Information processing device, sleep-awakening stage determination device, computer program, information processing method, and sleep-awakening stage determination method

Abstract

The present invention is provided with: an electroencephalogram processing unit that acquires the amplitude and phase of each of a plurality of prescribed frequency bands from training electroencephalogram data measured from a single living body of a mammalian type, and performs complexity analysis processing including entropy analysis on the amplitude and phase of each of the acquired frequency bands; and a machine learning unit that executes a machine learning process of a machine learning model for determining a sleep-awakening stage, using a result of the complexity analysis process and label data of the sleep-awakening stage corresponding to the training electroencephalogram data.

Description

Technical Field

[0001] This invention relates to an information processing device, a sleep-wake stage determination device, a computer program, an information processing method, and a sleep-wake stage determination method.

[0002] This application claims priority based on Japanese Patent Application No. 2023-111974, filed on July 7, 2023, the contents of which are incorporated herein by reference. Background Technology

[0003] Previously, sleep research has used mice to determine sleep-wake stages (wakefulness, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep). Regarding the automatic determination of sleep-wake stages, techniques for automatically determining sleep-wake stages based on mouse electroencephalogram (EEG) and electromyography (EMG) data are known (see, for example, Non-Patent Literature 1). Previous automatic sleep-wake stage determination techniques were based on the spectrum of EEG data and the amplitude of EMG data.

[0004] According to Non-Patent Document 2, the entropy obtained by decomposing the sample entropy (SampEn) of EEG into frequency band / phase / amplitude, i.e., the combination of decomposed entropy and deep learning, is useful for the diagnosis of Alzheimer's disease.

[0005] According to Non-Patent Literature 3, the combination of decomposition entropy obtained from EEG's expanded sample entropy (expSampEn) and deep learning is useful for neural decoding (such as predicting what a person is thinking). The expanded sample entropy is calculated using the autocorrelation of all time points in a time series with respect to a single point.

[0006] According to Non-Patent Literature 4, the temporal extended sample entropy of nighttime body movements (trunk acceleration) is useful for the diagnosis of autism spectrum disorder. Non-Patent Literature 4 does not utilize decomposed entropy or machine learning / deep learning.

[0007] [Background Technical Documents] Non-patent literature Non-patent literature 1: Masato Yamabe et al., "MC-SleepNet: Large-scale Sleep StageScoring in Mice by Deep Neural Networks", SCIENTIFIC REPORTS, 9:15793, 2019 Non-Patent Literature 2: Naoki Furutani et al., "Decomposed Temporal Complexity Analysis of Neural Oscillations and Machine Learning Applied to Alzheimer's Disease Diagnosis", Front. Psychiatry, 11:531801, September 3, 2020. Non-patent literature 3: Naoki Furutani et al., "Neural Decoding of Multi-Modal Imagery Behavior Focusing on Temporal Complexity", Front. Psychiatry, 11:746, July 30, 2020. Non-patent literature 4: Naoki Furutani et al., "Complexity of Body Movements during Sleep in Children with Autism Spectrum Disorder", Entropy, 23(4):418, March 31, 2021 Non-patent literature 5: Dorde Miladinovic et al., “SPINDLE: End-to-end learning from EEG / EMG to extrapolate animal sleep scoring across experimental settings, labs and species,” PLOS Computational Biology, 15(4):e1006968, April 18, 2019 Summary of the Invention [The problem the invention aims to solve] However, in the aforementioned conventional automatic sleep-wake stage determination technologies, the accuracy of the determination is difficult to improve. Furthermore, non-patent documents 2, 3, and 4 provide no explanation regarding the determination of the sleep-wake stage.

[0008] This invention was made in consideration of this situation, and its purpose is to provide a technique that helps improve the accuracy of automatic determination of the sleep-wake phase.

[0009] [Technical means to solve the problem] One aspect of the present invention is an information processing device comprising: an electroencephalogram (EEG) processing unit that acquires the amplitude and phase of each of a predetermined plurality of frequency bands from training EEG data measured from a live mammal, and performs complexity analysis processing, including entropy analysis, on the acquired amplitude and phase of each frequency band; and a machine learning unit that uses the results of the complexity analysis processing and label data of sleep-wake stages corresponding to the training EEG data to perform machine learning processing for determining a machine learning model of sleep-wake stages.

[0010] One aspect of the present invention is an information processing apparatus in which the EEG processing unit further extracts signals from the training EEG data from the plurality of frequency bands respectively, performs complexity analysis processing including entropy analysis on the extracted signals of each frequency band respectively, and the machine learning unit further uses the result of the complexity analysis processing to perform machine learning processing of the machine learning model.

[0011] One aspect of the present invention is an information processing apparatus in which the entropy analysis of the phase is performed by performing entropy analysis on the cosine of the phase to calculate the entropy of the cosine of the phase.

[0012] One aspect of the present invention is an information processing device in which, regarding entropy analysis, the electroencephalogram (EEG) processing unit uses time-extended sample entropy.

[0013] One aspect of the present invention is an information processing apparatus in which the machine learning unit expands the learning data by standardization or normalization based on multiple thresholds, or both standardization and normalization based on multiple thresholds, when generating learning data for use by a machine learning model.

[0014] One aspect of the present invention is an information processing device in which the living organism is a mouse.

[0015] One aspect of the present invention is a sleep-wake stage determination device, comprising: an electroencephalogram (EEG) processing unit that acquires the amplitude and phase of each of a predetermined plurality of frequency bands from EEG data measured from a live mammal, and performs complexity analysis processing including entropy analysis on the acquired amplitude and phase of each frequency band; and a determination unit that uses a determination model to determine the sleep-wake stage based on the result of the complexity analysis processing.

[0016] One aspect of the present invention is a sleep-wake stage determination device, wherein the EEG processing unit further extracts signals of the plurality of frequency bands from the EEG data, performs complexity analysis processing including entropy analysis on the extracted signals of each frequency band, and the determination unit further uses the result of the complexity analysis processing to determine the sleep-wake stage.

[0017] One aspect of the present invention is a sleep-wake stage determination device, in which the entropy analysis of the phase is performed by performing entropy analysis on the cosine of the phase to calculate the entropy of the cosine of the phase.

[0018] One aspect of the present invention is a sleep-wake stage determination device, wherein, regarding entropy analysis, the EEG processing unit uses time-extended sample entropy.

[0019] One aspect of the present invention is a sleep-wake stage determination device, which further includes an RSE method execution unit. The RSE method execution unit applies the RSE (Rescoring Short Epoch) method to the determination results of sleep-wake stages determined using a determination model. It removes epochs whose stage duration is determined to be shorter than a certain time from the stage duration epochs, starting from the epochs with shorter stage durations.

[0020] One aspect of the present invention is a sleep-wake stage determination device, which further includes a skipping unit that prevents the display of rounds with a confidence level of more than a certain degree in the determination results of the sleep-wake stage determined by using the determination model.

[0021] One aspect of the present invention is a sleep-wake stage determination device, which further includes a confidence display control unit that displays the confidence level of the determination relative to the determination result of the sleep-wake stage determined using a determination model.

[0022] One aspect of the present invention is a sleep-wake stage determination device, wherein the living organism is a mouse.

[0023] One aspect of the present invention is a computer program for causing a computer to perform: an electroencephalogram (EEG) processing step, acquiring the amplitude and phase of each of a specified plurality of frequency bands from training EEG data measured from a live mammal, and performing complexity analysis processing, including entropy analysis, on the acquired amplitude and phase of each frequency band; and a machine learning step, using the results of the complexity analysis processing and label data of sleep-wake stages corresponding to the training EEG data, performing machine learning processing for determining a machine learning model of sleep-wake stages.

[0024] One aspect of the present invention is a computer program for causing a computer to perform: an electroencephalogram (EEG) processing step, acquiring the amplitude and phase of each of a specified plurality of frequency bands from EEG data measured from a live mammal, and performing complexity analysis processing, including entropy analysis, on the acquired amplitude and phase of each frequency band; and a determination step, using a determination model, determining the sleep-wake stage based on the results of the complexity analysis processing.

[0025] One aspect of the present invention is an information processing method executed by an information processing device, comprising: an electroencephalogram (EEG) processing step, wherein the amplitude and phase of each of a plurality of frequency bands are obtained from training EEG data measured from a live mammal, and a complexity analysis process including entropy analysis is performed on the amplitude and phase of each obtained frequency band; and a machine learning step, wherein the results of the complexity analysis process and the label data of sleep-wake stages corresponding to the training EEG data are used to perform machine learning processing for determining a machine learning model for sleep-wake stages.

[0026] One aspect of the present invention is a method for determining sleep-wake stages, which is executed by a sleep-wake stage determination device and includes: an electroencephalogram (EEG) processing step, which involves acquiring the amplitude and phase of each of a specified plurality of frequency bands from EEG data measured from a live mammal, and performing complexity analysis processing, including entropy analysis, on the acquired amplitude and phase of each frequency band; and a determination step, which uses a determination model to determine the sleep-wake stage based on the result of the complexity analysis processing.

[0027] [The effects of the invention] According to the present invention, a technology that can help improve the accuracy of automatic determination of sleep-wake phases will be obtained. Attached Figure Description

[0028] Figure 1 This is a block diagram illustrating a configuration example of the information processing apparatus according to the first embodiment.

[0029] Figure 2 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the first embodiment.

[0030] Figure 3 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to the second embodiment.

[0031] Figure 4 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the second embodiment.

[0032] Figure 5 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to the third embodiment.

[0033] Figure 6 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the third embodiment.

[0034] Figure 7 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to the fourth embodiment.

[0035] Figure 8 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the fourth embodiment.

[0036] Figure 9 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to the fifth embodiment.

[0037] Figure 10 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the fifth embodiment.

[0038] Figure 11 This is a chart representing the evaluation results of the present invention.

[0039] Figure 12 This is a chart representing the evaluation results of the present invention.

[0040] Figure 13 This is a chart representing the evaluation results of the present invention. Detailed Implementation

[0041] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In this embodiment, the sleep-wake phase is determined in a live mammal. Examples of mammals include humans, apes, and mice. Hereinafter, a mouse will be used as an example of a mammal for explanation.

[0042] [First Implementation] Reference Figure 1 and Figure 2 The first embodiment will be described below. Figure 1 This is a block diagram illustrating a configuration example of the information processing apparatus according to the first embodiment. Figure 2 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the first embodiment.

