Blood glucose level measuring device and blood glucose level measuring method

Abstract

To provide a blood glucose level measuring device and a blood glucose level measuring method capable of accurately measuring a blood glucose level of a living body.SOLUTION: A blood glucose level measuring device 1 is a device for measuring a blood glucose level of a living body and includes an estimation unit 11 for estimating a blood glucose level using a learned machine learning model MO. The machine learning model MO is a model for outputting output data on a blood glucose level when a temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin of the blood of the living body is at least input as input parameters.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a blood glucose level measuring device and a blood glucose level measuring method. [Background technology]

[0002] Known technology relating to blood glucose level measuring devices that measure blood glucose levels in living organisms is, for example, the device described in Patent Document 1. The device described in Patent Document 1 predicts future glucose measurement values ​​by inputting past glucose measurement values ​​obtained by a continuous glucose monitoring system into a machine learning model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2023-529261 Summary of the Invention [Problem to be solved by the invention]

[0004] In recent years, with the advancement of machine learning technology, for example, development of blood glucose level measuring devices that measure blood glucose levels in living organisms using machine learning models has progressed. However, it is still difficult for such blood glucose level measuring devices to measure blood glucose levels with the same accuracy as, for example, medical devices.

[0005] Therefore, an object of the present disclosure is to provide a blood glucose level measuring device and a blood glucose level measuring method that are capable of measuring the blood glucose level of a living body with high accuracy. [Means for solving the problem]

[0006] As a result of extensive research, the present inventors have discovered that the temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin in a living body's blood (hereinafter also referred to as the "oxy-deoxy phase difference") correlates with the living body's blood glucose level. Specifically, they found that in a living body, blood glucose level correlates with local oxygen metabolism (oxygen consumption per cardiac beat), and local oxygen metabolism correlates with the oxy-deoxy phase difference. That is, for example, they found that when metabolism (cellular respiration) increases due to an increase in blood glucose level in a living body, oxygen consumption increases, and the oxy-deoxy phase difference also increases with this increase in oxygen consumption. By using a machine learning model trained based on these findings, they have been able to accurately estimate a living body's blood glucose level, thereby completing the present disclosure.

[0007] That is, the blood glucose measuring device of the present disclosure is [1] "a blood glucose measuring device that measures the blood glucose level of a living organism, and includes an estimation unit that estimates the blood glucose level using a trained machine learning model, and the machine learning model is a model that outputs output data related to the blood glucose level by inputting at least the temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin in the blood of the living organism as an input parameter."

[0008] The blood glucose measuring device described in [1] above can obtain output data related to the blood glucose level of a living body using a machine learning model with the oxy-deoxy phase difference as an input parameter. In this case, the above knowledge that the oxy-deoxy phase difference correlates with the blood glucose level of a living body can be used in the machine learning model, making it possible to measure the blood glucose level of a living body with high accuracy.

[0009] The blood glucose measuring device of the present disclosure may be [2] "the blood glucose measuring device according to the above [1], wherein the machine learning model is a model in which the total hemoglobin concentration of the blood of the living body is input as an input parameter." It has been found that in a living body, blood glucose levels correlate with local oxygen metabolism, and local oxygen metabolism also correlates with the total hemoglobin concentration of the blood. Therefore, the blood glucose measuring device described in the above [2] can use this knowledge in the machine learning model to more accurately measure the blood glucose level of the living body.

[0010] The blood glucose measuring device of the present disclosure may be [3] "the blood glucose measuring device according to the above [1] or [2], wherein the machine learning model is a model in which the oxygen saturation of the blood of the living body is input as an input parameter." It has been found that in a living body, blood glucose levels correlate with local oxygen metabolism, and local oxygen metabolism also correlates with blood oxygen saturation. Therefore, the blood glucose measuring device described in the above [3] can use this knowledge in the machine learning model to more accurately measure the blood glucose level of the living body.

[0011] The blood glucose measuring device of the present disclosure may be [4] "the blood glucose measuring device according to any one of [1] to [3] above, in which the machine learning model is a model in which the optical path length factor of the living body is input as an input parameter." It has been found that in a living body, blood glucose levels correlate with local oxygen metabolism, and local oxygen metabolism also correlates with the optical path length factor. Therefore, the blood glucose measuring device described in [4] above can use this knowledge in the machine learning model, making it possible to measure the blood glucose level of the living body with higher accuracy.

[0012] The blood glucose measuring device of the present disclosure may be [5] "the blood glucose measuring device according to any one of [1] to [4] above, wherein the machine learning model is a model in which at least one of the local surface body temperature of the living body, the deep body temperature of the living body, and the outside air temperature is input as an input parameter." It has been found that the local surface body temperature of the living body, the deep body temperature of the living body, and the outside air temperature affect the coefficient (metabolic coefficient) that correlates blood glucose levels with local oxygen metabolism. Therefore, the blood glucose measuring device described in [4] above can use this knowledge in the machine learning model, making it possible to measure the blood glucose level of a living body with higher accuracy.

[0013] The blood glucose measuring device of the present disclosure may be [6] "the blood glucose measuring device according to any one of [1] to [5] above, wherein the machine learning model is a model in which at least one of the altitude at which the living body is located, the sex of the living body, the physique of the living body, the type of the living body, and the age of the living body is input as input parameters." It has been found that the altitude, sex, physique, type, and age of a living body affect the metabolic coefficient that correlates blood glucose levels with local oxygen metabolism. Therefore, the blood glucose measuring device described in [6] above can use this knowledge in the machine learning model, making it possible to measure the blood glucose level of a living body with higher accuracy.

