Slow wave sleep closed-loop regulation system and method based on brain state critical point prediction

Abstract

The application discloses a slow wave sleep closed-loop regulation system and method based on brain state critical point prediction, relates to the technical field of sleep regulation, and comprises the following modules: a neural signal acquisition module, which is used for continuously collecting electroencephalogram signals and electromyogram signals of a subject; a critical point prediction module, which is used for performing real-time processing on the electroencephalogram signals and the electromyogram signals, extracting multi-dimensional features representing brain states, constructing a brain state Euclidean state space, calculating a wake distance of a current brain state to a wake state, fitting time evolution of the wake distance based on a dynamic model with a saddle-node bifurcation characteristic, combining a critical slowing down early warning index, determining a critical point of slow wave sleep to wake conversion and an intervention timing; and a closed-loop control module, which is used for triggering predictive closed-loop electrical stimulation applied to the facial nucleus in response to the intervention timing before the brain state evolves to the critical point of slow wave sleep to wake conversion, maintaining slow wave sleep by predicting the critical point in real time and triggering the electrical stimulation in advance.

Description

Technical Field

[0001] This invention relates to the field of sleep regulation technology, and in particular to a slow-wave sleep closed-loop regulation system and method based on brain state critical point prediction. Background Technology

[0002] Maintaining a certain duration of slow-wave sleep at night is a necessary prerequisite for achieving optimal physiological, psychological, and cognitive function. Animal and human studies have shown that activation of gamma-aminobutyric acid (GABA) neurons in the parafacial zone (PZ) of the mammalian brain promotes slow-wave sleep: activated parafacial zone neurons release GABA to neurons in the parabrachial nucleus (PB), which in turn project and release glutamate to cortical projection neurons in the basal forebrain, thereby regulating cortical brain activity and triggering slow-wave sleep. Slow-wave sleep (SWS) is characterized by predominance of delta frequency (0.5-4 Hz) activity and low muscle tone, while wakefulness is characterized by increased gamma frequency (30-80 Hz) activity and elevated muscle activity. The transition from slow-wave sleep to wakefulness (awakening) often occurs abruptly, interrupting the continuity of deep sleep. From a dynamic perspective, before the brain state transitions, it will experience a critical point (saddle node bifurcation) that transitions from a bistable to a monostable state, accompanied by early warning signals such as increased autocorrelation and variance in the wakefulness distance time series.

[0003] However, current sleep regulation technologies are mostly passive-response, meaning intervention is only applied after wakefulness or a transition has occurred. They lack the ability to predict the transition threshold in real time, making it difficult to intervene before the transition happens. This limits the effectiveness in maintaining slow-wave sleep continuity, and consequently, hinders the clinical translation of this target. Existing sleep regulation technologies have significant limitations: 1) Passive-response closed-loop deep brain stimulation (based on sleep stage classification): Electrodes are implanted in the PZ (target coordinates: AP=-5.3mm, ML=±0.7mm, DV=-3.4mm) of mice via stereotactic surgery. Whole-brain EEG signals are collected using anterior and posterior fontanelle screw electrodes. A convolutional neural network is used to classify the real-time EEG into sleep stages. Electrical stimulation is triggered when the stage is determined to be non-slow-wave sleep (wakeful or REM sleep). The limitation of this passive-response closed-loop stimulation is that intervention can only be applied after the transition has occurred (i.e., after entering wakeful or REM sleep), making it a reactive measure. It cannot provide predictive intervention before the critical point of slow-wave sleep-wake transition, making it difficult to prevent awakenings or maintain the continuity of deep sleep.

[0004] 2) Open-loop genetic regulation to induce slow-wave sleep: PZ neurons are activated through optogenetics (implanting an optical fiber above the PZ of VGAT-ChR2-EYFP mice and administering light pulses at a wavelength of 473 nm and a frequency of 40 Hz) or chemogenetics (expressing receptors in the PZ γ-aminobutyric acid (GABAergic) neurons of Vgat-IRES-Cre mice and intraperitoneally injecting clozapine-N-oxide). The drawbacks of this open-loop genetic regulation method are that it does not adjust based on real-time brain state and suffers from problems such as inactivation of light-sensitive channel proteins and slow onset of chemogenetics (requiring more than half an hour). Neither method allows for real-time prediction and timely intervention of the transition threshold.

