JP7204805B2 - Learning device, estimation device, learning method, estimation method and computer program - Google Patents

Description

The present invention relates to technology for estimating information about human cognitive ability.

In recent years, with the aging of society as a whole, suppression of deterioration in cognitive ability has become a social issue. In order to suppress the deterioration of cognitive ability, it is important to discover the tendency of deterioration of cognitive ability as early as possible. Therefore, techniques have been proposed for estimating whether or not a subject has dementia by performing learning processing (see, for example, Patent Literature 1).

JP 2020-149234 A

However, there is room for improving the accuracy of estimation in the conventional technique. There is also a demand for improving the accuracy of such estimates.
In view of the above circumstances, an object of the present invention is to provide a technology capable of estimating information on human cognitive ability with higher accuracy.

According to one aspect of the present invention, learning processing is performed using teacher data having unit teacher data relating to a plurality of people, in which biometric information of a person and information about the cognitive ability of the person are associated, a learning unit that obtains a trained model that outputs information on the cognitive ability of the estimation target by inputting information on the biometric information of the estimation target; and an output control unit that outputs data of the trained model. The learning device, wherein the unit teaching data includes sleep phase information, which is information indicating a time period during which the person slept in a day.

An aspect of the present invention is the learning device described above, wherein a common pattern is generated by performing principal component analysis on the teacher data, and a value indicating similarity of the teacher data to the common pattern is characterized. A learning data generation unit that generates learning data to be included as a quantity is further provided, and the learning unit performs learning processing using the learning data generated by the learning data generation unit.

According to one aspect of the present invention, an estimation process is performed using an information acquisition unit that acquires biometric information of an estimation target, a trained model generated by the learning device, and the biometric information of the estimation target. and an estimating device.

According to one aspect of the present invention, an information acquiring unit that acquires biometric information of an estimation target, a learned model generated by the learning device, and the estimation of the common pattern with respect to the biometric information of the estimation target. An estimation device comprising: estimation data including a value indicating similarity of biometric information of a subject; and an estimation unit that performs estimation processing using the estimated data.

According to one aspect of the present invention, learning processing is performed using teacher data having unit teacher data relating to a plurality of people, in which biometric information of a person and information about the cognitive ability of the person are associated, a learning step of acquiring a trained model that outputs information on the cognitive ability of the estimation target by inputting information on the biometric information of the estimation target; and an output control step of outputting data of the trained model. and wherein the unit teaching data includes sleep phase information, which is information indicating a time period during which the person slept in a day.

An aspect of the present invention performs an estimation process using an information acquisition step of acquiring biometric information of an estimation target, a trained model generated by the learning device, and the biometric information of the estimation target. and an estimation step.

One aspect of the present invention is a computer program for causing a computer to function as the learning device.

One aspect of the present invention is a computer program for causing a computer to function as the above estimation device.

INDUSTRIAL APPLICABILITY According to the present invention, it becomes possible to estimate information on human cognitive ability with higher accuracy.

1 is a schematic block diagram showing the functional configuration of a learning device 10 of the present invention; FIG. 2 is a schematic block diagram showing the functional configuration of an estimating device 20 of the present invention; FIG. 4 is a flowchart showing a specific example of the flow of processing of the learning device 10; 4 is a flow chart showing a specific example of the flow of processing of the estimation device 20. FIG.

Hereinafter, specific configuration examples of the present invention will be described with reference to the drawings.
FIG. 1 is a schematic block diagram showing the functional configuration of a learning device 10 of the present invention. The learning device 10 performs learning processing using teacher data including biometric information obtained from a plurality of persons. The learning device 10 generates a trained model for estimating information (hereinafter referred to as “cognitive information”) regarding the cognitive ability of the person to be presumed by executing the learning process. The learning device 10 is configured using an information processing device such as a personal computer or a server device. The learning device 10 includes an input unit 11 , an output unit 12 , a storage unit 13 and a control unit 14 .

The input unit 11 inputs teacher data to the learning device 10 . The input unit 11 may read teacher data recorded in a recording medium such as a CD-ROM or USB memory (Universal Serial Bus Memory) from the recording medium. Also, the input unit 11 may receive teacher data from another device via a network. The input unit 11 may be configured in a different manner as long as it is configured to receive input of teacher data.

