Sleep estimation system, sleep estimation method and program

Abstract

To easily estimate sleep stages.SOLUTION: A sleep estimation device includes: a first acquisition unit for acquiring blood flow data; a generation unit for generating a frequency spectrum of the blood flow data by executing frequency analysis processing on the blood flow data; and a first determination unit for determining the sleep stage of a subject on the basis of the frequency spectrum.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] FIELD The present disclosure relates to estimating a subject's sleep stage. [Background technology]

[0002] Patent Document 1 describes a technology for detecting sleep stages. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2018-161432 Summary of the Invention

[0004] A sleep estimation system according to one embodiment of the present disclosure includes a determination unit that determines a sleep stage of a subject based on a frequency spectrum of the subject's blood flow data, and a notification unit that executes notification processing based on the determination result of the determination unit, wherein the notification unit issues a notification to the subject to encourage awakening when the determination unit determines that the frequency spectrum includes a predetermined shape in a region that includes at least a frequency band of 0.2 Hz or more and 0.3 Hz or less, or after a predetermined time has elapsed since the determination, and the predetermined shape included in the region is an upwardly convex waveform shape that includes a first intensity that is the maximum intensity in the region.

[0005] Furthermore, a sleep estimation system according to one embodiment of the present disclosure includes a determination unit that determines a sleep stage of a subject based on a frequency spectrum of the subject's blood flow data, and a notification unit that executes notification processing based on the determination result of the determination unit, wherein the determination unit determines whether the frequency spectrum of the blood flow data corresponds to a correct answer when applied to a trained model that has been trained using training data in which a frequency spectrum having a predetermined shape in a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less is linked to a correct answer label based on biological information different from the blood flow data, and the notification unit issues a notification to the subject to encourage them to wake up when the determination unit determines that the frequency spectrum corresponds to a correct answer or after a predetermined time has elapsed since the determination, and the predetermined shape included in the region is an upwardly convex waveform shape that includes a first intensity that is the maximum intensity in the region, and the biological information is electroencephalogram data detected by an electroencephalograph.

[0006] Furthermore, a sleep estimation method according to one embodiment of the present disclosure is a sleep estimation method executed in a sleep estimation system, comprising: a determination step in which a determination unit included in the sleep estimation system determines a sleep stage of the subject based on a frequency spectrum of the subject's blood flow data; and a notification step in which a notification unit included in the sleep estimation system issues a notification to the subject to encourage awakening when the determination unit determines in the determination step that the frequency spectrum includes a predetermined shape in a region including at least a frequency band of 0.2 Hz or more and 0.3 Hz or less, or after a predetermined time has elapsed since the determination, wherein the predetermined shape included in the region is an upwardly convex waveform shape that includes a first intensity that is the maximum intensity in the region.

[0007] Furthermore, a sleep estimation method according to an aspect of the present disclosure is a sleep estimation method executed in a sleep estimation system, the sleep estimation method including: a determination step in which a determination unit included in the sleep estimation system determines a sleep stage of the subject based on a frequency spectrum of blood flow data of the subject; and a notification step in which a notification unit included in the sleep estimation system executes notification processing based on a determination result in the determination step, wherein the determination step determines whether the frequency spectrum of the blood flow data corresponds to a correct answer when applied to a trained model trained using teacher data in which a frequency spectrum having a predetermined shape in a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less is associated with a correct answer label based on biological information different from the blood flow data, and the notification step, when it is determined in the determination step that the frequency spectrum corresponds to a correct answer or after a predetermined time has elapsed since the determination, issues a notification to the subject to encourage awakening, and the predetermined shape included in the region is an upwardly convex waveform shape including a first intensity that is the maximum intensity in the region, The biological information is electroencephalogram data detected by an electroencephalograph. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a block diagram illustrating a schematic configuration example of a sleep estimation system according to a first embodiment. [Figure 2] 10 is a graph showing an example of blood flow waveform data detected by a blood flow meter. [Figure 3] 10 is a graph showing an example of a frequency spectrum generated by performing a Fourier transform process on blood flow waveform data. [Figure 4] 4 is a graph showing an example of electrocardiographic waveform data detected by an electrocardiograph. [Figure 5] 10 is a graph showing an example of a frequency spectrum generated by performing a Fourier transform process on electrocardiogram waveform data. [Figure 6] 10 is an image showing an example of the result of wavelet transform processing on blood flow waveform data. [Figure 7] 4 is a flowchart showing an example of a processing flow by the sleep estimation device of the first embodiment. [Figure 8] FIG. 10 is a block diagram showing a schematic configuration example of a sleep estimation system according to a second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] The following describes the determination (estimation) of a subject's sleep stage according to the present disclosure. First, the principle underlying the determination of a subject's sleep stage according to the present disclosure will be described. It should be noted that "A to B" in this specification means "A or higher and B or lower." Furthermore, this specification will use blood flow waveform data as an example of blood flow data used to determine a sleep stage.

[0010] 〔principle〕 FIG. 2 is a graph showing an example of blood flow waveform data detected by a blood flow meter. In FIG. 2, the vertical axis indicates a value proportional to the blood flow rate per unit time (unit: dimensionless), and the horizontal axis indicates the measurement time (unit: min). The blood flow meter can acquire raw waveform data W1 and processed waveform data W2 as blood flow waveform data, as shown in FIG. 2. The processed waveform data W2 is waveform data obtained by processing the raw waveform data W1 to make it easier to obtain the R-wave peak. The processed waveform data W2 is generated, for example, by performing a smoothing process on the raw waveform data W1. In the processed waveform data W2, the time interval between adjacent peaks (blood flow rates at positions indicated by inverted triangles in the figure) indicates the heartbeat interval (RRI: RR Interval). Similarly, in raw waveform data W12 of the heartbeat shown in FIG. 4, the time interval between adjacent peaks (blood flow rates at positions indicated by inverted triangles in the figure) indicates the heartbeat interval (RRI).

[0011] A blood flow meter for detecting blood flow waveform data is a sensor that can detect blood flow waveform data indicating the blood flow of a subject by receiving scattered light generated by irradiating light onto the blood vessels of the subject. The blood flow meter may include a light-emitting unit that irradiates light onto the blood vessels of the subject, and a light-receiving unit that receives the scattered light.

[0012] Generally, when a fluid is irradiated with laser light, the irradiated laser light is scattered by (i) scatterers contained in the fluid and moving with the fluid, and (ii) stationary objects such as pipes for flowing the fluid, generating scattered light. Generally, scatterers cause non-uniformity in the complex refractive index in the fluid.

[0013] The scattered light generated by scatterers moving with the fluid undergoes a wavelength shift due to the Doppler effect, which depends on the flow velocity of the scatterers. On the other hand, the scattered light generated by stationary objects does not undergo a wavelength shift. These scattered lights cause optical interference, resulting in the observation of optical beats.

[0014] The blood flow meter may be a sensor that utilizes this phenomenon, i.e., a laser Doppler blood flow meter that detects, as blood flow waveform data, an optical beat caused by scattered light generated in the blood as a fluid by irradiating a blood vessel of a subject with laser light.

[0015] More specifically, the acquired light receiving signal may be analyzed by a processor included in the blood flow meter, and frequency analysis data indicating the signal strength of each frequency of the light receiving signal may be calculated. As an example, the processor may analyze the acquired light receiving signal using a technique such as FFT (Fast Fourier Transformation).

[0016] The processor may further generate blood flow waveform data indicating a fluctuation pattern of the blood flow volume of the subject based on the frequency analysis data. As an example, the processor may calculate the first moment sum X of the acquired frequency analysis data as the blood flow waveform data. More specifically, the processor may calculate the first moment sum X of the acquired frequency analysis data using the following formula: The processor may calculate the first moment sum X in a certain frequency band (for example, 1 to 20 kHz) using the following formula: X=Σfx×P(fx) Here, "fx" is the frequency, and "P(fx)" is the value of the signal strength at frequency fx.