[0005] In summary, current technologies lack the ability to predict the critical point of transition from slow-wave sleep to wakefulness in real time, making it difficult to intervene before the transition occurs. This limits the effectiveness of maintaining the continuity of slow-wave sleep and hinders the clinical translation of this target. Therefore, there is an urgent need for a solution that can predict the critical point of transition from slow-wave sleep to wakefulness in real time and provide predictive intervention before the transition. Summary of the Invention

[0006] This invention aims to address the problem that existing technologies cannot predictively intervene before the transition from slow-wave sleep to wakefulness, making it difficult to maintain the continuity of slow-wave sleep. It provides a closed-loop regulation system and method for slow-wave sleep based on the prediction of brain state critical points. Without relying on genetic manipulation, it can maintain slow-wave sleep by predicting the critical point in real time and triggering electrical stimulation in advance.

[0007] This invention provides a slow-wave sleep closed-loop regulation system based on brain state critical point prediction, comprising: The neural signal acquisition module is used to continuously acquire electroencephalogram (EEG) and electromyogram (EMG) signals from the subjects and to apply electrical stimulation at the parafacial nucleus of the facial nerve. The critical point prediction module is used to process the EEG and EMG signals in real time, extract multidimensional features representing the brain state, construct the Euclidean state space of the brain state, calculate the awake distance from the current brain state to the awake state, and fit the time evolution of the awake distance based on the dynamic model with the bifurcation characteristics of the saddle node. Combined with the critical slowing warning index, the critical point of the transition from slow wave sleep to wakefulness and the timing of intervention are determined. A closed-loop control module is used to respond to the intervention timing by triggering predictive closed-loop electrical stimulation applied to the parafacial nucleus of the face before the brain state evolves to the critical point of transition from slow-wave sleep to wakefulness, in order to maintain the slow-wave sleep state.

[0008] Preferably, the neural signal acquisition module includes: The cranial nail electrode, the EEG acquisition electrode fixed to the anterior and posterior fontanelle regions of the skull of the subject, and the electromyography acquisition electrode fixed to the neck muscles, are used to acquire the EEG signals and electromyography signals. A deep electrode is implanted into the parafacial nucleus of the facial nerve in the subject via stereotactic positioning, and is used to apply the electrical stimulation under the control of the closed-loop control module.

[0009] Preferably, the critical point prediction module includes: The feature extraction unit is used to filter and preprocess the real-time acquired EEG and EMG signals, and segment them according to a preset time window to extract multidimensional features that characterize brain state. The state space mapping unit is used to normalize the multidimensional features, construct the Euclidean state space of the brain state with the feature centroid of the awake state of the subject as the reference point, and calculate the Euclidean distance from the brain state feature vector of the current time period to the feature centroid of the awake state as the awake distance. The dynamic analysis unit is used to construct and identify dynamic models with saddle node bifurcation characteristics based on the time series of historical wakefulness distances. By performing steady-state analysis on the dynamic model, the critical point for the transition from slow-wave sleep to wakefulness and the corresponding critical distance threshold are determined. The early warning determination unit is used to compare the real-time awakening distance with the critical distance threshold corresponding to the critical point. When the real-time awakening distance is less than the critical distance threshold and the duration reaches the preset duration, it determines that the intervention opportunity is met and outputs a trigger signal to the closed-loop control module.

[0010] Preferably, the feature extraction unit is configured as follows: The real-time acquired EEG and EMG signals are filtered by a first bandpass filter and a second bandpass filter corresponding to the physiological frequency band, respectively. The first bandpass filter and the second bandpass filter are zero-phase preset order Butterworth filters. The filtered EEG and EMG signals are segmented according to a preset time window, wherein adjacent preset time windows have a preset overlap rate. Calculate the root mean square amplitude of EEG and EMG signals within each preset time window, and identify preset time windows whose root mean square amplitude exceeds a preset multiple of the median root mean square amplitude of all preset time windows as artifact periods and remove them. The EEG and EMG signals within each preset time window after removing the artifact time periods are used to perform feature calculations to obtain a preset dimension feature vector for characterizing the brain state. The feature calculation includes performing spectral analysis on the EEG signal to obtain EEG frequency domain features, performing time domain analysis and frequency domain analysis on the electromyography signal to obtain electromyography features, and performing nonlinear dynamic analysis on at least one of the EEG signal and electromyography signal to obtain nonlinear dynamic features.

[0011] Preferably, the state space mapping unit is configured as follows: Obtain the feature vectors for each time period output by the feature extraction unit; Identify at least two consecutive wakefulness periods that first occur after a slow-wave sleep state, and determine the start time of the first wakefulness period among the at least two consecutive wakefulness periods as the wakefulness onset time; Using the moment of awakening as the time reference, feature vectors for each time period during the transition from slow-wave sleep to awakening are obtained, and global normalization is performed on the feature vectors for each time period. Construct a Euclidean state space with the characteristic centroid of the conscious state as the origin; The brain state feature vector of the current time period is mapped to the Euclidean state space, and the Euclidean distance from the brain state feature vector of the current time period to the origin is calculated. The Euclidean distance is defined as the awake distance to quantify the degree to which the current brain state deviates from the awake state.