[0043] Figure 1 The information processing device 1a shown generates data by performing machine learning on the machine learning model 210 to determine the sleep-wake stages (wake, non-rapid eye movement sleep, rapid eye movement sleep). Figure 2 The sleep-wake stage determination device 2a shown uses the determination model 210a. The machine learning algorithm is not particularly limited. For example, deep learning or LightGBM (Light Gradient Boosting Machine) can be used. Figure 2 The sleep-wake stage determination device 2a shown uses a determination model 210a generated by the information processing device 1a to determine the sleep-wake stage.

[0044] [Information processing device] Reference Figure 1 The information processing apparatus 1a of the first embodiment will now be described. Training EEG data A is input into the information processing apparatus 1a. Training EEG data A is time-series data of EEG measured from a single mouse (live animal), which is used by the machine learning model 210 for machine learning. The machine learning model 210 can use multiple training EEG data A. For example, multiple EEG data measured from multiple mice (live animals) can be used as training EEG data A respectively.

[0045] The information processing device 1a includes an EEG processing unit 100 and a machine learning unit 200. The EEG processing unit 100 includes a bandpass filter 110, a Hilbert transform unit 120, and a complexity analysis unit 130. The complexity analysis unit 130 includes an entropy analysis unit 131 and a fractal analysis unit 132.

[0046] The EEG processing unit 100 divides the training EEG data A into fixed intervals (epochs) according to the time sequence, and performs processing on each epoch. As an example of this embodiment, the epoch is 25 seconds.

[0047] The bandpass filter 110 extracts signals from a specified number of frequency bands from one round of training EEG data A. In this embodiment, as an example of multiple frequency bands, the bandpass filter 110 extracts signals from five frequency bands—δ, θ, α, β, and γ—from one round of training EEG data A (δ band signal B1, θ band signal B2, α band signal B3, β band signal B4, and γ band signal B5). Therefore, the bandpass filter 110 is composed of a bandpass filter supporting each of the δ, θ, α, β, and γ bands.

[0048] The Hilbert transform unit 120 performs a Hilbert transform on each of the delta-band signal B1, the θ-band signal B2, the α-band signal B3, the β-band signal B4, and the γ-band signal B5. Through this Hilbert transform, the Hilbert transform unit 120 obtains the amplitude "A(t)" (δ-band signal amplitude Ca1, θ-band signal amplitude Ca2, α-band signal amplitude Ca3, β-band signal amplitude Ca4, and γ-band signal amplitude Ca5) and phase "θ(t)" (δ-band signal phase Cp1, θ-band signal phase Cp2, α-band signal phase Cp3, β-band signal phase Cp4, and γ-band signal phase Cp5) of each of the delta-band signal B1, θ-band signal B2, α-band signal B3, β-band signal phase Cp4, and γ-band signal phase Cp5).

[0049] The complexity analysis unit 130 performs complexity analysis processing on each of the training EEG data A, and the amplitudes of the delta band signal Ca1, the theta band signal Ca2, the alpha band signal Ca3, the beta band signal Ca4, and the γ band signal Ca5. The complexity analysis processing includes entropy analysis performed by the entropy analysis unit 131 and fractal analysis performed by the fractal analysis unit 132.

[0050] The complexity analysis unit 130 performs complexity analysis processing on each of the δ-band signal B1, θ-band signal B2, α-band signal B3, β-band signal B4 and γ-band signal B5, and on each of the δ-band signal phase Cp1, θ-band signal phase Cp2, α-band signal phase Cp3, β-band signal phase Cp4 and γ-band signal phase Cp5. The complexity analysis processing includes entropy analysis performed by the entropy analysis unit 131.

[0051] Entropy analysis is the process of finding the value (hereinafter referred to as entropy) that represents the degree of irregularity in a time series of an object.

[0052] Fractal analysis is used to determine the degree of self-similarity of a time series represented by an object subjected to fractal analysis (hereinafter referred to as fractal).

[0053] The entropy analysis unit 131 performs entropy analysis on the training EEG data A. Through this entropy analysis, the entropy analysis unit 131 obtains the entropy (EEG entropy DA) of the training EEG data A.

[0054] The entropy analysis unit 131 performs entropy analysis on each of the δ-band signal B1, θ-band signal B2, α-band signal B3, β-band signal B4, and γ-band signal B5. Through this entropy analysis, the entropy analysis unit 131 obtains the entropy of each of the δ-band signal B1, θ-band signal B2, α-band signal B3, β-band signal B4, and γ-band signal B5 (δ-band signal entropy DB1, θ-band signal entropy DB2, α-band signal entropy DB3, β-band signal entropy DB4, and γ-band signal entropy DB5).

[0055] The entropy analysis unit 131 performs entropy analysis on each of the following signal amplitudes: δ-band signal amplitude Ca1, θ-band signal amplitude Ca2, α-band signal amplitude Ca3, β-band signal amplitude Ca4, and γ-band signal amplitude Ca5. Through this entropy analysis, the entropy analysis unit 131 obtains the entropy of each of the following signal amplitudes: δ-band signal amplitude modulation entropy Da1, θ-band signal amplitude modulation entropy Da2, α-band signal amplitude modulation entropy Da3, β-band signal amplitude modulation entropy Da4, and γ-band signal amplitude modulation entropy Da5.

[0056] The entropy analysis unit 131 performs entropy analysis on each of the δ-band signal phase Cp1, θ-band signal phase Cp2, α-band signal phase Cp3, β-band signal phase Cp4, and γ-band signal phase Cp5. Through this entropy analysis, the entropy analysis unit 131 obtains the entropy of each of the δ-band signal phase Cp1, θ-band signal phase Cp2, α-band signal phase Cp3, β-band signal phase Cp4, and γ-band signal phase Cp5 (δ-band signal phase modulation entropy Dp1, θ-band signal phase modulation entropy Dp2, α-band signal phase modulation entropy Dp3, β-band signal phase modulation entropy Dp4, and γ-band signal phase modulation entropy Dp5).

[0057] Furthermore, when the entropy analysis unit 131 performs entropy analysis on the phases (δ-band signal phase Cp1, θ-band signal phase Cp2, α-band signal phase Cp3, β-band signal phase Cp4, and γ-band signal phase Cp5), it performs entropy analysis on the cosine of the phase "θ(t)" "cosθ(t)". ​​This is because if the entropy analysis is performed directly on the phase "θ(t), discontinuities will appear at every 2π [rad]. By using the cosine of the phase "θ(t)" "cosθ(t)," the phase "θ(t)" can be treated as a continuous value. Therefore, the entropy analysis of the phase "θ(t)" is performed by calculating the entropy of the cosine of the phase "θ(t)" "cosθ(t)" through entropy analysis.

[0058] Furthermore, the entropy analysis of phase “θ(t)” is not limited to calculating the entropy of the cosine “cosθ(t)” of phase “θ(t)” by performing entropy analysis on the cosine “cosθ(t)” of phase “θ(t)”. Entropy analysis of phase “θ(t)” can also be performed by other methods.

[0059] The fractal analysis unit 132 performs fractal analysis on the training EEG data A. Through this fractal analysis, the fractal analysis unit 132 obtains the fractal of the training EEG data A (EEG fractal EA).

[0060] The fractal analysis unit 132 performs fractal analysis on each of the δ-band signal amplitude Ca1, θ-band signal amplitude Ca2, α-band signal amplitude Ca3, β-band signal amplitude Ca4, and γ-band signal amplitude Ca5. Through this fractal analysis, the fractal analysis unit 132 obtains the fractals of each of the δ-band signal amplitude Ca1, θ-band signal amplitude Ca2, α-band signal amplitude Ca3, β-band signal amplitude Ca4, and γ-band signal amplitude Ca5 (δ-band signal amplitude fractal Ea1, θ-band signal amplitude fractal Ea2, α-band signal amplitude fractal Ea3, β-band signal amplitude fractal Ea4, and γ-band signal amplitude fractal Ea5).

[0061] The results obtained from the complexity analysis processing by the EEG processing unit 100 (EEG entropy DA, delta band signal entropy DB1, θ band signal entropy DB2, α band signal entropy DB3, β band signal entropy DB4 and γ band signal entropy DB5, delta band signal amplitude modulation entropy Da1, θ band signal amplitude modulation entropy Da2, α band signal amplitude modulation entropy Da3, β band signal amplitude modulation entropy Da4 and γ band signal amplitude modulation entropy Da5, delta band signal phase modulation entropy Dp1, θ band signal phase modulation entropy Dp2, α band signal phase modulation entropy Dp3, β band signal phase modulation entropy Dp4 and γ band signal phase modulation entropy Dp5, EEG fractal EA, delta band signal amplitude fractal Ea1, θ band signal amplitude fractal Ea2, α band signal amplitude fractal Ea3, β band signal amplitude fractal Ea4 and γ band signal amplitude fractal Ea5) are sent to the machine learning unit 200.

[0062] The machine learning unit 200 uses the results of complexity analysis processing received from the EEG processing unit 100, as well as the label data Fa, to perform machine learning processing on the machine learning model 210 for determining the sleep-wake stage (wake, non-rapid eye movement sleep, rapid eye movement sleep).

[0063] Furthermore, in the machine learning processing of the machine learning model 210, the machine learning unit 200 can use all the results of the complexity analysis processing received from the EEG processing unit 100, or it can select a portion of the results of the complexity analysis processing received from the EEG processing unit 100.

[0064] The labeled data Fa is pre-generated for each round of the training EEG data A. The labeled data Fa has the correct labels, which represent the sleep-wake stage corresponding to each round of the training EEG data A: "wake," "non-rapid eye movement sleep," or "rapid eye movement sleep."

[0065] For example, the machine learning department 200 divides the results of complexity analysis into three types: training data, validation data, and test data. The machine learning model 210, which has been trained and validated using the training data, validation data, and corresponding label data Fa, uses the test data and corresponding label data Fa to confirm the accuracy of the judgment.

[0066] The machine learning model 210, which has undergone machine learning through the machine learning unit 200, is used as the determination model 210a in the sleep-wake stage determination device 2a.

[0067] [Sleep-wake stage determination device] Reference Figure 2 The sleep-wake stage determination device 2a of the first embodiment will be described. EEG data G is input into the sleep-wake stage determination device 2a. The EEG data G is time-series data of the EEG measured from a mouse (live animal), and is the object of sleep-wake stage determination.

[0068] The sleep-wake stage determination device 2a includes an electroencephalogram (EEG) processing unit 100 and a determination unit 300. The EEG processing unit 100 and the... Figure 1 The information processing device 1a is the same as the EEG processing unit 100, and its related description is omitted.

