diagnostic item, based on the second correspondence information, Multiplying the likelihood of each of the plurality of first options by a first weight based on the estimated at least one, and multiplying the likelihood of each of the plurality of second options by a second weight based on the estimated at least one. death,
The prescription candidate output unit includes a likelihood for each of the plurality of first options, the first weight, a likelihood for each of the plurality of second options, the second weight, and the second machine learning. updating the second correspondence information based on the Chinese herbal medicine candidate information output from the model;
The information processing device according to claim 11.
The second machine learning model has a herbal medicine contraindication filter that excludes contraindicated combinations of herbal medicines from among the one or more herbal medicine candidates indicated by the herbal medicine candidate information.
The information processing device according to claim 10.
The waveform of the pulse wave of the first subject is determined by changing the pressure with which a sensor for detecting the waveform of the pulse wave of the first subject is pressed against the first subject within a predetermined measurement time. is a waveform detected by the sensor while
The information processing device according to any one of claims 1 to 3.
Contains Chinese herbal medicine candidate information indicating a Chinese herbal medicine candidate to be prescribed to the first subject based on a first time frequency spectrum image indicating a first time frequency spectrum corresponding to the pulse wave waveform of the first subject. comprising a prescription candidate output step for outputting output information;
Information processing method.
to the computer,
Contains Chinese herbal medicine candidate information indicating a Chinese herbal medicine candidate to be prescribed to the first subject based on a first time frequency spectrum image indicating a first time frequency spectrum corresponding to the pulse wave waveform of the first subject. a prescription candidate output step for outputting output information;
A program to run.