The teacher data includes a plurality of unit teacher data obtained from individual persons. That is, the teacher data includes unit teacher data obtained by a plurality of persons. The unit training data consists of values of multiple data items and information on the cognitive ability of the person from whom the value of the data item was obtained (for example, classification as to whether it belongs to "healthy", "MCI", or "dementia"). (information indicating ) and information in association with each other. At least one of the data items included in the unit teaching data indicates information about a living body (hereinafter referred to as "biological information"). It is desirable that information regarding each person's cognitive ability be obtained based on the judgment of a physician.

A specific example of the data item of the biological information is information about sleep (hereinafter referred to as "sleep information"). The sleep information may be any information as long as it relates to sleep. For example, biological information in a period from going to bed to waking up (hereinafter referred to as "bedtime period") may be used as sleep information. More specifically, the sum of sleep time from going to bed to waking up, the number of awakenings from going to bed to waking up (hereinafter referred to as "mid-awakening"), each time of mid-awakening, total time of mid-awakening, Time-series information of heartbeat during sleep, time-series information of breathing frequency during sleep, time-series information of blood pressure during sleep, information on body movement during sleep, breathing interruption (number, time, interval) during sleep Specific examples of sleep information include time-series information.

Further, as another specific example of sleep information, there is information (hereinafter referred to as "sleep phase information") indicating a time period during which sleep was performed in a day. The sleep phase information is expressed, for example, as the median time from the time of the first falling asleep to the time of the last awakening in a series of sleeps performed in one day (for example, sleep performed between going to bed and waking up). good too. The sleep phase information may be represented, for example, as the time of the centroid value of one or more sleeping hours occurring in a series of sleeps performed in one day. The sleep phase information may be defined using other values as long as it is information indicating the time period during which sleep was performed in a day.

Not limited to sleep information, biometric information (breathing rate, heart rate, brain waves, etc.) acquired during the period when the user is active, activity information (action history [eating, drinking, toileting, bathing, going out, etc.], location information, etc.) ) may be used as biometric information. Biometric information may be acquired by any means. For example, biometric information may be generated using a device such as a sensor that acquires sleep information from a person.

The output unit 12 outputs information from the learning device 10 to the outside. The output unit 12 may record data of the learned model on a recording medium such as a CD-ROM or USB memory. Also, the output unit 12 may transmit data of the learned model to another device via the network. The output unit 12 may be configured in a different manner as long as it is configured to output the data of the trained model.

The storage unit 13 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device. The storage unit 13 stores data used by the control unit 14 and data generated by the control unit 14 . For example, the storage unit 13 may function as a teacher data storage unit 131 , a learning data storage unit 132 and a learned model storage unit 133 .

The teacher data storage unit 131 stores teacher data input from the input unit 11 . The learning data storage unit 132 stores learning data generated based on teacher data by processing of the control unit 14 . Learning data is data containing a plurality of feature values used in learning processing. The feature amount may be obtained, for example, as statistical values of training data, may be obtained as information indicating similarity to a plurality of common patterns, or may be generated using sparse coding. good. The learned model storage unit 133 stores data of a learned model obtained by learning processing using learning data.

The control unit 14 is configured using a processor such as a CPU (Central Processing Unit) and a memory. The control unit 14 functions as an information acquisition unit 141, a learning data generation unit 142, a learning unit 143, and an output control unit 144 by the processor executing programs. All or part of each function of the control unit 14 may be implemented using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, semiconductor storage devices (such as SSD: Solid State Drives), hard disks and semiconductor storage built into computer systems. It is a storage device such as a device. The above program may be transmitted via telecommunication lines.

The information acquisition unit 141 acquires teacher data via the input unit 11 . The information acquisition unit 141 records the acquired teacher data in the teacher data storage unit 131 .

The learning data generation unit 142 generates learning data having a plurality of feature amounts by performing predetermined processing on the teacher data recorded in the teacher data storage unit 131 . The learning data generation unit 142 records the generated learning data in the learning data storage unit 132 .

