Biological information calculation device, biological information calculation method, and program

Abstract

To provide a biological information calculation device that can appropriately calculate biological information for individual organisms. [Solution] The biological information calculation device comprises an acquisition unit that acquires biological signals about a living organism, and a biological information calculation unit that calculates predicted biological information values ​​for the biological information from the biological signals using a regression model corresponding to the biological signals and reference biological information indicating the biological information about the living organism.

Description

Technical Field

[0001] The present disclosure relates to a biological information calculation device, a biological information calculation method, and a program.

Background Art

[0002] Patent Document 1 discloses a technique for calculating the confidence level that the observed feature amount of a subject belongs to each class, obtaining a stress value using a stress estimation model for each class based on the observed feature amount, and calculating a stress estimation value obtained by integrating the stress values of each estimation model by a classification score.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the technique disclosed in Patent Document 1, a classification score is calculated using attribute information. The attribute information indicates characteristics such as personality, gender, occupation, race, age, height, weight, muscle mass, lifestyle, and exercise habits. However, even for living bodies belonging to the same attribute, the state of the living body indicated by blood pressure or the like varies depending on each individual living body. Therefore, there is a possibility that the biological information indicating the state of the living body cannot be correctly calculated by the technique disclosed in Patent Document 1. Accordingly, an object is to provide a biological information calculation device, a biological information calculation method, and a program that can appropriately calculate biological information for each individual living body.

Means for Solving the Problems

[0005] A biological information calculation device according to one embodiment of the present disclosure comprises an acquisition unit that acquires biological signals about a living organism, and a biological information calculation unit that calculates predicted biological information values ​​for the biological information from the biological signals using a regression model corresponding to the biological signals and reference biological information indicating biological information about the living organism.

[0006] A method for calculating biological information according to one embodiment of the present disclosure involves acquiring a biological signal of a living organism and using a regression model corresponding to the biological signal and reference biological information indicating the biological information of the living organism to calculate a predicted value of the biological information from the biological signal.

[0007] A program according to one form of this disclosure causes a computer to perform the following functions: acquiring a biological signal about a living organism; and calculating a predicted value of biological information from the biological signal using a regression model corresponding to the biological signal and reference biological information indicating the biological information about the living organism. [Brief explanation of the drawing]

[0008] [Figure 1] This is a block diagram showing an example of the configuration of a biological information calculation device. [Figure 2] This is a block diagram showing an example of the configuration of a learning device. [Figure 3] This is a flowchart showing an example of the operation of the biological information calculation device according to the first embodiment. [Figure 4] Figure 3 shows a flowchart illustrating an example of the operation of the biological information calculation device. [Figure 5] This graph shows an example of the relationship between a regression model and training data. [Figure 6] This graph shows an example of the relationship between biological attributes and training data. [Figure 7] This graph shows an example of a domain corresponding to a feature and reference biometric information. [Figure 8] This graph shows an example of the probability that a biological signal belongs to each domain. [Figure 9]An example of a Brand-Altman plot for a comparative example of the biometric information calculation device according to the first embodiment is shown. [Figure 10] An example of a Brand-Altman plot for a biometric information calculation device according to the first embodiment is shown. [Figure 11] This is a block diagram showing an example of the configuration of a biological information calculation device according to the second embodiment. [Figure 12] A flowchart showing an example of the operation of the biological information calculation device according to the second embodiment. [Figure 13] Figure 12 shows a flowchart illustrating an example of the operation of the biological information calculation device. [Modes for carrying out the invention]

[0009] (First Embodiment) The first embodiment will be described with reference to Figures 1 to 8. In the drawings, the same or similar elements are denoted by the same reference numerals, and redundant explanations are omitted.

[0010] Figure 1 is a block diagram showing an example of the configuration of a bio-information calculation device 100. The bio-information calculation device 100 acquires bio-signals 111 and calculates bio-information from the acquired bio-signals 111. For example, the bio-signal 111 is one of the following signals: pulse wave, electroencephalogram, electrocardiogram, electromyogram, or body movement. In this disclosure, a pulse wave is a time-series signal that shows changes in blood vessel volume, calculated from a time-series signal that shows the pixel values ​​of pixels included in an image for the same location on the body surface. For example, the bio-information is blood pressure. For example, the bio-information calculation device 100 is a PC (Personal Computer), smartphone, tablet terminal, dedicated bio-information calculation terminal, etc.

