Condition determination system, condition determination method, condition determination model, and condition determination model generation method

Abstract

To highly accurately determine a state of an analysis object from analysis object data.SOLUTION: The state determination system inputs analysis data of a new analysis target to the main determination model to acquire a first result regarding whether a state of the new analysis target corresponds to a first stage or a second stage, inputs the analysis data to the supplementary determination model to acquire a second result regarding whether the state of the new analysis target corresponds to the first stage or a third stage, and integrates the first result and the second result to calculate a determination result indicating whether the state of the new analysis target corresponds to the first stage or the second stage. The first stage, the second stage, and the third stage indicate the degree of the abnormal state, and the degree of the abnormal state in the second stage is located between the first stage and the third stage.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a condition determination system, a condition determination method, a condition determination model, and a condition determination model generation method. [Background technology]

[0002] JP 2022-139417 A (Patent Document 1) is a background technology of this technical field. This publication states that "a plurality of first element models having different specific features are prepared, and a second element model is generated by adjusting at least one of the plurality of first element models so as to adapt to adjustment data different from the training data of the model, a plurality of element models including at least one second element model are selected from a set of element models consisting of the first element model and the second element model, and an integrated model is generated by integrating the selected plurality of element models, and the integrated model outputs one of the classes into which input data is classified, and the class is a class into which the input data is classified based on the presence or absence of all specific features related to the selected plurality of element models" (see Abstract). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-139417 Summary of the Invention [Problem to be solved by the invention]

[0004] Although a technique for determining the state of an analysis target from data to be analyzed is known, such as the technique described in Patent Document 1, which outputs a class to which input data belongs from the input data, the accuracy of such determination is insufficient. In particular, there is a problem in that the accuracy of determining whether the state of a new analysis target corresponds to an intermediate stage among a plurality of stages is low.

[0005] Therefore, one aspect of the present invention is to determine the state of an analysis target from analysis target data with high accuracy. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, one embodiment of the present invention employs the following configuration: A condition discrimination system includes a processor and a memory, wherein the memory holds analytical data of a new analysis target, a main discriminant model that, when the analytical data of the new analysis target is input, outputs a first result as to whether the condition of the new analysis target corresponds to the first stage or the second stage, and a supplementary discriminant model that, when the analytical data of the new analysis target is input, outputs a second result as to whether the condition of the new analysis target corresponds to the first stage or the third stage, wherein the first stage, the second stage, and the third stage indicate degrees of the condition, and the degree of the condition in the second stage is located between the first stage and the third stage, and the main discriminant model divides a first analysis target group consisting of analysis targets whose states are the first stage and a supplementary discriminant model consisting of analysis targets whose states are the second stage. the supplementary discriminant model is generated by machine learning using analytical data for the first group of analysis subjects and a third group of analysis subjects consisting of analysis subjects whose state is the third stage; the processor inputs the analytical data of the new analysis subject into the main discriminant model to obtain a first result for the new analysis subject, inputs the analytical data of the new analysis subject into the supplementary discriminant model to obtain a second result for the new analysis subject, integrates the obtained first result and the obtained second result, calculates a discrimination result indicating whether the state of the new analysis subject corresponds to the first stage or the second stage, and generates data for displaying the discrimination result. [Effects of the Invention]

[0007] According to one aspect of the present invention, the state of an analysis target is determined with high accuracy from analysis target data.

[0008] Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a block diagram showing an example of the configuration of a human data measurement system according to a first embodiment. [Figure 2] 1 is a block diagram showing an example of the configuration of an MCI discrimination system according to a first embodiment. [Figure 3] 1 is a block diagram showing an example of the configuration of a measurement device according to a first embodiment. [Figure 4] FIG. 2 is a block diagram illustrating a configuration example of a terminal device according to the first embodiment. [Figure 5] 1 is a diagram showing a state in which a magnetic sensor, which is a motion sensor, is attached to a finger of a user in Example 1. FIG. [Figure 6] 2 is a diagram illustrating an example of a detailed configuration of a motion sensor control unit and the like of the measurement device in the first embodiment. FIG. [Figure 7] 10 is a flowchart illustrating an example of an integrated discriminant model generation process according to the first embodiment. [Figure 8] 1 is a flowchart showing an example of an MCI discrimination process for a new subject in Example 1. [Figure 9A] 10 is a graph showing a waveform signal of the distance between two fingers in Example 1. [Figure 9B] 10 is a graph showing waveform signals of the velocities of two fingers in Example 1. [Figure 9C] 10 is a graph showing waveform signals of accelerations of two fingers in Example 1. [Figure 9D] 4 is a graph showing an example of feature amounts in the first embodiment. [Figure 10] FIG. 4 is a diagram illustrating an example of a feature amount list according to the first embodiment. [Figure 11] FIG. 2 is an explanatory diagram illustrating an example of a main discriminant model, a supplementary discriminant model, and an integrated discriminant model according to the first embodiment. [Figure 12] FIG. 10 is a diagram illustrating an example of a search range of hyperparameters in the first embodiment. [Figure 13] FIG. 10 is a diagram showing a comparison result between the discrimination accuracy in MCI discrimination using only the main discriminant model in Example 1 and the discrimination accuracy in MCI discrimination using the integrated discriminant model of this example. [Figure 14] FIG. 10 is a diagram illustrating feature contributions of an integrated discriminant model in the first embodiment. [Figure 15] FIG. 2 is a diagram illustrating an example of a screen configuration of an initial screen in the first embodiment. [Figure 16] FIG. 10 is a diagram illustrating an example of a screen configuration of a task measurement screen in the first embodiment. [Figure 17] FIG. 10 is a diagram showing an example of a screen configuration of a result display screen in the first embodiment. [Figure 18] FIG. 10 is a diagram illustrating an example of a screen configuration of a basis presentation screen in the first embodiment. [Figure 19] FIG. 2 is a diagram illustrating an example of a screen configuration of a setting screen in the first embodiment. [Figure 20] FIG. 4 is a diagram illustrating an example of a data configuration of user information according to the first embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In this embodiment, the same components are generally designated by the same reference numerals, and repeated explanations will be omitted. It should be noted that this embodiment is merely an example for realizing the present invention, and does not limit the technical scope of the present invention.

[0011] As our society ages, the number of dementia patients is increasing, posing problems such as a decline in the quality of life (QOL) of patients and their families, as well as rising medical and nursing care costs. To mitigate these problems, simple screening tests are expected to enable early detection of dementia and enable early treatment to delay its progression. Conventional dementia screening tests, such as blood tests and olfactory tests, have been developed. However, while blood tests have high accuracy, they require medical professionals to draw blood, which is invasive and therefore difficult to perform. Furthermore, although olfactory tests are simple, they have drawbacks, such as insufficient accuracy and difficulty in performing them in patients whose sense of smell has deteriorated due to other diseases.

[0012] Furthermore, recently, early detection at the stage of Mild Cognitive Impairment (MCI), which is the precursor to dementia, has been attempted, with the aim of slowing the progression of dementia at an earlier stage by encouraging people to improve their lifestyles and seek medical treatment.

[0013] However, because MCI exhibits symptoms intermediate between those of dementia and healthy individuals, detecting MCI is more difficult than detecting dementia. There is technology to detect MCI using multiple types of data (location, sleep, home appliance usage, and conversation) and power consumption within elderly care facilities, but this technology requires a system to collect data within the facility or home, and requires continuous monitoring for a certain period of time, so the challenge is that it does not provide immediate results.

[0014] Therefore, in this embodiment, a simple screening test for cognitive decline based on finger movement measurement is used to distinguish between an MCI group and a healthy group. The measurement system of this embodiment measures finger-tapping movements, which involve repeatedly opening and closing the thumb and index finger, and calculates 44 types of feature quantities that represent various properties of the finger-tapping movements. The average value / variation of amplitude, the average value / variation of tap intervals, the total distance moved during the measurement time, and the number of small up and down movements are all examples of feature quantities that represent the properties of the finger-tapping movements.

[0015] Note that, if each of the above-mentioned features is used to distinguish between the MCI group and the healthy group, a certain degree of discrimination accuracy can be achieved, but since each feature only captures individual characteristics of the finger tapping movement and may not comprehensively evaluate various characteristics, there is room for improvement in accuracy. To achieve a highly reliable screening test, it is necessary to integrate various characteristics to create a discrimination model and achieve higher discrimination accuracy. Therefore, the MCI discrimination system of this embodiment uses machine learning to discriminate between the MCI group and the healthy group with high accuracy. [Example]

[0016] [Human data measurement system] FIG. 1 is a block diagram showing an example configuration of a human data measurement system including an MCI discrimination system. The human data measurement system is installed in a facility such as a hospital, a nursing home, or a user's home. The human data measurement system includes an MCI discrimination system 1 and a measurement system 2, which is a magnetic sensor-type finger tapping exercise system, which are connected to each other via a communication line. The measurement system 2 includes a measurement device 3 and a terminal device 4, which are connected to each other via a communication line. Multiple measurement systems 2 may be installed within a facility.

[0017] The measurement system 2 measures finger movements using a magnetic sensor-type motion sensor. The motion sensor is connected to the measurement device 3. The motion sensor is attached to the user's fingers. The measurement device 3 measures the finger movements through the motion sensor and obtains measurement data including time-series waveform signals. The terminal device 4 displays various information including the MCI discrimination results on a display screen and accepts operational inputs from the user. The terminal device 4 is configured by a computer such as a PC (Personal Computer), a tablet terminal, or a smartphone.