[0014] The blood glucose measuring device of the present disclosure may be [7] "the blood glucose measuring device according to any one of [1] to [6] above, wherein the machine learning model is a model in which at least one of the acceleration of the living body, the ambient light around the living body, the sleep stage of the living body, the SN ratio of the oxygenated hemoglobin, and the SN ratio of the deoxygenated hemoglobin is input as an input parameter." It has been found that the acceleration of the living body, the ambient light around the living body, the sleep stage of the living body, the SN ratio of the oxygenated hemoglobin, and the SN ratio of the deoxygenated hemoglobin affect a metabolic coefficient that correlates blood glucose levels with local oxygen metabolism. Therefore, the blood glucose measuring device described in [7] above can use this knowledge in the machine learning model, thereby enabling more accurate measurement of the blood glucose level of the living body.

[0015] The blood glucose level measuring device of the present disclosure may be [8] "the blood glucose level measuring device according to any one of [1] to [7] above, comprising a light output unit including a light source that outputs light to a living body, and a light detection unit that detects light output by the light output unit and transmitted through the living body, wherein the estimation unit calculates a temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin in the blood of the living body based on the detection result of the light detection unit." In the blood glucose level measuring device described in [8] above, light can be transmitted through the living body to calculate an oxy-deoxy phase difference.

[0016] The blood glucose measuring device of the present disclosure may be [9] "the blood glucose measuring device according to any one of [1] to [8] above, wherein the machine learning model is a model including a neural network." In the blood glucose measuring device according to [9] above, an algorithm suitable for measuring blood glucose levels can be used in the machine learning model.

[0017] The blood glucose measurement method of the present disclosure is

[10] "a blood glucose measurement method for measuring the blood glucose level of a living organism, comprising an estimation step of estimating the blood glucose level using a trained machine learning model, wherein the machine learning model is a model that outputs output data related to the blood glucose level when at least the temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin in the blood of the living organism is input as an input parameter." In the blood glucose measurement method described in

[10] above, as in the blood glucose measurement device described above, the above finding that the oxy-deoxy phase difference correlates with the blood glucose level of the living organism can be used in the machine learning model to accurately measure the blood glucose level of the living organism. [Effects of the Invention]

[0018] According to the present disclosure, it is possible to provide a blood glucose level measuring device and a blood glucose level measuring method that are capable of measuring the blood glucose level of a living body with high accuracy. [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 1 is a block diagram showing a blood glucose level measuring device according to an embodiment. [Figure 2] FIG. 2 is a graph showing an example of the detection result of the optical sensor of FIG. [Figure 3] FIG. 3 is a conceptual diagram illustrating the machine learning model of FIG. [Figure 4] FIG. 4 is a graph showing the waveform of oxygenated hemoglobin and the waveform of deoxygenated hemoglobin calculated from the detection results of FIG. [Figure 5] FIG. 5 is a schematic diagram showing the oxygenated hemoglobin waveform and the deoxygenated hemoglobin waveform of FIG. [Figure 6]FIG. 6 is a graph showing the spectra after fast Fourier transform of the oxygenated hemoglobin waveform and the deoxygenated hemoglobin waveform of FIG. [Figure 7] FIG. 7 is a flowchart showing the blood glucose level measurement method according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, the embodiments will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are designated by the same reference numerals, and redundant explanations will be omitted.

[0021] The blood glucose level measuring device 1 shown in Fig. 1 is, for example, a wearable device, a smartphone, or a pulse oximeter. Examples of wearable devices include a smart watch and a smart ring. The blood glucose level measuring device 1 measures the blood glucose level of a living organism. The living organism is, for example, a human body. The blood glucose level measuring device 1 includes a light output unit 2, a light sensor (light detection unit) 3, a temperature sensor 4, a GPS receiver (Global Positioning System) 5, a barometric pressure sensor 6, an input / output device 7, an acceleration sensor 8, an ambient light sensor 9, and an ECU (Electronic Control Unit) 10.

[0022] In this embodiment, the light output unit 2, the light sensor 3, the temperature sensor 4, the GPS receiver 5, the atmospheric pressure sensor 6, the input / output device 7, the acceleration sensor 8, the ambient light sensor 9, and the ECU 10 may all be mounted on a single device, or may be appropriately divided and mounted on multiple devices. Furthermore, at least some of the functions of the ECU 10 may be mounted on an external server that can communicate via a network.

[0023] The light output unit 2 includes a first light source 2A, a second light source 2B, and a third light source 2C that output light to the living body. The first light source 2A is a green light source. The first light source 2A outputs green light at a predetermined pulse interval in response to a control signal transmitted from the ECU 10. The green light is, for example, probe light having a central wavelength of approximately 515 nm. The second light source 2B is a red light source. The second light source 2B outputs red light at a predetermined pulse interval in response to a control signal transmitted from the ECU 10. The red light is, for example, probe light having a central wavelength of approximately 660 nm. The third light source 2C is an infrared light source. The third light source 2C outputs infrared light at a predetermined pulse interval in response to a control signal transmitted from the ECU 10. The infrared light is, for example, probe light having a central wavelength of approximately 880 nm or 910 nm. As the first light source 2A, the second light source 2B, and the third light source 2C, for example, a light emitting diode (LED), a laser diode (LD), or a super luminescent diode (SLD) is used.