[0011] The biometric information calculation device 100 comprises an acquisition unit 101, a personal identification unit 102, a biometric information conversion unit 103, a reference storage unit 104, a domain identification unit 105, a model selection unit 106, and a biometric information calculation unit 107. The acquisition unit 101, personal identification unit 102, biometric information conversion unit 103, domain identification unit 105, model selection unit 106, and biometric information calculation unit 107 are implemented by a processor such as a CPU (Central Processing Unit).

[0012] The acquisition unit 101 acquires biological signals 111 from a living organism. For example, the acquisition unit 101 is configured to include a visible light camera. The visible light camera is configured with a CCD (Charge Coupled Device) or CMOS (Complementary Metal Oxide Semiconductor) image sensor. The visible light camera may also be configured with an image sensor for a camera that includes RGB (Red Green Blue) filters. The visible light camera included in the acquisition unit 101 photographs the face of a living organism and acquires a face image. The acquisition unit 101 then calculates a pulse wave from the acquired face image. The acquisition unit 101 may also be configured to include a sensor that acquires biological signals 111. Note that the sensor can be any sensor capable of acquiring biological signals 111, and its details are not specified.

[0013] The personal identification unit 102 identifies the individual, which is the biological being measured. For example, if the acquisition unit 101 photographs the face of a living being and acquires a facial image, the personal identification unit 102 identifies the individual, which is the biological being, from the facial image.

[0014] The biological information conversion unit 103 includes a memory that stores the regression model 112 and an arithmetic device (not shown) that converts the biological signal 111 into the model output value 113. The biological information conversion unit 103 stores N regression models 112. N is an integer greater than or equal to 2. The regression model 112 takes as input the feature amount extracted from the biological signal 111 and outputs the model output value 113. The number of feature amounts input to the regression model 112 is not limited. For example, when the biological signal 111 represents a pulse wave, the regression model 112 receives as input the feature amount indicating the nature of the pulse wave and outputs the model output value 113 indicating blood pressure. The model output value 113 represents biological information. In the present disclosure, the group to which each of the N regression models 112 belongs is referred to as a domain. Details of the creation of the domain will be described later.

[0015] For example, the biological information conversion unit 103 uses the regression model 112 selected by the model selection unit 106 to convert the feature amount extracted from the biological signal 111 into the model output value 113. Alternatively, the biological information conversion unit 103 may use all the regression models 112 stored in the biological information conversion unit 103 to convert the feature amount extracted from the biological signal 111 into the model output value 113.

[0016] The reference storage unit 104 stores the reference biological information in association with the personal identification information indicating an individual. In the reference storage unit 104, the personal identification information, the pre-registered face image, and the reference biological information may be stored in association with each other. For example, the personal identification information indicates the name of the individual who is a living body. For example, the reference biological information indicates the normal blood pressure of the living body. The reference storage unit 104 is composed of an HDD (Hard Disk Drive), an SSD (Solid State Drive), a semiconductor memory, or the like.

[0017] The domain identification unit 105 identifies the domain from the biological signal 111 and the reference biological information. Specifically, the domain identification unit 105 identifies the domain from the biological data. The biological data is composed of a set in which the feature amount extracted from the biological signal 111 and the reference biological information are associated with each other.

[0018] The model selection unit 106 selects a regression model 112 corresponding to the identified domain from among multiple regression models 112 that have been pre-registered according to each of the multiple types of domains.

[0019] The biological information calculation unit 107 uses a regression model 112 that corresponds to the biological signal 111 and reference biological information indicating the biological information of the organism being measured, to calculate a predicted biological information value for the biological information from the biological signal 111. For example, if the type of biological information is blood pressure, the predicted biological information value will be blood pressure.

[0020] Figure 2 is a block diagram showing an example of the configuration of the learning device 200.

[0021] The learning device 200 is an information processing device that learns the regression model 112. For example, the learning device 200 is a different information processing device from the bio-information calculation device 100. Alternatively, the bio-information calculation device 100 may include the learning device 200.

[0022] In the following explanation, the data used for training will be referred to as training data. Furthermore, a collection of numerous training data will be referred to as a training dataset. Training data consists of pairs of feature vectors and corresponding ground truth values. Feature vectors are composed of multiple types of features. Ground truth values ​​represent the correct values ​​of biometric information corresponding to the features shown by the feature vectors.

