JP2022184311A - Core body temperature estimation device and method, and program - Google Patents

Abstract

To provide a core body temperature estimation device and method for highly accurately estimating a core body temperature by a simple method and configuration, and program.SOLUTION: A core body temperature estimation device includes: an input reception part 10 for receiving inputs of time series data of information relating to exercise loads of a measuring object person, information relating to skin surface temperatures and information relating to heart rates; an estimation part 11 for inputting the aforementioned time series data of the measuring object person to a learning model constructed by machine-learning, with time series data of information relating to exercise loads of subjects, information relating to skin surface temperatures and information relating to heart rates defined as input information, and time-series actual measurements of rectal temperatures of prescribed subjects defined as output information, and estimating a core body temperature of the measuring object person, and an output part 12 for outputting an estimated core body temperature of the measuring object person.SELECTED DRAWING: Figure 1

Description

The present invention relates to a core body temperature estimating device, method, and program for estimating core body temperature of a subject.

Heatstroke is categorized into heat syncope and heat cramps caused by lack of water in the body, heat exhaustion caused by lack of minerals (such as salt), and heat stroke caused by an increase in core body temperature. In heat stroke, the core body temperature rises to 40°C or higher, and symptoms such as disturbance of consciousness and cessation of sweating are exhibited. Therefore, prevention of heat stroke is considered particularly important.

To predict the onset of heat stroke, it is necessary to measure core body temperature, which is an index. Core body temperature can be obtained by measuring the temperature inside the esophagus and rectum. Therefore, the deep body temperature can be measured by inserting a special temperature sensor into the esophagus or the rectum, but this has the disadvantage of causing great discomfort to the subject and making it difficult to measure.

Therefore, for example, Non-Patent Document 1 proposes a deep body thermometer that can continuously measure the tissue temperature directly below the measurement part from the body surface, such as the forehead and chest, using a special sensor. Non-Patent Document 2 proposes a wearable core thermometer that measures the heat flux between two points by arranging temperature sensors at two locations and estimates the core body temperature based on the principle of the dual heat flow method. .

In addition, Patent Document 1 proposes a method of taking a thermal image of a person to be measured and estimating the core body temperature based on the thermal image.

Further, Patent Document 2 proposes a method of measuring the deep body temperature of a worker by providing a helmet used by the worker with a deep body temperature probe having a temperature sensor.

Further, Patent Document 3 proposes a deep body thermometer that non-invasively obtains the body temperature of a person to be measured.

In addition, in Patent Document 4, using a living body heat model, heat transfer in the body and heat exchange with the outside of the body are repeatedly calculated using a heat balance calculation formula from the heat production in the deep part obtained based on the measurement with the sensor. A method for estimating core body temperature during exercise has been proposed.

JP 2018-183564 A JP 2018-134137 A JP 2018-13395 A JP 2017-217224 A

BME Vol.2, No.3, 1988 Institute of Safety and Health News No.119 (2018-10-05)

However, the methods described in Non-Patent Documents 1 to 2 and Patent Documents 1 to 4 require a special sensor, or the information of the person to be measured must be measured at multiple points. However, there were problems such as the need to increase the size of the equipment and increase the size of the equipment, and the complexity of the measurement work.

In view of the above circumstances, it is an object of the present invention to provide a core body temperature estimating apparatus, method, and program capable of estimating core body temperature with high accuracy using a simple technique and configuration.

A core body temperature estimating apparatus of the present invention includes an input receiving unit that receives input of time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of a person whose core body temperature is to be measured; Learning constructed by machine learning with information on the subject's exercise load, information on the skin surface temperature, and time-series data on the heart rate as input information, and time-series actual measurements of the rectal temperature of a given subject as output information. Estimation for estimating core body temperature of the subject by inputting time-series data of the subject's exercise load information, skin surface temperature information, and heart rate information received by the input receiving unit into the model and an output unit that outputs the subject's core body temperature estimated by the estimation unit.

Further, in the core body temperature estimating apparatus of the present invention, the input information can further include information about the age of the subject, the input receiving unit can further receive information about the age of the subject, and the estimating unit can estimate the core body temperature of the subject by additionally inputting information about the subject's age to the learning model.