[0024] The optical sensor 3 is a sensor that detects light that is output by the optical output unit 2 and passes through a living body (hereinafter also referred to as "transmitted light"). The optical sensor 3 has, for example, one or more photodetection elements and a filter. The photodetection element is, for example, a photodiode (PD). The optical sensor 3 transmits a signal related to the intensity of the transmitted light to the ECU 10. The optical sensor 3 is not particularly limited, and various known sensors may be used.

[0025] As shown in FIG. 2, the optical sensor 3 detects transmitted light, for example, green light, and detects at least first pulse wave data D1 and second pulse wave data D2. The first pulse wave data D1 is a change in the intensity of transmitted light having a first wavelength over time. The second pulse wave data D2 is a change in the intensity of transmitted light having a second wavelength different from the first wavelength over time. The first pulse wave data D1 and the second pulse wave data D2 each vary approximately periodically over time. The periods of the first pulse wave data D1 and the second pulse wave data D2 approximately match the heartbeat period of the living body. Alternatively, the optical sensor 3 may detect transmitted red light or infrared light to detect the first pulse wave data D1 and the second pulse wave data D2.

[0026] The temperature sensors 4 include a surface temperature sensor 4A, a deep temperature sensor 4B, and an outside air temperature sensor 4C. The surface temperature sensor 4A detects the surface body temperature of the living body. The surface temperature sensor 4A transmits a signal related to the detected surface body temperature to the ECU 10. The deep body temperature sensor 4B detects the deep body temperature of the living body. The deep body temperature sensor 4B transmits a signal related to the detected deep body temperature to the ECU 10. The outside air temperature sensor 4C detects the outside air temperature around the living body. The outside air temperature sensor 4C transmits a signal related to the detected outside air temperature to the ECU 10. The temperature sensor 4 is not particularly limited, and various known sensors may be used. Note that the outside air temperature sensor 4C may not be provided, and the outside air temperature around the living body may be obtained via the Internet based on the current location of the living body measured by the GPS receiver 5, for example. Alternatively, the outside air temperature detected by the outside air temperature sensor 4C and the outside air temperature obtained via the Internet may be used in combination.

[0027] The GPS receiver 5 measures the current location of the living organism by receiving signals from three or more GPS satellites. The GPS receiver 5 transmits a signal related to the measured current location of the living organism to the ECU 10. The barometric pressure sensor 6 is a sensor that detects the barometric pressure around the living organism. The barometric pressure sensor 6 transmits a signal related to the detected barometric pressure to the ECU 10. The barometric pressure sensor 6 is not particularly limited, and various known sensors may be used. It is also possible to not include the barometric pressure sensor 6, and to obtain the barometric pressure around the living organism via the Internet, for example, based on the current location of the living organism measured by the GPS receiver 5. Alternatively, a combination of the barometric pressure detected by the barometric pressure sensor 6 and the barometric pressure obtained via the Internet may be used.

[0028] The input / output device 7 is a device capable of inputting and outputting biometric information (user information). The input / output device 7 includes, for example, a touch panel display. The input / output device 7 inputs and outputs various settings related to the biometric information through operations such as touching by the user. For example, if the living body is a human body, the biometric information includes at least one of the sex, physique, race, and age. The input / output device 7 outputs various information related to the measurement results of the blood glucose measuring device 1. The input / output device 7 communicates signals related to the biometric information and the measurement results with the ECU 10.

[0029] The acceleration sensor 8 detects the acceleration of the living body. The acceleration sensor 8 transmits a signal related to the detected acceleration to the ECU 10. The acceleration sensor 8 is not particularly limited, and various known sensors may be used. The ambient light sensor 9 is a sensor that detects ambient light around the living body. The ambient light sensor 9 has a light detection element such as a photodiode. The ambient light sensor 9 transmits a signal related to the detected ambient light to the ECU 10. The ambient light sensor 9 is not particularly limited, and various known sensors may be used.

[0030] The ECU 10 is an electronic control unit having, for example, a CPU (Central Processing Unit). The ECU 10 has, as its functional configuration, an estimation unit 11 and a memory unit 12. The estimation unit 11 estimates the blood glucose level of a living organism using a trained machine learning model MO. The memory unit 12 stores in advance the machine learning model MO used by the estimation unit 11. The ECU 10 outputs measurement results including the blood glucose level estimated by the estimation unit 11, for example, via the input / output device 7. As described above, at least some of the functions of the ECU 10 may be installed in an external server, and for example, the memory unit 12 may be provided in the external server. In this case, the device worn on the living organism does not need to store (hold) the machine learning model MO. The ECU 10 may be configured by a single electronic unit or may be configured by multiple electronic units that can communicate with each other.