[0017] The first moment sum X calculated by the processor based on the frequency analysis data can be a value proportional to the blood flow volume of the subject. The processor may generate pattern data indicating a fluctuation pattern of the blood flow volume of the subject over time by calculating the first moment sum X for each of the plurality of frequency analysis data. The processor may also generate blood flow waveform data using data included in a certain frequency band of the data included in the frequency analysis data. The processor can output the generated blood flow waveform data.

[0018] In addition to blood flow volume, the blood flow waveform data may include data on at least one of cardiac output and coefficient of variation of vasomotion. Cardiac output is the amount of blood pumped with one cardiac beat. Vasomotion is the spontaneous rhythmic contraction and expansion of blood vessels. The coefficient of variation of vasomotion is a value that indicates the variation in blood flow volume caused by vasomotion.

[0019] The blood flow waveform data may also include a pulse wave.

[0020] Fig. 3 is a graph showing an example of a frequency spectrum generated by performing a Fourier transform process on blood flow waveform data such as that shown in Fig. 2. The vertical axis represents the intensity (unit: dB) of the frequency spectrum, and the horizontal axis represents the frequency (unit: Hz). The Fourier transform process is an example of a frequency analysis process, and is a process for generating a frequency spectrum of waveform data that does not include temporal changes.

[0021] 3 shows frequency spectra FW1 and FW2 corresponding to each sleep stage. The frequency spectrum FW1 is a frequency spectrum generated by performing a Fourier transform on the raw waveform data W1. The frequency spectrum FW2 is a frequency spectrum generated by performing a Fourier transform on the beat-to-beat interval (RRI) of the processed waveform data W2.

[0022] The sleep stages can be classified into three stages: wakefulness, REM sleep, and non-REM sleep. Non-REM sleep can be further classified into stage 1 (N1), stage 2 (N2), and stage 3 (N3), in order of lightest sleep. REM sleep is sleep accompanied by rapid eye movement (REM). Non-REM sleep is sleep not accompanied by rapid eye movement.

[0023] This classification is performed based on electroencephalogram data detected by an electroencephalograph attached to the subject. Electroencephalograms are divided into four types, in order of decreasing wavelength: beta waves, alpha waves, theta waves, and delta waves. Beta waves are brain waves with a frequency of, for example, about 38 to 14 Hz. Alpha waves are brain waves with a frequency of, for example, about 14 to 8 Hz. Theta waves are brain waves with a frequency of, for example, about 8 to 4 Hz. Delta waves are brain waves with a frequency of, for example, about 4 to 0.5 Hz.

[0024] A person is asleep when theta and delta waves are dominant compared to beta and alpha waves. Here, "dominant" refers to a certain wave being present in a larger proportion of measured electroencephalograms. It is known that dominant electroencephalograms periodically vary between theta and delta waves during sleep. Furthermore, when the proportion of theta waves in the electroencephalograms is less than a predetermined value, the person is in REM sleep. When the proportion of theta waves is greater than a predetermined value and delta waves are dominant, the person is in non-REM sleep. Stage 1, for example, is a state in which alpha waves are 50% or less and various low-amplitude frequencies are present. Stage 2, for example, is a state in which low-amplitude theta and delta waves appear irregularly but high-amplitude slow waves are absent. Stage 3, for example, is a state in which the frequency is 2 Hz or less and 75 μV slow waves account for 20% or more. A state in which the frequency is 2 Hz or less and 75 μV slow waves account for 50% or more may be referred to as stage 4.

[0025] In Figure 3, the sleep stages are indicated as follows: wakefulness is indicated as "WK," REM sleep is indicated as "RM," stage 1 of non-REM sleep is indicated as "N1," stage 2 of non-REM sleep is indicated as "N2," and stage 3 of non-REM sleep is indicated as "N3."

[0026] As shown in the frequency spectra FW1 and FW2 in Fig. 3, a significant change in intensity is observed in the frequency band of 0.2 to 0.3 Hz (predetermined frequency band) in the frequency spectrum obtained from a subject in sleep stages 2 and 3. In other words, in the frequency spectrum, the intensity in a first range R1, which is a part of the frequency band of 0.2 to 0.3 Hz, is greater than the intensity in a second range R2 other than the first range R1 by a predetermined value or more. Hereinafter, the intensity in the first range R1 will be referred to as the "first intensity," and the intensity in the second range R2 will be referred to as the "second intensity."

[0027] The first intensity may be, for example, the maximum intensity in a frequency band of 0.2 to 0.3 Hz. The second intensity may be, for example, the maximum intensity in a second range R2 outside a first range R1 including the maximum intensity (e.g., approximately ±0.02 Hz around the maximum intensity, but within the frequency band of 0.2 to 0.3 Hz). The predetermined value may be set to a magnitude sufficient to identify, through experimentation, a characteristic waveform Sh that is not present in adjacent frequency bands in the 0.2 to 0.3 Hz frequency band. The characteristic waveform Sh may be, for example, an upwardly convex waveform with a relatively broad waveform (e.g., a waveform with a full half-width of 0.03 Hz or greater). Figure 3 shows an example of the first range R1 and the second range R2 in the frequency spectra FW1 and FW2 corresponding to stage 3.

[0028] On the other hand, in the frequency spectra obtained from subjects in wakefulness, REM sleep, and stage 1 sleep, no significant intensity changes were observed in the frequency band of 0.2 to 0.3 Hz, and the characteristic waveform Sh as described above was also not observed.

[0029] After extensive research, the inventors have found that when a frequency spectrum in which a significant intensity change (characteristic waveform Sh) is observed in the frequency band of 0.2 to 0.3 Hz is obtained, there is a high possibility that the subject is in stage 2 or 3 sleep. In other words, the inventors have found that when a frequency spectrum having a characteristic waveform Sh is obtained, there is a high possibility that the subject is in stage 2 or 3 sleep. The inventors have also found that, particularly in the frequency spectrum of blood flow waveform data detected by the above-mentioned blood flow meter (e.g., laser Doppler blood flow meter), there is a significant intensity change in the frequency band of 0.2 to 0.3 Hz. Based on these findings, the inventors have developed a sleep estimation device that can improve the accuracy of determining a subject's sleep stage.

[0030] (Comparison with electrocardiogram waveform data) Furthermore, there is a difference between electrocardiogram waveform data (electrocardiogram) detected by an electrocardiograph and blood flow waveform data detected by a blood flowmeter, as explained below.

[0031] Fig. 4 is a graph showing an example of electrocardiographic waveform data detected by an electrocardiograph. In Fig. 4, the vertical axis represents the intensity of the heartbeat (unit: dB) and the horizontal axis represents the measurement time (unit: min). Fig. 4 shows raw waveform data W11 and W12 of the heartbeat as electrocardiographic waveform data.

[0032] 5 is a graph showing an example of a frequency spectrum generated by performing a Fourier transform on electrocardiogram waveform data, where the vertical axis represents the intensity (unit: dB) of the frequency spectrum and the horizontal axis represents the frequency (unit: Hz).

[0033] 5 shows frequency spectra FW11 and FW12 corresponding to each sleep stage. The frequency spectrum FW11 is a frequency spectrum generated by performing a Fourier transform on raw waveform data W11. The frequency spectrum FW12 is a frequency spectrum generated by performing a Fourier transform on the heart rate interval (RRI). The sleep stages corresponding to the frequency spectra FW11 and FW12 are identified based on electroencephalogram data acquired from an electroencephalograph attached to the subject.