Further, in the core body temperature estimating apparatus of the present invention, the input information can further include information about the height and weight of the subject, and the input reception unit can further receive information about the height and weight of the subject. The estimating unit can estimate the core body temperature of the subject by further inputting information about the subject's height and weight to the learning model.

Further, in the core body temperature estimating apparatus of the present invention, the input information may further include information regarding the subject's measurement environment conditions, and the input receiving unit may further receive information regarding the subject's measurement environment conditions. The estimating unit can estimate the core body temperature of the subject by further inputting information about the measurement environment conditions of the subject to the learning model.

Further, in the core body temperature estimating apparatus of the present invention, the input information can further include information about the subject's sex, the input receiving unit can further receive information about the subject's sex, and the estimating unit can can estimate the core body temperature of the subject by additionally inputting information about the subject's gender to the learning model.

Further, the core body temperature estimating apparatus of the present invention can include an exercise load information detection unit that detects information regarding the exercise load of the person being measured.

In addition, the core body temperature estimating device of the present invention can include a skin surface temperature detection unit that detects information about the skin surface temperature of the person being measured.

Further, the core body temperature estimating apparatus of the present invention can include a heartbeat information detection section for detecting information about the heartbeat of the person being measured.

Also, in the core body temperature estimating apparatus of the present invention, the estimating section can construct a learning model using an RNN (Recurrent Neural Network).

Further, in the core body temperature estimation device of the present invention, the RNN can be LSTM (Long Short Term Memory).

The method for estimating core body temperature of the present invention accepts input of time-series data on exercise load of a subject whose core body temperature is to be measured, information on skin surface temperature, and information on heart rate. information, skin surface temperature, and heart rate are used as input information, and time-series actual measurements of rectal temperature of a given subject are used as output information for a learning model constructed by machine learning. , estimating the core body temperature of the subject by inputting time-series data of received information on the exercise load of the subject, information on the skin surface temperature, and information on the heart rate, and calculating the estimated core body temperature of the subject to output

A program for estimating core body temperature of the present invention includes steps of accepting input of time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of a subject whose core body temperature is to be measured; A learning model constructed by machine learning with information on exercise load, information on skin surface temperature, and time-series data on information on heart rate as input information, and time-series actual measurements of rectal temperature of a given subject as output information. On the other hand, a step of estimating the body temperature of the person to be measured by inputting time-series data of received information on the exercise load of the person to be measured, information on the skin surface temperature, and information on the heart rate; and outputting the core body temperature of the person.

According to the core body temperature estimating apparatus, method, and program of the present invention, time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of a predetermined subject are used as input information, and the rectal temperature of the predetermined subject is calculated. By inputting time-series data of information on the subject's exercise load, skin surface temperature, and heart rate into a learning model constructed by machine learning using time-series measured values as output information. A core body temperature of a person to be measured is estimated, and the estimated core body temperature of the person to be measured is output.

As a result, the core body temperature can be estimated with high accuracy using a simple technique and configuration. In particular, according to the core body temperature estimating apparatus, method, and program of the present invention, time-series data are used for the exercise load information, the skin surface temperature information, and the heart rate information. That is, since the above information obtained from body measurement is regarded as a continuous change in state, a learning model is constructed to estimate the core body temperature. The core body temperature can be estimated, and the estimation accuracy of the core body temperature can be improved.

FIG. 1 is a block diagram showing a schematic configuration of a first embodiment of a core body temperature estimating device of the present invention; 3 is a flow chart for explaining the processing of the core body temperature estimation device of the first embodiment of the core body temperature estimation device of the present invention; A block diagram showing a schematic configuration of a second embodiment of a core body temperature estimating device of the present invention. A diagram showing the flow of measuring the number of steps taken by a subject, the surface temperature of the skin on the back of the hand, the heart rate, and the rectal temperature. A diagram showing an example of the passage of time between the predicted value of rectal temperature using the learning model and the measured value of rectal temperature A diagram showing another example of the time course of the predicted value of the rectal temperature using the learning model and the measured value of the rectal temperature

A first embodiment of a core body temperature estimating device of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a schematic configuration diagram of a core body temperature estimating device 1 of this embodiment.