[0031] The machine learning model MO of this embodiment will be specifically described. The machine learning model MO is, for example, a deoxyhemoglobin concentration amplitude C deoxy,AC This model is machine-learned based on the knowledge that the deoxyhemoglobin concentration amplitude C deoxy,AC (t) is a parameter related to local oxygen metabolism and corresponds to the oxygen consumption per cardiac beat. The optical path length factor is a parameter obtained by dividing the net optical path length AC component [cm] by the net optical path length DC component [cm] and corresponds to cardiac output. The machine learning model MO is a model including a neural network. The "net" optical path length refers to the optical path length related to the contribution of hemoglobin, which causes dominant light absorption, and is used to distinguish it from the geometric apparent optical path length. In this respect, the optical path length factor may also be referred to as the net optical path length factor. Incidentally, for example, the hemoglobin volume in the probing optical path may approximate the net optical path length.

number

[0032] For example, it has been found that when metabolism (cellular respiration) increases due to an increase in blood glucose levels, the increased metabolism leads to increased oxygen consumption, and the increased oxygen consumption leads to an increase in the temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin (hereinafter also referred to as the "oxy-deoxy phase difference"). Therefore, by adopting the oxy-deoxy phase difference as an input parameter for the machine learning model MO, the machine learning model MO can accurately estimate blood glucose levels. Similarly, it has been found that increased oxygen consumption leads to increases in total hemoglobin concentration, arterial blood oxygen saturation, and optical path length factor. Therefore, by adopting at least one of these as an input parameter for the machine learning model MO, the machine learning model MO can accurately estimate blood glucose levels.

[0033] Furthermore, the machine learning model MO is a model that has been machine-learned based on the knowledge that, for example, when the living body is a human body, the outside temperature (ambient temperature), the living body's core body temperature, surface body temperature (local surface body temperature), altitude, gender, physique, race, age, acceleration, ambient light, sleep stage, SN ratio of oxygenated hemoglobin, and SN ratio of deoxygenated hemoglobin affect the metabolic coefficient for local oxygen metabolism (association with blood glucose levels).

[0034] In this machine learning model MO, blood glucose levels are based on the hierarchy shown in Figure 3. In other words, in the case of a human body, blood glucose levels in the machine learning model MO are dependent on local oxygen metabolism, and the metabolic coefficient for local oxygen metabolism is affected by ambient temperature, core body temperature, surface body temperature, altitude, gender, physique, race, age, acceleration, ambient light, sleep stage, SN ratio of oxygenated hemoglobin, and SN ratio of deoxygenated hemoglobin. As shown in equations (1) and (2) above, local oxygen metabolism is a parameter correlated with the oxy-deoxy phase difference, total hemoglobin concentration, arterial oxygen saturation, and optical path length factor.

[0035] That is, the machine learning model MO of this embodiment is a model that outputs output data related to blood glucose levels by inputting at least the oxy-deoxy phase difference of a living body as an input parameter. The machine learning model MO is a model that inputs the total hemoglobin concentration of the blood of a living body as an input parameter. The machine learning model MO is a model that inputs the oxygen saturation of the blood of a living body as an input parameter. The machine learning model MO is a model that inputs the optical path length factor of a living body as an input parameter.

[0036] The machine learning model MO is a model in which at least one of the local surface body temperature of the living body, the deep body temperature of the living body, and the outside air temperature is input as an input parameter. The machine learning model MO is a model in which at least one of the altitude where the living body is located, the sex of the living body, the physique of the living body, the type of the living body, and the age of the living body is input as an input parameter. The machine learning model MO is a model in which at least one of the acceleration of the living body, the ambient light around the living body, the sleep stage of the living body, the SN ratio of oxygenated hemoglobin, and the SN ratio of deoxygenated hemoglobin is input as an input parameter.

[0037] The oxy-deoxy phase difference is an index that significantly affects local oxygen metabolism. By using the oxy-deoxy phase difference as an input parameter for the machine learning model MO, it becomes possible to practically estimate blood glucose levels using the machine learning model MO. Total hemoglobin concentration is an index that affects local oxygen metabolism. By using total hemoglobin concentration as an input parameter for the machine learning model MO, it becomes possible to correct estimated blood glucose levels based on the degree of anemia. Total hemoglobin concentration varies by about two and a half times between individuals, which has a significant impact on metabolism-based blood glucose estimation.

[0038] Arterial blood oxygen saturation is an index that affects local oxygen metabolism. By using arterial blood oxygen saturation as an input parameter of the machine learning model MO, it is possible to correct the estimated blood glucose level from the perspective of oxygen saturation. The optical path length factor is an index that affects local oxygen metabolism. By using the optical path length factor as an input parameter of the machine learning model MO, it is possible to correct the estimated blood glucose level from the perspective of the net optical path length. The net optical path length where light absorption occurs due to oxygenated hemoglobin and deoxygenated hemoglobin can vary by up to 10 times depending on how the blood glucose measuring device 1 is attached to the living body and the living body's blood circulation state. Therefore, this correction is particularly effective.

[0039] Surface body temperature is an index that affects local metabolism. By using surface body temperature as an input parameter for the machine learning model MO, it is possible to correct estimated blood glucose levels from a metabolic perspective. This correction is effective because, even for the same blood glucose level, oxygen consumption differs depending on the surface body temperature (at low temperatures, more oxygen is burned to generate heat). Core body temperature is an index that affects local metabolism. By using core body temperature as an input parameter for the machine learning model MO, it is possible to correct estimated blood glucose levels from a metabolic perspective. This correction is effective because, even for the same blood glucose level, oxygen consumption differs depending on the core body temperature. Core body temperature more accurately reflects the metabolic state inside the body than surface body temperature.

[0040] The outside temperature is an indicator that affects local metabolism. By using the outside temperature as an input parameter for the machine learning model MO, it is possible to correct the estimated blood glucose level from a metabolic perspective. This correction is effective because oxygen consumption varies depending on the outside temperature even for the same blood glucose level. Furthermore, by combining the outside temperature and surface body temperature, it is possible to estimate core body temperature.