[0069] In the sleep-wake stage determination device 2a, the EEG processing unit 100 processes the results of complex analysis derived from the EEG data G (EEG entropy DA, delta band signal entropy DB1, theta band signal entropy DB2, the alpha band signal entropy DB3, the beta band signal entropy DB4 and the gamma band signal entropy DB5, the delta band signal amplitude modulation entropy Da1, the theta band signal amplitude modulation entropy Da2, the alpha band signal amplitude modulation entropy Da3, the beta band signal amplitude modulation entropy Da4 and the gamma band signal entropy Da5). The amplitude modulation entropy Da5, the phase modulation entropy of the delta band signal Dp1, the phase modulation entropy of the θ band signal Dp2, the phase modulation entropy of the α band signal Dp3, the phase modulation entropy of the β band signal Dp4 and the phase modulation entropy of the γ band signal Dp5, the EEG fractal EA, the amplitude fractal of the delta band signal Ea1, the amplitude fractal of the θ band signal Ea2, the amplitude fractal of the α band signal Ea3, the amplitude fractal of the β band signal Ea4 and the amplitude fractal of the γ band signal Ea5) are sent to the determination unit 300.

[0070] The determination unit 300, based on the complexity analysis and processing results DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, and Ea1-Ea5 received from the EEG processing unit 100, uses the determination model 210a to determine the sleep-wake stage (wakefulness, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep). The determination unit 300 outputs the sleep-wake stage determination result H. The sleep-wake stage determination result H can be information determining which stage it belongs to among "wakefulness," "NREM sleep," or "REM sleep," or it can be the confidence level of each of "wakefulness," "NREM sleep," and "REM sleep."

[0071] According to this embodiment, using the results of complexity analysis of the amplitude and phase of each of the specified multiple frequency bands included in the EEG of a living mammal, and the corresponding labeled data of sleep-wake stages, machine learning processing is performed on a machine learning model for determining sleep-wake stages (wakefulness, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep). Here, in the EEG, each frequency band, each amplitude, and each phase has different physiological significance. Generally, the determination of sleep-wake stages is based on the amplitude of the delta (δ) or theta (θ) frequency bands, but differences between stages are also reported in the high-frequency bands. Therefore, by performing complexity analysis of the amplitude and phase of each of the specified multiple frequency bands included in the EEG, and using the results of this complexity analysis and the corresponding labeled data of sleep-wake stages, machine learning processing is performed on a machine learning model for determining sleep-wake stages (wakefulness, NREM sleep, and REM sleep), thus improving the performance of the determination model. This will help improve the accuracy of automatic detection of the sleep-wake phase.

[0072] In the example described, regarding complexity analysis processing, the EEG processing unit 100 can... Perform entropy analysis on the following elements respectively: • Training EEG Data A • Delta band signal B1, theta band signal B2, thea band signal B3, theta band signal B4, and the gamma band signal B5 • Amplitudes of the delta band signal Ca1, the theta band signal Ca2, the alpha band signal Ca3, the beta band signal Ca4, and the γ band signal Ca5 • The phase of the δ-band signal Cp1, the phase of the θ-band signal Cp2, the phase of the α-band signal Cp3, the phase of the β-band signal Cp4, and the phase of the γ-band signal Cp5; Perform fractal analysis on the following elements respectively: • Training EEG Data A • The amplitudes of the δ-band signal (Ca1), θ-band signal (Ca2), α-band signal (Ca3), β-band signal (Ca4), and γ-band signal (Ca5).

[0073] The Machine Learning Department 200 uses the results of this complexity analysis for machine learning.

[0074] In another example, regarding complexity analysis processing, the EEG processing unit 100 can also... Perform entropy analysis on the following elements respectively: • Amplitudes of the delta band signal Ca1, the theta band signal Ca2, the alpha band signal Ca3, the beta band signal Ca4, and the γ band signal Ca5 • The phase of the δ-band signal Cp1, the phase of the θ-band signal Cp2, the phase of the α-band signal Cp3, the phase of the β-band signal Cp4, and the phase of the γ-band signal Cp5; Perform fractal analysis on the following elements respectively: • The amplitudes of the δ-band signal (Ca1), θ-band signal (Ca2), α-band signal (Ca3), β-band signal (Ca4), and γ-band signal (Ca5).

[0075] The Machine Learning Department 200 uses the results of this complexity analysis for machine learning.

[0076] In another example, regarding complexity analysis processing, the EEG processing unit 100 can also... Perform entropy analysis on the following elements respectively: • Delta band signal B1, theta band signal B2, thea band signal B3, theta band signal B4, and the gamma band signal B5 • Amplitudes of the delta band signal Ca1, the theta band signal Ca2, the alpha band signal Ca3, the beta band signal Ca4, and the γ band signal Ca5 • The phase of the δ-band signal Cp1, the phase of the θ-band signal Cp2, the phase of the α-band signal Cp3, the phase of the β-band signal Cp4, and the phase of the γ-band signal Cp5; Perform fractal analysis on the following elements respectively: • The amplitudes of the δ-band signal (Ca1), θ-band signal (Ca2), α-band signal (Ca3), β-band signal (Ca4), and γ-band signal (Ca5).

[0077] The Machine Learning Department 200 uses the results of this complexity analysis for machine learning.

[0078] Furthermore, regarding complexity analysis processing, the EEG processing unit 100 can also perform at least entropy analysis, which is part of entropy analysis and fractal analysis. Therefore, regarding complexity analysis processing, the EEG processing unit 100 may not need to perform fractal analysis. In the case where fractal analysis is not performed regarding complexity analysis processing, only the result of entropy analysis is sent to the machine learning unit 200 for machine learning use.

[0079] [Second Implementation] Reference Figure 3 and Figure 4 The second embodiment will be described below. Figure 3 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to the second embodiment. Figure 4 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the second embodiment. Figure 3 , Figure 4 In the middle, to and Figure 1 , Figure 2 The corresponding parts of each part are marked with the same symbol, and their related explanations are omitted.

[0080] [Information processing device] exist Figure 3 In comparison, information processing device 1b Figure 1 The information processing device 1a also includes a frequency analysis unit 400. The frequency analysis unit 400 transforms the spectrum J for each round of the training EEG data A. The rounds in the EEG processing unit 100 and the frequency analysis unit 400 are common. The frequency analysis unit 400 sends the spectrum J to the machine learning unit 200.

[0081] The machine learning unit 200 uses the results of complexity analysis processing received from the EEG processing unit 100 (DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, Ea1-Ea5), the spectrum J received from the frequency analysis unit 400, and the tag data Fa to perform machine learning processing on the machine learning model 210 for determining the sleep-wake stage (wake, non-rapid eye movement sleep, rapid eye movement sleep).

[0082] Furthermore, in the machine learning processing of the machine learning model 210, the machine learning unit 200 can use both the results of the complexity analysis processing received from the EEG processing unit 100 and the spectrum received from the frequency analysis unit 400, or it can select a portion of the results of the complexity analysis processing received from the EEG processing unit 100 and the spectrum received from the frequency analysis unit 400.

[0083] The labeled data Fa is pre-generated for each round of the training EEG data A. The labeled data Fa has the correct labels, which represent the sleep-wake stage corresponding to each round of the training EEG data A: "wake," "non-rapid eye movement sleep," or "rapid eye movement sleep."

[0084] The machine learning model 210, which has undergone machine learning through the machine learning unit 200, is used as the determination model 210b in the sleep-wake stage determination device 2b.

[0085] [Sleep-wake stage determination device] exist Figure 4 In comparison, the sleep-wake stage determination device 2b Figure 2 The sleep-wake stage determination device 2a also includes a frequency analysis unit 400. The frequency analysis unit 400 transforms the spectrum J for each round of EEG data G. The rounds in the EEG processing unit 100 and the frequency analysis unit 400 are common. The frequency analysis unit 400 sends the spectrum J to the determination unit 300.

[0086] The determination unit 300 uses the determination model 210b to determine the sleep-wake stage (wake, non-rapid eye movement (NREM) sleep, and based on the complexity analysis processing results DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, Ea1-Ea5 received from the EEG processing unit 100, and the spectrum J received from the frequency analysis unit 400. The determination unit 300 outputs the sleep-wake stage determination result H. The sleep-wake stage determination result H can be information determining which stage of "wake," "NREM sleep," or "REM sleep" it belongs to, or it can be the confidence level for each of "wake," "NREM sleep," and "REM sleep."

[0087] According to this embodiment, compared to the first embodiment, the spectrum of electroencephalography (EEG) is used to perform machine learning processing on the machine learning model. As a result, the machine learning result of the model is obtained, indicating a further improvement in the model's performance.

[0088] [Third Implementation] Reference Figure 5 and Figure 6 The third embodiment will be described. Figure 5 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to the third embodiment. Figure 6 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the third embodiment. Figure 5 , Figure 6 In the middle, to and Figures 1-4 The corresponding parts of each part are marked with the same symbol, and their related explanations are omitted.

[0089] [Information processing device] exist Figure 5 In comparison, information processing device 1c Figure 3 The information processing device 1b also includes an amplitude analysis unit 500. The amplitude analysis unit 500 acquires the amplitude L for each round of training electromyography (EMG) data K. The training EMG data K is time-series data of EMG measured from a mouse (live animal) for which training EEG data A was previously measured, and is used for machine learning by the machine learning model 210. The rounds of the amplitude analysis unit 500 are common to those in the EEG processing unit 100 and the frequency analysis unit 400. The amplitude analysis unit 500 sends the amplitude L to the machine learning unit 200.

[0090] The machine learning unit 200 uses the results of complexity analysis processing received from the EEG processing unit 100 (DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, Ea1-Ea5), the spectrum J received from the frequency analysis unit 400, the amplitude L received from the amplitude analysis unit 500, and the label data Fc to perform machine learning processing on the machine learning model 210 for determining sleep-wake stages (wake, non-rapid eye movement sleep, rapid eye movement sleep).

[0091] Furthermore, in the machine learning processing of the machine learning model 210, the machine learning unit 200 can use all of the results of the complexity analysis processing received from the EEG processing unit 100, the spectrum received from the frequency analysis unit 400, and the amplitude received from the amplitude analysis unit 500, or it can select a portion of the results of the complexity analysis processing received from the EEG processing unit 100, the spectrum received from the frequency analysis unit 400, and the amplitude received from the amplitude analysis unit 500.