[0018] The MCI discrimination system 1 has a function of providing an MCI discrimination service as an information processing service. The MCI discrimination system 1 has an MCI discrimination function as one of its functions. The MCI discrimination function uses measurement data measured by the measurement system 2 to determine whether a user has MCI or is a healthy individual. Healthy, MCI, and dementia are examples of stages of the degree of the user's abnormal state. Furthermore, healthy, MCI, and dementia may be examples of the first, second, and third stages, respectively, or may be examples of the third, second, and first stages, respectively. In other words, MCI, which is an example of the second stage, is located between the first and third stages in terms of the degree of the abnormality.

[0019] The MCI discrimination system 1 receives, as input data from the measurement system 2, for example, periodic time-series data of finger tapping movements. The MCI discrimination system 1 outputs, as output data to the measurement system 2, for example, an MCI discrimination result. The MCI discrimination result may include, in addition to the discrimination result as to whether the user is MCI-suffering or a healthy individual, information on the feature quantities that contributed to the discrimination result and information on the calculation process of the discrimination result (the basis for obtaining the discrimination result), for example.

[0020] The human data measurement system of this embodiment is not limited to facilities such as hospitals and elderly care facilities and their subjects, but can be widely applied to general facilities and people. The measurement device 3 and the terminal device 4 may be configured as an integrated measurement system. The measurement system 2 and the MCI discrimination system 1 may be configured as an integrated device. The terminal device 4 and the MCI discrimination system 1 may be configured as an integrated device. The measurement device 3 and the MCI discrimination system 1 may be configured as an integrated device. Note that communication between integrated devices or systems, which will be described later, is omitted.

[0021] [MCI Detection System 1] 2 is a block diagram showing an example of the configuration of the MCI discrimination system 1. The MCI discrimination system 1 is configured by a computer having a control unit 101, a storage unit 102, an input unit 103, an output unit 104, and a communication unit 105, which are connected to each other via an internal communication line such as a bus.

[0022] The MCI discrimination system 1 is a computer system configured on one physical computer or on multiple logically or physically configured computers, and may operate in separate threads on the same computer, or on a virtual computer constructed on multiple physical computer resources.

[0023] The input unit 103 accepts operational inputs from an administrator or the like of the MCI discrimination system 1. The input unit 103 is constituted by an input interface device connected to, for example, a keyboard, a mouse, a touch panel, or the like. The output unit 104 outputs the results of program execution in a format (for example, a screen display, etc.) that can be viewed by the administrator or the like of the MCI discrimination system 1. The output unit 104 is constituted by, for example, an output interface device connected to a display device, a printer, or the like.

[0024] The communication unit 105 controls communication between the MCI discrimination system 1 and other devices (the measuring device 3 and the terminal device 4) using a predetermined protocol. The communication unit 105 is configured by, for example, a network interface device or a serial interface device such as a USB (Universal Serial Bus).

[0025] The control unit 101 controls the entire MCI discrimination system 1. The control unit 101 is configured with a CPU (Central Processing Unit), which is an example of a processor, and a ROM (Read Only Memory) and a RAM (Random Access Memory), both of which are examples of memory. The control unit 101 realizes a data processing unit that performs MCI discrimination and the like based on software program processing.

[0026] The CPU constituting the control unit 101 executes programs stored in the memory constituting the control unit 101. The ROM is a non-volatile storage element, and the RAM is a volatile storage element. The ROM stores unchanging programs (e.g., BIOS (Basic Input / Output System)). The RAM is a high-speed, volatile storage element such as a DRAM (Dynamic Random Access Memory), and temporarily stores programs executed by the CPU and data used when the programs are executed.

[0027] The storage unit 102 is configured by, for example, a large-capacity, non-volatile storage device (auxiliary storage device) such as a magnetic storage device (HDD (Hard Disk Drive)) and a flash memory (SSD (Solid State Drive)). The storage unit 102 stores programs executed by the CPU constituting the control unit 101 and data used when the programs are executed. That is, the programs are read from the auxiliary storage device constituting the storage unit 102, loaded into the memory constituting the control unit 101, and executed by the CPU constituting the control unit 101.

[0028] Some or all of the programs executed by the control unit 101 may be provided to the MCI discrimination system 1 from removable media (CD-ROM, flash memory, etc.) which are non-transitory storage media, or from an external computer equipped with a non-transitory storage device via a network, and stored in a non-volatile storage device (auxiliary storage device constituting the storage unit 102) which is also a non-transitory storage medium. For this reason, the MCI discrimination system 1 should preferably have an interface for reading data from removable media.

[0029] The data processing unit of the control unit 101 has, for example, a user information management unit 110, a feature calculation unit 120, a model generation unit 130, an MCI discrimination unit 140, and a result output unit 150, which are all functional units. The control unit 101 also realizes functions such as a function to input measurement data from the measurement device 3, a function to process and analyze the measurement data, a function to output control instructions to the measurement device 3 and the terminal device 4, a function to output display data to the terminal device 4, and a function to store data input from the input unit 103, the measurement device 3, and the terminal device 4 in a DB (DataBase) 40 of the storage unit 102.

[0030] In addition, DB40 of the storage unit 102 stores, for example, user information 410, measurement data 420, feature list 430, feature data 440, subject DB 450, main discriminant model 461, supplementary discriminant model 462, integrated discriminant model constant table 463, integrated discriminant model 466, and MCI discrimination result 470.

[0031] 2, some or all of the information stored in the storage unit 102 may be stored in a memory constituting the control unit 101, or may be stored in an external database connected to the MCI discrimination system 1. In this embodiment, the information used by the human data measurement system may be expressed in any data structure independent of the data structure. For example, the information may be stored in a data structure appropriately selected from a table, a list, a database, or a queue.

[0032] The user information management unit 110 performs processes such as registering and managing user information input by a user in user information 410 in DB 40, and checking the user information 410 in DB 40 when the user uses a service. The user information 410 includes attribute values ​​and measurement task information for each user. The attribute values ​​include, for example, gender and age. The measurement task information includes the date and time when a task for analyzing and evaluating motor function, etc., was measured, and the type of the measured task. In this embodiment, the task is a predetermined finger movement, i.e., the finger tapping described above.

[0033] The feature calculation unit 120 calculates the features indicated by the feature list 430 from the measurement data 420 (periodic information indicating the state of the living body) of the task of the new subject (new user) measured for a predetermined time by the measurement device 3, and stores the calculated features in the DB 40 as feature data 440.

[0034] The model generation unit 130 generates various discriminant models using a subject DB 450 indicating feature quantities (an example of analysis data) as learning data (teaching data) given in advance to a subject (an example of an analysis target) and an integrated discriminant model constant table 463 indicating hyperparameters of the integrated discriminant model. In other words, the MCI discrimination system 1 also functions as a state discrimination model generation system that generates various discriminant models.

[0035] The model generation unit 130 has a main discriminant model generation unit 131, a supplementary discriminant model generation unit 132, and an integrated discriminant model generation unit 133, which are all functional units. The subject DB 450 includes a healthy group DB 451, a pre-disease group DB 452, and a disease group DB 453.

[0036] The healthy group DB 451 holds feature quantities of the type indicated by the feature quantity list 430, obtained from measurement data of each task of a subject (healthy subject) to whom "healthy" is assigned as the correct label. The pre-disease group DB 452 holds feature quantities of the type indicated by the feature quantity list 430, obtained from measurement data of each task of a subject (MCI patient) to whom "pre-disease" is assigned as the correct label. The disease group DB 453 holds feature quantities of the type indicated by the feature quantity list 430, obtained from measurement data of each task of a subject (dementia patient) to whom "disease" is assigned as the correct label.

[0037] The main discriminant model generation unit 131 generates a main discriminant model 461 through supervised learning using the healthy group DB 451 and the pre-disease group DB 452. When a feature of a type indicated by the feature list 430 is input to the main discriminant model 461, the main discriminant model 461 outputs a discrimination result indicating whether a subject corresponding to the feature is healthy or has MCI. The supplemental discriminant model generation unit 132 generates a supplemental discriminant model 462 through supervised learning using the healthy group DB 451 and the disease group DB 453. When a feature of a type indicated by the feature list 430 is input to the supplemental discriminant model 462, the supplemental discriminant model 462 outputs a discrimination result indicating whether a subject corresponding to the feature is healthy or has dementia.

[0038] The integrated discriminant model generation unit 133 generates an integrated discriminant model 466 by integrating the main discriminant model 461 and the supplemental discriminant model 462, using a positive detection constant 464 and a supplemental model influence constant 465 included in an integrated discriminant model constant table 463. When a feature of the type indicated by the feature list 430 is input, the integrated discriminant model 466 is a model that outputs a discrimination result indicating whether the subject corresponding to the feature corresponds to a healthy state or to MCI.

[0039] Although details will be described later, the positive detection constant 464 is, for example, a constant for reflecting a value contributing to a discrimination result indicating dementia in the supplementary discriminant model 462 in a value contributing to a discrimination result indicating MCI in the main discriminant model 461. The positive detection constant 464 is an example of a first weight. Furthermore, the supplementary model influence constant 465 is, for example, a constant for adjusting the magnitude of the influence of the supplementary discriminant model 462 on the main discriminant model 461. The supplementary model influence constant 465 is an example of a second weight.

[0040] The MCI discrimination section 140 inputs the feature of the new subject included in the feature data 440 into the integrated discrimination model 466, obtains a discrimination result indicating whether the new subject corresponds to healthy or MCI, and stores the obtained discrimination result in the DB 40 as an MCI discrimination result 470. Note that the MCI discrimination section 140 may also include information indicating the basis for obtaining the discrimination result and information on the feature that contributed to the discrimination result, etc., in the MCI discrimination result 470. The result output section 150 outputs information indicating the MCI discrimination result 470 to the terminal device 4.