[0023] The learning device 200 comprises a storage unit 202 and a control unit 201. The storage unit 202 is a storage medium capable of storing programs, various data, etc. The learning dataset is stored in the storage unit 202. The storage unit 202 is composed of an HDD, SSD, semiconductor memory, etc.

[0024] The control unit 201 executes various processes according to the program stored in the memory unit 202. For example, the control unit 201 is composed of a processor such as a CPU.

[0025] The control unit 201 learns a regression model 112 from the training dataset. Specifically, the control unit 201 extracts a subset from the training dataset. Then, the control unit 201 learns the regression model 112 using the training data included in the extracted subset. The regression model 112 outputs the correct values ​​from the features indicated by the feature vectors that make up the training data. For example, the control unit 201 learns the regression model 112 from the features by performing multiple linear regression. Alternatively, the control unit 201 may learn the regression model 112 using machine learning.

[0026] Figure 3 is a flowchart illustrating an example of the operation of the biological information calculation device 100 according to this embodiment. In this example, the acquisition unit 101 is equipped with a visible light camera.

[0027] In step S301, the acquisition unit 101 captures a photograph of the living person's face and acquires a facial image. Specifically, the visible light camera included in the acquisition unit 101 captures a photograph of the living person's face and acquires a facial image.

[0028] In step S302, the acquisition unit 101 acquires the biosignal 111. For example, the acquisition unit 101 calculates a pulse wave as the biosignal 111 from the facial image acquired in step S301.

[0029] In step S303, the acquisition unit 101 extracts feature quantities from the biological signal 111 acquired in step S301.

[0030] In step S304, the personal identification unit 102 determines whether or not an individual can be identified from the face image acquired in step S301. Specifically, the personal identification unit 102 determines whether or not an individual can be identified by comparing the acquired face image with the face image stored in the reference storage unit 104.

[0031] If, in step S304, an individual cannot be identified from the facial image acquired in step S301, in step S305, the acquisition unit 101 acquires the reference biometric information entered by the user and the personal identification information entered by the user. For example, if the predicted biometric information value calculated by the biometric information calculation unit 107 indicates blood pressure, the reference biometric information indicates the normal blood pressure of the organism being measured. For example, if the personal identification information entered by the user is stored in the reference storage unit 104, the acquisition unit 101 acquires the personal identification information stored in the reference storage unit 104. On the other hand, if the personal identification information is not stored in the reference storage unit 104, the acquisition unit 101 outputs a message to the display device (not shown) prompting the user to input personal identification information. The acquisition unit 101 then acquires the personal identification information entered by the user.

[0032] For example, the acquisition unit 101 outputs a message to a display device (not shown) to input reference biometric information. When the operation unit (not shown) of the biometric information calculation device 100 receives an operation from the user to input reference biometric information, the acquisition unit 101 acquires the reference biometric information input by the user. For example, the operation unit may be a keyboard, a touch panel, etc. Alternatively, the acquisition unit 101 may acquire the reference biometric information input by the user from an information processing device different from the biometric information calculation device 100.

[0033] In step S306, the acquisition unit 101 links the acquired personal identification information with the reference biometric information acquired in step S304 and stores it in the reference storage unit 104. Alternatively, the acquisition unit 101 may store the personal identification information, the acquired face image, and the reference biometric information in the reference storage unit 104. Then, the biometric information calculation device 100 proceeds to step S401, which is illustrated in Figure 3.

[0034] On the other hand, if an individual can be identified from the facial image acquired in step S301 in step S304, the personal identification unit 102 acquires reference biometric information in step S307. Specifically, the personal identification unit 102 identifies personal identification information to be stored in the reference storage unit 104, which is linked to the authenticated facial image by being compared with the acquired facial image. The personal identification unit 102 then acquires the reference biometric information linked to the identified personal identification information in the reference storage unit 104. The biometric information calculation device 100 then proceeds to step S401, which is illustrated in Figure 3.

[0035] Figure 4 is a flowchart illustrating an example of the operation of the biological information calculation device 100, following Figure 3.