The deep body temperature estimation device 1 is a device that estimates the deep body temperature of a subject. The core body temperature estimating device 1 of the present embodiment receives six preset items of information about a person whose core body temperature is to be measured, and applies the received information to a learning model constructed in advance by machine learning. By inputting, the core body temperature of the person to be measured is estimated. The six preset items are information on the subject's exercise load, information on skin surface temperature, information on heart rate, information on age, information on weight and height, and information on measurement environment conditions.

The core body temperature estimation device 1 specifically includes an input reception unit 10, an estimation unit 11, and an output unit 12, as shown in FIG.

The input reception unit 10 receives input of information on the subject's exercise load, information on the skin surface temperature, time-series data on information on the heart rate, information on age, information on weight and height, and information on the measurement environment. Time-series data is a group of data continuously measured at predetermined time intervals.

The information about the exercise load is information indicating the state of the exercise load of the person being measured, and in this embodiment, it is the average number of steps per minute of the person being measured. That is, the input receiving unit 10 of the present embodiment receives input of time-series data of the average number of steps per minute of the subject. However, the information on the exercise load is not limited to this. When no exercise load is imposed on the person, input of numerical information of "0" may be accepted as information on the exercise load. Further, the information about the exercise load is not limited to these numerical information, and may be other information as long as it indicates the status of the exercise load of the person being measured.

As the information on age, for example, the age of the person to be measured is used as it is. Information about the age is set and input using, for example, the input device 2, which will be described later, and the input reception unit 10 receives the set and input age.

The information on the weight and height may be any information as long as it is information on the weight and height of the person to be measured. accepts the input of As for the BMI value, for example, the input receiving unit 10 may directly receive the value output from the BMI measuring device, or the BMI value previously measured for the person to be measured may be set using the input device 2. The input reception unit 10 may receive the set input value.

The information on the skin surface temperature may be any information as long as it is information on the skin surface temperature of the subject. receives input of time-series data of skin surface temperature. The chest or back of the hand is desirable as a measurement target because the skin surface temperature is stable compared to other regions.

Regarding the value of the skin surface temperature, for example, the time-series data output from a temperature sensor or the like may be directly received by the input receiving unit 10. However, the present invention is not limited to this. You may accept it. Alternatively, the time-series data of the skin surface temperature measured in advance for the person to be measured may be set and input using the input device 2, and the input receiving unit 10 may receive the set and input values.

The information on the heart rate may be any information as long as it is information on the heart rate of the person being measured. Accepts raw input.

As for the heart rate value, for example, the input receiving unit 10 may directly receive time-series data output from a heart rate measuring device or the like. The time-series data may be set and input using the input device 2, and the input receiving unit 10 may receive the set and input values.

As a device for measuring heart rate, for example, a transmission pulse wave measuring device or a reflection pulse wave measuring device using photoplethysmography can be used. A device that measures changes in vascular pressure using a sphygmomanometer, a device that measures sound using a phonocardiography, and the like can be used.

The measurement environmental conditions are the environmental conditions under which the subject and the person to be measured are placed when measuring the actual rectal temperature of the subject and estimating the core body temperature of the subject. Environmental temperature (eg, room temperature) that is thought to affect the subject's core body temperature can be used. By including such measurement environment conditions when estimating the core body temperature of the subject, the environment in which the subject is placed can be taken into account, so the accuracy of estimating the core body temperature can be further improved. .

The environmental conditions for measurement are not limited to the environmental temperature described above, but may be, for example, a heat index (WBGT (Wet Bulb Globe Temperature)). Also, the dry-bulb temperature, wet-bulb temperature, radiation temperature, humidity, etc. used when calculating the heat index may be used singly or in combination.

As for the measurement environment conditions, for example, the input reception unit 10 may directly receive values output from a device that measures the measurement environment temperature, or the measurement environment temperature that has been measured in advance may be set using the input device 2. The input reception unit 10 may receive the set input value. Also, the input of the time-series data of the measured environmental temperature may be accepted.