[0092] The labeled data Fc is pre-generated for each round common to both training EEG data A and training EMG data K. The labeled data Fc has the correct label, which represents the sleep-wake stage corresponding to each round common to both training EEG data A and training EMG data K: "wake," "non-rapid eye movement (NREM) sleep," or "rapid eye movement (REM) sleep."

[0093] The machine learning model 210, which has undergone machine learning through the machine learning unit 200, is used as the determination model 210c in the sleep-wake stage determination device 2c.

[0094] [Sleep-wake stage determination device] exist Figure 6 In comparison, the sleep-wake stage determination device 2c Figure 4The sleep-wake stage determination device 2b also includes an amplitude analysis unit 500. The amplitude analysis unit 500 acquires the amplitude L for each cycle of the electromyographic data P. The electromyographic data P is time-series data of electromyography measured from a mouse (live animal) whose electroencephalogram (EEG) data G was previously measured, and is the object of sleep-wake stage determination. The cycles of the amplitude analysis unit 500 are common to those in the EEG processing unit 100 and the frequency analysis unit 400. The amplitude analysis unit 500 sends the amplitude L to the determination unit 300.

[0095] The determination unit 300 determines the sleep-wake stage (wake, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep) based on the complexity analysis results DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, Ea1-Ea5 received from the EEG processing unit 100, the spectrum J received from the frequency analysis unit 400, and the amplitude L received from the amplitude analysis unit 500, using the determination model 210c. The determination unit 300 outputs the sleep-wake stage determination result H. The sleep-wake stage determination result H can be information determining which stage of "wake," "NREM sleep," or "REM sleep" it belongs to, or it can be the confidence level of each of "wake," "NREM sleep," and "REM sleep."

[0096] According to this embodiment, compared to the second embodiment, the amplitude of electromyography is used to perform machine learning processing on the machine learning model. As a result, the machine learning result of the model is obtained, indicating a further improvement in the model's performance.

[0097] [Fourth Implementation] Reference Figure 7 and Figure 8 The fourth embodiment will be described. Figure 7 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to the fourth embodiment. Figure 8 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the fourth embodiment. Figure 7 , Figure 8 In the middle, to and Figures 1-6 The corresponding parts of each part are marked with the same symbol, and their related explanations are omitted.

[0098] [Information processing device] exist Figure 7 In the middle, information processing device 1D comparison Figure 5The information processing device 1c also includes a complexity analysis unit 510. The complexity analysis unit 510 includes an entropy analysis unit 511 and a fractal analysis unit 512. For each round of training electromyographic data K, the complexity analysis unit 510 performs complexity analysis processing, which includes entropy analysis performed by the entropy analysis unit 511 and fractal analysis performed by the fractal analysis unit 512. The complexity analysis unit 510 shares the same rounds as those in the electroencephalogram processing unit 100, the frequency analysis unit 400, and the amplitude analysis unit 500.

[0099] The entropy analysis unit 511 performs entropy analysis on the training electromyography data K. Through this entropy analysis, the entropy analysis unit 511 obtains the entropy (electromyography entropy M) of the training electromyography data K.

[0100] The fractal analysis unit 512 performs fractal analysis on the training electromyography data K. Through this fractal analysis, the fractal analysis unit 512 obtains the fractal (electromyography fractal N) of the training electromyography data K.

[0101] The complexity analysis unit 510 sends the results of the complexity analysis processing (electromyographic entropy M, electromyographic fractal N) to the machine learning unit 200.

[0102] The machine learning unit 200 uses the results of complexity analysis processing received from the EEG processing unit 100 (DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, Ea1-Ea5), the spectrum J received from the frequency analysis unit 400, the amplitude L received from the amplitude analysis unit 500, the results of complexity analysis processing received from the complexity analysis unit 510 (M, N), and the label data Fc) to perform machine learning processing on the machine learning model 210 for determining sleep-wake stages (wakefulness, non-rapid eye movement sleep, rapid eye movement sleep).

[0103] Furthermore, in the machine learning processing of the machine learning model 210, the machine learning unit 200 can use all of the following four data points: the result of complexity analysis processing received from the EEG processing unit 100, the spectrum received from the frequency analysis unit 400, the amplitude received from the amplitude analysis unit 500, and the result of complexity analysis processing received from the complexity analysis unit 510. Alternatively, it can select a portion of these data points for use.

[0104] The labeled data Fc is pre-generated for each round common to both training EEG data A and training EMG data K. The labeled data Fc has the correct label, which represents the sleep-wake stage corresponding to each round common to both training EEG data A and training EMG data K: "wake," "non-rapid eye movement (NREM) sleep," or "rapid eye movement (REM) sleep."

[0105] The machine learning model 210, which has undergone machine learning through the machine learning unit 200, is used as the determination model 210d in the sleep-wake stage determination device 2d.

[0106] [Sleep-wake stage determination device] exist Figure 8 In China, a 2D comparison of sleep-wake stage determination devices. Figure 6 The sleep-wake stage determination device 2c also includes a complexity analysis unit 510. For each round of electromyographic data P, the complexity analysis unit 510 acquires electromyographic entropy M and electromyographic fractal N. The rounds of the complexity analysis unit 510 are common to those in the electroencephalogram (EEG) processing unit 100, the frequency analysis unit 400, and the amplitude analysis unit 500. The complexity analysis unit 510 sends the results of the complexity analysis processing (electromyographic entropy M, electromyographic fractal N) to the determination unit 300.

[0107] The determination unit 300 determines the sleep-wake stage (wake, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep) based on the complexity analysis processing results DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, Ea1-Ea5 received from the EEG processing unit 100, the spectrum J received from the frequency analysis unit 400, the amplitude L received from the amplitude analysis unit 500, and the complexity analysis processing results M and N received from the complexity analysis unit 510, using the determination model 210d. The determination unit 300 outputs the sleep-wake stage determination result H. The sleep-wake stage determination result H can be information determining which stage of "wake," "NREM sleep," or "REM sleep" it belongs to, or it can be the confidence level of each of "wake," "NREM sleep," and "REM sleep."

[0108] According to this embodiment, compared to the third embodiment, the results of complex analysis of electromyography are used to perform machine learning processing on the machine learning model. This results in a further improvement in the performance of the machine learning model. According to this fourth embodiment, experimental verification shows that the accuracy of automatic determination of the mouse sleep-wake phase, previously around 96%, can be increased to over 98%.

[0109] The embodiments described use mice as an example of mammals, but the same applies to other mammals such as humans and apes.

[0110] Furthermore, in human sleep analysis, polysomnography is typically performed, thus providing access to a wide range of data besides electroencephalography (EEG) and electromyography (EMG). However, for mice, the available data is largely limited to EEG and EMG. Therefore, in the case of mice, the results obtained from complex analysis of EEG and EMG using the described implementation method are particularly effective.

[0111] Furthermore, the functions of information processing devices 1a-1d are implemented by computer hardware such as a CPU (Central Processing Unit) and memory, with the CPU executing computer programs stored in the memory. Additionally, information processing devices 1a-1d can be constructed using general-purpose computer devices or dedicated hardware devices. For example, information processing devices 1a-1d can also be constructed using server computers connected to communication networks such as the Internet. Furthermore, the functions of information processing devices 1a-1d can also be implemented through cloud computing. Alternatively, information processing devices 1a-1d can be implemented using a single computer, or their functions can be distributed across multiple computers. Furthermore, information processing devices 1a-1d can also be configured to operate websites using systems such as the World Wide Web.

[0112] Furthermore, the functions of the sleep-wake stage determination devices 2a-2d are implemented by the fact that the sleep-wake stage determination devices 2a-2d are equipped with computer hardware such as a CPU and memory, and the CPU executes the computer program stored in the memory. In addition, the sleep-wake stage determination devices 2a-2d can be constructed using general-purpose computer devices or dedicated hardware devices. For example, the sleep-wake stage determination devices 2a-2d can also be constructed using a server computer connected to a communication network such as the Internet. Furthermore, the functions of the sleep-wake stage determination devices 2a-2d can also be implemented through cloud computing. Alternatively, the sleep-wake stage determination devices 2a-2d can be implemented using a single computer, or their functions can be distributed across multiple computers. Furthermore, the sleep-wake stage determination devices 2a-2d can also be configured to operate a website using a WWW system, for example.

[0113] In addition, the information processing device and the sleep-wake stage determination device can be configured as different devices or as the same information processing device.

[0114] Furthermore, in the embodiments described above, entropy analysis, fractal analysis, and frequency analysis are used as methods for analyzing EEG data, and amplitude analysis, entropy analysis, and fractal analysis are used as methods for analyzing EMG data. However, analysis methods other than these can also be applied.

[0115] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the specific configuration is not limited to these embodiments, and also includes design changes that do not depart from the spirit of the present invention.

[0116] For example, in the above embodiments, the sleep-wake stage determination devices 2a to 2d use determination models 210a to 210d generated by the information processing devices 1a to 1d through machine learning, but are not limited to this.

[0117] For example, Figure 2 The sleep-wake stage determination device 2a only needs to have a determination model 210a that determines the sleep-wake stage (wake, non-rapid eye movement sleep, rapid eye movement sleep) based on the results of complex analysis and processing received from the EEG processing unit 100, namely DA, DB1~DB5, EA, Da1~Da5, Dp1~Dp5, Ea1~Ea5. The method of generating the determination model 210a is not limited.

[0118] For example, Figure 4 The sleep-wake stage determination device 2b only needs to have a determination model 210b that determines the sleep-wake stage (wake, non-rapid eye movement sleep, rapid eye movement sleep) based on the results of complex analysis and processing received from the EEG processing unit 100 (DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, Ea1-Ea5) and the spectrum J received from the frequency analysis unit 400. The method of generating the determination model 210b is not limited.

[0119] For example, Figure 6 The sleep-wake stage determination device 2c only needs to have a determination model 210c that determines the sleep-wake stage (wake, non-rapid eye movement sleep, rapid eye movement sleep) based on the results of complex analysis and processing received from the EEG processing unit 100, DA, DB1 to DB5, EA, Da1 to Da5, Dp1 to Dp5, Ea1 to Ea5, the spectrum J received from the frequency analysis unit 400, and the amplitude L received from the amplitude analysis unit 500. The method of generating the determination model 210c is not limited.