[0041] Altitude is an indicator that affects basal oxygen consumption. By using altitude as an input parameter for the machine learning model MO, it is possible to correct estimated blood glucose levels from the perspective of altitude difference (high altitude correction). Gender, physique, race, and age are indicators that affect basal metabolic rate. By using gender, physique, race, and age as input parameters for the machine learning model MO, it is possible to correct estimated blood glucose levels from the perspective of metabolism. In general, women have less muscle mass and a lower basal metabolic rate. There are also significant racial differences in basal metabolic rate. Basal metabolic rate declines with age.

[0042] Acceleration is an indicator that affects data reliability. Using acceleration as an input parameter for the machine learning model MO makes it possible to reject low-reliability data. By assigning a negative weight to acceleration, estimated blood glucose levels can become negative under certain conditions, which can be treated as an error.

[0043] Ambient light is an index that affects data reliability. Using ambient light as an input parameter of the machine learning model MO makes it possible to reject low-reliability data. Because the blood glucose measuring device 1 is sensitive to ambient light noise, rejection processing when ambient light is above a threshold is effective. Sleep stage, SN ratio of oxygenated hemoglobin, and SN ratio of deoxygenated hemoglobin are indexes that affect data reliability. Sleep stages include REM sleep stages and non-REM sleep stages. SN ratio is the signal-to-noise ratio, which is the ratio of the amount of effective signal components to the amount of noise components. Using sleep stage, SN ratio of oxygenated hemoglobin, and SN ratio of deoxygenated hemoglobin as input parameters of the machine learning model MO makes it possible to label data as highly reliable.

[0044] The following describes in detail the processing of the estimation unit 11. As shown in Fig. 1, the estimation unit 11 acquires each input parameter of the machine learning model MO based on each input from the light output unit 2, the light sensor 3, the temperature sensor 4, the GPS receiver 5, the barometric pressure sensor 6, the input / output device 7, the acceleration sensor 8, and the ambient light sensor 9.

[0045] Specifically, the estimation unit 11 calculates an oxy-deoxy phase difference based on the detection result of the optical sensor 3. The estimation unit 11 calculates an oxygenated hemoglobin waveform and a deoxygenated hemoglobin waveform by performing spectroscopic calculation processing based on, for example, the Modified Beer-Lambert (MBL) method on the first pulse wave data D1 and second pulse wave data D2 (see FIG. 2) detected by the optical sensor 3. The oxygenated hemoglobin waveform is data related to the oxygenated hemoglobin (OHb) concentration in the blood of the living body. The deoxygenated hemoglobin waveform is data related to the deoxygenated hemoglobin (OHb) concentration in the blood of the living body.

[0046] More specifically, the estimation unit 11 calculates the relative change in oxygenated hemoglobin over time (ΔO2Hb) and the relative change in deoxygenated hemoglobin over time (ΔHHb) based on the difference between the intensity of the first pulse wave data D1 at the first time and the intensity of the first pulse wave data D1 at the second time (the amount of change over time in the intensity of the first pulse wave data D1), the difference between the intensity of the second pulse wave data D2 at the first time and the intensity of the second pulse wave data D2 at the second time (the amount of change over time in the intensity of the second pulse wave data D2), the absorption coefficients of oxygenated hemoglobin and deoxygenated hemoglobin for the first pulse wave data D1, and the absorption coefficients of O2Hb and HHb for the second pulse wave data D2. The estimation unit 11 continues to calculate ΔO2Hb and ΔHHb at predetermined time intervals (for example, approximately 16 milliseconds). The change in ΔO2Hb over time is the oxygenated hemoglobin waveform P1 shown in FIG. 4, and the change in ΔHHb over time is the deoxygenated hemoglobin waveform P2 shown in FIG.

[0047] As shown in FIG. 5 , the estimation unit 11 calculates the oxy-deoxy phase difference Δθ as the time difference between a first feature point C1 of the oxygenated hemoglobin waveform P1 and a second feature point C2 of the deoxygenated hemoglobin waveform P2. In this embodiment, the first feature point C1 is the bottom point of the oxygenated hemoglobin waveform P1, and the second feature point C2 is the bottom point of the deoxygenated hemoglobin waveform P2. The first feature point C1 may be, for example, a peak point or a notch point of the oxygenated hemoglobin waveform P1, and the second feature point C2 may be, for example, a peak point or a notch point of the deoxygenated hemoglobin waveform P2. In this embodiment, the oxy-deoxy phase difference Δθ may be calculated using, for example, the methods disclosed in Japanese Patent No. 6846152.

[0048] The estimation unit 11 calculates the total hemoglobin concentration based on the detection results (e.g., the intensities of green light, red light, and infrared light) of the optical sensor 3. The estimation unit 11 calculates the arterial blood oxygen saturation based on the detection results (e.g., the intensities of red light and infrared light) of the optical sensor 3. The estimation unit 11 calculates the optical path length factor based on the detection results (e.g., the intensities of red light and infrared light) of the optical sensor 3. Known methods are used to calculate the total hemoglobin concentration, arterial blood oxygen saturation, and optical path length factor.