[0036] In step S401, the domain identification unit 105 calculates the probability that the biological signal 111 belongs to each of several types of domains, based on the feature quantities extracted in step S303 as illustrated in Figure 3 and the reference biological information obtained in step S305 or step S307 as illustrated in Figure 3. In other words, the domain identification unit 105 calculates the probability that the biological signal 111 belongs to each of several types of domains, based on the biological signal 111 and the reference biological information.

[0037] For example, the domain identification unit 105 uses the conditions under which the learning device 200 created the domain to identify the domain to which the biosignal 111 belongs, based on the features extracted from the acquired biosignal 111 and the reference biometric information. The domain identification unit 105 then calculates the probability that the biosignal 111 belongs to each domain based on the features and the reference biometric information.

[0038] In step S402, the domain identification unit 105 identifies a domain based on the probability calculated in step S401. In step S403, the model selection unit 106 selects a regression model 112 corresponding to the domain identified in step S402. For example, the model selection unit 106 selects one regression model 112 corresponding to the identified domain. Alternatively, the model selection unit 106 may select multiple regression models 112 corresponding to the identified domain. Also, for example, the model selection unit 106 may determine the number of regression models 112 to select depending on the identified domain. Alternatively, if the probability that the biosignal 111 belongs to each domain is shown by a vector, the model selection unit 106 selects a regression model 112 based on that probability. For example, the model selection unit 106 may select a regression model 112 to use from a set of regression models 112 corresponding to each of several types of domains for which the probability calculated in step S401 is not zero.

[0039] Furthermore, even if the biological signals 111 are obtained from the same living organism, differences may occur in the biological signals 111 depending on the measurement conditions under which the biological signals 111 are measured. In such cases, differences will also occur in the feature quantities extracted from the biological signals 111 and the model output values ​​113 due to the influence of the measurement conditions. However, the biological information calculation device 100 according to this embodiment identifies domains from the biological signals 111 and reference biological information, and selects a regression model corresponding to the identified domains. As a result, the biological information calculation device 100 according to this embodiment can select a regression model according to the measurement conditions, and can suppress a decrease in the accuracy of biological information calculation due to the influence of the measurement conditions.

[0040] In step S404, the model selection unit 106 calculates the importance of the regression model 112 selected in step S402. Specifically, the model selection unit 106 calculates the importance of the regression model 112 based on the biosignal 111 and the reference bioinformation. For example, the model selection unit 106 calculates the importance of the regression model 112 based on the probability calculated in step S401. For example, the model selection unit 106 calculates the importance of the regression model 112 such that the greater the calculated probability for a domain, the higher the importance of the regression model 112 corresponding to that domain. Furthermore, if multiple regression models 112 are selected, the model selection unit 106 calculates the importance of each regression model 112.

[0041] In step S405, the bioinformation conversion unit 103 uses the selected regression model 112 to convert the features extracted from the biosignal 111 into model output values ​​113. For example, the bioinformation conversion unit 103 converts the features used by the domain identification unit 105 to identify a domain into model output values ​​113. In this case, the acquisition unit 101 performs the process of extracting features from the biosignal 111 once, and the bioinformation conversion unit 103 and the domain identification unit 105 use these features. Alternatively, the bioinformation conversion unit 103 may convert features different from those used by the domain identification unit 105 to identify the identified domain into model output values ​​113.

[0042] The bio-information calculation device 100 converts the features extracted from the bio-signal 111 into multiple model output values ​​113 using the selected regression model 112, thereby reducing the computational load compared to converting the features extracted from the bio-signal 111 into model output values ​​113 using all the regression models 112 stored in the bio-information conversion unit 103.

[0043] In step S406, the bio-information calculation unit 107 calculates a predicted bio-information value from the model output value 113 converted in step S404 and the importance calculated in step S403. In other words, the bio-information calculation unit 107 calculates a predicted bio-information value based on the model output value 113 obtained by converting the bio-signal 111 using the selected regression model 112.

[0044] For example, if the model selection unit 106 selects multiple regression models 112, the bio-information calculation unit 107 uses the importance of each regression model 112 to synthesize the model output values ​​113 calculated by each regression model 112. For example, the bio-information calculation unit 107 synthesizes the model output values ​​113 by calculating a weighted average of the model output values ​​113 according to the importance of each regression model 112. Then, the bio-information calculation unit 107 calculates a bio-information prediction value from the synthesized model output values ​​113.