[0034] As shown in Figure 5, in the frequency spectra FW11 and FW12 obtained by converting electrocardiogram waveform data detected from a subject in sleep stages 2 and 3, no significant intensity change (characteristic waveform Sh) is observed in the frequency band of 0.2 to 0.3 Hz. The frequency spectrum FW12 corresponding to sleep stages 2 and 3 has an upward convex shape in the frequency band of 0.2 to 0.3 Hz. However, the frequency spectrum FW12 corresponding to sleep stage 1 also has a similar shape to the frequency spectrum FW12 corresponding to stages 2 and 3. Therefore, it is not recognized that the frequency spectrum FW12 corresponding to stages 2 and 3 has a significant intensity change in the frequency band of 0.2 to 0.3 Hz.

[0035] After extensive research, the inventors discovered that the significant intensity change observed in the frequency band of 0.2 to 0.3 Hz is a phenomenon specific to the frequency spectrum of blood flow waveform data. As a result, the inventors discovered that by determining a subject's sleep stage using a frequency spectrum converted from blood flow waveform data rather than electrocardiogram waveform data, it is highly likely that it will be possible to accurately determine whether the subject is in stage 2 or 3 sleep.

[0036] (Regarding frequency bands) The frequency band in which the above-mentioned significant intensity changes are observed may vary slightly depending on the blood flow meter used and individual differences between subjects. Considering this, it is quite possible that significant intensity changes not observed in the electrocardiogram waveform data may be observed in the frequency band of, for example, 0.15 to 0.4 Hz in the frequency spectra FW1 and FW2 corresponding to stage 2 or 3. In the following explanation, the frequency band in which the above-mentioned significant intensity changes are observed will be described as being 0.2 to 0.3 Hz.

[0037] [Wavelet transform processing] The above principle has been described using a frequency spectrum obtained by performing a Fourier transform process as the frequency analysis process. However, the sleep stage of a subject may be determined based on a frequency spectrum obtained by performing a wavelet transform process as the frequency analysis process. The wavelet transform process is an example of a time-frequency analysis process. The time-frequency analysis process is a process for generating a frequency spectrum of waveform data including temporal changes. The wavelet transform process is a process for generating a frequency spectrum of waveform data using a mother wavelet, which is an arbitrary reference waveform.

[0038] The mother wavelet used in the wavelet transform process is defined as follows: In the following equation, "t" is the time variable, "a" is the scale parameter (a parameter that expands or contracts the mother wavelet in the time axis direction), and "b" is the translate parameter (a parameter that translates the mother wavelet in the time axis direction).

[0039]

number

[0040] The function for performing wavelet transform processing is defined as follows: In the following equation, "f(t)" represents waveform data, and "*" represents a complex conjugate number. By substituting the mother wavelet with the adjusted values ​​of "a" and "b" into the following equation, the frequency spectrum of the waveform data can be generated.

[0041]

number

[0042] By using wavelet transform processing, the intensity in the frequency band of 0.2 to 0.3 Hz (hereinafter referred to as target intensity) can be emphasized relatively more than other frequency bands. As explained in the above principle, when a frequency spectrum having a characteristic waveform Sh is obtained in the frequency band of 0.2 to 0.3 Hz, there is a high possibility that the subject is in stage 2 or 3 sleep. Therefore, by using wavelet transform processing to emphasize the target intensity, it is possible to increase the possibility of more accurately determining whether the subject is in stage 2 or 3 sleep.

[0043] In the wavelet transform process, a mother wavelet set to enhance the target intensity as described above may be used. A mother wavelet with enhanced target intensity can be set by adjusting the values ​​of "a" and "b" described above. Morlet may also be used as the mother wavelet. In this case, the scale parameter "a" has the relationship "ω = 2π / a" and represents the local angular frequency. Since the angular frequency "ω" can be expressed as "ω = 2πf", the wavelet transform process may be performed with this "f (frequency)" portion set to 0.2 to 0.3 Hz. The target intensity may be the intensity of the entire frequency band from 0.2 to 0.3 Hz, or the intensity within a partial range of the frequency band (e.g., the first range R1).

[0044] Furthermore, as a result of the wavelet transform processing of the blood flow waveform data, intensity change data can be generated that indicates the change in intensity over time in each frequency band within a predetermined time. The predetermined time may be set, for example, through experiments, to a time that allows accurate determination of the subject's sleep stage. In this embodiment, the predetermined time may be set, for example, to 2.5 minutes.

[0045] Unlike Fourier transform processing, wavelet transform processing can generate a frequency spectrum that includes temporal changes in intensity, allowing for a larger amount of data than Fourier transform processing. Generally, in generating a trained model described below, the greater the amount of data (the more features to be trained), the more accurate the trained model that can output output data can be generated. Therefore, using intensity change data is effective when generating a trained model.

[0046] FIG. 6 is an image showing an example of the results of wavelet transform processing on blood flow waveform data. FIG. 6 shows an image generated by performing wavelet transform processing on blood flow waveform data (raw waveform data W1) that is the source of frequency spectrum FW1 of stage 3 shown in FIG. 3. This image is an example of intensity change data in which target intensities are emphasized. Hereinafter, an image showing intensity change data will be referred to as a wavelet image.

[0047] In Fig. 6, the vertical axis represents frequency (unit: Hz) and the horizontal axis represents time (unit: min). The shading in the wavelet image in Fig. 6 represents intensity (unit: dB). In other words, the wavelet image is data showing the intensity distribution of the frequency spectrum on a plane defined by frequency and time.

[0048] In the wavelet image of this embodiment, frequency bands (intensity) may be represented by color gradations. In the wavelet image, for example, low frequency bands may be represented by cool colors, and high frequency bands may be represented by warm colors. Specifically, the frequency bands may be represented in descending order by dark blue, blue, light blue, yellow-green, light yellow-green, yellow, orange, and red. As long as the intensity distribution is visible in the wavelet image, the frequency bands may be represented by other colors or by grayscale. FIG. 6 shows an example of a grayscale image obtained by converting the wavelet image represented by the above colors.

[0049] In the wavelet image of FIG. 6, a first region AR1 distributed along the time axis at and around 0.2 Hz exhibits higher intensity than the intensity in the frequency bands adjacent to the first region AR1. Specifically, in the frequency band of approximately 0.2 Hz ± approximately 0.05 Hz in the first region AR1, a red intensity band is distributed along the time axis, and intensity regions shown in orange, yellow, light yellow-green, and yellow-green are distributed around this intensity band. In FIG. 6, a portion of the red intensity band is indicated by reference numeral 101. Furthermore, a portion of the orange, yellow, light yellow-green, and yellow-green intensity regions is indicated by reference numeral 102. Meanwhile, in the adjacent frequency band, intensity regions shown in light blue, blue, and dark blue are mainly distributed, but the red, orange, and yellow intensity regions are not distributed. In FIG. 6, a portion of the light blue, blue, and dark blue intensity regions is indicated by reference numeral 103.

[0050] Furthermore, as shown in Fig. 6, in the wavelet image, a sawtooth intensity band is formed along the time axis in frequency bands higher than the 0.2 to 0.3 Hz frequency band. In the wavelet image of Fig. 6, a sawtooth intensity band is formed in a second region AR2 (a frequency band of approximately 0.7 Hz or higher). The second region AR2 has a lower intensity than the first region AR1 in a frequency band of approximately 0.9 to 1.0 Hz, and the intensity gradually decreases toward frequency bands below 0.9 Hz and frequency bands of 1.0 Hz or higher.

[0051] The sawtooth-shaped intensity bands indicate an intensity distribution corresponding to heartbeats. When asleep, the intensity distribution takes the shape of a band along the time axis, and the lighter the sleep, the more distorted the band shape becomes. The intensity distribution corresponding to heartbeats is an intensity distribution that cannot be obtained by Fourier transform processing. By using wavelet images to generate the trained model described below, it is possible to generate a trained model that also takes heartbeats into account.

[0052] FIG. 6 shows an example of a wavelet image in which the object intensities are enhanced, but it should be noted that even in a wavelet image in which the object intensities are not enhanced, the intensity of the first region AR1 will be higher than the intensity in the adjacent frequency bands.