[0041] For example, the processor constituting the control unit 101 operates in accordance with a user information program loaded into a memory constituting the control unit 101 to function as a user information management unit 110, and operates in accordance with a feature calculation program loaded into the memory to function as a feature calculation unit 120. This also applies to other functional units included in the control unit 101 and other functions realized by the control unit 101.

[0042] Note that some or all of the functions realized by the control unit 101 may be realized by hardware such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field-Programmable Gate Array).

[0043] [Measuring equipment] FIG. 3 is a block diagram showing an example configuration of the measurement device 3. The measurement device 3 includes, for example, a motion sensor 20, a housing unit 301, a measurement unit 302, and a communication unit 303. The housing unit 301 includes a motion sensor interface unit 311 to which the motion sensor 20 is connected, and a motion sensor control unit 312 that controls the motion sensor 20. The measurement unit 302 measures waveform signals via the motion sensor 20 and the housing unit 301 and outputs the measured data. The measurement unit 302 includes a task measurement unit 321 that obtains the measured data. The communication unit 303 has a communication interface and communicates with the MCI discrimination system 1 to transmit the measured data to the MCI discrimination system 1. The motion sensor interface unit 311 includes an analog-to-digital conversion circuit and converts the analog waveform signal detected by the motion sensor 20 into a digital waveform signal by sampling. The digital waveform signal is input to the motion sensor control unit 312.

[0044] The measurement device 3 may store each measurement data in a storage means, or the measurement device 3 may not store each measurement data, and only the MCI discrimination system 1 may store the measurement data.

[0045] [Terminal device 4] FIG. 4 is a block diagram showing an example of the configuration of the terminal device 4. The terminal device 4 includes, for example, a control unit 401, a storage unit 402, a communication unit 403, an input device 404, and a display device 405. The control unit 401 performs control processing based on software program processing, such as displaying the MCI discrimination result. The storage unit 402 stores user information, measurement data (periodic time-series data), feature data, the MCI discrimination result, and the like obtained from the MCI discrimination system 1. The communication unit 403 has a communication interface and communicates with the MCI discrimination system 1 to receive various data from the MCI discrimination system 1 and transmit user instruction input information and the like to the MCI discrimination system 1. The input device 404 includes a keyboard, a mouse, and the like. The display device 405 displays various information on a display screen 406. The display device 405 may be a touch panel including the display screen 406 and a touch sensor 407.

[0046] [Hand, motion sensor 20, finger tap measurement] FIG. 5 is a diagram showing a state in which a magnetic sensor, which is a motion sensor 20, is attached to the fingers of a user. The motion sensor 20 has a pair of coil sections, a transmitter coil section 21 and a receiver coil section 22, which are connected via a signal line 23 to the measurement device 3. The transmitter coil section 21 generates a magnetic field, and the receiver coil section 22 detects that magnetic field. In the example of FIG. 5, on the user's right hand, the transmitter coil section 21 is attached near the nail of the thumb, and the receiver coil section 22 is attached near the nail of the index finger. The fingers on which the sensors are attached can be changed. The attachment location is not limited to near the nails.

[0047] As shown in FIG. 5 , the motion sensor 20 is attached to the user's target fingers, for example, the thumb and index finger of the left hand. In this state, the user performs finger tapping, which is a repeated movement of opening and closing the two fingers. In finger tapping, the user transitions between a closed state of the two fingers, i.e., a state in which the fingertips of the two fingers are touching, and an open state of the two fingers, i.e., a state in which the fingertips of the two fingers are open. As this movement occurs, the distance between the coil sections of the transmitter coil section 21 and the receiver coil section 22, which corresponds to the distance between the fingertips of the two fingers, changes. The measurement device 3 measures a waveform signal corresponding to the change in the magnetic field between the transmitter coil section 21 and the receiver coil section 22 of the motion sensor 20.

[0048] That is, when a current flows through the transmitter coil section 21, a magnetic field is generated, and the receiver coil section 22 detects this magnetic field. The magnitude of the magnetic field detected by the receiver coil section 22 changes depending on the distance between the transmitter coil section 21 and the receiver coil section 22, so the distance between the fingertips of two fingers can be estimated from the change in the magnetic field. This distance between the fingertips of two fingers is measured as time-series data sampled at a predetermined interval (for example, every 10 msec).

[0049] The motion sensor 20 may be a sensor other than a magnetic sensor as long as it can measure the distance between two fingers. For example, a distance waveform of two fingers may be obtained by placing two fingers on a tablet terminal or a touch panel PC and repeatedly opening and closing the device. Alternatively, a distance waveform of two fingers may be obtained by detecting the shape of the hand and the positions of the fingertips using an infrared sensor.

[0050] The tasks in this embodiment include four types of finger taps: left-hand finger tap, right-hand finger tap, simultaneous finger tap on both hands, and alternating finger tap on both hands. Left-hand finger tap is a task in which the subject repeatedly opens and closes two fingers of the left hand as quickly as possible. Right-hand finger tap is a task in which the subject repeatedly opens and closes two fingers of the right hand as quickly as possible. Simultaneous finger tap on both hands is a task in which the subject repeatedly opens and closes two fingers of the left hand and two fingers of the right hand as quickly as possible with the same timing. Alternating finger tap on both hands is a task in which the subject repeatedly opens and closes two fingers of the left hand and two fingers of the right hand as quickly as possible with alternating timing. When measuring the four types of finger taps described above, the subject is instructed to keep the amplitude of the finger tap (the distance between the pads of the two fingers during the finger tap; it is 0 when the fingers are closed) between 30 and 40 mm. The measurement time for each of the four types of finger taps is a predetermined time, such as 15 seconds.

[0051] [Motion sensor control unit 312 and finger tap measurement] 6 is a diagram showing an example of the detailed configuration of the motion sensor control unit 312 etc. of the measurement device 3. Distance D indicates the distance between the transmitting coil unit 21 and the receiving coil unit 22 in the motion sensor 20. The motion sensor control unit 312 has an AC generating circuit 312a, a current generating amplifier circuit 312b, a preamplifier circuit 312c, a detection circuit 312d, an LPF (Low Pass Filter) circuit 312e, a phase adjustment circuit 312f, an amplifier circuit 312g, and an output signal terminal 312h.

[0052] A current generating amplifier circuit 312b and a phase adjustment circuit 312f are connected to the AC generating circuit 312a. The transmitting coil section 21 is connected to the current generating amplifier circuit 312b via a signal line 23. The receiving coil section 22 is connected to the preamplifier circuit 312c via a signal line 23. The detection circuit 312d, LPF circuit 312e, amplifier circuit 312g, and output signal terminal 312h are connected in this order downstream of the preamplifier circuit 312c. The detection circuit 312d is connected to the phase adjustment circuit 312f.

[0053] The AC generating circuit 312a generates an AC voltage signal of a predetermined frequency. The current generating amplifier circuit 312b converts the AC voltage signal into an AC current of a predetermined frequency and outputs it to the transmitting coil section 21. The transmitting coil section 21 generates a magnetic field by the AC current. This magnetic field generates an induced electromotive force in the receiving coil section 22. The receiving coil section 22 outputs the AC current generated by the induced electromotive force. This AC current has the same frequency as the predetermined frequency of the AC voltage signal generated by the AC generating circuit 312a.

[0054] The preamplifier circuit 312c amplifies the detected AC current. The detector circuit 312d detects the amplified signal based on a reference signal 312i from a phase adjuster circuit 312f. The phase adjuster circuit 312f adjusts the phase of the AC voltage signal of a predetermined frequency or double the frequency from the AC generator circuit 312a and outputs it as a reference signal 312i. The LPF circuit 312e band-limits the detected signal and outputs it, and the amplifier circuit 312g amplifies the signal to a predetermined voltage. An output signal corresponding to the measured waveform signal is then output from an output signal terminal 312h.

[0055] The output waveform signal has a voltage value representing the distance D between the two fingers. The distance D and the voltage value can be converted based on a predetermined formula. This formula can also be obtained by calibration. In calibration, for example, measurements are taken while the user holds a block of a predetermined length between the two fingers of the target hand. From a data set of the voltage and distance values ​​in the measured values, a predetermined formula is obtained as an approximate curve that minimizes the error. Furthermore, the size of the user's hand can be determined by calibration and used for normalizing the feature values. In this embodiment, the above-mentioned magnetic sensor is used as the motion sensor 20, and a measurement means corresponding to the magnetic sensor is used. However, other detection and measurement means, such as an acceleration sensor, a strain gauge, or a high-speed camera, may also be used.

[0056] [Integrated discriminant model generation process] 7 is a flowchart showing an example of the integrated discriminant model generation process. It is assumed that various information is stored in the subject DB 450 before the integrated discriminant model generation process starts. The main discriminant model generation unit 131 generates a main discriminant model 461 by supervised learning using the healthy group DB 451 and the pre-disease group DB 452, and stores the main discriminant model 461 in the DB 40 (S701). The supplemental discriminant model generation unit 132 generates a supplemental discriminant model 462 by supervised learning using the healthy group DB 451 and the disease group DB 453, and stores the supplemental discriminant model 462 in the DB 40.

[0057] In the following, this embodiment will be described as an example in which the main discriminant model 461 is a random forest classifier (which outputs a discrimination result based on a plurality of branching conditions based on the feature) that, when a feature of a type indicated by the feature list 430 is input, classifies whether a subject corresponding to the feature corresponds to a healthy state or an MCI state, and the supplementary discriminant model 462 is a random forest classifier that, when a feature of a type indicated by the feature list 430 is input, classifies whether a subject corresponding to the feature corresponds to a healthy state or an MCI state. In this case, for example, it can be assumed that the first stage corresponds to a state corresponding to a healthy state, the second stage corresponds to a state corresponding to an MCI state, and the third stage corresponds to a state corresponding to a dementia state.