[0045] For example, if the biosignal 111 indicates a pulse wave and the reference bioinformation indicates the normal blood pressure of the organism, the selected regression model 112 takes features representing the properties of the pulse wave as input and outputs a model output value 113 for blood pressure. In this case, the predicted bioinformation value will be blood pressure.

[0046] The biometric information calculated from the biosignal 111 cannot be classified according to attributes such as gender, age, height, and weight, and fluctuates due to internal and external factors. Internal factors include physical condition and mental state. External factors include temperature and atmospheric pressure. Therefore, if a domain is identified based on attributes such as gender, age, height, and weight, there is a risk that the appropriate domain will not be identified. As a result, there is a risk that the biometric information will not be calculated using an appropriate regression model 112. However, the biometric information calculation device 100 according to this embodiment identifies a domain from the biosignal 111 and reference biometric information. In other words, the biometric information calculation device 100 according to this embodiment can identify an appropriate domain for each individual organism. As a result, the biometric information calculation device 100 according to this embodiment can calculate predicted biometric information values ​​using an appropriate regression model 112 for each individual organism.

[0047] Refer to Figures 5 and 6 for a detailed explanation of how to train regression model 112 and how to create the domain.

[0048] First, we will explain in detail one method for training regression model 112 and creating the domain.

[0049] The control unit 201 performs the process of extracting subsets from the training dataset N times, where N is a natural number greater than or equal to 1. For example, the N extracted subsets are independent and contain different training data. Alternatively, the N extracted subsets may include subsets containing overlapping training data. In the following description, each of the N regression models 112 will be denoted as regression model 112n, where n is a natural number greater than or equal to 1 and less than or equal to N.

[0050] The control unit 201 may extract subsets based on at least one of the environmental information and attribute information from which the training data was acquired. Environmental information indicates information about the surrounding environment when the training data was acquired. For example, environmental information indicates the temperature, humidity, etc., when the training data was acquired. Attribute information indicates the attributes of a living organism that correspond to the features indicated by the feature vectors included in the training data. For example, the attributes of a living organism include gender, age, etc. For example, the control unit 201 extracts subsets from the training data for each age group and trains the regression model 112.

[0051] The control unit 201 creates domains such that the regression model 112 with the smallest error in the training data is the same. For example, the control unit 201 creates domains using methods such as least squares or machine learning. As a result, the control unit 201 creates a model that identifies domains from features and ground truth values.

[0052] When the control unit 201 learns N regression models 112, the output value from each regression model 112 differs from the ground truth value. Therefore, the control unit 201 creates a discriminant model to identify the regression model 112 in which the error between the feature and the ground truth value is relatively small.

[0053] Figure 5 is a graph showing an example of the relationship between the regression model 112 and the training data. In Figure 5, the horizontal axis represents the feature vector and the vertical axis represents the ground truth value of the biometric information. Figure 5 shows a graph plotted for each regression model 112 that minimizes the error for each training data set. Specifically, Figure 5 shows a graph plotted for three regression models 112 with N=3. The areas separated by dashed lines in the example in Figure 5 represent domains. Note that while Figure 5 shows an example of a model that identifies a domain from one feature vector and its ground truth value, the control unit 201 may identify a domain from multiple feature vectors and their ground truth values.

[0054] Next, we will describe in detail the training of regression model 112 and other methods for domain creation.

[0055] Figure 6 is a graph illustrating an example of the relationship between biological attributes and training data. Specifically, Figure 6 shows graphs plotting training data for women aged 40 and over, training data for women under 40, training data for men aged 40 and over, and training data for men under 40. In Figure 6, the horizontal axis represents features, and the vertical axis represents the ground truth values ​​of biological information.

[0056] The control unit 201 creates a domain according to the error of the regression model 112. The control unit 201 creates a domain defined by the feature amount and the correct value. For example, the control unit 201 creates a domain based on at least one of the environmental information and the attribute information from which the learning data is acquired. Then, the control unit 201 extracts subsets for each domain and learns the regression model 112. For example, as illustrated in FIG. 6, assume that the control unit 201 creates a domain. In that case, the domain identification unit 105 identifies the domain to which the biological signal 111 belongs among the plurality of types of domains illustrated in FIG. 6 from the feature amount extracted from the biological signal 111 and the reference biological information.