The estimating unit 11 estimates the core body temperature of the subject. The estimating unit 11 of the present embodiment estimates the core body temperature of the subject by inputting the six items of information received by the input receiving unit 10 into a learning model constructed in advance by performing machine learning. . The learning model may be stored in estimation unit 11, or may be stored in a server device or the like connected to core body temperature estimation device 1 via a communication network, for example.

The learning model includes time-series data of the information on the exercise load, the information on the skin surface temperature, and the information on the heart rate of the predetermined subject, the information on the age, the information on the weight and height, and the measurement environment conditions. It was obtained by machine learning based on the relationship between the information and the time-series data of the measured rectal temperature of the subject. That is, the information on the exercise load, the information on the skin surface temperature, and the time-series data on the information on the heart rate of the predetermined subject, the information on the age, the information on the weight and height, and the information on the measurement environment conditions. It was obtained by machine learning using time-series data of the measured values of the rectal temperature of the subject as input information and output information.

As a machine learning method, in this embodiment, LSTM (Long Short Term Memory), which is one of RNNs (Recurrent Neural Networks) and learns time-series data while weighting past data, is used. LSTM has the characteristic of being able to analyze the time-series data obtained from body measurements as continuous changes in the state. . In particular, since the evaluation of the data is slanted so that the influence becomes weaker as the evaluation of the data goes back in time, the temporal change in the rectal temperature prediction value changes at a value close to the actual measurement value.

However, the machine learning method is not limited to LSTM, and any method may be used as long as it accepts input of time-series data and performs machine learning while weighting past data. A method of constructing a learning model will be described in detail later.

The output unit 12 outputs the subject's core body temperature estimated by the estimation unit 11 . Specifically, the output unit 12 of the present embodiment displays the core body temperature of the person to be measured estimated by the estimation unit 11 on the display device 3 (to be described later) connected to the core body temperature estimation device 1 . Note that the output destination of the output unit 12 of the present embodiment is not limited to the display device 3, and may be output to a computer, server device, or the like connected to the core body temperature estimation device 1 via a communication network. Alternatively, it may be output to a printing device such as a printer.

Further, when the core body temperature of the subject estimated by the estimation unit 11 or the increase in core body temperature exceeds a preset threshold value, the output unit 12 causes the display device 3 to display a warning message. , or an alarm sound may be emitted. As a result, it is possible to prevent the subject from developing heat stroke such as heat stroke.

The core body temperature estimation device 1 is composed of a computer, and includes a CPU (Central Processing Unit) or GPU (Graphics Processing Unit), semiconductor memories such as ROM (Read Only Memory) and RAM (Random Access Memory), storage such as a hard disk, and A communication I/F (Interface) and the like are provided.

A semiconductor memory or hard disk of the core body temperature estimation device 1 is installed with a core body temperature estimation program according to an embodiment of the present invention. When the CPU or GPU executes this core body temperature estimation program, the input reception unit 10, the estimation unit 11 and the output unit 12 shown in FIG. 1 function.

That is, the core body temperature estimation program includes information on the exercise load of the subject whose core body temperature is to be measured, information on the skin surface temperature, time-series data on the heart rate, information on the age, information on the weight and height, and a step of accepting input of information about measurement environment conditions; a step of estimating the core body temperature of the person to be measured by inputting the information accepted in the receiving step into the learning model; and the estimated person to be measured. outputting the core body temperature of the computer.

Note that the hardware configuration of the core body temperature estimation device 1 is not limited to the configuration described above.

Further, in the present embodiment, all the functions of the input receiving unit 10, the estimating unit 11, and the output unit 12 described above are executed by the core body temperature estimating program. It may be configured from hardware such as ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and other electric circuits.

As shown in FIG. 1, an input device 2 and a display device 3 are connected to the core body temperature estimating device 1 by wire or wirelessly. The input device 2 is configured to be able to set and input various information as described above, and includes, for example, a keyboard and a mouse. Moreover, the display device 3 includes, for example, a liquid crystal display. Note that the deep body temperature estimation device 1 may be configured by a tablet terminal or a wearable terminal. A detailed configuration when the deep body temperature estimating device 1 is a wearable terminal will be described later.