[0120] For example, Figure 8The sleep-wake stage determination device 2d only needs to have a determination model 210d for determining the sleep-wake stage (wake, non-rapid eye movement sleep, rapid eye movement sleep) based on the results of complex analysis processing received from the EEG processing unit 100 (DA, DB1-DB5, EA, Da1-Da5, Dp1-Dp5, Ea1-Ea5), the spectrum J received from the frequency analysis unit 400, the amplitude L received from the amplitude analysis unit 500, and the results of complex analysis processing received from the complexity analysis unit 510 (M, N). The method for generating the determination model 210d is not limited.

[0121] In addition, in the above embodiments, the training EEG data A and EEG data G can also be obtained through spline interpolation. Spline interpolation allows for the periodic application of EEG data measured under various conditions without relying on the specimenization of the EEG device.

[0122] Alternatively, in the aforementioned embodiments, moving averages can also be used for entropy analysis. Specifically, when calculating the value (entropy) representing the degree of irregularity of the time series being analyzed, a fixed-width window is moved while the average of the time series within the window is used as the object to calculate autocorrelation. By using moving averages for entropy analysis, the accuracy of short-term data analysis can be improved.

[0123] Alternatively, in the aforementioned embodiments, expanded sample entropy (expSampEn) can also be used for entropy analysis. Expanded sample entropy calculates the autocorrelation of all time points in a time series with respect to 1 when solving for the entropy value (entropy) representing the degree of irregularity of the time series being analyzed. By using expanded sample entropy, the temporal resolution is improved, enabling high-precision entropy analysis.

[0124] In addition, for entropy analysis, sample entropy (SampEn) and approximate entropy (ApEn) can also be used. Among them, time-extended sample entropy can speed up entropy analysis compared to sample entropy and approximate entropy.

[0125] Furthermore, in the aforementioned embodiment, when generating learning data for use by the machine learning model 210, the machine learning unit 200 may also expand the learning data through standardization or normalization based on multiple thresholds. The machine learning unit 200 may use only standardization, only normalization based on multiple thresholds, or both.

[0126] Furthermore, in the aforementioned implementation, entropy analysis is a logarithmic nonlinear process, thus the efficient and rapid convergence of machine learning can be expected.

[0127] Alternatively, in the aforementioned embodiment, the Rescoring Short Epoch (RSE) method can be applied to the determination results (sleep wakefulness stage determination results H) of the sleep wakefulness stages (wakefulness, non-rapid eye movement sleep, rapid eye movement sleep) determined using a determination model. In this case, the sleep wakefulness stage determination device includes an RSE method execution unit that performs the RSE method. When the sleep wakefulness stage determiner visually judges the sleep wakefulness stage determination results H to make the final determination of the sleep wakefulness stage, the visual determination of the sleep wakefulness stage generally uses the rule that the same stage lasts between 8 and 20 seconds. However, in fuzzy rounds such as stage transition points, the predictions made by machine learning can easily become unstable, and stages with durations shorter than a certain time may occur. Regarding the algorithm that removes rounds of sleep wakefulness stage determination results that are judged to have a stage duration shorter than a certain time by sequentially decreasing the priority from rounds with shorter stage durations, the RSE method execution unit utilizes the RSE method. By applying the Reactive Sequence of Effect (RSE) method to the sleep-wake stage determination results H, the workload of the sleep-wake stage determiner in visually assessing the sleep-wake stage can be significantly reduced. This results in an improvement in the final determination accuracy of the sleep-wake stage. In the experiment, the final determination accuracy of the sleep-wake stage using the RSE method improved by approximately 1.0% from 0.1% compared to not using the RSE method.

[0128] Furthermore, if time variations are learned as in well-known Hidden Markov Models (HMMs) and Recurrent Neural Networks (RNNs), the accuracy of sleep-wake phase determination in disease model mice will be reduced. However, when applying the RSE method, time variations are not learned when correcting the sleep-wake phase determination results, thus maintaining applicability to sleep structure abnormalities.

[0129] Furthermore, in the aforementioned embodiment, the rounds with a confidence level exceeding a certain threshold in the determination results (sleep-wake stage determination results H) for the sleep-wake stages (wakefulness, non-REM sleep, REM sleep) determined using the determination model may not be displayed. In this case, the sleep-wake stage determination device includes a skipping section that prevents the display of such rounds. This significantly reduces the workload of the person determining the sleep-wake stage visually.

[0130] Additionally, the confidence level of the determination can be displayed relative to the determination results of the sleep-wake stages (wake, non-rapid eye movement sleep, rapid eye movement sleep) determined using the determination model (sleep-wake stage determination result H).

[0131] In this case, the sleep-wake stage determination device includes a confidence display control unit that displays the confidence level. This allows the visual determination criteria for the sleep-wake stage to be shown to the person determining the sleep-wake stage, thus helping to improve the efficiency of visual determination of the sleep-wake stage.

[0132] Furthermore, in the above embodiments, deep learning can also be applied to machine learning. Therefore, in this embodiment, machine learning can be deep learning, or it can be other types of machine learning besides deep learning.

[0133] [Fifth Implementation] Reference Figure 9 and Figure 10 The fifth embodiment will now be described. Figure 9 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to the fifth embodiment. Figure 10 This is a block diagram illustrating an example of the configuration of the sleep-wake stage determination device according to the fifth embodiment.

[0134] [Information processing device] Reference Figure 9 The information processing apparatus 1e of the fifth embodiment will now be described. Training data A1000 is input into the information processing apparatus 1e. The training data A1000 may consist only of training electroencephalogram (EEG) data A, only of training electromyography (EMG) data K, or both training EEG data A and training EMG data K. The choice of which data to use as the training data A1000 is predetermined. Furthermore, the training EEG data A, the training EMG data K, or both training EEG data A and training EMG data K may be obtained through spline interpolation.

[0135] The information processing device 1e includes an amplitude / phase entropy processing unit 1000, a narrowband entropy processing unit 1100, an analysis processing unit 1200, and a machine learning unit 200.

[0136] The amplitude / phase entropy processing unit 1000 acquires the amplitude and phase of each of the specified multiple frequency bands from the training data A1000, and performs entropy analysis on the acquired amplitude and phase of each frequency band. An example of the specified multiple frequency bands is the δ band, θ band, α band, β band, and γ band.

[0137] When only training EEG data A is available in training data A1000, the amplitude / phase entropy processing unit 1000 obtains the amplitude and phase of each of the specified multiple frequency bands from the training EEG data A, and performs entropy analysis on the amplitude and phase of each obtained frequency band.

[0138] When the training data A1000 only has the training electromyography data K, the amplitude / phase entropy processing unit 1000 obtains the amplitude and phase of each of the specified multiple frequency bands from the training electromyography data K, and performs entropy analysis on the amplitude and phase of each obtained frequency band.

[0139] When the training data A1000 includes both training EEG data A and training EMG data K, the amplitude / phase entropy processing unit 1000 acquires the amplitude and phase of each of the specified multiple frequency bands from the training EEG data A, and performs entropy analysis on the acquired amplitude and phase of each frequency band. Furthermore, the amplitude / phase entropy processing unit 1000 acquires the amplitude and phase of each of the specified multiple frequency bands from the training EMG data K, and performs entropy analysis on the acquired amplitude and phase of each frequency band.

[0140] The result of entropy analysis by the amplitude / phase entropy processing unit 1000, namely entropy D1000, is sent to the machine learning unit 200.

[0141] Furthermore, when the amplitude / phase entropy processing unit 1000 performs entropy analysis on the phase, it can also perform entropy analysis on the cosine of the phase. Therefore, the entropy analysis of the phase is performed by performing entropy analysis on the cosine of the phase to calculate the entropy of the cosine of the phase.

[0142] The narrowband entropy processing unit 1100 extracts signals from multiple specified frequency bands from the training data A1000, and performs entropy analysis on the signals of each extracted frequency band. An example of the multiple specified frequency bands is the δ band, θ band, α band, β band, and γ band.

[0143] When the training data A1000 only contains training EEG data A, the narrowband entropy processing unit 1100 extracts signals from multiple specified frequency bands from the training EEG data A, and performs entropy analysis on the extracted signals of each frequency band.

[0144] When the training data A1000 only contains the training electromyography data K, the narrowband entropy processing unit 1100 extracts signals from multiple specified frequency bands from the training electromyography data K, and performs entropy analysis on the extracted signals of each frequency band.

[0145] When the training data A1000 includes both training EEG data A and training EMG data K, the narrowband entropy processing unit 1100 extracts signals from multiple predetermined frequency bands from the training EEG data A and performs entropy analysis on each extracted frequency band signal. Furthermore, the narrowband entropy processing unit 1100 extracts signals from multiple predetermined frequency bands from the training EMG data K and performs entropy analysis on each extracted frequency band signal.

[0146] The result of entropy analysis by the narrowband entropy processing unit 1100, namely entropy D1100, is sent to the machine learning unit 200.

[0147] Furthermore, the amplitude / phase entropy processing unit 1000 and the narrowband entropy processing unit 1100 can also perform entropy analysis on the data obtained by moving average of the input data. Additionally, regarding entropy analysis, the amplitude / phase entropy processing unit 1000 and the narrowband entropy processing unit 1100 can, for example, use time-extended sample entropy, sample entropy, or approximate entropy. Among these, time-extended sample entropy enables faster entropy analysis compared to sample entropy and approximate entropy.

[0148] The analysis and processing unit 1200 performs prescribed analysis and processing on the training data A1000. The analysis and processing performed by the analysis and processing unit 1200 may be, for example, one or more of the following: frequency analysis of electroencephalogram (EEG), amplitude analysis of electromyography (EMG), fractal analysis, and broadband entropy analysis. In broadband entropy analysis, unlike the amplitude / phase entropy processing unit 1000 and the narrowband entropy processing unit 1100, which extract signals from multiple prescribed frequency bands separately from the training data A1000, entropy analysis is performed directly on the training data A1000.

[0149] The analysis results obtained by the analysis processing unit 1200, namely the analysis data D1200, are sent to the machine learning unit 200.

[0150] The machine learning unit 200 uses entropy D1000, entropy D1100, analysis data D1200, and label data Fe to perform machine learning processing on machine learning model 210 for determining sleep wakefulness stages (wakefulness, non-rapid eye movement sleep, rapid eye movement sleep).

[0151] The label data Fe is pre-generated for each round of the training data A1000.

[0152] In the case where only training EEG data A is available in training data A1000, the labeled data Fe has the correct label, which represents the sleep-wake stage "wake", "non-rapid eye movement sleep" or "rapid eye movement sleep" corresponding to each round of training EEG data A.