[0049] The estimation unit 11 acquires the surface body temperature of the living organism based on the detection result of the surface temperature sensor 4A. The estimation unit 11 acquires the deep body temperature of the living organism based on the detection result of the deep temperature sensor 4B. The estimation unit 11 acquires the outside air temperature based on the detection result of the outside air temperature sensor 4C. The estimation unit 11 acquires the altitude at which the living organism is located based on the output from the GPS receiver 5 and the detection result of the barometric pressure sensor 6. A known method is used to acquire the altitude.

[0050] The estimation unit 11 acquires the sex, physique, race, and age of the living body to be measured based on the biological information input via the input / output device 7. The estimation unit 11 acquires the acceleration of the living body based on the detection result of the acceleration sensor 8. The estimation unit 11 acquires the ambient light based on the detection result of the ambient light sensor 9. The estimation unit 11 calculates and acquires the sleep stage based on the detection result of the light sensor 3 (e.g., the intensity of green light) and the detection result of the acceleration sensor 8. A known method is used to acquire the sleep stage.

[0051] The estimation unit 11 calculates and acquires the S / N ratio of oxygenated hemoglobin and the S / N ratio of deoxygenated hemoglobin based on the spectrum after fast Fourier transform (FFT) of the detection result of the optical sensor 3. Specifically, as shown in Fig. 6, the estimation unit 11 performs fast Fourier transform on each of the first parameter P1 and the second parameter P2, and acquires a first spectrum F1 which is the power spectrum after fast Fourier transform of the first parameter P1, and a second spectrum F2 which is the power spectrum after fast Fourier transform of the second parameter P2.

[0052] The first spectrum F1 includes a first main peak Pm1 and a first noise floor N1. The first main peak Pm1 is the peak with the smallest frequency among the main peaks in the first spectrum F1. The first main peak Pm1 is a peak corresponding to the pulse of a living body (heartbeat frequency peak). The first noise floor N1 is a region in the first spectrum F1 that has a frequency higher than the frequency of the first main peak Pm1 and an intensity lower than the intensity of the first main peak Pm1. The intensity of the first noise floor N1 is, for example, the average intensity of a frequency range in the first spectrum F1 that is considered to be noise.

[0053] The second spectrum F2 includes a second main peak Pm2 and a second noise floor N2. The second main peak Pm2 is the peak with the smallest frequency among the main peaks of the second spectrum F2. The second main peak Pm2 is a peak corresponding to the pulse of a living body (heartbeat frequency peak). The second noise floor N2 is a region of the second spectrum F2 that has a frequency higher than the frequency of the second main peak Pm2 and an intensity lower than the intensity of the second main peak Pm2. The intensity of the second noise floor N2 is, for example, the average intensity of a frequency range of the second spectrum F2 that is considered to be noise. The first spectrum F1 and the second spectrum F2 shown in FIG. 7 are normalized so that the first main peak Pm1 of the first spectrum F1 and the second main peak Pm2 of the second spectrum F2 coincide with each other. The estimation unit 11 recognizes the first difference W1 between the first main peak Pm1 and the first noise floor N1 as the S / N ratio of the first spectrum F1 (the S / N ratio of oxygenated hemoglobin), and recognizes the second difference W2 between the second main peak Pm2 and the second noise floor N2 as the S / N ratio of the second spectrum F2 (the S / N ratio of deoxygenated hemoglobin).

[0054] The estimation unit 11 inputs the input parameters acquired as described above into the machine learning model MO. As a result, the blood glucose level is estimated in the machine learning model MO, and output data related to the blood glucose level is output from the machine learning model MO. The type and format of the output data are not particularly limited as long as it is data related to the blood glucose level.

[0055] Next, we will explain the blood glucose level measurement method performed by the blood glucose level measuring device 1. As shown in Fig. 7, first, light is output from the light output unit 2 to a living body, and the transmitted light that has passed through the living body is detected by the optical sensor 3 (step S1). As a result, first pulse wave data D1 and second pulse wave data D2 are obtained (step S2).

[0056] The estimation unit 11 calculates an oxygenated hemoglobin waveform P1 and a deoxygenated hemoglobin waveform P2 based on the first pulse wave data D1 and the second pulse wave data D2 (step S3). The estimation unit 11 obtains an oxy-deoxy phase difference based on the calculated oxygenated hemoglobin waveform P1 and deoxygenated hemoglobin waveform P2 (step S4). The estimation unit 11 obtains a total hemoglobin concentration based on the detection result of the optical sensor 3 (step S5).

[0057] The estimation unit 11 acquires the arterial blood oxygen saturation based on the detection result of the optical sensor 3 (step S6). The estimation unit 11 acquires the optical path length factor based on the detection result of the optical sensor 3 (step S7). The estimation unit 11 acquires the surface body temperature, core body temperature, and outside air temperature of the living body based on the detection result of the temperature sensor 4 (step S8). The estimation unit 11 acquires the altitude based on the output from the GPS receiver 5 and the detection result of the barometric pressure sensor 6 (step S9). The estimation unit 11 acquires the sex, physique, race, and age of the living body based on the biometric information input via the input / output device 7 (step S10).

[0058] The estimation unit 11 acquires the acceleration of the living body based on the detection result of the acceleration sensor 8 (step S11). The estimation unit 11 acquires the ambient light based on the detection result of the ambient light sensor 9 (step S12). The estimation unit 11 acquires the sleep stage based on the detection result of the light sensor 3 and the detection result of the acceleration sensor 8 (step S13). The estimation unit 11 acquires the SN ratio of oxygenated hemoglobin and the SN ratio of deoxygenated hemoglobin based on the detection result of the light sensor 3 (step S14).