[0057] FIG. 7 is a graph showing an example of a domain corresponding to the feature amount and the reference biological information. In FIG. 7, the feature amount is taken on the x-axis which is the horizontal axis, and the reference biological information is taken on the y-axis which is the vertical axis. In this example, the feature amount is denoted as x, and the value indicated by the reference biological information is denoted as y.

[0058] Assume that a domain is created as illustrated in the graph illustrated in FIG. 7. In that case, when x ≦ y ≦ 20 / x, the domain identification unit 105 identifies that the biological signal 111 belongs to domain 1. When x ≦ y and 20 / x < y, the domain identification unit 105 identifies that the biological signal 111 belongs to domain 2. When y < x and y ≦ 20 / x, the domain identification unit 105 identifies that the biological signal 111 belongs to domain 3. When 20 / x < y < x, the domain identification unit 105 identifies that the biological signal 111 belongs to domain 4.

[0059] FIG. 8 is a graph showing an example of the probability that the biological signal 111 belongs to each domain. In FIG. 8, the feature amount is taken on the x-axis which is the horizontal axis, and the reference biological information is taken on the y-axis which is the vertical axis. In this example, the feature amount is denoted as x, and the value indicated by the reference biological information is denoted as y.

[0060] Assume that the domains are created as illustrated in the graph shown in Figure 8. In this case, if xy < 10, the domain identification unit 105 calculates the probability that the biological signal 111 belongs to domain 1 as 1 and the probability that the biological signal 111 belongs to domain 2 as 0. If 10 ≤ xy < 30, the domain identification unit 105 calculates the probability that the biological signal 111 belongs to domain 1 as 1.5 - xy / 20 and the probability that the biological signal 111 belongs to domain 2 as xy / 20 - 0.5. If xy ≥ 30, the domain identification unit 105 calculates the probability that the biological signal 111 belongs to domain 1 as 0 and the probability that the biological signal 111 belongs to domain 2 as 1.

[0061] Figure 9 shows an example of a Brand-Altman plot for a comparative example of the biometric information calculation device 100 according to this embodiment. Figure 9 shows a Brand-Altman plot when blood pressure is calculated from pulse waves using a single regression model, regardless of the domain. In Figure 9, the horizontal axis is the mean and the vertical axis is the error. As illustrated in Figure 9, when blood pressure is calculated from pulse waves using a single regression model, the range of the error, expressed as ±1.96 × standard deviation, is 49.2.

[0062] Figure 10 shows an example of a Brand-Altman plot for the biometric information calculation device 100 according to this embodiment. Figure 10 shows a Brand-Altman plot when the biometric information calculation device 100 according to this embodiment calculates a predicted biometric information value, such as blood pressure, from a biometric signal, using a regression model corresponding to a biometric signal representing a pulse wave and reference biometric information. In Figure 10, the horizontal axis is the mean value and the vertical axis is the error. As illustrated in Figure 10, when the biometric information calculation device 100 according to this embodiment calculates a predicted biometric information value from a biometric signal, using a regression model corresponding to a biometric signal and reference biometric information, the range of error, expressed as ±1.96 × standard deviation, is 25.4. In other words, compared to the case where a predicted biometric information value is calculated using a single regression model regardless of the domain, the biometric information calculation device 100 according to this embodiment can reduce the range of error.

[0063] For example, as illustrated in Figure 9, if a single regression model is used to calculate predicted biological information regardless of the domain, there is a risk that the predicted biological information may be calculated using a regression model unsuitable for the organism being measured. On the other hand, the biological information calculation device 100 according to this embodiment can identify the domain to which a regression model suitable for the organism being measured belongs by using reference biological information. As a result, as illustrated in Figure 10, the biological information calculation device 100 according to this embodiment can improve the accuracy of calculating predicted biological information by using a regression model suitable for the organism being measured. Furthermore, since the biological information calculation device 100 according to this embodiment only requires the input of reference biological information and does not need to require the user to input various attribute information, it can accurately calculate predicted biological information without burdening the user.

[0064] (Second embodiment) The second embodiment will be described with reference to Figures 11 to 13. In the drawings, the same or equivalent elements are denoted by the same reference numerals, and redundant explanations are omitted.