The learning data generation unit 142, for example, acquires statistical values for time-series values of biometric data obtained as each unit teacher data, and generates learning data by using the acquired statistical values as feature amounts. good too. Examples of such statistics include values such as mean, standard deviation, variance, minimum, maximum, first quartile, third quartile, and interquartile range. The learning data generating unit 142 may, for example, divide the time-series values for 24 hours into 6-hour intervals and acquire statistical values for each interval.

The learning data generation unit 142 generates learning data by, for example, generating a common pattern common to a plurality of unit teaching data obtained as teaching data, and acquiring a value indicating similarity to the common pattern as a feature quantity. You may A specific example of a method for obtaining such a feature amount will be described below. In the following example, it is assumed that the time-series values included in the unit teaching data are values for 24 hours.

First, the time-series values of each data item are divided into predetermined intervals (for example, 1 hour, 2 hours, etc.), and statistical values are obtained for each interval. Specific examples of statistical values are as described above. For example, if there are 7 types of data items and the statistical values are obtained by dividing the time-series values for 24 hours into hourly intervals, a 7×24=168-dimensional vector can be obtained from one unit of teacher data. . A vector obtained in this way is called a unit teacher vector in the following description.

When unit teaching data are obtained for a plurality of days, a matrix is generated by arranging each unit teaching vector. Eigenvectors are obtained by connecting matrices of each person obtained as training data and executing principal component analysis. Some of the acquired eigenvectors may not be used according to a predetermined criterion, the designer's judgment, or the like. The set of eigenvectors used is used as a plurality of common patterns.

The learning data generation unit 142 acquires a value indicating similarity for each unit teacher vector with respect to the plurality of common patterns (plurality of eigenvectors) thus obtained. For example, the learning data generation unit 142 may acquire feature amounts by acquiring inner products for each combination of eigenvectors and unit teacher vectors. Also, the learning data generation unit 142 may acquire the average and variance (or standard deviation) of each inner product as feature amounts. In this case, the closer the obtained feature amount is to "0", the more the unit teacher vector deviates from the common pattern, and the closer the obtained feature amount is to "1", the more similar the unit teacher vector is to the common pattern. By using such a feature amount, it becomes possible to use a value indicating which of a plurality of common patterns is similar to which common pattern as the feature amount.

The learning data generation unit 142 may generate learning data using feature amounts obtained using sparse coding. In this case, the common pattern (set of eigenvectors) obtained by the principal component analysis described above may be used for the dictionary (set of basis vectors) used for sparse coding.

The learning data generation unit 142 may generate learning data obtained by combining a plurality of types of feature amounts among the feature amounts described above. The designer may predetermine which features are used as learning data. The learning data generation unit 142 records the generated learning data in the learning data storage unit 132 .

The learning unit 143 uses learning data generated by the learning data generation unit 142 (including learning data recorded in the learning data storage unit 132) to execute learning processing according to a predetermined learning model. A learning model used by the learning unit 143 is determined in advance. The learning model used by the learning unit 143 may be, for example, a decision tree, a support vector machine, or other algorithms. By the learning unit 143 performing the learning process, the input data obtained by performing the same process as the learning data generation process performed by the learning data generation unit 142 is used to obtain the input data. A trained model is generated for estimating information about the person's cognitive abilities. The learning unit 143 records the generated trained model in the trained model storage unit 133 .

The output control unit 144 outputs the learned model generated by the learning unit 143 (including the learned model stored in the learned model storage unit 133 ) through the output unit 12 . The trained model output by the output control unit 144 is used by the estimation device 20, which will be described later.

FIG. 2 is a schematic block diagram showing the functional configuration of the estimation device 20 of the present invention. The estimation device 20 performs estimation processing using estimation target data including biometric information obtained from an estimation target person. The estimating device 20 estimates information (cognitive information) about the cognitive ability of the person to be presumed by executing the estimating process. The estimation device 20 is configured using an information processing device such as a personal computer or a server device. The estimation device 20 includes an input unit 21 , an output unit 22 , a storage unit 23 and a control unit 24 .