[0053] [Generating a trained model] In determining the sleep stage of a subject, a trained model (approximator) for determining the sleep stage of the subject can be used. The trained model is a mathematical model (a neural network including an input layer, a hidden layer, and an output layer) that mimics the neurons of the human nervous system and is trained to be able to determine the sleep stage of a user. Any mathematical model may be used as long as it can generate a trained model that can determine the sleep stage of a subject. The mathematical model may be, for example, a convolutional neural network (CNN), a recurrent neural network (RNN), or a long short-term memory (LSTM).

[0054] Learning refers to adjusting the strength of connections between units, bias in connections, and the like so that correct calculation results are output from the output layer. In this embodiment, when learning is performed, learning data is input to the input layer. In the hidden layer, calculations based on calculation data are performed on the learning data, and the calculation results in the hidden layer are output as output data from the output layer. The teacher data and the output data are compared, and the calculation data is adjusted to reduce errors. This process is repeatedly performed for each of multiple learning data, thereby generating a trained model in which the calculation data has been adjusted. In other words, in this embodiment, the trained model may be generated by so-called supervised learning using the training data and teacher data. The sleep estimation device 51, which will be described later, can determine the sleep stage of a subject by using the trained model generated in this way.

[0055] The training data is data that serves as an example for generating a trained model. The training data may be a frequency spectrum generated from blood flow waveform data. In this embodiment, a wavelet image is used. The wavelet image may be one in which the target intensity is emphasized or one in which the target intensity is not emphasized. The training data may be data that exhibits different behaviors during wakefulness and sleep, or data that exhibits behaviors that change depending on the depth of sleep. The training data may be various types of data (e.g., frequency spectra of blood flow waveform data that exhibit different waveforms).

[0056] The training data is data in which a correct label is associated with the training data. For example, data in which the sleep stage of a person from whom blood flow waveform data was acquired is associated as a correct label with a frequency spectrum as the training data may be used as the training data. As described above, the sleep stage of a subject may be identified based on electroencephalogram data detected by an electroencephalograph. A code indicating each sleep stage may be used as the correct label. Alternatively, a code indicating a correct answer may be used for a specific sleep stage (e.g., stage 2 or 3) and a code indicating an incorrect answer may be used for other sleep stages. In this embodiment, an example of the training data may be data in which a correct label is associated with a wavelet image known to correspond to stage 2 or 3 (a wavelet image identified as stage 2 or 3 based on electroencephalogram data).

[0057] The calculation data is data related to calculations for generating a trained model, including data such as calculation formulas, variables of the calculation formulas (e.g., biases and weights), and activation functions. The biases and weights define the strength of connections between units. By adjusting the biases and weights, the accuracy of the trained model can be improved. As a method for adjusting the calculation data, for example, backpropagation and gradient descent may be adopted.

[0058] [Embodiment 1] An example of a sleep estimation system 1 that is constructed based on the above principle and is capable of determining the sleep stage of a subject will be described below. The sleep estimation system 1 of this embodiment may be a system that is capable of determining the sleep stage of a subject using the trained model.

[0059] <Sleep stage estimation system> Fig. 1 is a block diagram showing a schematic configuration example of a sleep estimation system 1 according to embodiment 1. As shown in Fig. 1, the sleep estimation system 1 includes an accelerometer 2, a blood flow meter 3, and a mobile terminal 5. In the mobile terminal 5, as part of the function of a control unit that comprehensively controls each component of the mobile terminal 5, a sleep estimation device 51 that determines the sleep stage of a subject by executing, for example, an application that can determine the sleep stage is implemented.

[0060] <Accelerometer> The accelerometer 2 is a sensor that can detect acceleration caused by the movement of the subject. The accelerometer 2 may transmit the detected acceleration as acceleration data to the sleep estimation device 51 via wireless or wired communication. The accelerometer 2 is attached to a part of the subject's body, such as the head or a finger. As the accelerometer 2, a known sensor such as a frequency change type, a piezoelectric type, a piezo-resistive type, or a capacitance type may be used.

[0061] <Blood flow meter> The blood flow meter 3 may be the blood flow meter described in the above principle. The blood flow meter 3 may be, for example, a laser Doppler blood flow meter. In this embodiment, the blood flow meter 3 may transmit raw waveform data W1 as blood flow waveform data to the sleep estimation device 51. The blood flow meter 3 may transmit processed waveform data W2 instead of the raw waveform data W1 to the sleep estimation device 51. The blood flow meter 3 may not generate the processed waveform data W2, but the sleep estimation device 51 may generate the processed waveform data W2. The blood flow meter 3 is attached to a part of the subject's body, such as an ear, a finger, a wrist, an arm, a forehead, a nose, or a neck.

[0062] <Mobile device> The mobile terminal 5 may be any terminal capable of data communication with at least the accelerometer 2 and the blood flow meter 3. The mobile terminal 5 may be, for example, a smartphone or a tablet. The mobile terminal 5 includes a sleep estimation device 51, a storage unit 52, and a notification unit 53.

[0063] The storage unit 52 can store programs and data used by the control unit (particularly, the sleep estimation device 51). The storage unit 52 can store, for example, the trained model generated as described above and a threshold value for determining whether the subject is stationary or not.

[0064] The notification unit 53 can notify various types of information to those around the mobile terminal 5 (e.g., the subject). In this embodiment, the notification unit 53 can notify various types of information in accordance with a notification instruction from the sleep estimation device 51. The notification unit 53 may be at least one of a sound output device that outputs sound, a vibration device that vibrates the mobile terminal 5, and a display device that displays an image.

[0065] (Sleep estimation device) The sleep estimation device 51 can determine the sleep stage of a subject wearing the accelerometer 2 and the blood flow meter 3. The sleep estimation device 51 may include a second acquisition unit 11, a second determination unit 12, a first acquisition unit 13 (acquisition unit), a generation unit 14, a first determination unit 15 (determination unit), and a notification unit 16.

[0066] The second acquisition unit 11 can acquire acceleration data from the accelerometer 2. The second determination unit 12 can determine whether the subject is stationary or not based on the acceleration data acquired by the second acquisition unit 11. For example, the second determination unit 12 may determine that the subject is stationary when the acceleration indicated by the acceleration data is less than a threshold value stored in the storage unit 52. The second determination unit 12 may transmit determination result data to the generation unit 14.

[0067] The first acquisition unit 13 can acquire blood flow waveform data (raw waveform data W1) from the blood flow meter 3. The generation unit 14 can generate a frequency spectrum of the blood flow waveform data by performing frequency analysis processing on the blood flow waveform data acquired by the first acquisition unit 13. The first determination unit 15 can determine the sleep stage of the subject based on the frequency spectrum generated by the generation unit 14. In other words, the first determination unit 15 can determine the transition of the depth of sleep of the subject based on the frequency spectrum.

[0068] In this embodiment, the generator 14 generates a wavelet image by performing a wavelet transform process as a frequency analysis process, in which a target intensity is relatively emphasized compared to other frequency bands. The wavelet image is processed data that indicates the result of a time-frequency analysis process of blood flow waveform data. The first determination unit 15 can determine the sleep stage of the subject based on the wavelet image generated by the generator 14.

[0069] In the present embodiment, when the first determination unit 15 determines that the characteristic waveform Sh is included in the frequency band of 0.2 to 0.3 Hz of the frequency spectrum, it can determine that the subject's sleep stage is stage 2 or 3. In this case, the first determination unit 15 may determine that the subject's sleep stage has transitioned from stage 1 to stage 2 or 3. On the other hand, when the first determination unit 15 determines that the characteristic waveform Sh is not included in the frequency band of 0.2 to 0.3 Hz, it can determine that the subject's sleep stage is a sleep stage other than stage 2 or 3. When the first determination unit 15 determines that the subject is in stage 2 or 3 sleep and then determines that the characteristic waveform Sh is not included in the frequency band of 0.2 to 0.3 Hz, it can determine that the subject's sleep stage has transitioned from stage 2 or 3 to stage 1. The determination of whether the characteristic waveform Sh is included can also be made, for example, by determining whether the first intensity is greater than the second intensity by a predetermined value or more.