Next, the processing flow of the core body temperature estimation device 1 of this embodiment will be described with reference to the flowchart shown in FIG.

First, using the input device 2, the user inputs information on the subject's exercise load, information on the skin surface temperature, time-series data on heart rate, information on age, information on weight and height, and measurement environment conditions. (S10).

The above six items of information set and input by the user are received by the input receiving unit 10 of the core body temperature estimating device 1 (S12).

The six items of information received by the input receiving unit 10 are input to the estimating unit 11 . The estimating unit 11 estimates the core body temperature of the subject by inputting the six items of information received by the input receiving unit 10 into the learning model described above (S14).

The core body temperature of the subject estimated by the estimation section 11 is output to the output section 12, and the output section 12 displays the input core body temperature of the subject on the display device 3 (S16).

In the deep body temperature estimating apparatus 1 of the above-described embodiment, the above-described six items of information are used as the input information for the learning model. It is also possible to use a learning model that has undergone machine learning. Then, the estimating unit 11 may estimate the deep body temperature of the person to be measured by inputting the six items of information on the person to be measured and the information on the sex of the person to be measured into the learning model.

As information on gender, for example, if the gender is male, "1" and if the gender is female, "2" may be input to the learning model. Core body temperature may differ depending on gender even if the person being measured is in the same environment. Therefore, when estimating core body temperature, including information on gender will further improve the estimation accuracy of core body temperature. can be done.

Next, the wearable terminal 5 provided with the second embodiment of the core body temperature estimating device of the present invention will be described.

The basic configuration of the wearable terminal 5 for estimating the core body temperature is the same as that of the core body temperature estimation device 1 of the first embodiment described above, but information on the exercise load of the person to be measured and the skin surface temperature The method of acquiring time-series data for information and heart rate information is different.

The form of the wearable terminal 5 may be a wristband type (wristwatch type), a glasses type, or a clip type. Preferably.

FIG. 3 is a block diagram showing a schematic configuration of the wearable terminal 5 of this embodiment. The wearable terminal 5 of this embodiment includes a main body 50, an exercise load information detector 55, a skin surface temperature detector 56, a heartbeat information detector 57, and an environmental temperature measurer 58, as shown in FIG.

The body portion 50 includes an input reception portion 51 , an estimation portion 52 , an output portion 53 and a display portion 54 . The basic functions of the input receiving unit 51, the estimating unit 52, and the output unit 53 of the main unit 50 are the same as those of the input receiving unit 10, the estimating unit 11, and the output unit 12 of the core body temperature estimating device 1 of the above-described first embodiment. It has the same configuration as

Display unit 54 has, for example, a liquid crystal display, and displays the core body temperature output from output unit 12 . Further, the display unit 54 may be configured by a touch panel so that the core body temperature is displayed on the display unit 54 and the setting input of various information is received on the display unit 54 .

The exercise load information detection unit 55 detects information about the exercise load of the subject. A pedometer, for example, can be used as the exercise load information detector 55 . For example, the input reception unit 51 receives the number of steps measured by a pedometer, and the estimation unit 52 calculates the time-series data of the above-described average number of steps per minute as information related to exercise load based on the input number of steps, This should be input to the learning model.

The skin surface temperature detector 56 detects information about the skin surface temperature of the subject. A temperature sensor, for example, can be used as the skin surface temperature detector 56 . For example, if the wearable terminal 5 is a wristband type, the temperature sensor can measure the skin surface temperature of the part near the back of the hand, so the accuracy of estimating the core body temperature can be improved.

The heartbeat information detector 57 detects information about the heartbeat of the person being measured. As heartbeat information detector 57, for example, a reflective pulse wave sensor can be used. The input reception unit 51 receives the heart rate measured by the reflective pulse wave sensor.

The environmental temperature measuring unit 58 measures the measured environmental temperature of the subject. A temperature sensor, for example, can be used as the environmental temperature measurement unit 58 . By measuring the measured environmental temperature in real time by the environmental temperature measurement unit 58, it is possible to receive the input of the time-series data of the measured environmental temperature and improve the accuracy of estimating the core body temperature.