[0065] Figure 11 is a block diagram showing an example of the configuration of the biological information calculation device 100 according to this embodiment. The difference between the biological information calculation device 100 illustrated in Figure 11 and the biological information calculation device 100 illustrated in Figure 1 is that the biological information calculation device 100 illustrated in Figure 11 is equipped with a domain storage unit 1101 instead of a reference storage unit 104.

[0066] The domain storage unit 1101 stores identification domain information associated with personal identification information. Alternatively, the domain storage unit 1101 may store identification domain information associated with both the face image and the personal identification information. The identification domain information indicates the domain identified by the domain identification unit 105. For example, the identification domain information indicates the identification information of the domain to which the biosignal 111 belongs. Alternatively, the identification domain information may represent the probability that the biosignal 111 belongs to each domain using a vector.

[0067] In this embodiment, the domain identification unit 105 identifies a domain from the biosignal 111 and reference biosignal information if no identification domain information is stored in the domain storage unit 1101.

[0068] In this embodiment, the model selection unit 106 selects a regression model 112 that corresponds to a domain indicated by the identification domain information stored in the domain storage unit 1101.

[0069] Figure 12 is a flowchart showing an example of the operation of the biological information calculation device 100 according to this embodiment. The processes of steps S1201 to S1203 illustrated in Figure 12 are the same as the processes of steps S301 to S304 illustrated in Figure 3, so a detailed explanation is omitted.

[0070] If, in step S1204, an individual cannot be identified from the facial image acquired in step S1201, the acquisition unit 101 acquires the reference biometric information and personal identification information entered by the user in step S1205. The process in step S1205 is the same as the process in step S305 illustrated in Figure 3, so a detailed explanation is omitted.

[0071] In step S1206, the domain identification unit 105 calculates the probability that the biological signal 111 belongs to each domain based on the features extracted in step S1203 and the reference biological information obtained in step S1204.

[0072] In step S1207, the domain identification unit 105 identifies the domain to which the biosignal 111 belongs based on the probability calculated in step S1206.

[0073] In step S1208, the domain identification unit 105 links the acquired personal identification information with the identification domain information indicating the identified domain and stores it in the domain storage unit 1101. Alternatively, the acquisition unit 101 may link the personal identification information, the acquired face image, and the identification domain information and store them in the domain storage unit 1101. Then, the biometric information calculation device 100 proceeds to step S1301, which is illustrated in Figure 13.

[0074] On the other hand, if an individual can be identified from the facial image in step S1204, the personal identification unit 102 acquires identification domain information in step S1209. Specifically, the personal identification unit 102 identifies personal identification information that is linked to the authenticated facial image by being compared with the acquired facial image and stored in the domain storage unit 1101. The personal identification unit 102 then acquires the identification domain information associated with the personal identification information identified in step S1204 from the domain storage unit 1101. The biometric information calculation device 100 then proceeds to step S1301, which is illustrated in Figure 13.

[0075] Figure 13 is a flowchart illustrating an example of the operation of the biological information calculation device 100, following Figure 12.

[0076] In step S1301, the model selection unit 106 selects a regression model 112 corresponding to the domain obtained in step S1207 as illustrated in Figure 12, or the domain indicated by the identification domain information obtained in step S1209 as illustrated in Figure 12. In step S1302, the model selection unit 106 calculates the importance of the regression model 112 selected in step S1301. In step S1303, the bioinformation conversion unit 103 uses the selected regression model 112 to convert the features extracted from the biosignal 111 into a model output value 113. In step S1304, the bioinformation calculation unit 107 calculates a bioinformation prediction value from the model output value 113 converted in step S404 and the importance calculated in step S1302. The processing in steps S1301 to S1304 is the same as the processing in steps S403 to S407 as illustrated in Figure 4, so a detailed explanation is omitted.

[0077] As described above, the biological information calculation device 100 according to this embodiment stores domain identification information that indicates a domain corresponding to each individual biological organism, eliminating the need to repeat the domain identification process when calculating predicted biological information values ​​for the same biological organism multiple times. Therefore, when acquiring biological signals from the same biological organism multiple times, the biological information calculation device 100 according to this embodiment can efficiently calculate predicted biological information values ​​while appropriately calculating biological information for that organism.