[0059] The estimation unit 11 inputs each acquired input parameter into the machine learning model MO (step S15). As a result, the blood glucose level of the living body is estimated by the machine learning model MO, and output data related to the blood glucose level is output from the machine learning model MO (step S16). Note that the above steps S2 to S16 constitute an estimation step that estimates the blood glucose level using the machine learning model MO. The order in which the processes of the above steps S4 to S14 are performed is not particularly limited and may be in any order.

[0060] As described above, it has been found that blood glucose levels correlate with local oxygen metabolism in living organisms, and that local oxygen metabolism correlates with the oxy-deoxy phase difference. That is, for example, when metabolism (cellular respiration) increases due to an increase in blood glucose levels in living organisms, oxygen consumption increases, and the oxy-deoxy phase difference also increases in accordance with this increase in oxygen consumption. Therefore, the blood glucose measuring device 1 can obtain output data related to the blood glucose level of a living organism using a machine learning model MO with the oxy-deoxy phase difference as an input parameter. By utilizing the above finding that the oxy-deoxy phase difference correlates with the blood glucose level of a living organism in the machine learning model, it becomes possible to accurately measure the blood glucose level of a living organism.

[0061] In the blood glucose level measuring device 1, the machine learning model MO is a model in which the total hemoglobin concentration of the blood of a living organism is input as an input parameter. As described above, it has been found that the blood glucose level in a living organism correlates with local oxygen metabolism, and that local oxygen metabolism also correlates with the total hemoglobin concentration of the blood. Therefore, in this case, this knowledge can be used in the machine learning model MO to more accurately measure the blood glucose level of the living organism.

[0062] In the blood glucose level measuring device 1, the machine learning model MO is a model in which the oxygen saturation of the blood of a living organism is input as an input parameter. As described above, it has been found that the blood glucose level in a living organism correlates with local oxygen metabolism, and that local oxygen metabolism also correlates with blood oxygen saturation. Therefore, in this case, this knowledge can be used in the machine learning model MO to more accurately measure the blood glucose level of a living organism.

[0063] In the blood glucose level measuring device 1, the machine learning model MO is a model in which the optical path length factor of a living organism is input as an input parameter. As described above, it has been found that the blood glucose level in a living organism correlates with local oxygen metabolism, and that local oxygen metabolism also correlates with the optical path length factor. Therefore, in this case, this knowledge can be used in the machine learning model MO to more accurately measure the blood glucose level of a living organism.

[0064] In the blood glucose level measuring device 1, the machine learning model MO is a model in which at least one of the local surface body temperature of a living body, the deep body temperature of a living body, and the outside air temperature is input as an input parameter. As described above, it has been found that the local surface body temperature of a living body, the deep body temperature of a living body, and the outside air temperature affect the metabolic coefficient that correlates the blood glucose level with local oxygen metabolism. Therefore, in this case, this knowledge can be used in the machine learning model to more accurately measure the blood glucose level of a living body.

[0065] In the blood glucose measuring device 1, the machine learning model MO is a model in which at least one of the altitude at which the living body is located, the sex of the living body, the physique of the living body, the type of living body, and the age of the living body are input as input parameters. As described above, it has been found that the altitude, sex, physique, type, and age of the living body affect the metabolic coefficient that correlates the blood glucose level with local oxygen metabolism. Therefore, in this case, this knowledge can be used in the machine learning model to more accurately measure the blood glucose level of the living body.

[0066] In the blood glucose level measuring device 1, the machine learning model MO is a model in which at least one of the acceleration of the living body, the ambient light around the living body, the sleep stage of the living body, the SN ratio of oxygenated hemoglobin, and the SN ratio of deoxygenated hemoglobin is input as input parameters. As described above, it has been found that the acceleration of the living body, the ambient light around the living body, the sleep stage of the living body, the SN ratio of oxygenated hemoglobin, and the SN ratio of deoxygenated hemoglobin affect the metabolic coefficient that correlates blood glucose levels with local oxygen metabolism. Therefore, in this case, this knowledge can be used in the machine learning model to more accurately measure the blood glucose level of the living body.

[0067] The blood glucose level measuring device 1 includes a light output unit 2 including a first light source 2A, a second light source 2B, and a third light source 2C that output light to a living body, and an optical sensor 3 that detects the transmitted light output by the light output unit 2 and transmitted through the living body. The estimation unit 11 calculates an oxy-deoxy phase difference based on the detection result of the optical sensor 3. In this case, the oxy-deoxy phase difference can be calculated by transmitting light through the living body.

[0068] The machine learning model MO of the blood glucose level measuring device 1 is a model including a neural network. In this case, an algorithm suitable for measuring blood glucose levels can be used for the machine learning model MO.

[0069] The blood glucose measurement method is a blood glucose measurement method for measuring the blood glucose level of a living organism, and includes an estimation step of estimating the blood glucose level using a machine learning model MO. In this blood glucose measurement method, as in the above-mentioned blood glucose measurement device 1, the above-mentioned knowledge that the oxy-deoxy phase difference correlates with the blood glucose level of a living organism is utilized in the machine learning model MO, making it possible to accurately measure the blood glucose level of a living organism.

[0070] As described above, one aspect of the present disclosure is not limited to the above embodiment.