In the wearable terminal 5, the information about age and the information about weight and height may be set and input on the touch panel if the display unit 54 is composed of a touch panel, for example.

Alternatively, the communication I/F of the main unit 50 may be connected to another communication terminal device via a communication network, and the communication terminal device may set and input information on age and information on weight and height.

Other communication terminal devices include smartphones and tablet terminals. For example, a dedicated application may be installed on a smartphone or tablet terminal, and setting inputs of information regarding age and information regarding weight and height may be accepted on the dedicated application. Further, the core body temperature output from the output unit 53 may be displayed on the dedicated application.

Also, in the wearable terminal 5, as in the core body temperature estimating device 1 of the above-described embodiment, information regarding the sex of the subject and the person to be measured may be input to the learning model. Information about the subject's sex may be set and input on the display unit 54 or the dedicated application described above. In addition, the measurement environment conditions for the subject and the person to be measured may also be set and input on the display unit 54 or the dedicated application described above.

Other actions of the wearable terminal 5 are the same as those of the core body temperature estimation device 1 of the above embodiment.

A more specific embodiment of the core body temperature estimating device of the present invention will be described below. First, an example of a method of generating a learning model used for estimating core body temperature will be described.

In this example, 12 subjects (subject numbers 1 to 12, average age 24.6 ± 5.8 years old) were subjected to exercise load in an artificial climate room, information on exercise load, information on skin surface temperature and Time-series data of information on heart rate were acquired. The measurement environment conditions of the artificial climate chamber are as follows: Low-risk caution level room temperature 25°C, humidity 50%, and WBGT 22°C as measurement environment condition 1; Humidity was 50% and WBGT was 30°C. Under both measurement environment condition 1 and measurement environment condition 2, exercise load was performed using a treadmill 80w.

FIG. 4 shows the measurement flow of measuring the subject's number of steps, surface temperature of the skin on the back of the hand, heart rate, and rectal temperature (these four items are hereinafter referred to as biological information). As shown in FIG. 4, after resting for 10 minutes and setting a preparatory period, biological information was collected in a no-load state for 6 minutes. Next, an exercise load of 80 w was performed on a treadmill for 18 minutes, which was defined as "exercise load 1". After exercise load 1, a rest period of 12 minutes was provided, and then exercise load was further performed for 18 minutes to obtain "exercise load 2". After that, there was a 12-minute break, and as a post-treatment, biological information was measured in a no-load state for 10 minutes, and the experiment ended. Measurement of biological information was started after the preparatory period and continued until the end of rest after exercise load 2, and the total time of measurement was 66 minutes. Biological information was measured every 10 seconds.

Rectal temperature and dorsal skin surface temperature were measured using a thermocouple sensor. Heart rate was measured using Nihon Kohden BSM-2401. Regarding the measurement environmental conditions in the artificial climate room, the environmental temperature (room temperature) was measured using a VAISALA thermometer. In addition, the number of steps taken by the subject was measured using a Fujitsu vital sensing band, and the average number of steps for one minute output every 10 seconds was obtained.

These measured values were recorded with a real-time data recording device NR-500 of Keyence data logger.

In addition, during the preparation period described above, the subject's height and precise weight were measured, and the BMI was obtained from the measured values.

In this example, as described above, biological information was measured for 12 subjects under two measurement environment conditions, so 24 time-series data sets of 12 people × 2 were obtained. One person (Subject No. 2) had no rectal temperature measured, and one person (Subject No. 8) only measured at an environmental temperature (room temperature) of 25°C, and at an environmental temperature (room temperature) of 35°C, rectal temperature As the measurement became impossible halfway through, finally, 21 time-series data sets were obtained for 11 subjects (mean age 24.6±6.0 years). Data set numbers 1 to 21 are assigned to these time-series data sets.

Then, the number of steps of the subject, the time-series data measured every 10 seconds for the skin surface temperature of the back of the hand, and the heart rate, the age and BMI of the subject, and the environmental temperature (room temperature) were used as independent variables. A learning model was constructed by performing deep learning using AI using the actual value of temperature as teacher data, and prediction of core body temperature was performed using the learning model.