The input unit 21 inputs the estimation target data to the estimation device 20 . The input unit 21 may read the estimation target data recorded in a recording medium such as a CD-ROM or a USB memory from the recording medium. The input unit 21 may also receive estimation target data from another device via a network. The input unit 21 may be configured in a different manner as long as it is configured to receive input of estimation target data.

The estimation target data has values of data items similar to the teacher data used in the learning data. When the estimation target data includes sleep information, for example, the estimation target data may be configured using data obtained during sleep of the estimation target by a sensor used by the estimation target.

The output unit 22 outputs information from the estimation device 20 to the outside. The output unit 22 may record data indicating the estimation result on a recording medium such as a CD-ROM or a USB memory. Also, the output unit 22 may transmit data indicating the estimation result to another device via the network. The output unit 22 may be configured in a different manner as long as it is configured to output data indicating the estimation result.

The storage unit 23 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device. The storage unit 23 stores data used by the control unit 24 and data generated by the control unit 24 . For example, the storage unit 23 may function as a trained model storage unit 231 , an estimation target data storage unit 232 , an estimated data storage unit 233 and an estimation result data storage unit 234 .

The trained model storage unit 231 stores data of trained models. The learned model data is, for example, the learned model data generated by the learning device 10 . The learned model data may be recorded in the learned model storage unit 231 in advance, or may be input from the input unit 21 . The estimation target data storage unit 232 stores the estimation target data input from the input unit 21 . The estimated data storage unit 233 stores estimated data generated based on the estimation target data by the processing of the control unit 24 . Estimation data is data containing a plurality of feature values used in estimation processing. The feature amount included in the estimated data is obtained by subjecting the estimation target data to the same processing as the feature amount used for the learning data in the learning device 10 . The estimation result data storage unit 234 stores data indicating estimation results obtained by estimation processing using estimation data.

The control unit 24 is configured using a processor such as a CPU and a memory. The control unit 24 functions as an information acquiring unit 241, an estimated data generating unit 242, an estimating unit 243, and an output control unit 244 by the processor executing programs. All or part of each function of the control unit 24 may be realized using hardware such as ASIC, PLD, FPGA, or the like. The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, semiconductor storage devices (such as SSD), and storage devices such as hard disks and semiconductor storage devices built into computer systems. It is a device. The above program may be transmitted via telecommunication lines.

The information acquisition unit 241 acquires estimation target data via the input unit 21 . The information acquisition unit 241 records the acquired estimation target data in the estimation target data storage unit 232 . When acquiring the data of the learned model via the input unit 21 , the information acquisition unit 241 records the acquired data of the learned model in the learned model storage unit 231 .

The estimated data generator 242 generates estimated data by performing the same processing as the learning data generator 142 of the learning device 10 on the estimation target data. Therefore, the generated estimation data has the same configuration as the learning data. That is, the number of data items (number of dimensions) and the type of each data item (acquisition method) included in the estimated data match the number of data items and the type of each data item in the learning data. For example, if a common pattern is used in generating training data, the same common pattern is used in generating estimated data. The estimated data generation unit 242 records the generated estimated data in the estimated data storage unit 233 .

The estimation unit 243 performs estimation processing by using the estimated data generated by the estimated data generation unit 242 (including the estimated data stored in the estimated data storage unit 233) as input values for the trained model. The estimation unit 243 acquires information about the cognitive ability of the person to be estimated by executing the estimation process. For example, the estimation unit 243 determines which of the predetermined classifications (for example, the classification of MCI and dementia and others, the classification of MCI, dementia, or healthy, etc.) to which the estimation subject may fall. You may acquire the estimation result which shows. The estimation unit 243 records data indicating the acquired estimation result in the estimation result data storage unit 234 .

The output control unit 244 outputs data indicating the estimation result obtained by the estimation unit 243 (including data stored in the estimation result data storage unit 234 ) via the output unit 22 . The output control unit 244 may transmit the estimation result data to, for example, a terminal device used by the person to be presumed or a terminal device of a person related to the person to be presumed (for example, a relative or a doctor in charge).

FIG. 3 is a flowchart showing a specific example of the processing flow of the learning device 10. As shown in FIG. The learning data generation unit 142 generates learning data by performing preprocessing on the acquired teacher data (step S101). The learning unit 143 executes learning processing using the learning data (step S102). The learning unit 143 records the data of the learned model obtained by executing the learning process in the learned model storage unit 133 (step S103). The output control unit 144 outputs data of the trained model (step S104).