[0153] In the case where only training EMG data K is available in training data A1000, the labeled data Fe has the correct label, which represents the sleep-wake stage "wake", "non-rapid eye movement sleep" or "rapid eye movement sleep" corresponding to each round of training EMG data K.

[0154] In the case where training data A1000 has both training EEG data A and training EMG data K, the label data Fe has the correct label, which represents the sleep-wake stage "wake", "non-rapid eye movement sleep" or "rapid eye movement sleep" corresponding to each round common in both training EEG data A and training EMG data K.

[0155] Furthermore, in the machine learning processing of the machine learning model 210, the machine learning unit 200 can use all of the three entropy D1000, entropy D1100 and analysis data D1200, or it can choose a portion of entropy D1000, entropy D1100 and analysis data D1200.

[0156] Furthermore, when generating learning data for the machine learning model 210, the machine learning unit 200 can expand the learning data using multiple rescaling methods, such as standardization or normalization based on multiple thresholds. The machine learning unit 200 may use only standardization, only normalization based on multiple thresholds, or both.

[0157] The machine learning model 210, which has undergone machine learning through the machine learning unit 200, is used as the determination model 210e in the sleep-wake stage determination device 2e.

[0158] In addition, Figure 9 In the example configuration, the information processing device 1e includes an amplitude / phase entropy processing unit 1000, a narrowband entropy processing unit 1100, and an analysis processing unit 1200. However, the information processing device 1e may also include any one or both of the amplitude / phase entropy processing unit 1000, the narrowband entropy processing unit 1100, and the analysis processing unit 1200.

[0159] [Sleep-wake stage determination device] Reference Figure 10The sleep-wake stage determination device 2e of the fifth embodiment will be described. The sleep-wake stage determination object, i.e., determination data A2000, is input into the sleep-wake stage determination device 2e. The determination data A2000 is matched with the training data A1000 used by the information processing device 1e. If the training data A1000 only contains training EEG data A, the determination data A2000 only contains EEG data G. If the training data A1000 only contains training EMG data K, the determination data A2000 only contains EMG data P. If the training data A1000 contains both training EEG data A and training EMG data K, the determination data A2000 contains both EEG data G and EMG data P. Furthermore, the EEG data G may also be data obtained through spline interpolation.

[0160] The sleep-wake stage determination device 2e includes an amplitude / phase entropy processing unit 1000, a narrowband entropy processing unit 1100, an analysis processing unit 1200, a determination unit 300, and a determination post-processing unit 2000. The amplitude / phase entropy processing unit 1000, the narrowband entropy processing unit 1100, and the analysis processing unit 1200 are connected to the... Figure 9 The information processing device 1e has the same components 1000, 1100, and 1200, so their descriptions are omitted.

[0161] In the sleep-wake stage determination device 2e, the amplitude / phase entropy processing unit 1000 sends the result of entropy analysis of the determination data A2000, namely entropy D1000, to the determination unit 300.

[0162] In the sleep-wake stage determination device 2e, the narrowband entropy processing unit 1100 sends the result of entropy analysis of the determination data A2000, namely entropy D1100, to the determination unit 300.

[0163] In the sleep-wake stage determination device 2e, the analysis and processing unit 1200 sends the result of analyzing the determination data A2000, namely the analysis data D1200, to the determination unit 300.

[0164] The determination unit 300 uses the determination model 210e to determine the sleep-wake stage (wakefulness, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep) based on entropy D1000, entropy D1100, and analysis data D1200. The determination unit 300 outputs the sleep-wake stage determination result H. The sleep-wake stage determination result H can be information determining which stage of "wakefulness," "NREM sleep," or "REM sleep" it belongs to, or it can be the confidence level of each of "wakefulness," "NREM sleep," and "REM sleep."

[0165] The post-processing unit 2000 includes an RSE execution unit 2100 and a skipping unit 2200.

[0166] The RSE method execution unit 2100 performs the following processing: it applies the RSE method to the sleep-wake stage (wake, non-rapid eye movement sleep, rapid eye movement sleep) determination results (i.e., sleep-wake stage determination results H) determined using the determination model 210e, and removes the rounds in the sleep-wake stage determination results H that are determined to have a stage duration of less than a certain time by sequentially reducing the priority of the rounds with shorter stage durations.

[0167] Skip section 2200 performs the following processing, namely, it prevents the display of the sleep-wake stage (wake, non-rapid eye movement sleep, rapid eye movement sleep) determination results determined by determination model 210e, namely, the rounds with a confidence level of a certain degree or above in the sleep-wake stage determination results H.

[0168] The post-processing unit 2000 outputs the result obtained by the RSE method execution unit 2100 and the skipping unit 2200, which is the judgment output data H2000.

[0169] In addition, Figure 10 In the configuration example, the post-processing unit 2000 has both the RSE method execution unit 2100 and the skipping unit 2200, but in fact, the post-processing unit 2000 may also have only either the RSE method execution unit 2100 or the skipping unit 2200.

[0170] Furthermore, the post-processing unit 2000 is not essential, and the sleep-wake stage determination device 2e may not have the post-processing unit 2000.

[0171] The above is a description of the fifth embodiment.

[0172] Figure 11 , Figure 12 and Figure 13 This is a chart representing the evaluation results of the present invention.

[0173] Figure 11 , Figure 12 and Figure 13 The evaluation results were obtained through analysis using "5-fold cross-validation".

[0174] also, Figure 12 The numerical values ​​for the "MC SleepNet" item are cited from non-patent literature 1 for comparison with this invention. Additionally, Figure 13 The blank space is because it has not been evaluated at the time this application was submitted.

[0175] right Figure 11 , Figure 12 and Figure 13 Explain the terms used in the text.

[0176] "TF" indicates that time-frequency analysis was performed by the analysis and processing unit 1200. If only TF is used, it will become a model similar to SPINDLE described in Non-Patent Document 5.

[0177] “MSE(x)” indicates that a multi-scale entropy analysis was performed on the time series data of x.

[0178] “bb” indicates broadband EEG.

[0179] “band” refers to narrowband EEG.

[0180] “phase / amp” indicates the phase / amplitude of narrowband EEG.

[0181] "EMG" stands for electromyography.

[0182] "DFA" stands for Detrended Fractal Analysis.

[0183] “expMSE(x)” indicates that time-spread entropy analysis was performed on the time series data of x.

[0184] “MC SleepNet” refers to the model described in Non-Patent Document 1.

[0185] "multiple_rescaling" means "multiple rescaling".

[0186] “[RSE]” indicates that the RSE method was executed by the post-processing unit 2000.

[0187] "[HMM]" indicates that the post-processing unit 2000 performed the same hidden Markov model as in non-patent document 5, instead of the RSE method.

[0188] "Precision" refers to the accuracy rate.

[0189] “Recall” refers to the recall rate.

[0190] "Accuracy" refers to the accuracy rate.

[0191] “Kappa” represents Cohen’s Kappa coefficient.

[0192] "Performance vs TF+MSE(bb+band+phase / amp+EMG)+DFA [RSE] (McNemartest, p<0.001)" indicates that when comparing the correctness of the model matching this row with the correctness of the model using "TF+MSE(bb+band+phase / amp+EMG)+DFA [RSE]" as a form of the present invention using the McNemar test, a threshold of "p<0.001" is used to determine whether the performance of the matching model is higher or lower than that of "TF+MSE(bb+band+phase / amp+EMG)+DFA [RSE]".

[0193] For example, such as Figure 12 As shown in the shaded area Figure 11 , Figure 12 and Figure 13 The evaluation results show that the overall accuracy of sleep-wake stage determination in this invention is improved. In particular, the accuracy of REM (rapid eye movement sleep), which was previously difficult to achieve, is significantly improved, and this effect is particularly significant in the sleep-wake stage determination of this invention.

[0194] Figure 12 The evaluation results show that as the entropy analysis is decomposed into features such as “broadband” → “broadband + narrowband” → “broadband + narrowband + amplitude / phase”, the accuracy of sleep-wake stage determination improves little by little overall (significantly in the McNemar test).

[0195] Furthermore, in the past, for example as shown in Non-Patent Document 1, there was an unevenness in the sleep-wake phase (less REM sleep), so improving the accuracy of WAKE (wakefulness) and NREM (non-rapid eye movement sleep) determination is beneficial to improving the overall accuracy of sleep-wake phase determination. Therefore, the difficulty in improving the accuracy of REM determination has become a problem. However, according to the method of the present invention, as Figure 12 The evaluation results show that, compared with models similar to SPINDLE, the overall accuracy was improved by 0.5%, with a significant improvement of up to 2.3% in the REM recall evaluation.

[0196] In addition, such as Figure 12 The evaluation results show that even the model with reduced computational load (which calculates expMSE only in the middle 1 second of a 25-second window, thus speeding up the process by more than 7 times) achieved roughly the same decision accuracy as the model using MSE.

[0197] Alternatively, the computer program used to implement the functions of the aforementioned devices can be recorded on a computer-readable recording medium, and the program recorded on the recording medium can be read into a computer system for execution. Furthermore, the term "computer system" as used here may also include hardware such as an operating system (OS) and peripheral devices.

[0198] In addition, the term "computer-readable recording media" refers to non-volatile memory devices that can be written, such as floppy disks, magneto-optical disks, ROM (Read-Only Memory), and flash memory, as well as portable media such as DVDs (Digital Versatile Discs) and hard drives built into computer systems.

[0199] Furthermore, the term "computer-readable recording medium" also includes media that store programs for a certain period of time, such as servers that send programs via networks such as the Internet or communication lines such as telephone lines, and volatile memory (e.g., DRAM (Dynamic Random Access Memory)) inside a computer system that serves as the user end.

[0200] Alternatively, the program can also be transmitted from a computer system storing the program in a storage device or the like to other computer systems via a transmission medium or by transmission waves in the transmission medium. Here, the "transmission medium" for transmitting the program refers to a medium with information transmission capabilities, such as a network (communication network) like the Internet or a communication line (communication line) like a telephone line.

[0201] Alternatively, the program can be a part of a program used to implement the function. Furthermore, it can be a so-called differential file (differential program) that combines the function with a program already recorded in the computer system.

[0202] The following is a description of the technical aspects of this invention.