[0070] In the present embodiment, the first determination unit 15 may determine the sleep stage of the subject by using a trained model. In this case, the first determination unit 15 can provide the wavelet image generated by the generation unit 14 as input data to the input layer of the trained model, and output the determination result of the sleep stage of the subject from the output layer of the trained model.

[0071] As described above, the trained model is generated using, as an example of training data, data associated as a correct label with a wavelet image known to correspond to stage 2 or 3. Therefore, by providing the trained model with the wavelet image generated by the generation unit 14, the first determination unit 15 can determine whether the subject's sleep stage is stage 2 or 3, and whether the subject has transitioned from stage 1 to stage 2 or 3. When the first determination unit 15 provides the trained model with a frequency spectrum containing a waveform Sh characteristic of the 0.2 to 0.3 Hz frequency band, in particular, the first determination unit 15 may be able to accurately determine whether the subject's sleep stage is stage 2 or 3.

[0072] Here, when estimating sleep stages using blood flow waveform data, the sleep estimation device may not be able to determine whether the subject is asleep or not based on the blood flow waveform data alone. For example, in the case of a subject whose blood flow is stable even when awake, there may be no significant difference between the blood flow waveform data when awake and the blood flow waveform data when asleep. In this case, the sleep estimation device may not be able to determine whether the subject is asleep or not. In this embodiment, when the second determination unit 12 determines that the subject is stationary, the generation unit 14 performs frequency analysis processing on the blood flow waveform data, thereby enabling the first determination unit 15 to perform determination processing in a state where the subject is likely to have fallen asleep.

[0073] The notification unit 16 can execute notification processing based on the determination result of the first determination unit 15. The notification unit 16 may transmit a notification instruction according to the notification processing to the alarm unit 53. This allows the alarm unit 53 to notify the surroundings of the mobile terminal 5 of information according to the notification processing. The notification unit 16 may include a first notification unit 161 and a second notification unit 162.

[0074] As the notification process, the first notification unit 161 can execute the first notification process after a predetermined time has elapsed since the first determination unit 15 detected a transition from stage 1 to stage 2 or 3. The first notification process is a notification process accompanying the detection of the transition, and may be, for example, an alarm process for urging the subject to wake up, or a process for notifying the subject of the detection of the transition. The predetermined time may be appropriately set, for example, through experiments, depending on the purpose of the notification. In this embodiment, the predetermined time may be set to, for example, a time at which the subject is estimated to be likely to wake up when timed from the time of the transition from stage 1 to stage 2 or 3. The first notification process allows, for example, the subject to wake up at an appropriate time after falling asleep. The first notification unit 161 may execute the first notification process when the first determination unit 15 detects a transition from stage 1 to stage 2 or 3.

[0075] As the notification process, the second notification unit 162 can execute the second notification process after a predetermined time has elapsed since the first determination unit 15 detected a transition from Stage 2 or 3 to Stage 1. The second notification process is a notification process accompanying the detection of the transition, and may be, for example, an alarm process to prompt the subject to wake up, or a process to notify the subject of the detection of the transition. The predetermined time may be appropriately set, for example, through experiments, depending on the purpose of the notification. In this embodiment, the predetermined time may be set to, for example, a time estimated to be a time at which the subject is likely to wake up when timed from the transition from Stage 2 or 3 to Stage 1. The second notification process allows, for example, the subject to wake up at an appropriate time after falling asleep. The second notification unit 162 may execute the second notification process when the first determination unit 15 detects a transition from Stage 2 or 3 to Stage 1.

[0076] [Processing flow] 7 is a flowchart showing an example of a processing flow (sleep estimation method) by the sleep estimation device 51. When the sleep estimation device 51 determines the sleep stage of a subject, the blood flow meter 3 may start detecting blood flow waveform data after the blood flow meter 3 is attached to the subject.

[0077] As shown in FIG. 7, in the sleep estimation device 51, the first acquisition unit 13 can acquire blood flow waveform data from the blood flow meter 3 (S1: first acquisition step, acquisition step). The generation unit 14 can generate a frequency spectrum of the blood flow waveform data by performing frequency analysis processing on the blood flow waveform data. In this embodiment, the generation unit 14 can generate a wavelet image in which target intensities are emphasized by performing wavelet transform processing on the blood flow waveform data (S2: generation step). The first determination unit 15 can determine the sleep stage of the subject based on the frequency spectrum. In this embodiment, the first determination unit 15 can input the wavelet image in which target intensities are emphasized into a trained model (S3) to determine the sleep stage of the subject (S4: first determination step, determination step).

[0078] The first determination unit 15 can determine whether or not there has been a transition from stage 1 to stage 2 or 3 based on the sleep stage determination result (S5). When the first determination unit 15 determines that there has been a transition from stage 1 to stage 2 or 3 (YES in S5), the first notification unit 161 can determine whether or not a predetermined time has elapsed since that determination (S6). When the first notification unit 161 determines that the predetermined time has elapsed (YES in S6), it can execute, as the first notification process, for example, alarm processing to issue an alarm sound from the notification unit 53 (S7). The notification unit 53 can issue an alarm sound upon receiving a notification instruction from the first notification unit 161.

[0079] If the result of S5 is NO, the process may return to S1. If the result of S6 is NO, the process of S6 may be repeated. Furthermore, if the first determination unit 15 determines in S5 that the state has transitioned from stage 2 or 3 to stage 1, the second notification unit 162 may execute the second notification process, for example, after a predetermined time has elapsed.

[0080] [Problems with the prior art and effects of the sleep estimation device according to the present disclosure] Patent Document 1 discloses a method for detecting non-REM sleep, which includes the following steps 1 to 4. Step 1: A step of generating time series data of the subject's cardiac intervals. Step 2: A step of setting a window of a predetermined time length that moves along the time axis of the time series data, and performing spectral analysis on the time series data within the window that includes each of a plurality of determination points on the time axis. Step 3: Calculate the power concentration of high frequency components of heart rate variability from the spectrum of each window. A step of determining whether or not the sleep is non-REM sleep based on the calculated concentration level.

[0081] In the above method, the device for detecting the time-series data in step 1 is, for example, a pulse wave monitor or an electrocardiograph.

[0082] However, operating an electroencephalograph and acquiring brain waves requires highly specialized knowledge, and attaching the electroencephalograph is cumbersome. Therefore, it is difficult for a subject to easily acquire brain waves and easily grasp his or her own sleep stage.

[0083] Furthermore, as described above, the inventors have discovered that if a significant change in intensity is observed in the frequency band of 0.2 to 0.3 Hz in the frequency spectrum of blood flow waveform data, there is a high possibility that the subject is in stage 2 or 3 sleep.

[0084] The sleep estimation device 51 of the present disclosure can determine the sleep stage of a subject using blood flow waveform data. Handling a blood flow meter and acquiring blood flow waveform data does not require advanced specialized knowledge compared to an electroencephalograph. Furthermore, attaching a blood flow meter is easier compared to an electroencephalograph. In other words, the sleep estimation device 51 can acquire blood flow waveform data of a subject relatively easily, and therefore can determine the sleep stage of the subject relatively easily. Furthermore, the sleep estimation device 51 allows the subject to easily grasp his or her own sleep stage.

[0085] Furthermore, when a frequency spectrum having the above-described significant intensity change not observed in electrocardiogram waveform data is obtained, the sleep estimation device 51 of the present disclosure can determine that the subject's sleep stage is stage 2 or 3. Therefore, the sleep estimation device 51 can increase the possibility of accurately estimating that the subject's sleep stage is stage 2 or 3.