[0071] In the above embodiment, the input parameters of the machine learning model MO are not particularly limited, and other parameters may be used instead of or in addition to the above-mentioned parameters. For example, the machine learning model MO may be a model in which at least one of the water content of a living body, the magnetic force around the living body, and the wearing method (wearing posture) of the blood glucose measuring device 1 is input as input parameters.

[0072] Moisture content is an index that affects local metabolism. By using moisture content as an input parameter of the machine learning model MO, it is possible to correct the estimated blood glucose level from the perspective of metabolism according to moisture content. This is because the metabolic level is likely to fluctuate depending on factors such as the degree of thirst. Moisture content can be obtained by a known method using, for example, a bioelectrical impedance analysis sensor.

[0073] Magnetic force is an index that affects data reliability. By using magnetic force as an input parameter for the machine learning model MO, it becomes possible to reject data obtained under inappropriate measurement conditions. This is required to detect an inappropriate condition in which the blood glucose measuring device 1 is connected to a charger (with a built-in magnet) rather than to an arm or finger. Magnetic force can be obtained by a known method using, for example, a Hall sensor.

[0074] The wearing method is an indicator that affects the reliability of data. By using the wearing method as an input parameter of the machine learning model MO, it becomes possible to reject data with low reliability. It is possible to determine whether the blood glucose measuring device 1 is worn in a specified position and orientation, and whether the user is in an appropriate posture, and if the posture is inappropriate, the data can be rejected. The wearing method can be acquired by a known method using, for example, a geomagnetic sensor, an acceleration sensor 8, and a GPS receiver 5.

[0075] In the above embodiment, the estimation unit 11 calculates the S / N ratio by performing a fast Fourier transform on each of the oxygenated hemoglobin waveform P1 and the deoxygenated hemoglobin waveform P2, but the S / N ratio may be calculated by performing a fast Fourier transform on each of the first pulse wave data D1 and the second pulse wave data D2. In the above embodiment, a gyro sensor may be used instead of the acceleration sensor 8. In the above embodiment, an illuminance sensor may be used instead of the ambient light sensor 9.

[0076] In the above embodiment, the algorithm used in the machine learning model MO is not particularly limited. The machine learning model MO may be, for example, a linear regression model or a random forest model. The neural network model of the machine learning model MO may be a deep learning model with multiple intermediate layers.

[0077] In the above embodiment, the light output unit 2 may include a broadband light source instead of or in addition to at least one of the first light source 2A, the second light source 2B, and the third light source 2C. As the broadband light source, for example, a white LED or the like is used. When the light output unit 2 includes a broadband light source in this way, the optical sensor 3 may include a spectroscope.

[0078] The components in the above-described embodiment and modified examples are not limited to the materials and shapes described above, and various materials and shapes can be applied. Furthermore, the components in the above-described embodiment and modified examples can be arbitrarily applied to the components in other embodiments or modified examples. [Explanation of symbols]

[0079] 1...blood glucose measuring device, 2...light output unit, 2A...first light source, 2B...second light source, 2C...third light source, 3...light sensor (light detection unit), 11...estimation unit, 12...memory unit, MO...machine learning model.

Claims

1. A blood glucose level measuring device for measuring a blood glucose level of a living body, an estimation unit that estimates the blood glucose level using a trained machine learning model; A blood glucose level measuring device, wherein the machine learning model is a model that outputs output data regarding the blood glucose level when at least the temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin in the blood of the living body is input as an input parameter.

2. The blood glucose measuring device according to claim 1 , wherein the machine learning model is a model in which the total hemoglobin concentration of the blood of the living body is input as an input parameter.

3. The blood glucose measuring device according to claim 1 , wherein the machine learning model is a model in which the oxygen saturation of the blood of the living body is input as an input parameter.

4. The blood glucose measuring device according to claim 1 , wherein the machine learning model is a model in which an optical path length factor of the living body is input as an input parameter.

5. The blood glucose measuring device according to claim 1 , wherein the machine learning model is a model in which at least one of the local surface body temperature of the living body, the deep body temperature of the living body, and the outside air temperature is input as an input parameter.

6. The blood glucose measuring device of claim 1, wherein the machine learning model is a model in which at least one of the altitude at which the living body is located, the gender of the living body, the physique of the living body, the type of the living body, and the age of the living body is input as input parameters.

7. 2. The blood glucose measuring device of claim 1, wherein the machine learning model is a model in which at least one of the acceleration of the living body, the ambient light around the living body, the sleep stage of the living body, the signal-to-noise ratio of the oxygenated hemoglobin, and the signal-to-noise ratio of the deoxygenated hemoglobin is input as input parameters.

8. a light output unit including a light source that outputs light to a living body; a light detection unit that detects light output from the light output unit and transmitted through the living body, The blood glucose level measuring device according to claim 1 , wherein the estimation unit calculates a temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin in the blood of the living body based on the detection result of the light detection unit.

9. The blood glucose measuring device according to claim 1 , wherein the machine learning model is a model including a neural network.

10. A blood glucose measurement method for measuring a blood glucose level of a living body, comprising: an estimation step of estimating the blood glucose level using a trained machine learning model, A blood glucose measurement method, wherein the machine learning model is a model that outputs output data regarding the blood glucose level by inputting at least the temporal phase difference between oxygenated hemoglobin and deoxygenated hemoglobin in the blood of the living body as an input parameter.

Images