The learning model of deep learning used LSTM (Long Short Term Memory).

The AI platform is Anaconda (registered trademark) ver. 4.4.0, and the program is Python (registered trademark) ver. 3.7, and Tensorflow (registered trademark) ver. 2.1, Keras (registered trademark), etc. were implemented.

The LSTM settings are such that the transfer function is a fully connected neural network with one layer, the activation function is linear (normalized linear function, normalization range is -1 to +1), and the optimization function is Adam (Adaptive Moment Estimation, probability Gradient descent method), the error function uses the least squares method, and the layers (layers) are 256 intermediate layers (including 36 hidden layers) + 1 output layer (fully connected neural network) = 257 layers, Machine learning was performed with the number of nodes set to 32, the number of epocks (the number of repetitions of learning) set to 50, and the batch size set to 32.

In this embodiment, since the biological information is measured every 10 seconds, by setting the hidden layer of the LSTM to 36 layers, using the time-series data set from about 6 minutes ago to the present time, The rectal temperature (core body temperature) after 10 seconds was predicted. That is, the predicted value of the current rectal temperature (core body temperature) was obtained using the time-series data set up to 10 seconds ago. The LSTM is referred to by increasing the influence (heavy weight) on the predicted value as it is closer to the current data set and decreasing (lighter weight) the influence on the predicted value as it is further away. In this embodiment, the data set from 6 minutes to 10 seconds ago was used to obtain the current prediction value, so the hidden layer was set to 36 layers. The number of hidden layers may be arbitrarily changed depending on the interval and how many minutes before the data set is used.

Then, the prediction accuracy was evaluated by comparing the predicted value and the measured value. The prediction accuracy was evaluated by the k-fold cross validation method. This method extracts one test data from k sample data, obtains a learning result (learning model) from the remaining (k-1) data sets, and performs this operation in order, resulting in ( This is a method of evaluating the performance of a learning model from the average prediction accuracy of k−1) learning models.

In this embodiment, as described above, 21 time-series data sets are obtained from 11 subjects, so one time-series data set is excluded as test data, and the remaining 20 time-series data sets A learning model was constructed by machine learning, and the above test data was verified using the constructed learning model. Then, this operation was sequentially repeated 20 times to construct 20 learning models, and each learning model was evaluated by verifying test data using each learning model.

FIG. 5 is a diagram showing the time course of the predicted rectal temperature and the measured rectal temperature obtained by inputting the time-series data set (test data) of subject number 1 at an environmental temperature of 25° C. into the learning model. be. In FIG. 5, the predicted values are indicated by dashed lines, and the measured values are indicated by solid lines.

In the example shown in FIG. 5, all remaining time-series data sets except for the time-series data set of data set number 1 are trained to predict rectal temperature as described above, and the resulting learning model is used to , predicting rectal temperature for data set number 1. When the predicted value and the actual value are compared, the matching rate is 74% when the error is allowed within ±0.1°C, and the matching rate is 95% when the error is allowed within ±0.2°C. %Met.

FIG. 6 is a diagram summarizing predicted values and measured values when the same operation as for data set number 1 is performed on 20 time-series data sets with data set numbers 2 to 21, respectively. Table 1 below is a table showing the rate of agreement between predicted values and measured values for data set numbers 1 to 21. As shown in FIG. 6 and Table 1 below, the agreement rate between the measured values and the predicted values is high. In particular, the agreement rate of data set No. 18 is the best, and the rectal temperature prediction is almost within ±0.2 ° C. 100% match. Also, even in the case of data set number 14, which has the lowest matching rate, judging from the results of FIGS. It was a possible result.

In the above example, a learning model is constructed using six items: the number of steps of the subject, the skin surface temperature of the back of the hand, and the time series data of the heart rate, the age and BMI of the subject, and the environmental temperature (room temperature). , the rectal temperature (deep body temperature) was estimated by inputting the above six items of the subject into the learning model. ), a learning model was constructed in the same manner as in the above example, and the concordance rate between the predicted value and the measured value when the rectal temperature was estimated is shown in Table 2 below. As shown in Table 2 below, even in the case of the above 4 items, the same accuracy as in the case of the above 6 items could be obtained.