[0078] (modified version) As a modification of the biometric information calculation device 100 according to this embodiment, the biometric information calculation device 100 may store selected regression models 112. For example, the biometric information calculation device 100 according to this modification stores the selected regression model in association with personal identification information. Alternatively, the biometric information calculation device 100 according to this modification may store the importance of each regression model 112. For example, the biometric information calculation device 100 according to this modification stores the importance of each regression model in association with personal identification information. As a result, when calculating predicted biometric information for the same organism multiple times, the biometric information calculation device 100 according to this modification can efficiently calculate predicted biometric information while appropriately calculating the biometric information for that organism.

[0079] The processes performed in the above embodiments are not limited to the processing modes exemplified in each embodiment. The functional blocks described above may be implemented using either logic circuits (hardware) formed on an integrated circuit or the like, or software using a CPU. The processes performed in the above embodiments may be executed on multiple computers.

[0080] This disclosure is not limited to the embodiments described above, and may be replaced with configurations substantially identical to those shown in the embodiments, configurations that produce the same effects, or configurations that can achieve the same objectives. This disclosure also includes embodiments obtained by appropriately combining the technical means disclosed in different embodiments. Furthermore, new technical features can be formed by combining the technical means disclosed in each embodiment. [Explanation of Symbols]

[0081] 1 Regression model, 100 Biological information calculation device, 101 Acquisition unit, 102 Personal identification unit, 103 Biological information conversion unit, 104 Reference storage unit, 105 Domain identification unit, 106 Model selection unit, 107 Biological information calculation unit, 111 Biological signal, 112 Regression model, 113 Model output value, 200 Learning device, 201 Control unit, 202 Storage unit, 1101 Domain storage unit

Claims

1. An acquisition unit that acquires biological signals from living organisms, A biological information calculation unit that calculates a predicted biological information value for the biological information from the biological signal using a regression model corresponding to the biological signal and reference biological information indicating the biological information of the biological organism, A biological information calculation device equipped with the following features.

2. A domain identification unit that identifies a domain from the biological signal and the reference biological information, A model selection unit that selects the regression model corresponding to the identified domain from a plurality of regression models that have been pre-registered according to each of the multiple types of domains, Furthermore, The biological information calculation unit calculates the predicted biological information value based on the model output value obtained by converting the biological signal using the selected regression model. The biological information calculation device according to claim 1.

3. The system further comprises a personal identification unit for identifying the aforementioned biological individual. A biological information calculation device according to claim 1 or 2.

4. The aforementioned biosignal shows a pulse wave, The acquisition unit acquires the biological signal from the facial image of the living organism, The personal identification unit identifies the personal from the facial image. The biological information calculation device according to claim 3.

5. The system further comprises a reference storage unit that stores the reference biometric information linked to the personal identification information representing the aforementioned individual. The biological information calculation device according to claim 3.

6. The system further comprises a domain storage unit that stores identification domain information indicating the domain in association with the personal identification information indicating the biological organism. The biological information calculation device according to claim 2.

7. The model selection unit calculates the importance of the regression model based on the biological signal and the reference biological information. The biological information calculation unit uses the importance level to synthesize multiple model output values ​​and calculates the biological information prediction value from the synthesized model output values. The biological information calculation device according to claim 2.

8. The domain identification unit calculates the probability that the biological signal belongs to each of the multiple types of domains from the biological signal and the reference biological information, The model selection unit calculates the importance from the probability. The biological information calculation device according to claim 7.

9. The acquisition unit acquires the reference biometric information entered by the user. A biological information calculation device according to claim 1 or 2.

10. The aforementioned biological signal is one of the following signals: pulse wave, electroencephalogram, electrocardiogram, electromyogram, or body movement. A biological information calculation device according to claim 1 or 2.

11. The aforementioned predicted biological information value indicates blood pressure. A biological information calculation device according to claim 1 or 2.

12. We acquire biosignals from living organisms, Using a regression model that corresponds to the aforementioned biological signal and reference biological information representing the biological information of the organism, predictive values ​​of the biological information are calculated from the biological signal. Method for calculating biological information.

13. On the computer, Functions for acquiring biological signals from living organisms, A function to calculate predicted biological information values ​​for the biological information from the biological signal using a regression model corresponding to the biological signal and reference biological information indicating the biological information of the biological organism, A program that executes the command.

Images