[0086] [Embodiment 2] Another embodiment of the present disclosure will be described below. For ease of explanation, the same reference numerals are used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated. Fig. 8 is a block diagram showing a schematic configuration example of a sleep estimation system 1A according to a second embodiment.

[0087] In the sleep estimation system 1 of the first embodiment, the mobile terminal 5 can acquire blood flow waveform data from the blood flow meter 3 attached to the subject, for example, by wireless communication. Then, the sleep estimation device 51 constructed in the mobile terminal 5 can determine the sleep stage of the subject based on the blood flow waveform data.

[0088] On the other hand, as shown in FIG. 8, in the sleep estimation system 1A of the second embodiment, the wearable device 20 may be provided with a blood flow meter 3. In addition, the wearable device 20 may be configured with a sleep estimation device 51 as part of the functions of a control unit that comprehensively controls each component of the wearable device 20. That is, in the sleep estimation system 1A, the sleep estimation device 51 may be implemented in the wearable device 20 together with the blood flow meter 3. Therefore, a single device can perform the process of acquiring blood flow waveform data and the process of determining sleep stages based on the blood flow waveform data. Furthermore, various devices or components that are required for wireless or wired communication between the two devices are not required. The wearable device 20 may be worn at a position similar to the position where the blood flow meter 3 is worn on the subject.

[0089] [Modification] The invention according to the present disclosure has been described above based on the drawings and examples. However, the invention according to the present disclosure is not limited to the above-described embodiments. In other words, the invention according to the present disclosure can be modified in various ways within the scope of the present disclosure, and embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the invention according to the present disclosure. In other words, it should be noted that a person skilled in the art can easily make various modifications or corrections based on the present disclosure. It should also be noted that these modifications or corrections are included in the scope of the present disclosure.

[0090] (Modification of frequency analysis processing) For example, the generating unit 14 may perform wavelet transform processing on blood flow waveform data to generate a wavelet transformed image, and does not necessarily need to perform wavelet transform processing so as to enhance the target intensity.

[0091] The generating unit 14 may also perform time-frequency analysis processing other than wavelet transform processing on the blood flow waveform data. The generating unit 14 may also perform, for example, short-time Fourier transform processing on the blood flow waveform data. The short-time Fourier transform processing is a processing in which a Fourier transform is performed on each of multiple waveform data extracted along the time axis using a window function. In this case, the generating unit 14 may perform short-time Fourier transform processing in which the target intensity is relatively emphasized compared to other frequency bands, thereby generating an image (intensity change data) similar to a wavelet image as processed data.

[0092] Furthermore, the generation unit 14 may execute a process other than time-frequency analysis as the frequency analysis process. The generation unit 14 may execute, for example, a Fourier transform process. When the generation unit 14 executes a Fourier transform process, for example, the frequency spectrum FW1 or FW2 shown in Fig. 3 can be generated. Furthermore, the generation unit 14 may execute a short-time Fourier transform process to generate a frequency spectrum (waveform) such as that shown in Fig. 3.

[0093] In this way, the generation unit 14 can generate various frequency spectra. Therefore, the first determination unit 15 can determine the sleep stage of the subject by inputting various frequency spectra into the trained model. However, the trained model is generated using the same type of frequency spectrum as the frequency spectrum generated by the generation unit 14 as training data and teacher data.

[0094] (Modification of the Determination Process of the First Determination Unit) The trained model does not have to be stored in the storage unit 52 of the mobile terminal 5 or the wearable device 20. In other words, the sleep estimation system 1 or 1A does not have to use the trained model to determine the sleep stage of the subject.

[0095] For example, when the first determination unit 15 determines that the first intensity is greater than the second intensity by a predetermined value or more in the frequency spectrum generated by the generation unit 14, the first determination unit 15 may determine that the sleep stage of the subject is stage 2 or 3. The storage unit 52 may store a predetermined value instead of a trained model.

[0096] Furthermore, when determining the sleep stage of a subject using the frequency spectrum FW1, the first determination unit 15 may determine that the subject's sleep stage is stage 2 or 3 if a characteristic waveform Sh can be extracted in a frequency band of 0.2 to 0.3 Hz. In this case, the storage unit 52 may store a reference waveform from which a characteristic waveform Sh can be extracted in the frequency spectrum FW1, instead of a trained model. When the first determination unit 15 determines that a waveform matching the reference waveform exists in the frequency band of 0.2 to 0.3 Hz of the frequency spectrum FW1, the first determination unit 15 may determine that a characteristic waveform Sh has been extracted in the frequency band. When the degree of match between a waveform included in the frequency band of 0.2 to 0.3 Hz and the reference waveform is equal to or greater than a threshold value set by experiment, for example, the first determination unit 15 may determine that a waveform matching the reference waveform exists in the frequency band of 0.2 to 0.3 Hz. The first determination unit 15 may also determine whether or not the characteristic waveform Sh is included in the frequency band of 0.2 to 0.3 Hz using other indicators (for example, the degree of change in the slope of the waveform).Furthermore, the same determination as for the frequency spectrum FW1 may be made for the frequency spectrum FW2.

[0097] (Modification of Sleep Estimation System 1 or 1A) It is not necessary for the accelerometer 2 to detect acceleration caused by the subject's movement and the sleep estimation device 51 to determine the subject's sleep state based on the detected acceleration. In this case, the sleep estimation system 1 or 1A does not need to include the accelerometer 2, and the sleep estimation device 51 does not need to include the second acquisition unit 11 and the second determination unit 12.

[0098] [Software implementation example] The control block of the sleep estimation device 51 may be realized by a logic circuit (hardware) formed on an integrated circuit (IC chip) or the like, or may be realized by software.

[0099] In the latter case, the sleep estimation device 51 includes a computer that executes instructions of a program, which is software that realizes each function. The computer includes, for example, at least one processor (control device) and at least one computer-readable recording medium storing the program. The object of the present disclosure is achieved by the processor reading and executing the program from the recording medium in the computer. The processor may be, for example, a CPU (Central Processing Unit). The recording medium may be a "non-transitory tangible medium," such as a ROM (Read Only Memory), a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The sleep estimation device 51 may also include a RAM (Random Access Memory) for expanding the program. The program may be supplied to the computer via any transmission medium capable of transmitting the program (such as a communication network or broadcast waves). Note that one aspect of the present disclosure may also be realized in the form of a data signal embedded in a carrier wave, in which the program is embodied by electronic transmission.

[0100] [Expression examples of the present disclosure] Furthermore, one aspect of the present disclosure may be expressed as follows.

[0101] A sleep estimation device according to one embodiment of the present disclosure includes a first acquisition unit that acquires blood flow data indicating blood flow in a subject, a generation unit that generates a frequency spectrum of the blood flow data by performing a frequency analysis process on the blood flow data, and a first determination unit that determines the sleep stage of the subject based on the frequency spectrum.

[0102] Furthermore, a sleep estimation device according to one aspect of the present disclosure includes an acquisition unit that acquires blood flow data indicating the blood flow of a subject, a generation unit that generates processed data indicating the results of a time-frequency analysis process of the blood flow data by performing a wavelet transform process or a short-time Fourier transform process on the blood flow data in which the intensity in a predetermined frequency band is relatively emphasized compared to other frequency bands, and a determination unit that determines the sleep stage of the subject based on the processed data.

[0103] Furthermore, a sleep estimation method according to one embodiment of the present disclosure includes a first acquisition step of acquiring blood flow data indicating blood flow in a subject, a generation step of generating a frequency spectrum of the blood flow data by performing a frequency analysis process on the blood flow data, and a first determination step of determining the sleep stage of the subject based on the frequency spectrum.