[0058] The hyperparameters (depth of decision trees and number of decision trees in the random forest) of the main discriminant model 461 generated in step S701 and the integrated discriminant model 466 generated in step S702 may be, for example, predetermined or may be input via the input unit 103 by an administrator of the MCI discrimination system 1, etc.

[0059] The integrated discriminant model generation unit 133 sets a positive detection constant α and a supplementary model influence constant β and stores them in the integrated discriminant model constant table 463 (S703). The integrated discriminant model generation unit 133 sets the positive detection constant α and the supplementary model influence constant β input via the input unit 103 by, for example, an administrator of the MCI discrimination system 1. The positive detection constant α and the supplementary model influence constant β may also be determined in advance. Note that, in step S703, the integrated discriminant model generation unit 133 may display a setting screen on the terminal device 4 for setting the positive detection constant α and the supplementary model influence constant β. A specific example of the setting screen will be described later with reference to FIG. 19 .

[0060] Furthermore, since the positive detection constant α, the supplementary model influence constant β, the hyperparameters of the random forest that is the main discriminant model 461 generated in step S701, and the hyperparameters of the random forest that is the supplementary discriminant model 462 generated in step S702 are all included in the hyperparameters of the integrated discriminant model 466, the integrated discriminant model generation unit 133 may search for and set the hyperparameters of the integrated discriminant model 466 using a predetermined algorithm such as grid search.

[0061] The integrated discriminant model generation unit 133 generates an integrated discriminant model 466 using the main discriminant model 461 generated in step S701, the supplemental discriminant model 462 generated in step S702, and the positive detection constant α and supplemental model influence constant β set in step S703, and stores the generated integrated discriminant model 466 in DB 40 (S704). A specific example of the integrated discriminant model 466 will be described later with reference to FIG. 11. In this way, the MCI discrimination system 1 also functions as a state discriminant model generation system (state discriminant model learning system) that generates the integrated discriminant model 466.

[0062] [MCI discrimination processing] Fig. 8 is a flowchart showing an example of the MCI discrimination process for a new subject. It is assumed that the integrated discriminant model generation process in Fig. 7 has been completed before the MCI discrimination process in Fig. 8 starts.

[0063] The MCI discriminator 140 displays an initial screen on the terminal device 4 and receives an instruction to measure finger movements (finger tapping) from a new subject who is a user via the initial screen (S801). A specific example of the initial screen will be described later with reference to FIG. 15. The MCI discriminator 140 may also acquire user information of the user via the initial screen and store it in the DB 40.

[0064] The MCI discrimination unit 140 displays a task measurement screen for measuring the user's finger movement task, the measurement device 3 measures the user's finger movement task, and the control unit 101 receives the measurement results from the measurement device 3 and stores them in the DB 40 as measurement data 420 (S802). A specific example of the task measurement screen will be described later with reference to Fig. 16. The feature calculation unit 120 calculates the feature amounts indicated by the feature list 430 from the user's measurement data 420 acquired in step S802, and stores them in the DB 40 as feature data 440 (an example of analysis data) (S803).

[0065] The MCI discriminator 140 inputs the feature amount calculated in step S803 into the integrated discriminant model 466 to obtain a classification result for the user, and stores the result in the DB 40 as an MCI discrimination result 470 (S804).

[0066] Furthermore, the MCI discriminator 140 may further include information indicating the basis on which the classification result is obtained in the MCI discrimination result 470. The information indicating the basis on which the MCI discrimination result 470 is obtained may be, for example, the number of votes N for MCI obtained from the main discriminant model 461. 1,MCI and the number of votes for healthy people, N 1,NC , the number of votes for dementia obtained from the supplementary discriminant model 462, N 2,AD and the number of votes for healthy people, N 2,NC , the positive detection constant α and the supplementary model influence constant β, and the number of votes for MCI obtained from the integrated discriminant model 466, N MCI and the number of votes for healthy people, N NC Details of each vote count will be explained later.

[0067] Furthermore, the MCI discrimination section 140 may calculate the feature contribution of each feature corresponding to the user to the classification result obtained in step S804 (i.e., the determination result of whether the user is healthy or has MCI) using a predetermined algorithm such as SHAP (SHapley Additive exPlanations), and may further include the calculated feature contribution in the MCI discrimination result 470.

[0068] The result output unit 150 displays a result display screen indicating the MCI discrimination result 470 and the contribution of the feature on the terminal device 4 (S805). A specific example of the result display screen will be described later with reference to FIG. 17. The result output unit 150 may also display a basis presentation screen on the terminal device 4, which displays information indicating the basis on which the classification result obtained by the integrated discriminant model 466 was obtained. A specific example of the basis presentation screen will be described later with reference to FIG. 18.

[0069] [Waveform signal of feature quantity] 9A to 9D show examples of waveform signals of feature quantities. FIG. 9A is a graph showing a waveform signal of the distance D between two fingers. FIG. 9B is a graph showing a waveform signal of the velocity of two fingers. FIG. 9C is a graph showing a waveform signal of the acceleration of two fingers. The velocity in FIG. 9B is obtained by time differentiation of the waveform signal of the distance in FIG. 9A. The acceleration in FIG. 9C is obtained by time differentiation of the waveform signal of the velocity in FIG. 9B. The feature quantity calculation unit 120 obtains a waveform signal of a predetermined feature quantity as in this example from the waveform signal of the measurement data 420 based on calculations such as differentiation and integration. Furthermore, the feature quantity calculation unit 120 obtains a value from the feature quantity by a predetermined calculation.

[0070] FIG. 9D is an enlarged graph of FIG. 9A showing example features. It shows the maximum value Dmax of the finger tap distance D, the tap interval TI, and so on. The horizontal dashed line indicates the average value Dav of the distance D over the entire measurement time. The maximum value Dmax indicates the maximum value of the distance D over the entire measurement time. The tap interval TI is the time corresponding to the period TC of one finger tap, and in particular indicates the time from the minimum point Pmin to the next minimum point Pmin. It also shows the maximum point Pmax and minimum point Pmin within one period of the distance D, the time T1 for the opening operation, and the time T2 for the closing operation, which will be described later.

[0071] Further detailed examples of feature quantities are shown below. In this embodiment, a plurality of feature quantities obtained from the waveforms of the distance, velocity, and acceleration are used. Note that only some of the following plurality of feature quantities may be used, or other feature quantities may be used, and the definition of the feature quantities is not limited in detail.

[0072] Fig. 10 is a diagram showing an example of a feature list 430. This association setting is an example and can be changed. The feature list 430 in Fig. 10 has columns for feature category, feature number, and feature parameter (feature name and unit). The feature category includes [distance], [velocity], [acceleration], [tap interval], and [phase difference].

[0073] For example, the feature category [distance] has multiple feature parameters identified by feature numbers (A1) to (A7). The brackets [ ] around the feature parameters indicate units. (A1) "Maximum distance amplitude" [mm] is the difference between the maximum and minimum amplitude values ​​in the distance waveform (Figure 9A). (A2) "Total distance traveled" [mm] is the sum of the absolute values ​​of the distance change amount over the entire measurement time of one measurement.

[0074] (A3) "Average of maximum distance values" [mm] is the average of the maximum amplitude values ​​for each cycle. (A4) "Standard deviation of maximum distance values" [mm] is the standard deviation for the above values. (A5) "Slope of approximate curve of maximum distance points" (decay rate) [mm / sec] is the slope of the curve approximating the maximum amplitude points. This parameter mainly represents the change in amplitude due to fatigue during the measurement period.

[0075] (A6) "Coefficient of variation of the maximum distance value" is the coefficient of variation of the maximum amplitude value, and its unit is a dimensionless quantity (indicated by [-]). This parameter is a value obtained by normalizing the standard deviation by the mean, and therefore, it can eliminate individual differences in finger length. (A7) "Standard deviation of the local maximum distance value" [mm] is the standard deviation of the maximum amplitude values ​​at three adjacent locations. This parameter is used to evaluate the degree of local variation in amplitude over a short period of time.

[0076] The feature category [velocity] has the feature parameters indicated by the following feature numbers (A8) to (A22). (A8) "Maximum velocity amplitude" [m / sec] is the difference between the maximum and minimum velocity values ​​in the velocity waveform (Figure 9B). (A9) "Average opening velocity maximum value" [m / sec] is the average of the maximum velocity values ​​during the opening motion of each finger tap waveform. The opening motion is the motion of moving two fingers from a closed state to a maximum open state (Figure 9D). (A10) "Average closing velocity minimum value" [m / sec] is the average of the minimum velocity values ​​during the closing motion. The closing motion is the motion of moving two fingers from a maximum open state to a closed state. (A11) "Standard deviation of opening velocity maximum value" [m / sec] is the standard deviation of the maximum velocity values ​​during the opening motion.

[0077] (A12) "Standard deviation of minimum closing speed" [m / sec] is the standard deviation of minimum speed during the closing motion. (A13) "Energy balance" [-] is the ratio of the sum of squares of the speed during the opening motion to the sum of squares of the speed during the closing motion. (A14) "Total energy" [m 2 / sec 2] is the sum of squares of the speed during the entire measurement time. (A15) "Coefficient of variation of the maximum value of opening speed" [-] is the coefficient of variation of the maximum value of speed during the opening movement, and is the value obtained by normalizing the standard deviation by the mean.

[0078] (A16) "Coefficient of variation of maximum closing speed" [-] is the coefficient of variation of the maximum speed during the closing motion. (A17) "Number of tremors" [-] is the number of times the speed waveform changes from positive to negative minus the number of large opening and closing finger taps. (A18) "Average distance ratio at peak opening speed" [-] is the average value of the ratio of the distance at the maximum speed during the opening motion, assuming the amplitude of the finger tap is 1.0. (A19) "Average distance ratio at peak closing speed" [-] is the average of the similar ratio of the distance at the minimum speed during the closing motion.