FIG. 4 is a flowchart showing a specific example of the processing flow of the estimation device 20. As shown in FIG. The estimated data generation unit 242 generates estimated data by performing preprocessing on the acquired estimation target data (step S201). The estimation unit 243 executes estimation processing using the estimation data (step S202). The estimation unit 243 records the data of the estimation result obtained by executing the estimation process in the estimation result data storage unit 234 (step S203). The output control unit 244 outputs the data of the estimation result (step S204).

With the learning device 10 and the estimation device 20 configured in this way, it is possible to estimate information on human cognitive ability with higher accuracy.

For example, the learning device 10 may perform learning processing using sleep phase information to generate a trained model. The sleep phase information expresses what time period the person is sleeping during the day. It is generally known that sleep-related behavior and biological information are correlated with cognitive ability. Therefore, by using such a novel feature amount related to sleep, it is possible to estimate information related to human cognitive ability with higher accuracy.

Also, in the learning device 10, a value indicating similarity to a common pattern is used as a feature amount. It is generally known that sleep-related behavior and biological information are correlated with cognitive ability. Therefore, by using such new feature amounts related to sleep and biometric information that have not existed in the past, information related to human cognitive ability can be estimated with higher accuracy.

Although the embodiment of the present invention has been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and design and the like are included within the scope of the gist of the present invention.

DESCRIPTION OF SYMBOLS 10... Learning apparatus, 11... Input part, 12... Output part, 13... Storage part, 131... Teacher data storage part, 132... Learning data storage part, 133... Trained model storage part, 14... Control part, 141... Information Acquisition unit 142 Learning data generation unit 143 Learning unit 144 Output control unit 20 Estimation device 21 Input unit 22 Output unit 23 Storage unit 231 Learned model storage unit 232 Estimated data storage unit 233 Estimated data storage unit 234 Estimation result data storage unit 24 Control unit 241 Information acquisition unit 242 Estimated data generation unit 243 Estimation unit 244 Output control unit

Claims (8)

By performing a learning process using teacher data that has unit teacher data relating to a plurality of people in which biometric information of a person and information about the cognitive ability of the person are associated, a learning unit that acquires a trained model that outputs information about the cognitive ability of the person to be estimated by inputting information;
an output control unit that outputs data of the trained model,
The unit teaching data includes sleep phase information, which is information indicating the time period during which the person slept in a day,
learning device. a learning data generating unit that generates a common pattern by performing principal component analysis on the teacher data, and generates learning data that includes, as a feature quantity, a value indicating similarity of the teacher data to the common pattern; ,
2. The learning device according to claim 1, wherein said learning unit performs learning processing using learning data generated by said learning data generation unit. an information acquisition unit that acquires biometric information of an estimation target;
An estimating unit that performs an estimating process using a trained model generated by the learning device according to claim 1 or 2 and the biometric information of the person to be estimated;
An estimating device comprising: an information acquisition unit that acquires biometric information of an estimation target;
3. A trained model generated by the learning device according to claim 2; estimated data including a value indicating similarity of the biometric information of the estimation target for the common pattern with respect to the biometric information of the estimation target; an estimation unit that performs estimation processing using
An estimating device comprising: A computer performs a learning process using teacher data that has unit teacher data that associates a person's biometric information and information about the person's cognitive ability with respect to a plurality of people, so that the estimation target person A learning step of acquiring a trained model that outputs information on the cognitive ability of the subject by inputting information on biometric information;
an output control step in which the computer outputs data of the trained model;
The unit teaching data includes sleep phase information, which is information indicating the time period during which the person slept in a day,
learning method. an information acquisition step in which the computer acquires the biometric information of the presumed subject;
an estimation step in which a computer performs an estimation process using the trained model generated by the learning device according to claim 1 or 2 and the biometric information of the estimation target;
An estimation method with A computer program for causing a computer to function as the learning device according to claim 1 or 2. A computer program for causing a computer to function as the estimation device according to claim 3 or 4.

Images