[0203] [Time Series Analysis] The experimental data of this invention were obtained under various conditions, thus the conditions were standardized through preprocessing. EEG and EMG signals were bandpass filtered from 0.5 to 50 Hz and transformed into a time series with a sampling frequency of 128 Hz using cubic spline interpolation. This time series is referred to as the broadband (bb) time series. Finite impulse response filters were used to decompose these signals into different frequency bands: 2–4 Hz (δ), 4–8 Hz (θ), 8–13 Hz (α), 13–25 Hz (β), and 25–45 Hz (γ). This time series was divided into 25-second cycles using a 1-second sliding window.

[0204] Next, irregularities and unpredictability are quantified using Sample Entropy (SampEn), an example of entropy analysis, and fractals are quantified using Detrended Fractal Analysis (DFA). Regarding these complexity analyses, previously, to calculate SampEn, a tolerance r of 0.1–0.25 SD standard deviation of the original time series before coarsening was often chosen, with an embedding dimension m = 2. To study the vibrational distribution, a value of r = 0.2 SD was used; for machine learning, three different parameters, r = 0.1 SD, 0.2 SD, and 0.4 SD, were used to expand the feature set. Furthermore, in Multiscale Entropy (MSE), i.e., the distribution of SampEn spanning multiple time scales, scale factors of 1, 2, 4, 8, 16, 32, and 64 were chosen, and their time delays were set to 7.8, 15.6, 31.3, 62.5, 125, 250, and 500 ms. To ensure accuracy, it is recommended to use at least several hundred data points for SampEn calculation. Typically, MSE calculations use coarse-grained methods, but this can lead to data loss as the time scale increases. For example, when performing entropy analysis on 1 second of data sampled at 100 Hz, there might be 100 data points on a 0.01-second time scale, but only 50 on a 0.02-second time scale, and only 25 on a 0.04-second time scale. Such coarse-grained calculation results in fewer data points for entropy calculation, thus weakening noise resistance. Therefore, a sliding window method is used to preserve data points, and averaging within the window stabilizes the value. A moving average, also known as a sliding window averaging, is used. Thus, regardless of the scale, 3200 data points (128 Hz × 25 seconds) are used to calculate SampEn. Furthermore, SampEn for bb and frequency-decomposed time series is investigated, as well as SampEn (phase and amplitude MSE) for narrowband (δ to γ ​​band) phase modulation and amplitude modulation. To avoid discontinuities in the time series, the phase SampEn is calculated as the cosine of the phase time series. Then, the DFA is calculated using time windows n = 3, 4, 7, 11, 18, 29, 46, 73, 116, 184, 290, 459, 726, 1147 (corresponding time scales = 23, 31, 55, 86, 141, 227, 359, 570, 906, 1438, 2266, 3586, 5672, 8961 ms) for the normalized bb and narrowband time series amplitudes. Each narrowband amplitude exhibits two distinct scaling regions, therefore their exponential values ​​α are investigated. The SampEn and DFA described here are calculated using the Python module AntroPy.

[0205] Furthermore, following the methodology used in SPINDLE, EEG spectra and EMG amplitudes are employed for machine learning applications. In the EEG, a high-speed Fourier transform (FFT) is performed on the bb data from 2-second cycles, averaging adjacent frequencies to create 24 frequencies (1 to 24 Hz) in 1 Hz steps, prepared with a 0.125-second sliding window. In the EMG, an FFT is applied to the 2-second cycles of the EMG data, integrating frequencies from 0.5 to 30 Hz, and repeated 24 times in conjunction with the EEG. The logarithms of these values ​​are normalized over 24 hours, creating an input array of 24 frequencies × 160 time points × EEG / EMG for each cycle.

[0206] [Machine Learning] To automatically score sleep, machine learning techniques were applied, leveraging complex features. Data from a total of 72 WT mice was used during the training, validation, and testing phases. The sleep data from the machine learning dataset... The wakefulness stage was scored by an expert. Each sleep stage must last at least 12 seconds; stages that fall below this threshold are adjusted based on the preceding and following stages.

[0207] The SAC model comprises three convolutional blocks for extracting features from temporal frequency analysis (TF), MSE, and DFA, with their outputs integrated through fully connected layers. TF analysis follows the SPINDLE algorithm, processing an input of 24 frequencies × 160 time points × EEG / EMG using convolutional blocks. These convolutional blocks include: a max-pooling layer with a filter size and stride of (2, 3); a convolutional layer with 50 filters, a filter size of (3, 3), and a stride of (1, 1); and another max-pooling layer with a filter size and stride of (2, 2). In the SPINDLE-based experiments, two EEG channels and one EMG channel were used, while in the experiments of this invention, the model was trained with one EEG channel and one EMG channel. On the other hand, for MSE and DFA, four-dimensional inputs were prepared using multiple parameters and rescaling of dimensions. MSE analysis was performed on 17 time series across bb, EMG, and 5 narrowbands, each decomposed into 3 categories (normal / phase / amplitude) and analyzed across 7 time scales and 3 tolerance levels. DFA was applied to 7 time series (bb, 5 narrowband amplitudes, and EMG) spanning 14 time scales. Various dimensional rescaling methods were applied to all mice, standardizing by excluding the 1%, 3%, 5%, 7%, and 9% percentiles within each mouse, resulting in 7 dimensional rescaling inputs (typically, the best one is used, but considering the different optimal rescaling benchmarks for different features and the advantage of data expansion, all the above rescaling results were used as machine learning input). Similar to TF blocks, the dimensions of EMG were aligned with those of EEG for integration. Through these steps, a four-dimensional array was obtained. The MSE input contained 16 time series × 7 time scales × 7 dimensional rescalings × 6 (3 tolerances × EEG / EMG), and the DFA input contained 6 time series × 14 time scales × 7 dimensional rescalings × EEG / EMG. For local feature extraction, MSE used 50 filters, and DFA used 30 filters in a convolutional layer with filter sizes of (2, 2, 2) and strides of (1, 1, 1). Features extracted from TF, MSE, and DFA blocks were integrated through a two-layer fully connected network (1000 neurons in the first layer and 500 neurons in the second layer) with 50% Dropout and batch normalization. The output was then processed through the softmax function for the WAKE, REM, and NREM stages. In contrast, the SPINDLE model integrated features through only two fully connected layers (each with 1000 neurons) with 50% Dropout.

[0208] In the SPINDLE experiment, an Hidden Markov Model (HMM) algorithm, which incorporates transition probabilities, was used to improve prediction results through a Convolutional Neural Network (CNN). However, this method may fail to detect specific features (e.g., SOREM) in mice with abnormal sleep structures, such as the OXKO mouse. To avoid this drawback, the Rescoring Short Epoch (RSE) method was introduced as a constraint technique. Consistent with expert scoring rules (the minimum duration of each phase is 12 seconds), phases shorter than 12 seconds were replaced with the phase with the second highest probability (i.e., the second largest softmax value). Since the shortest REM phase in the sleep-wake cycle usually lasts for tens of seconds, a 12-second threshold was considered reasonable.

[0209] The following is a note on the invention described in the claims originally appended to Japanese Patent Application No. 2023-111974. The appendix number is the same as the number in the claims originally appended to Japanese Patent Application No. 2023-111974.

[0210] (Note 1) An information processing device comprising: The EEG processing unit acquires the amplitude and phase of each of several specified frequency bands from training EEG data measured from a live mammal, and performs complexity analysis processing, including entropy analysis, on the acquired amplitude and phase of each frequency band; and The machine learning unit uses the results of the complexity analysis and the labeled data of sleep-wake stages corresponding to the training EEG data to perform machine learning processing for determining the sleep-wake stage of the machine learning model.

[0211] (Note 2) According to the information processing apparatus described in Appendix 1, wherein The EEG processing unit then extracts signals from the multiple frequency bands from the training EEG data, and performs complexity analysis processing, including entropy analysis, on each extracted frequency band signal. The machine learning unit then uses the results of the complexity analysis to perform machine learning processing on the machine learning model.

[0212] (Note 3) According to the information processing apparatus described in Appendix 2, wherein The EEG processing unit then performs complexity analysis processing, including entropy analysis, on the training EEG data, and The machine learning unit then uses the results of the complexity analysis to perform machine learning processing on the machine learning model.

[0213] (Note 4) According to the information processing apparatus described in Appendix 1, wherein The EEG processing unit then performs complex analysis processing, including fractal analysis, on the amplitudes of each acquired frequency band, and The machine learning unit then uses the results of the complexity analysis to perform machine learning processing on the machine learning model.

[0214] (Note 5) According to the information processing apparatus described in Appendix 4, wherein The EEG processing unit then extracts signals from the multiple frequency bands from the training EEG data, and performs complexity analysis processing, including entropy analysis, on each extracted frequency band signal. The machine learning unit then uses the results of the complexity analysis to perform machine learning processing on the machine learning model.

[0215] (Note 6) According to the information processing apparatus described in Appendix 5, wherein The EEG processing unit then performs complexity analysis processing, including entropy analysis and fractal analysis, on the training EEG data, and The machine learning unit then uses the results of the complexity analysis to perform machine learning processing on the machine learning model.

[0216] (Note 7) The information processing apparatus according to any one of Appendices 1 to 6, wherein It also includes a frequency analysis unit that transforms the training EEG data into a spectrum; and The machine learning unit then uses the spectrum transformed from the training EEG data to perform machine learning processing on the machine learning model.

[0217] (Postscript 8) According to the information processing apparatus described in Appendix 7, wherein It also includes an amplitude analysis unit that acquires amplitude from training electromyography data measured from the living organism; and The machine learning unit then uses the amplitude obtained from the training electromyography data to perform machine learning processing on the machine learning model. The labeled data is, in turn, labeled data of the sleep-wake phase corresponding to the electromyographic data used for training.

[0218] (Note 9) According to the information processing apparatus described in Appendix 8, wherein It also includes a complexity analysis unit that performs complexity analysis processing, including entropy analysis and fractal analysis, on the training electromyography data; and The machine learning unit then uses the results of complexity analysis of the training electromyography data to perform machine learning processing on the machine learning model.

[0219] (Postscript 10) The information processing apparatus according to any one of Appendices 1 to 6, wherein The multiple frequency bands are the δ band, θ band, α band, β band, and γ band.

[0220] (Postscript 11) The information processing apparatus according to any one of Appendices 1 to 6, wherein The living organism in question is a mouse.

[0221] (Postscript 12) A sleep-wake stage determination device, comprising: The EEG processing unit acquires the amplitude and phase of each of several specified frequency bands from EEG data measured from a live mammal, and performs complexity analysis processing, including entropy analysis, on the acquired amplitude and phase of each frequency band; and The determination unit uses a determination model to determine the sleep-wake stage based on the results of the complexity analysis.