[0104] Furthermore, a sleep estimation method according to one aspect of the present disclosure includes an acquisition step of acquiring blood flow data indicating blood flow in a subject, a generation step of generating processed data indicating the results of a time-frequency analysis process of the blood flow data by performing a wavelet transform process or a short-time Fourier transform process on the blood flow data, in which the intensity in a predetermined frequency band is relatively emphasized compared to other frequency bands, and a determination step of determining the sleep stage of the subject based on the processed data.

[0105] Furthermore, a sleep estimation system according to one embodiment of the present disclosure includes a first acquisition unit that acquires blood flow data indicating blood flow in a subject, a generation unit that generates a frequency spectrum of the blood flow data by performing a frequency analysis process on the blood flow data, and a first judgment unit that judges the sleep stage of the subject based on the frequency spectrum.When the sleep stages in non-REM sleep are classified as stages 1, 2, and 3, in order from lightest to lightest sleep, the first judgment unit judges that the sleep stage of the subject has transitioned from stage 1 to stage 2 or 3 when a frequency spectrum is generated in which a first intensity, which is the maximum intensity in a first range that is a part of a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less, is greater than a second intensity in a second range that is a part of the region other than the first range.

[0106] Furthermore, a sleep estimation system according to one embodiment of the present disclosure may include a first notification unit that executes a first notification process when the first determination unit determines that a transition has occurred from stage 1 to stage 2 or 3, or after a predetermined time has elapsed since it was determined that a transition has occurred from stage 1 to stage 2 or 3.

[0107] Furthermore, in a sleep estimation system according to one aspect of the present disclosure, when the first determination unit determines that the subject's sleep stage has transitioned from stage 1 to stage 2 or 3, the frequency spectrum generated by the generation unit may include a broad, upwardly convex waveform centered on the first intensity in the frequency band.

[0108] Furthermore, a sleep estimation system according to one embodiment of the present disclosure includes a first acquisition unit that acquires blood flow data indicating blood flow in a subject, a generation unit that generates a frequency spectrum of the blood flow data by performing a frequency analysis process on the blood flow data, and a first determination unit that determines the sleep stage of the subject based on the frequency spectrum. When the sleep stages in non-REM sleep are classified as stages 1, 2, and 3, in order from lightest to lightest sleep, the generation unit performs a time-frequency analysis process as the frequency analysis process to generate intensity change data as the frequency spectrum, indicating changes in intensity over time in each frequency band within a predetermined time. The first determination unit determines that the sleep stage of the subject has transitioned from stage 1 to stage 2 or 3 when the intensity change data including the frequency spectrum having the predetermined shape generated by the generation unit is given to a trained model trained using, as training data, intensity change data including a frequency spectrum having the predetermined shape in a frequency band of 0.2 Hz or more and 0.3 Hz or less, which has been identified as corresponding to stage 2 or 3 based on biological information other than the blood flow data.

[0109] Furthermore, a sleep estimation system according to one embodiment of the present disclosure includes a first acquisition unit that acquires blood flow data indicating blood flow in a subject; a generation unit that generates a frequency spectrum of the blood flow data by performing a frequency analysis process on the blood flow data; a first determination unit that, when sleep stages in non-REM sleep are stages 1, 2, and 3, in order from lightest to lightest sleep, determines that the subject's sleep stage has transitioned from stage 1 to stage 2 or 3 if it determines that the frequency spectrum has a predetermined shape in a frequency band of 0.2 Hz or more and 0.3 Hz or less; a second acquisition unit that acquires acceleration data indicating acceleration caused by movement of the subject; and a second determination unit that determines whether the subject is stationary based on the acceleration data, and the generation unit performs the frequency analysis process if the second determination unit determines that the subject is stationary.

[0110] Furthermore, a sleep estimation system according to an aspect of the present disclosure includes an acquisition unit that acquires blood flow data indicating blood flow in a subject, a generation unit that generates processed data indicating a result of time-frequency analysis processing of the blood flow data by executing processing on the blood flow data to relatively emphasize intensity in a frequency band equal to or greater than 0.2 Hz and equal to or less than 0.3 Hz compared to other frequency bands, and a determination unit that determines a sleep stage of the subject based on the processed data, wherein the generation unit generates intensity change data indicating a change over time in intensity in each frequency band within a predetermined time as the processed data, and determines a sleep stage of the subject. When the sleep stages in a sleep pattern are defined as stages 1, 2, and 3, in order from lightest to lightest, the determination unit determines that the sleep stage of the subject has transitioned from stage 1 to stage 2 or 3 when the determination unit provides the intensity change data having the predetermined shape generated by the generation unit to a trained model that has been trained using, as training data, intensity change data including a frequency spectrum having a predetermined shape in a frequency band of 0.2 Hz or more and 0.3 Hz or less, which has been identified as corresponding to stage 2 or 3 based on biological information different from the blood flow data.

[0111] Furthermore, in the sleep estimation system according to an aspect of the present disclosure, the generation unit may generate the processed data by performing wavelet transform processing or short-time Fourier transform processing as the enhancement processing.

[0112] Furthermore, in the sleep estimation system according to the aspect of the present disclosure, the biological information may be electroencephalogram data detected by an electroencephalograph.

[0113] Furthermore, in the sleep estimation system according to the aspect of the present disclosure, the predetermined shape may be a broad waveform that is convex upward.

[0114] Furthermore, the sleep estimation system according to one aspect of the present disclosure may include a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating light onto the blood vessels of the subject.

[0115] Furthermore, the sleep estimation system according to one aspect of the present disclosure may include a wearable device having a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating light onto the blood vessels of the subject.

[0116] Furthermore, a sleep estimation method according to one embodiment of the present disclosure includes a first acquisition step of acquiring blood flow data indicating blood flow in a subject, a generation step of generating a frequency spectrum of the blood flow data by performing a frequency analysis process on the blood flow data, and a first determination step of determining the sleep stage of the subject based on the frequency spectrum.When the sleep stages in non-REM sleep are defined as stages 1, 2, and 3, in order from lightest to lightest sleep, in the first determination step, if a frequency spectrum is generated as the frequency spectrum, in which a first intensity, which is the maximum intensity in a first range that is a part of a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less, is greater than a second intensity in a second range that is a part of the region other than the first range, it is determined that the sleep stage of the subject has transitioned from stage 1 to stage 2 or 3.

[0117] Furthermore, a sleep estimation method according to one embodiment of the present disclosure includes an acquisition step of acquiring blood flow data indicating blood flow in a subject; a generation step of generating processed data indicating the results of a time-frequency analysis process of the blood flow data by performing a process on the blood flow data to relatively emphasize the intensity in a frequency band of 0.2 Hz or more and 0.3 Hz or less compared to other frequency bands; and a determination step of determining the sleep stage of the subject based on the processed data. In the generation step, intensity change data indicating the change in intensity over time in each frequency band within a predetermined time is generated as the processed data. When the sleep stages of non-REM sleep are classified as stages 1, 2, and 3, in order from lightest to lightest sleep, in the determination step, when the intensity change data having the predetermined shape generated in the generation step is given to a trained model trained using as training data the intensity change data including a frequency spectrum having a predetermined shape in the frequency band of 0.2 to 0.3 Hz, which has been identified as corresponding to stage 2 or 3 based on biological information other than the blood flow data, it is determined that the sleep stage of the subject has transitioned from stage 1 to stage 2 or 3.

[0118] Furthermore, a sleep estimation system according to one embodiment of the present disclosure includes a determination unit that determines the sleep stage of the subject based on the frequency spectrum of the subject's blood flow data. When the sleep stages of non-REM sleep are classified as stages 1, 2, and 3, in order from lightest to lightest, the determination unit determines that the subject's sleep stage has transitioned from stage 1 to stage 2 or 3 if the frequency spectrum contains a predetermined shape in at least a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less, and the predetermined shape contained in the region is an upwardly convex waveform shape that includes a first intensity, which is the maximum intensity in the region.