[0079] (A20) "Ratio of distance ratio at peak speed" [-] is the ratio of the value of (A22) to the value of (A23). (A21) "Standard deviation of distance ratio at peak opening speed" [-] is the standard deviation of the ratio of the distance at the maximum speed during the opening movement, when the amplitude of the finger tap is 1.0. (A22) "Standard deviation of distance ratio at peak closing speed" [-] is the standard deviation of the same ratio of the distance at the minimum speed during the closing movement.

[0080] The feature category [acceleration] has the feature parameters indicated by the following feature numbers (A23) to (A32): (A23) "Maximum amplitude of acceleration" [m / s 2 ] is the difference between the maximum and minimum values ​​of acceleration in the acceleration waveform (Figure 9C). (A24) "Average of the opening acceleration maximum value" [m / s 2 ] is the average of the maximum acceleration values ​​during the opening movement, and is the first of four extreme values ​​that appear during one cycle of finger tapping. (A25) "Average of the minimum acceleration values ​​of the opening movement" [m / s 2 ] is the average of the acceleration minima during the opening movement, and is the second of the four extreme values.

[0081] (A26) "Average of the maximum value of closing acceleration" [m / s 2 ] is the average of the maximum acceleration values ​​during the closing operation, and is the third of the four extreme values. (A27) "Average of the minimum closing acceleration values" [m / s 2 ] is the average of the minimum acceleration values ​​during the closing motion, and is the fourth of the four extreme values.

[0082] (A28) "Average contact time" [seconds] is the average contact time when two fingers are closed. (A29) "Standard deviation of contact time" [seconds] is the standard deviation of the above contact times. (A30) "Coefficient of variation of contact time" [-] is the coefficient of variation of the above contact times. (A31) "Number of zero crossings of acceleration" [-] is the average number of times the acceleration changes from positive to negative during one cycle of finger tapping. Ideally, this value should be 2 times. (A32) "Number of freezing movements" [-] is the value obtained by subtracting the number of finger taps with large opening and closing movements from the number of times the acceleration changes from positive to negative during one cycle of finger tapping.

[0083] The feature category [Tap Interval] has the following feature parameters indicated by feature numbers (A33) to (A40). (A33) "Number of Taps" [-] is the number of finger taps during the entire measurement time of one measurement. (A34) "Tap Interval Average" [sec] is the average for the tap intervals mentioned above (Figure 9D) in the distance waveform. (A35) "Tap Frequency" [Hz] is the frequency at which the spectrum is maximized when the distance waveform is Fourier transformed.

[0084] (A36) "Tap interval standard deviation" [seconds] is the standard deviation of the tap interval. (A37) "Tap interval variation coefficient" [-] is the variation coefficient of the tap interval, which is the value obtained by normalizing the standard deviation by the mean value. (A38) "Tap interval variation" [mm 2 ] is the integrated value for frequencies between 0.2 and 2.0 Hz when the tap interval is spectrally analyzed.

[0085] (A39) "Skewness of tap interval distribution" [-] is the skewness of the frequency distribution of tap intervals, and indicates the degree to which the frequency distribution is skewed compared to a normal distribution. (A40) "Standard deviation of local tap intervals" [seconds] is the standard deviation of three adjacent tap intervals.

[0086] The feature category [phase difference] has the following feature parameters indicated by feature numbers (A41) to (A44). (A41) "Average phase difference" [degrees] is the average phase difference between the distance waveforms of both hands. The phase difference is an index value that represents the difference in angle between the finger tap of the left hand and the right hand, assuming one cycle of the finger tap of the right hand is 360 degrees. When there is no difference, it is set to 0 degrees. (A42) "Standard deviation of phase difference" [degrees] is the standard deviation of the phase difference. The larger the values ​​of (A41) and (A42), the larger and more unstable the difference between the two hands is. (A43) "Similarity between both hands" [-] is a value that represents the correlation when the time difference is 0 when a cross-correlation function is applied to the waveforms of the left and right hands. (A44) "Time difference at which the similarity between both hands is maximum" [seconds] is a value that represents the time difference at which the correlation of (A43) is maximum.

[0087] Since the features (A1) through (A40) can be calculated for both left-hand and right-hand finger taps, a total of 80 features can be calculated for both left-hand and right-hand finger taps. For simultaneous finger taps on both hands, a total of 80 features (A1) through (A40) can be calculated for each hand, and four features (A41) through (A44) can be calculated for each hand, for a total of 84 features. For alternate finger taps on both hands, a total of 80 features (A1) through (A40) can be calculated for each hand, and four features (A41) through (A44) can be calculated for each hand, for a total of 84 features. Therefore, a total of 248 features shown in the feature list 430 of FIG. 10 can be calculated for the four types of finger taps: left-hand finger taps, right-hand finger taps, simultaneous finger taps on both hands, and alternate finger taps on both hands.

[0088] All of these 248 features may be calculated and input to the main discriminant model 461, the supplementary discriminant model 462, and the integrated discriminant model 466. Alternatively, for example, features whose F1 value, which is an example of a discrimination accuracy index described later, is a predetermined value or greater (e.g., 0.5 or greater) may be selected from these 248 features, and only the selected features may be calculated and input to the main discriminant model 461, the supplementary discriminant model 462, and the integrated discriminant model 466.

[0089] [Specific example of a discriminant model] FIG. 11 is an explanatory diagram showing an example of the main discriminant model 461, the supplementary discriminant model 462, and the integrated discriminant model 466. MCI is a precursor to dementia and exhibits symptoms intermediate between those of a healthy individual and those of dementia. Therefore, motor function in finger movements also deteriorates in the following order: healthy, MCI, and dementia. Therefore, by shifting the threshold value in the supplementary discriminant model 462 for discriminating between healthy and dementia closer to the healthy individual side, MCI, a precursor to dementia, can be detected. In other words, the integrated discriminant model 466, which supplementarily integrates the main discriminant model 461 for discriminating between healthy and MCI, can discriminate between healthy and MCI with high accuracy. This allows one embodiment of the present invention to accurately discriminate whether the state of a new analysis target corresponds to an intermediate stage among abnormal states having multiple stages.

[0090] The primary discriminant model 461 includes a set of N1 decision trees generated by a random forest. When the features of a new subject are input to the primary discriminant model 461, each of the N1 decision trees outputs a judgment result (voting result) as to whether the new subject is healthy or MCI. Therefore, the number of decision trees that have judged the new subject to have MCI in the primary discriminant model 461 (i.e., the number of votes for MCI) N 1,MCI and the number of decision trees that determined that the new subject was healthy in the primary discriminant model 461 (i.e., the number of votes for healthy) N 1,NC and we get N 1,MCI And, N 1,NC and are examples of the first result, and N 1,NC is an example of the first discriminant value, and N 1,MCI is an example of a second discriminant value.

[0091] If the new subject is determined to be healthy or MCI using only the main discriminant model 461, for example, N 1,MCI >N 1,NC If so, the new subject is diagnosed as having MCI, and N 1,MCI ≦N 1,NC If so, the new subject is determined to be healthy.

[0092] The supplementary discriminant model 462 includes a set of N2 decision trees generated by a random forest. When the feature values ​​of a new subject are input to the supplementary discriminant model 462, a judgment result (voting result) of whether the new subject is healthy or has dementia (AD) is output from each of the N2 decision trees. Therefore, the number of decision trees that have judged the new subject to have dementia in the supplementary discriminant model 462 (i.e., the number of votes for dementia) N 2,AD and the number of decision trees that determined the tree to be healthy in the supplementary discriminant model 462 (i.e., the number of votes for healthy) N 2,NC and we get N 2,AD And, N 2,NC and are examples of the second result, and N 2,NC is an example of the third discriminant value, and N 2,AD is an example of a fourth discriminant value.

[0093] If the supplementary discriminant model 462 is used to determine whether a new subject is healthy or has dementia, for example, N 2,AD >N 2,NC If so, the new subject is determined to have dementia, and N 2,AD ≦N 2,NC If so, the new subject is determined to be healthy.

[0094] The integrated discriminant model 466 includes a set of N1 decision trees of the main discriminant model 461 and a set of N2 decision trees of the supplemental discriminant model 462. The integrated discriminant model 466 in the example of Fig. 11 can also be called an integrated random forest in which the random forest, which is the main discriminant model 461, and the random forest, which is the supplemental discriminant model 462, are integrated.

[0095] When the feature quantities of a new subject are input to the integrated discriminant model 466, a determination result (voting result) as to whether the new subject is healthy or has MCI is output from each of the N1 decision trees in the main discriminant model 461, and a determination result (voting result) as to whether the new subject is healthy or has dementia is output from each of the N2 decision trees in the supplemental discriminant model 462. As described above, the integrated discriminant model 466 is set with a positive detection constant α>1.0 and a supplemental model influence constant β>0.

[0096] The integrated discriminant model 466 is the number of votes for MCI by the primary discriminant model 461, N 1,MCI is the number of votes for dementia N by the supplementary discriminant model 462 2,AD N corrected by MCI =N 1,MCI +αβN 2,AD and the number of votes for healthy people according to the primal discriminant model 461, N 1,NC is the number of votes for healthy people N by the supplementary discriminant model 462 2,NC N corrected by NC =N 1,NC +βN 2,NC The MCI discriminator 140 outputs, for example, N MCI >N NC If so, the new subject is determined to have MCI, and N MCI ≦N NC If so, the new subject is determined to be healthy.