[0222] (Postscript 13) According to the sleep-wake stage determination device described in Appendix 12, wherein The EEG processing unit then extracts signals from the multiple frequency bands from the EEG data, and performs complexity analysis processing, including entropy analysis, on each extracted frequency band signal. The determination unit then uses the results of the complexity analysis process to determine the sleep-wake stage.

[0223] (Postscript 14) According to the sleep-wake stage determination device described in Appendix 13, wherein The EEG processing unit then performs complexity analysis processing, including entropy analysis, on the EEG data, and The determination unit then uses the results of the complexity analysis process to determine the sleep-wake stage.

[0224] (Postscript 15) According to the sleep-wake stage determination device described in Appendix 12, wherein The EEG processing unit then performs complex analysis processing, including fractal analysis, on the amplitudes of each acquired frequency band, and The determination unit then uses the results of the complexity analysis process to determine the sleep-wake stage.

[0225] (Postscript 16) According to the sleep-wake stage determination device described in Appendix 15, wherein The EEG processing unit then extracts signals from the multiple frequency bands from the EEG data, and performs complexity analysis processing, including entropy analysis, on each extracted frequency band signal. The determination unit then uses the results of the complexity analysis process to determine the sleep-wake stage.

[0226] (Postscript 17) According to the sleep-wake stage determination device described in Appendix 16, wherein The EEG processing unit then performs complexity analysis processing on the EEG data, including entropy analysis and fractal analysis, and The determination unit then uses the results of the complexity analysis process to determine the sleep-wake stage.

[0227] (Postscript 18) According to any one of Appendices 12 to 17, the sleep-wake stage determination device, wherein It also includes a frequency analysis unit that transforms the EEG data into a spectrum; and The determination unit then uses the spectrum converted from the EEG data to determine the sleep-wake phase.

[0228] (Postscript 19) According to the sleep-wake stage determination device described in Appendix 18, wherein It also includes an amplitude analysis unit that acquires the amplitude from electromyographic data measured from the living organism; and The determination unit then uses the amplitude obtained from the electromyographic data to determine the sleep-wake phase.

[0229] (Postscript 20) According to the sleep-wake stage determination device described in Appendix 19, wherein It also includes a complexity analysis unit that performs complexity analysis processing on the electromyographic data, including entropy analysis and fractal analysis; and The determination unit then uses the results of complex analysis of the electromyographic data to determine the sleep-wake phase.

[0230] (Postscript 21) According to any one of Appendices 12 to 17, the sleep-wake stage determination device, wherein The multiple frequency bands are the δ band, θ band, α band, β band, and γ band.

[0231] (Postscript 22) According to any one of Appendices 12 to 17, the sleep-wake stage determination device, wherein The living organism in question is a mouse.

[0232] (Postscript 23) A computer program that causes a computer to perform: The EEG processing steps involve acquiring the amplitude and phase of each of several specified frequency bands from training EEG data measured from a live mammal, and performing complexity analysis, including entropy analysis, on the acquired amplitude and phase of each frequency band; and The machine learning step uses the results of the complexity analysis and the labeled data of sleep-wake stages corresponding to the training EEG data to perform machine learning processing for determining the sleep-wake stage of the machine learning model.

[0233] (Postscript 24) A computer program that causes a computer to perform: The EEG processing steps involve acquiring the amplitude and phase of each of several specified frequency bands from EEG data measured in a live mammal, and performing complexity analysis, including entropy analysis, on the acquired amplitude and phase of each frequency band; and The determination step involves using a determination model to determine the sleep-wake stage based on the results of the complexity analysis.

[0234] (Postscript 25) An information processing method, executed by an information processing device, includes: The EEG processing steps involve acquiring the amplitude and phase of each of several specified frequency bands from training EEG data measured from a live mammal, and performing complexity analysis, including entropy analysis, on the acquired amplitude and phase of each frequency band; and The machine learning step uses the results of the complexity analysis and the labeled data of sleep-wake stages corresponding to the training EEG data to perform machine learning processing for determining the sleep-wake stage of the machine learning model.

[0235] (Postscript 26) A method for determining sleep-wake stages, executed by a sleep-wake stage determination device, includes: The EEG processing steps involve acquiring the amplitude and phase of each of several specified frequency bands from EEG data measured in a live mammal, and performing complexity analysis, including entropy analysis, on the acquired amplitude and phase of each frequency band; and The determination step involves using a determination model to determine the sleep-wake stage based on the results of the complexity analysis.

[0236] Symbol Explanation 1a, 1b, 1c, 1d Information processing devices 100 EEG Processing Department 200 Machine Learning Department 110 bandpass filter 120 Hilbert Transformer Complexity Analysis Department 130, 510 131, 511 Entropy Analysis Department 132,512 Fractal Analysis Section 210 Machine Learning Model 2a, 2b, 2c, 2d Sleep-wake stage determination device 300 Judgment Department 210a, 210b, 210c, 210d, 210e decision models 400 Frequency Analysis Department 500 Amplitude Analysis Department 1000 Amplitude / Phase Entropy Processing Unit 1100 Narrowband Entropy Processing Unit 1200 Analysis and Processing Department 2000 Post-Processing Department 2100RSE Enforcement Department Skip section 2200.

Claims

1. An information processing apparatus comprising: an electroencephalogram processing section that acquires an amplitude and a phase of each of a plurality of prescribed frequency bands from training electroencephalogram data measured from one living body of a mammal, and separately performs a complexity analysis process including entropy analysis on the acquired amplitude and phase of each frequency band; and a machine learning section that performs a machine learning process of a machine learning model for determining a sleep-wake stage using a result of the complexity analysis process and label data of a sleep-wake stage corresponding to the training electroencephalogram data. 2.The information processing apparatus according to claim 1, wherein the electroencephalogram processing section further extracts signals of the plurality of frequency bands from the training electroencephalogram data, respectively, and separately performs a complexity analysis process including entropy analysis on the extracted signals of each frequency band, and the machine learning section further performs the machine learning process of the machine learning model using a result of the complexity analysis process. 3.The information processing apparatus according to claim 1, wherein the entropy analysis of the phase is entropy of a cosine of the phase calculated by entropy analysis of the cosine of the phase. 4.The information processing apparatus according to claim 1, wherein the electroencephalogram processing section uses time-expanded sample entropy as the entropy analysis. 5.The information processing apparatus according to claim 1, wherein the machine learning section, when generating learning data for machine learning of the machine learning model, expands the learning data by standardization or normalization based on a plurality of thresholds, or both of the standardization and the normalization based on a plurality of thresholds. 6.The information processing apparatus according to any one of claims 1 to 5, wherein the living body is a mouse. 7.A sleep-wake stage determination apparatus comprising: an electroencephalogram processing section that acquires an amplitude and a phase of each of a plurality of prescribed frequency bands from electroencephalogram data measured from one living body of a mammal, and separately performs a complexity analysis process including entropy analysis on the acquired amplitude and phase of each frequency band; and a determination section that determines a sleep-wake stage based on a result of the complexity analysis process using a determination model. 8.The sleep-wake stage determination apparatus according to claim 7, wherein the electroencephalogram processing section further extracts signals of the plurality of frequency bands from the electroencephalogram data, respectively, and separately performs a complexity analysis process including entropy analysis on the extracted signals of each frequency band, and the determination section further determines the sleep-wake stage based on a result of the complexity analysis process. 9.The sleep-wake stage determination apparatus according to claim 7, wherein the entropy analysis of the phase is entropy of a cosine of the phase calculated by entropy analysis of the cosine of the phase. 10.The sleep-wake stage determination apparatus according to claim 7, wherein the electroencephalogram processing section uses time-expanded sample entropy for the entropy analysis. 11.The sleep-wake stage determination apparatus according to claim 7, wherein Further provided is an RSE method execution unit that applies an RSE (Rescoring Short Epoch) method to the determination results of the sleep-wake stages determined using the determination model, removes the epochs of the sleep-wake stages determined to have a stage duration shorter than a certain time, and sequentially lowers the priority order from the epochs having a shorter stage duration.

12. The sleep-wake stage determination device according to claim 7, wherein Further provided is a skip unit that does not cause the epochs of the determination results of the sleep-wake stages determined using the determination model to be displayed when the confidence level is equal to or higher than a certain level.

13. The sleep-wake stage determination device according to claim 7, wherein Further provided is a confidence level display control unit that causes the confidence level of the determination to be displayed for the determination results of the sleep-wake stages determined using the determination model.

14. The sleep-wake stage determination device according to any one of claims 7 to 13, wherein The living body is a mouse.

15. A computer program for causing a computer to execute: an electroencephalogram processing step of acquiring an amplitude and a phase of each of a plurality of prescribed frequency bands from training electroencephalogram data measured from a living body of a mammal, and performing a complexity analysis process including entropy analysis on the acquired amplitude and phase of each frequency band; and a machine learning step of performing a machine learning process of a machine learning model for determining sleep-wake stages using a result of the complexity analysis process and label data of sleep-wake stages corresponding to the training electroencephalogram data.

16. A computer program for causing a computer to execute: an electroencephalogram processing step of acquiring an amplitude and a phase of each of a plurality of prescribed frequency bands from electroencephalogram data measured from a living body of a mammal, and performing a complexity analysis process including entropy analysis on the acquired amplitude and phase of each frequency band; and a determination step of performing determination of sleep-wake stages using a determination model based on a result of the complexity analysis process.

17. An information processing method executed by an information processing device, and including: an electroencephalogram processing step of acquiring an amplitude and a phase of each of a plurality of prescribed frequency bands from training electroencephalogram data measured from a living body of a mammal, and performing a complexity analysis process including entropy analysis on the acquired amplitude and phase of each frequency band; and a machine learning step of performing a machine learning process of a machine learning model for determining sleep-wake stages using a result of the complexity analysis process and label data of sleep-wake stages corresponding to the training electroencephalogram data.

18. A sleep-wake stage determination method executed by a sleep-wake stage determination device, and including: an electroencephalogram processing step of acquiring an amplitude and a phase of each of a plurality of prescribed frequency bands from electroencephalogram data measured from a living body of a mammal, and performing a complexity analysis process including entropy analysis on the acquired amplitude and phase of each frequency band; and a determination step of performing determination of sleep-wake stages using a determination model based on a result of the complexity analysis process. determining step of determining the sleep-wake stage using a determination model based on the results of the complexity analysis processing.

Images