[0119] Furthermore, in a sleep estimation system according to one aspect of the present disclosure, the determination unit may determine that the frequency spectrum includes the specified shape when the first intensity in a first range that is a part of the frequency band is greater than the second intensity in a second range other than the first range.

[0120] Furthermore, a sleep estimation system according to one embodiment of the present disclosure may include a first notification unit that executes a first notification process when the determination unit determines that a transition has occurred from stage 1 to stage 2 or 3, or after a predetermined time has elapsed since it was determined that a transition has occurred from stage 1 to stage 2 or 3.

[0121] Furthermore, the sleep estimation system according to one aspect of the present disclosure may include a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating light onto the blood vessels of the subject.

[0122] Furthermore, a sleep estimation system according to one aspect of the present disclosure may include a wearable device having a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating light onto the subject's blood vessels.

[0123] Furthermore, a sleep estimation system according to one embodiment of the present disclosure includes a determination unit that determines the sleep stage of a subject based on the frequency spectrum of the subject's blood flow data. When the sleep stages of non-REM sleep are defined as stages 1, 2, and 3, in order from lightest to lightest, the determination unit determines whether the subject's sleep stage has transitioned from stage 1 to stage 2 or 3 when the frequency spectrum of the blood flow data is applied to a trained model that has been trained using, as training data, a frequency spectrum having a predetermined shape in a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less, which has been identified as corresponding to stage 2 or 3 based on biological information other than the blood flow data. The predetermined shape included in the region is an upwardly convex waveform shape that includes a first intensity, which is the maximum intensity in the region, and the biological information is electroencephalogram data detected by an electroencephalograph.

[0124] Furthermore, a sleep estimation system according to one embodiment of the present disclosure may include a first notification unit that executes a first notification process when the determination unit determines that a transition has occurred from stage 1 to stage 2 or 3, or after a predetermined time has elapsed since it was determined that a transition has occurred from stage 1 to stage 2 or 3.

[0125] Furthermore, the sleep estimation system according to one aspect of the present disclosure may include a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating light onto the blood vessels of the subject.

[0126] Furthermore, a sleep estimation system according to one aspect of the present disclosure may include a wearable device having a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating light onto the subject's blood vessels.

[0127] Furthermore, a sleep estimation method according to one embodiment of the present disclosure is a sleep estimation method executed in a sleep estimation system, wherein a judgment unit included in the sleep estimation system includes a judgment step of judging the sleep stage of the subject based on the frequency spectrum of the subject's blood flow data, and when the sleep stages in non-REM sleep are classified as stages 1, 2, and 3, in order from lightest to lightest, the judgment step judges that the subject's sleep stage has transitioned from stage 1 to stage 2 or 3 if the frequency spectrum includes a predetermined shape in at least a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less, and the predetermined shape included in the region is an upwardly convex waveform shape that includes a first intensity, which is the maximum intensity in the region.

[0128] Furthermore, a sleep estimation method according to one embodiment of the present disclosure is a sleep estimation method executed in a sleep estimation system, in which a judgment unit included in the sleep estimation system includes a judgment step of judging the sleep stage of the subject based on the frequency spectrum of the subject's blood flow data, and when the sleep stages of non-REM sleep are stages 1, 2, and 3, in order from lightest to lightest, the judgment step determines whether the subject's sleep stage has transitioned from stage 1 to stage 2 or 3 when the frequency spectrum of the blood flow data is applied to a trained model trained using, as training data, a frequency spectrum having a predetermined shape in a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less, which has been identified as corresponding to stage 2 or 3 based on biological information other than the blood flow data, and the predetermined shape included in the region is an upwardly convex waveform shape including a first intensity, which is the maximum intensity in the region, and the biological information may be electroencephalogram data detected by an electroencephalograph. [Explanation of symbols]

[0129] 1. 1A Sleep Estimation System 3 Blood flow meter 11 Second acquisition part 12 Second Judgment Section 13 1st Acquisition Department (Acquisition Department) 14 Generation part 15 1st judgment section (judgment section) 20 Wearable Devices 51 Sleep estimation device 161 First Notification Department 162 2nd Notification Department

Claims

1. a determination unit that determines a sleep stage of a subject based on a frequency spectrum of blood flow data of the subject; a notification unit that executes notification processing based on the determination result of the determination unit, the notifying unit issues a notification to the subject to encourage awakening when the determining unit determines that the frequency spectrum includes a predetermined shape in at least a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less, or after a predetermined time has elapsed since the determination, A sleep estimation system, wherein the predetermined shape included in the region is an upwardly convex waveform shape that includes a first intensity that is the maximum intensity in the region.

2. 2. The sleep estimation system of claim 1, wherein the determination unit determines that the frequency spectrum includes the predetermined shape when the first intensity in a first range that is a part of the frequency band is greater than the second intensity in a second range other than the first range.

3. 3. The sleep estimation system according to claim 1, further comprising a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating blood vessels of the subject with light.

4. The sleep estimation system according to claim 1 or 2, further comprising a wearable device including a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating light onto the blood vessels of the subject.

5. a determination unit that determines a sleep stage of a subject based on a frequency spectrum of blood flow data of the subject; and a notification unit that executes notification processing based on a determination result of the determination unit, the determination unit determines whether a frequency spectrum of the blood flow data corresponds to a correct answer when applied to a trained model trained using training data in which a frequency spectrum having a predetermined shape in a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less is associated with a correct answer label based on biological information different from the blood flow data; the notification unit issues a notification to the subject to encourage awakening when the determination unit determines that the answer is correct or after a predetermined time has elapsed since the determination unit determined that the answer is correct, the predetermined shape included in the region is an upwardly convex wave shape that includes a first intensity that is a maximum intensity in the region, A sleep estimation system, wherein the biological information is electroencephalogram data detected by an electroencephalograph.

6. The sleep estimation system according to claim 5 , further comprising a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating blood vessels of the subject with light.

7. The sleep estimation system according to claim 5 , further comprising a wearable device including a blood flow meter that detects the blood flow data by receiving scattered light generated by irradiating light onto the blood vessels of the subject.

8. A sleep estimation method executed in a sleep estimation system, comprising: a determination step in which a determination unit included in the sleep estimation system determines a sleep stage of the subject based on a frequency spectrum of blood flow data of the subject; a notification step in which a notification unit included in the sleep estimation system notifies the subject to wake up when it is determined in the determination step that the frequency spectrum includes a predetermined shape in an area including at least a frequency band of 0.2 Hz or more and 0.3 Hz or less, or after a predetermined time has elapsed since the determination, A sleep estimation method, wherein the predetermined shape included in the region is an upwardly convex waveform shape that includes a first intensity that is the maximum intensity in the region.

9. A sleep estimation method executed in a sleep estimation system, comprising: a determination step in which a determination unit included in the sleep estimation system determines a sleep stage of the subject based on a frequency spectrum of blood flow data of the subject; a notification step in which a notification unit included in the sleep estimation system executes a notification process based on a determination result in the determination step, In the determination step, when the frequency spectrum of the blood flow data is applied to a trained model trained using training data in which a frequency spectrum having a predetermined shape in a region including a frequency band of 0.2 Hz or more and 0.3 Hz or less is associated with a correct answer label based on biological information different from the blood flow data, it is determined whether the frequency spectrum corresponds to a correct answer; In the notification step, when the determination step determines that the answer is correct, or after a predetermined time has elapsed since the determination step, a notification is given to the subject to encourage them to wake up; the predetermined shape included in the region is an upwardly convex wave shape that includes a first intensity that is a maximum intensity in the region, A sleep estimation method, wherein the biological information is electroencephalogram data detected by an electroencephalograph.

10. A program for causing a computer to function as the sleep estimation system according to claim 1 , the program causing the computer to function as the determination unit and the notification unit.

11. A program for causing a computer to function as the sleep estimation system according to claim 5, the program causing the computer to function as the determination unit and the notification unit.

Images