[0097] Here, the supplementary discriminant model 462 is a model for discriminating between healthy and dementia. In order to more strongly reflect the influence of the voting results for dementia (worse symptoms than MCI) in the supplementary discriminant model 462 on the voting results for MCI in the integrated discriminant model 466, the integrated discriminant model 466 is set to N as described above. 2,AD is multiplied by a positive detection constant α>1.0, and N 2,NC is not multiplied by the positive detection constant α > 1.0.

[0098] In addition, in the integrated discriminant model 466, in order to adjust the magnitude of the influence of the supplementary discriminant model 462 on the main discriminant model 461, N2,AD and N 2,NC Both are multiplied by a supplementary model influence constant β>0.

[0099] It is desirable that the positive detection constant α and the supplementary model influence constant β are set to a predetermined upper limit value or less (the upper limit value for α and the upper limit value for β may be different). For example, if the positive detection constant α is set to be extremely large compared to N1 and N2, N 2,AD The number of votes other than those will no longer contribute to the discrimination result (N 2,AD If >0 is satisfied, then N MCI >N NC ), the discrimination accuracy will decrease. In addition, for example, if the supplementary model influence constant β is set to be extremely large compared to N1 and N2, the number of votes N by the main discriminant model 461 will 1,MCI and N 1,NC will no longer contribute to the discrimination result (i.e., the discrimination result will be determined only by the voting result of the supplementary discriminant model 462 and α), resulting in a decrease in discrimination accuracy. In order to allow each vote number to contribute to the discrimination result, it is desirable that the upper limits for α and β be determined based on the total number N1 of decision trees in the main discriminant model 461 and the total number N2 of decision trees in the supplementary discriminant model 462.

[0100] In the above example, the model generation unit 130 generates the integrated discriminant model 466 and stores it in the DB 40, but the processing of step S704 may be omitted. In this case, in step S804, the MCI discrimination unit 140 may acquire an MCI discrimination result by integrating the voting results obtained by inputting the feature amounts calculated in step S803 into the main discriminant model 461 and the supplementary discriminant model 462. That is, the MCI discrimination unit 140 acquires an MCI discrimination result by integrating the voting results N 1,MCI and N 1,NC , and the voting result N by the supplementary discriminant model 462 2,NC and N 2,AD , to N MCI =N 1,MCI +αβN 2,AD And, N NC =N 1,NC +βN 2,NCCalculate and N MCI and N NC The MCI discrimination result may be calculated by comparing the above.

[0101] [Comparison of discrimination accuracy] Below, we will explain a comparison between the discrimination accuracy in MCI discrimination (discriminating whether a subject is healthy or has MCI) using only the main discriminant model 461 and the discrimination accuracy in MCI discrimination using the integrated discriminant model 466 of this embodiment.

[0102] Fig. 12 is a diagram showing an example of search ranges for hyperparameters. Fig. 12 shows the search ranges for hyperparameters (i.e., the maximum depth of the decision tree and the number of decision trees in the main discriminant model 461) when MCI discrimination is performed using only the main discriminant model 461 (comparative example), and the search ranges for hyperparameters (i.e., the maximum depth of the decision tree and the number of decision trees in the main discriminant model 461 and the supplementary discriminant model 462, as well as the positive detection constant α and the supplementary model influence constant β) when MCI discrimination is performed using the integrated discriminant model 466. These search ranges are determined in advance.

[0103] FIG. 13 is a diagram showing the results of comparison between the discrimination accuracy in MCI discrimination using only the main discriminant model 461 and the discrimination accuracy in MCI discrimination using the integrated discriminant model 466 of this embodiment.

[0104] As shown in Figure 13, each discrimination accuracy is evaluated using the F1 score. The F1 score is the harmonic mean of the recall and precision. Precision is the proportion of those predicted as positive (MCI in this case) that were actually correctly predicted. Recall (also called sensitivity) is the proportion of those predicted as positive out of all positives. The F1 score ranges from 0 to 1, with 1 being the highest accuracy and 0 being the lowest accuracy.

[0105] Note that both the main discriminant model 461 and the integrated discriminant model 466 have hyperparameters, and therefore, hyperparameter tuning is required. However, if hyperparameter tuning and discriminant method selection are performed simultaneously using ordinary cross validation (CV), the hyperparameters may be adapted to the validation data, making it impossible to accurately evaluate accuracy, and there is a risk that the discriminant accuracy for unknown data may decrease.

[0106] To prevent such a decline in generalization performance, the hyperparameter values ​​and the discrimination accuracy values ​​of each discriminant model in Fig. 13 are calculated using nCV (nested cross validation), which can perform both hyperparameter tuning and discrimination accuracy comparison while separating them. nCV is also called double cross validation, and is performed using a doubly nested CV structure.

[0107] In nCV, the inner CV performs hyperparameter tuning to select the optimal hyperparameters for each discrimination method. In nCV, the outer CV evaluates the discrimination accuracy using the hyperparameters selected in the inner CV and selects the method with the highest accuracy from multiple discrimination methods.

[0108] In this way, nCV prevents the above-mentioned degradation in generalization performance by preventing the mixing of validation data used for hyperparameter tuning and validation data used for discrimination accuracy evaluation. When selecting hyperparameters in the inner CV, nCV selects hyperparameters that minimize the generalization error (the difference between the discrimination accuracy for the training data and the discrimination accuracy for the validation data) from the hyperparameter candidates that maximize the discrimination accuracy in each fold divided by the CV. nCV also selects the discrimination method that minimizes the above-mentioned generalization error in the outer CV. In the example in Figure 13, the number of divisions in the inner CV is 3, and the number of divisions in the outer CV is 4. Stratified k-fold cross-validation is used so that the healthy group and the MCI group are evenly divided into each fold.

[0109] The subjects of evaluation in the example of FIG. 13 were 210 patients diagnosed by a doctor with Alzheimer's disease, 182 patients diagnosed with MCI, and 352 healthy individuals.

[0110] 13 , the F1 value for the MCI discrimination using only the main discriminant model 461 (i.e., the normal random forest) is 0.737, and the F1 value for the MCI discrimination using the integrated discriminant model 466 is 0.795. Thus, the integrated discriminant model 466 has higher discrimination accuracy than the main discriminant model 461.

[0111] The integrated discriminant model 466 uses not only the main discriminant model 461 but also the supplementary discriminant model 462, and therefore some MCI patients who would have been judged as healthy and missed detection by the main discriminant model 461 alone can be detected by supplementally using the results of the supplementary discriminant model 462, resulting in higher discrimination accuracy.

[0112] As a result of hyperparameter tuning of the integrated discriminant model 466, the positive detection constant α was determined to be 10.0, and the supplementary model influence constant β was determined to be 0.1. β = 0.1 means that the influence of the supplementary discriminant model 462 is limited to one-tenth of the influence of the main discriminant model 461. In other words, in the integrated discriminant model 466, while the main discriminant model 461 created from the MCI group and the healthy group is still the main focus, the supplementary discriminant model 462 picks up at least some of the MCI patients that cannot be detected by the main discriminant model 461. α = 10.0 indicates that in order to use the supplementary discriminant model 462, which discriminates between healthy and dementia, for MCI detection, it is necessary for the supplementary discriminant model 462 to make a positive determination 10 times more sensitively.

[0113] FIG. 14 is a diagram showing the feature contribution of the integrated discriminant model 466. The feature contribution is an index showing the magnitude of influence on the MCI discrimination result using the integrated discriminant model 466. SHAP (SHapley Additive exPlanations) was used to calculate the feature contribution shown in FIG. 14. SHAP is a method for interpreting the predictions of a machine learning model, and indicates the degree of contribution of each feature to the model prediction. The final integrated discriminant model 466 was generated using the hyperparameters obtained by nCV and all data of the healthy group and the MCI group, and SHAP was applied.

[0114] For 100 samples sampled from all the data, the SHAP value of each feature of each subject included in the sample was calculated, and the average SHAP value for each feature was used as the feature contribution for that feature. Figure 14 shows the top 20 features and their feature contributions in descending order.

[0115] As shown in Fig. 14, the top 20 features, when classified by task type, consist of six features for left-hand finger tapping, one feature for right-hand finger tapping, three features for both-hand simultaneous finger tapping, and ten features for both-hand alternate finger tapping. Also, as shown in Fig. 14, the top 20 features, when classified by the hand being measured, consist of 16 features related to the left hand and four features related to the right hand (features related to both hands are not included).

[0116] [Various display screens] Various display screens displayed on the display screen of the terminal device 4 will be described below.

[0117] 15 is a diagram showing an example of the screen configuration of the initial screen displayed in step S801. The initial screen 1500 includes, for example, a user information column 1501 and an operation menu column 1502.

[0118] The user information field 1501 is an area for accepting input of user information by the user, and when the "Set" button is selected, the information input into the user information field 1501 is registered in the DB 40 as user information 410. Note that if user information has already been input into an electronic medical record or the like, the MCI discrimination system 1 may be configured to link with that user information.

[0119] Examples of user information that can be input include a user ID, name, date of birth or age, gender, dominant hand, disease / symptoms, and notes. Dominant hand can be selected and input from right hand, left hand, both hands, unknown, etc. Disease / symptoms can be selected and input from options in a list box, for example, or can be input as arbitrary text. When the human data measurement system is used in a hospital or the like, a doctor or the like can input user information on behalf of the user, rather than the user. The MCI discrimination system 1 can also be applied when user information is not registered.

[0120] The operation menu field 1502 displays operation items for the functions provided by the service. The operation items include "calibration," "measurement of finger movements," "mild cognitive impairment risk assessment," and "end." When "calibration" is selected, the above-mentioned calibration, i.e., processing related to adjustment of the movement sensor 20 to the user's fingers, is performed. The status of whether or not adjustment has been completed is also displayed.