The above embodiment is an embodiment of the core body temperature estimating device of the present invention, and the core body temperature estimating device of the present invention is not limited to the above embodiment.

1 deep body temperature estimation device 2 input device 3 display device 5 wearable terminal 10 input reception unit 11 estimation unit 12 output unit 50 body unit 51 input reception unit 52 estimation unit 53 output unit 54 display unit 55 exercise load information detection unit 56 skin surface temperature Detector 57 Heartbeat information detector 58 Environmental temperature measurement unit

Claims (12)

an input reception unit that receives input of time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of a subject whose core body temperature is to be measured;
It is constructed by machine learning using time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of a given subject as input information, and time-series actual measured values of rectal temperature of a given subject as output information. The deep body temperature of the subject is calculated by inputting time-series data of the information regarding the subject's exercise load, the information regarding the skin surface temperature, and the information regarding the heart rate received by the input receiving unit into the learning model. an estimating unit for estimating;
and an output unit for outputting the core body temperature of the person to be measured estimated by the estimation unit. the input information further includes information about the age of the subject;
The input reception unit further receives information regarding the age of the person to be measured,
2. The core body temperature estimating device according to claim 1, wherein the estimation unit estimates the core body temperature of the person to be measured by further inputting information about the age of the person to be measured to the learning model. the input information further includes information about height and weight of the subject;
the input reception unit further receives information about the height and weight of the person to be measured;
3. The deep body temperature estimating device according to claim 1, wherein the estimation unit estimates the deep body temperature of the person to be measured by further inputting information about the height and weight of the person to be measured to the learning model. . the input information further includes information about the subject's measurement environment conditions;
the input reception unit further receives information about the measurement environment conditions of the person to be measured;

4. The estimating unit according to any one of claims 1 to 3, wherein the estimating unit estimates the core body temperature of the person to be measured by further inputting information about the measurement environment conditions of the person to be measured to the learning model. Core body temperature estimator. the input information further includes information about the sex of the subject;
the input reception unit further receives information about the sex of the person to be measured;
The core body temperature according to any one of claims 1 to 4, wherein the estimating unit estimates the core body temperature of the subject by further inputting information about the sex of the subject to the learning model. estimation device. 6. The core body temperature estimating device according to any one of claims 1 to 5, further comprising an exercise load information detection unit that detects information about the exercise load of the person to be measured. 7. The deep body temperature estimating device according to any one of claims 1 to 6, further comprising a skin surface temperature detection unit that detects information about the skin surface temperature of the person to be measured. 8. The core body temperature estimating device according to any one of claims 1 to 7, further comprising a heartbeat information detector for detecting information about the heartbeat of the subject. The deep body temperature estimating device according to any one of claims 1 to 8, wherein the estimating unit constructs a learning model using RNN (Recurrent Neural Network). 10. The core body temperature estimating device according to claim 9, wherein said RNN is LSTM (Long Short Term Memory). Receiving input of time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of the subject whose core body temperature is to be measured,
It is constructed by machine learning using time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of a given subject as input information, and time-series actual measured values of rectal temperature of a given subject as output information. estimating the body temperature of the person to be measured by inputting time-series data of the received information on the exercise load of the person to be measured, information on the skin surface temperature, and information on the heart rate to the learning model,
A core body temperature estimation method for outputting the estimated core body temperature of a subject. a step of accepting input of time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of a subject whose core body temperature is to be measured;
It is constructed by machine learning using time-series data of information on exercise load, information on skin surface temperature, and information on heart rate of a given subject as input information, and time-series actual measured values of rectal temperature of a given subject as output information. estimating the deep body temperature of the subject by inputting the received time-series data of the information on the exercise load of the subject, the information on the skin surface temperature, and the information on the heart rate to the learning model;
A core body temperature estimation program causing a computer to execute the step of outputting the estimated core body temperature of the subject.

Images