[0121] When "Measure finger movement" is selected, the screen transitions to a task measurement screen (Fig. 16) for measuring finger movement tasks such as finger tapping. When "Mild cognitive impairment risk assessment" is selected, the processing of steps S803 to S805 is executed for the measured data, and the screen transitions to a result display screen (Fig. 17) for displaying the MCI discrimination result 470. When "Exit" is selected, the service ends.

[0122] 16 is a diagram showing an example of the screen configuration of the task measurement screen displayed in step S802. The task measurement screen 1600 displays task information. The task measurement screen 1600 includes a graph display area 1601 for displaying a measurement waveform 1602, for example, for each of the left and right hands, with the horizontal axis representing time and the vertical axis representing the distance between two fingers. The task measurement screen 1600 may further display other instruction information for explaining the content of the task. Specifically, for example, the task measurement screen 1600 may be provided with a video area for explaining the content of the task using video and audio.

[0123] The task measurement screen 1600 includes operation buttons such as "Start measurement," "Retry measurement," "End measurement," and "Save (register)." The user selects "Start measurement" according to the task information on the screen and performs the task exercise. The measurement device 3 measures the task exercise and obtains a waveform signal.

[0124] The terminal device 4 displays in real time on the graph display area 1601 a measured waveform 1602 corresponding to the waveform signal being measured. After exercising, the user selects "End measurement," and to confirm, selects "Save (register)." The measurement device 3 transmits the measurement data to the MCI discrimination system 1, which registers it in the DB 40 as measurement data 420.

[0125] 17 is a diagram showing an example of the screen configuration of the result display screen displayed in step S805. The result display screen 1700 includes, for example, a user information field 1701, a discrimination result field 1702, and a feature contribution field 1703. The user information field 1701 displays the user information entered on the initial screen 1500.

[0126] The discrimination result column 1702 displays the MCI discrimination result by the integrated discriminant model 466. When the discrimination result by the integrated discriminant model 466 is MCI, "Yes" is displayed in the "Mild cognitive impairment risk judgment" in the discrimination result column 1702, and when the discrimination result by the integrated discriminant model 466 is healthy, "No" is displayed in the "Mild cognitive impairment risk judgment" in the discrimination result column 1702.

[0127] The feature contribution column 1703 displays the feature contribution (calculated using, for example, SHAP) of the user to the MCI discrimination result of the individual integrated discriminant model 466. In the example of Fig. 17 , the feature contribution column 1703 displays features with high feature contribution (for example, a predetermined number of feature values ​​with the highest SHAP values, or feature values ​​with SHAP values ​​equal to or greater than a predetermined value).

[0128] The result display screen 1700 includes operation buttons such as "Display basis for judgment" and "End." When "Display basis for judgment" is selected, the screen transitions to the basis presentation screen. When "End" is selected, the MCI determination process ends.

[0129] 18 is a diagram showing an example of the screen configuration of the evidence presentation screen. The evidence presentation screen 1800 includes, for example, a main discriminant model voting result field 1801, a supplementary discriminant model voting result field 1802, a hyperparameter display field 1803, and an integrated information display field 1804. The main discriminant model voting result field 1801 displays voting results obtained by inputting features into the main discriminant model 461. The supplementary discriminant model voting result field 1802 displays voting results obtained by inputting features into the supplementary discriminant model 462.

[0130] The hyperparameter display field 1803 displays the values ​​of the hyperparameters of the integrated discriminant model 466. In the example of Fig. 18, the hyperparameter display field 1803 displays only the values ​​of some of the hyperparameters (the positive detection constant α and the supplementary model influence constant β), but the values ​​of other hyperparameters (the number of decision trees and the maximum depth of the decision tree) may also be displayed.

[0131] The integrated information display field 1804 displays information indicating the basis for the determination by the integrated discriminant model 466. Specifically, for example, the integrated information display field 1804 displays information indicating the process of integrating the voting results indicated in the supplemental discriminant model voting result field 1802 with the voting results indicated in the main discriminant model voting result field 1801 using the positive detection constant α and the supplemental model influence constant β indicated in the hyperparameter display field 1803, and the basis for obtaining the discrimination result by the integrated discriminant model 466.

[0132] The basis presentation screen 1800 includes operation buttons such as "Back." When "Back" is selected, the screen transitions to the result display screen 1700.

[0133] 19 is a diagram showing an example of the screen configuration of the setting screen displayed in step S703. The setting screen 1900 includes, for example, a parameter setting field 1901. The parameter setting field 1901 accepts the selection of a discrimination method (an algorithm used in the main discriminant model 461 and the supplemental discriminant model 462). The parameter setting field 1901 also accepts input of a positive detection constant α and a supplemental model influence constant β.

[0134] The number of decision trees and the maximum depth of the decision trees of the integrated discriminant model 466 (i.e., the main discriminant model 461 and the supplementary discriminant model 462) may be determined by a predetermined algorithm such as a grid search, or may be set in the parameter setting field 1901 by displaying the setting screen 1900 before steps S701 and S702.

[0135] The setting screen 1900 includes operation buttons such as "Settings." When "Settings" is selected, the processing of step S703 ends.

[0136] [User Information 410] 20 is a diagram showing an example of the data configuration of user information 410. The user information 410 indicates, for example, a user ID, gender, age, and measurement task information. The user ID is unique identification information for a user in the human data measurement system. The measurement task information indicates the date and time when the user measured a task using the measurement device 3, the type of the measured task, and the like.

[0137] [Application example of the integrated discrimination model 466] In the above example, the MCI discrimination system 1 discriminates whether a new subject is healthy or has MCI, but the MCI discrimination system 1 of this embodiment can also be applied to cases where a new subject needs to be discriminated whether they have dementia or MCI. In this case, the integrated discriminant model 466 and the main discriminant model 461 can be used as models for discriminating between a dementia group and an MCI group.

[0138] Furthermore, the MCI discrimination system 1 of this embodiment can be applied to discriminating between stages of symptoms not only for MCI and dementia but also for any disease that has similar properties but different stages of symptoms (e.g., mild and severe, pre-disease and disease). For example, the MCI discrimination system 1 of this embodiment can be applied to discriminating between high blood pressure and hypertension (a case in which healthy and hypertension are discriminated, or hypertension and hypertension are discriminated), pre-diabetes and diabetes (a case in which healthy and pre-diabetes are discriminated, or pre-diabetes and diabetes). In these examples, MCI is a pre-disease of dementia, which is a disease, hypertension is a pre-disease of hypertension, which is a disease, and pre-diabetes is a pre-disease of diabetes, which is a disease.

[0139] Furthermore, the MCI discrimination system 1 of this embodiment can be applied to discriminating not only diseases but also stages indicating the degree of an abnormal state. Specifically, for example, when the stages of an abnormal state of a product to be analyzed produced in a factory include normal, mildly abnormal, and severely abnormal, the MCI discrimination system 1 of this embodiment can be applied to discriminating whether the product to be analyzed is normal or mildly abnormal, or whether the product to be analyzed is mildly abnormal or severely abnormal.

[0140] In addition, there are N levels (N is a natural number of 4 or more) of states in which the degree of the abnormal state differs step by step, and it is determined whether the analysis target is in the L-th level (L is a natural number of 1 or more and N - 1 or less) or the M-th level (M is a natural number of 2 or more and N or less, and further L < M). The MCI discrimination system 1 of this embodiment can also be applied to such an example.

[0141] In this case, for example, the main discrimination model 461 is constituted by a random forest that outputs whether the state of the analysis target corresponds to the L-th level or the M-th level when the feature amount of the analysis target is input. In addition, the supplementary discrimination model 462 is constituted by a random forest that outputs, for example, whether the state of the analysis target corresponds to the X-th level (X is a natural number of 1 or more and L or less) or the Y-th level (Y is a natural number of M or more and N or less) (however, at least one of X ≠ L and Y ≠ M is satisfied) when the feature amount of the analysis target is input. Note that a plurality of supplementary discrimination models 462 satisfying the above conditions regarding X and Y may be generated, or one supplementary discrimination model 462 may be generated.

[0142] For example, for each of the above-described supplementary discrimination models 462, a positive detection constant α i > 1.0 and a supplementary model influence constant β i > 0 and (i is a natural number of 1 or more and less than or equal to the number of supplementary discrimination models 462) are set. Note that the larger the difference between L and X, and the larger the difference between M and Y, the smaller the positive detection constant α i and the supplementary model influence constant β i for each of the supplementary discrimination models 462 are desirably small values.

[0143] In the integrated discrimination model 466, for example, the number of votes for the L-th level in the main discrimination model 461 is added to the product of the number of votes for the X-th level in the supplementary discrimination model 462 and the positive detection constant and the supplementary model influence constant (corresponding to the supplementary discrimination model 462), and the number of votes for the M-th level in the main discrimination model 461 is added to the product of the number of votes for the Y-th level in the supplementary discrimination model 462 and the supplementary model influence constant (corresponding to the supplementary discrimination model 462).

[0144] In the above example, a random forest is used for the main discriminant model 461, the supplementary discriminant model 462, and the integrated discriminant model 466. However, a discriminant method other than a random forest may be used. Specifically, for example, a discriminant method such as XGBoost (eXtreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine), or SVM (Support Vector Machine) may be applied to the main discriminant model 461, the supplementary discriminant model 462, and the integrated discriminant model 466.

[0145] For example, when two-class discrimination is performed using XGBoost and LightGBM, the probability of class classification is output. In addition, in SVM, a classification score is calculated, and the probability of class classification can be calculated based on the classification score. Therefore, when one of these discrimination methods is used, when the feature to be analyzed is input to the main discriminant model 461, the probability of class classification corresponding to MCI P M,MCI (and the probability of classification corresponding to normality P M,NC =1-P M,MCI ) is output, and when the feature to be analyzed is input to the supplementary discriminant model 462, the classification probability P S,AD (and the probability of classification corresponding to normality P S,NC =1-P S,AD ) is output.

[0146] In the integrated discriminant model 466, the supplementary model influence constant β (where 1.0 > β > 0.0) is used to calculate P M,MCI and P S,AD The combined probability of classification into MCI, P I,MCI =(1.0-β)P M,MCI +βP S,AD and the probability of classification corresponding to normality P I,NC =1-P I,MCI and is output, and P I,MCI ≧P I,NC If the condition is met, the patient is judged to have MCI, and if the condition is met, the patient is judged to have MCI. I,MCI <P I,NCIn order to more sensitively detect MCI, whose symptoms are closer to normal than dementia, the probability of classification into dementia, P S,AD may be multiplied by a positive detection constant α>1.0, that is, in the integrated discriminant model 466, the class classification probability corresponding to MCI is P I,MCI =(1.0-β)P M,MCI +αβP S,AD It may also be calculated by:

[0147] The present invention is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those including all of the described configurations. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

[0148] Furthermore, the above-described configurations, functions, processing units, processing means, etc. may be partially or entirely implemented in hardware, for example, by designing them as integrated circuits. The above-described configurations, functions, etc. may also be implemented in software, with a processor interpreting and executing a program that implements each function. Information such as the programs, tables, and files that implement each function can be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

[0149] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the product. In reality, it can be assumed that almost all components are interconnected. [Explanation of symbols]

[0150] 1 MCI discrimination system, 40 DB, 101 control unit, 102 memory unit, 103 input unit, 104 output unit, 105 communication unit, 120 feature calculation unit, 130 model generation unit, 140 MCI discrimination unit, 150 result output unit, 440 feature data, 450 subject DB, 461 main discriminant model, 462 supplementary discriminant model, 464 positive detection constant, 465 supplementary model influence constant, 466 integrated discriminant model, 1700 result display screen, 1800 evidence presentation screen

Claims

1. A state determination system, a processor and a memory, The memory includes: Analysis data for new analysis targets, a primary discriminant model that, when analytical data of the new analysis target is input, outputs a first result regarding whether the state of the new analysis target corresponds to the first stage or the second stage; a supplementary discrimination model that, when analysis data of the new analysis target is input, outputs a second result regarding whether the state of the new analysis target corresponds to the first stage or the third stage; The first stage, the second stage, and the third stage indicate the degree of the condition, the degree of the condition in the second stage is between the degree of the condition in the first stage and the degree of the condition in the third stage; the primary discriminant model is generated by machine learning using analysis data of a first analysis object group consisting of analysis objects whose state is the first stage and a second analysis object group consisting of analysis objects whose state is the second stage, the supplementary discrimination model is generated by machine learning using analysis data of the first analysis object group and a third analysis object group consisting of analysis objects whose state is the third stage, The processor: inputting analytical data of the new analyte into the primary discriminant model to obtain a first result for the new analyte; inputting analytical data of the new analyte into the supplemental discriminant model to obtain a second result for the new analyte; integrating the acquired first result and the acquired second result to calculate a discrimination result indicating whether the state of the new analysis target corresponds to the first stage or the second stage; a condition determination system that generates data for displaying the determination result;

2. The condition determination system according to claim 1, when analysis data of the new analysis object is input, the primary discriminant model outputs a first discriminant value indicating whether the state of the new analysis object corresponds to the first stage and a second discriminant value indicating whether the state of the new analysis object corresponds to the second stage; when analysis data of the analysis object is input, the supplemental discriminant model outputs a third discriminant value relating to whether the state of the new analysis object corresponds to the first stage and a fourth discriminant value relating to whether the state of the new analysis object corresponds to the third stage; The processor: obtaining the first discriminant value and the second discriminant value for the new analysis subject as the first result for the new analysis subject; obtaining the third discriminant value and the fourth discriminant value for the new analyte as the second result for the new analyte; In integrating the acquired first result and the acquired second result, a condition discrimination system that calculates the discrimination result by comparing a value obtained by reflecting the third discriminant value in the first discriminant value with a value obtained by reflecting the fourth discriminant value in the second discriminant value.

3. 3. The condition determination system according to claim 2, the primary discriminant model outputs the first result based on a plurality of branching conditions based on values ​​of the analytical data; The supplemental discrimination model outputs the second result based on a plurality of branching conditions based on the values ​​of the analytical data.

4. 3. The condition determination system according to claim 2, The processor: In integrating the acquired first result and the acquired second result, a condition discrimination system that reflects the third discrimination value weighted by a predetermined first weight on the first discrimination value;

5. 3. The condition determination system according to claim 2, The processor: In integrating the acquired first result and the acquired second result, reflecting the third discriminant value weighted by a predetermined second weight on the first discriminant value; a condition discrimination system that reflects the fourth discrimination value weighted by the second weight on the second discrimination value;

6. 3. The condition determination system according to claim 2, The processor generates data for displaying the first discriminant value, the second discriminant value, the third discriminant value, and the fourth discriminant value for the new analyte.

7. The condition determination system according to claim 1, the first stage, the second stage, and the third stage indicate the degree of the abnormal state; A condition discrimination system, wherein the degree of the abnormal condition in the second stage is between the first stage and the third stage.

8. The state determination system according to claim 7, the analyte and the new analyte are subjects; the analytical data is data related to the subject's health; A condition determination system, wherein the degrees of the abnormal condition in the first stage, the second stage, and the third stage are healthy, pre-disease, and disease, respectively.

9. The state determination system according to claim 8, the pre-disease is mild cognitive impairment; A condition determination system, wherein the disease is dementia.

10. The state determination system according to claim 9, A state determination system, wherein the analysis data includes a feature based on the distance between two fingers during finger tapping motion by the subject.

11. A state determination method using a state determination system, comprising: the condition determination system includes a processor and a memory; The memory includes: Analysis data for new analysis targets, a primary discriminant model that, when analytical data of the new analysis target is input, outputs a first result regarding whether the state of the new analysis target corresponds to the first stage or the second stage; a supplementary discrimination model that, when analysis data of the new analysis target is input, outputs a second result regarding whether the state of the new analysis target corresponds to the first stage or the third stage; The first stage, the second stage, and the third stage indicate the degree of the condition, the degree of the condition in the second stage is between the degree of the condition in the first stage and the degree of the condition in the third stage; the primary discriminant model is generated by machine learning using analysis data of a first analysis object group consisting of analysis objects whose state is the first stage and a second analysis object group consisting of analysis objects whose state is the second stage, the supplementary discrimination model is generated by machine learning using analysis data of the first analysis object group and a third analysis object group consisting of analysis objects whose state is the third stage, The state determination method includes: the processor inputs analytical data for the new analyte into the primary discriminant model to obtain a first result for the new analyte; the processor inputs analytical data for the new analyte into the supplemental discriminant model to obtain a second result for the new analyte; the processor integrates the acquired first result and the acquired second result to calculate a discrimination result indicating whether the state of the new analysis target corresponds to the first stage or the second stage; The processor generates data for displaying the determination result.

12. A state discrimination model that causes a computer to calculate a discrimination result indicating whether the state of a new analysis target corresponds to a first stage or a second stage, the computer holds analytical data of the new analysis target; The state discrimination model is a primary discriminant model that, when analytical data of the new analysis target is input, outputs a first result regarding whether the state of the new analysis target corresponds to the first stage or the second stage; a supplementary discrimination model that, when analysis data of the new analysis target is input, outputs a second result regarding whether the state of the new analysis target corresponds to the first stage or the third stage; inputting analytical data of the new analyte into the primary discriminant model to obtain a first result for the new analyte; inputting analytical data of the new analyte into the supplemental discriminant model to obtain a second result for the new analyte; a process of integrating the acquired first result and the acquired second result to calculate the discrimination result; The first stage, the second stage, and the third stage indicate the degree of the condition, the degree of the condition in the second stage is between the degree of the condition in the first stage and the degree of the condition in the third stage; the primary discriminant model is generated by machine learning using the analysis data for a first analysis object group consisting of analysis objects whose state is the first stage and a second analysis object group consisting of analysis objects whose state is the second stage, A state discrimination model, wherein the supplementary discrimination model is generated by machine learning using the analysis data for the first analysis object group and a third analysis object group consisting of analysis objects whose state is the third stage.

13. A state discrimination model generation method by a state discrimination model generation system, the condition discrimination model generation system includes a processor and a memory; the memory holds analysis data for a first analysis target group consisting of analysis targets whose state is a first stage, a second analysis target group consisting of analysis targets whose state is a second stage, and a third analysis target group consisting of analysis targets whose state is a third stage; The first stage, the second stage, and the third stage indicate the degree of the condition, the degree of the condition in the second stage is between the degree of the condition in the first stage and the degree of the condition in the third stage; The state discrimination model generation method includes: the processor generates, by machine learning using the analytical data for the first analytical target group and the second analytical target group, a primary discriminant model that outputs, when analytical data for a new analytical target is input, a first result regarding whether the state of the new analytical target corresponds to the first stage or the second stage; the processor generates, by machine learning using the analysis data of the first analysis target group and the third analysis target group, a supplementary discriminant model that outputs, when the analysis data of the new analysis target is input, a second result regarding whether the state of the new analysis target corresponds to the first stage or the third stage; the processor integrates the main discriminant model and the supplementary discriminant model to generate an integrated discriminant model that outputs a discrimination result indicating whether the state of the new analysis target corresponds to the first stage or the second stage when the analysis data of the new analysis target is input.

Images