[0012] Preferably, the dynamic analysis unit is configured as follows: Obtain the wakefulness distance time series output by the state space mapping unit; A dynamic model with saddle-node bifurcation characteristics is constructed to describe the temporal evolution of the wakefulness distance during the transition from slow-wave sleep to wakefulness, and the wakefulness distance time series is used to identify the parameters of the dynamic model to obtain the model parameters. Steady-state analysis is performed on the dynamic model based on the model parameters to solve for the bifurcation point where the system changes from bistable to monostable. The control parameters corresponding to the bifurcation point are determined as the critical control parameters for the transition from slow-wave sleep to wakefulness. The target equilibrium point is calculated based on the critical control parameters, and the sobriety distance value corresponding to the target equilibrium point is determined as the critical distance threshold.

[0013] Preferably, the dynamic analysis unit is configured to perform parameter identification using a two-stage optimization strategy, the two-stage optimization strategy including: A coarse-grid search is performed within a preset parameter space to obtain candidate parameter combinations; Based on the candidate parameter combination, a local refinement search is performed to determine the model parameters of the dynamic model by optimizing the joint objective function of maximizing the goodness of fit and minimizing the root mean square error. The dynamic model is solved using numerical integration, and the fitting accuracy of the dynamic model to the wakefulness distance time series is evaluated based on the coefficient of determination and root mean square error. Preferably, the warning determination unit is configured as follows: Obtain the current sobriety distance output by the state space mapping unit and the critical distance threshold output by the dynamic analysis unit; Causal median filtering is applied to the current lucidity distance to suppress temporal fluctuation noise; The filtered sobriety distance is compared with the critical distance threshold. When the filtered distance to wakefulness is less than the critical distance threshold and the state continues for a predetermined time, it is determined that the intervention opportunity is met, and a trigger signal is output to the closed-loop control module. Monitoring will continue if the filtered sobriety distance is not less than the critical distance threshold, or if the duration for which the filtered sobriety distance is less than the critical distance threshold does not reach the predetermined time.

[0014] Preferably, the closed-loop control module is configured as follows: Receive the trigger signal output by the critical point prediction module; In response to the trigger signal, an electrical stimulation pulse sequence is applied to the parafacial nucleus of the facial nerve before the brain state evolves to the critical point; During the application of the electrical stimulation pulse sequence, EEG and EMG signals are continuously acquired, and brain state monitoring results generated by the critical point prediction module based on the continuously acquired EEG and EMG signals are obtained. When the brain state monitoring results indicate that the current awake distance is greater than the critical distance threshold and the state continues for a preset duration, the application of the electrical stimulation pulse sequence is terminated.

[0015] This invention also provides a method for slow-wave sleep closed-loop regulation based on brain state critical point prediction, comprising: Continuous acquisition of electroencephalogram (EEG) and electromyogram (EMG) signals from the subjects; The electroencephalogram (EEG) and electromyogram (EMG) signals are processed in real time to extract multidimensional features representing brain states, construct a Euclidean state space for brain states, and calculate the awakening distance from the current brain state to the awake state. The temporal evolution of the wakefulness distance is fitted based on a dynamic model with saddle-node bifurcation characteristics. Combined with critical slowing warning indicators, the critical point of transition from slow-wave sleep to wakefulness and the timing of intervention are determined. In response to the timing of the intervention, a predictive closed-loop electrical stimulation of the parafacial nucleus is triggered before the brain state evolves to the critical point of transition from slow-wave sleep to wakefulness, in order to maintain the slow-wave sleep state.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a slow-wave sleep closed-loop regulation system and method based on brain state critical point prediction. By predicting the critical point of transition from slow-wave sleep to wakefulness in real time, the system can trigger electrical stimulation of the parafacial nucleus before the actual transition occurs, achieving predictive intervention. The wakefulness distance quantified by the bifurcation dynamics model and the critical warning signal together provide early perception of the transition trend, enabling objective judgment on the timing and necessity of intervention. Thus, intervention can be carried out before wakefulness occurs, avoiding the occurrence of wakefulness and maintaining the continuity of slow-wave sleep. Compared with passive response closed-loop stimulation, it significantly improves the effect of maintaining deep sleep. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the slow-wave sleep closed-loop regulation system based on brain state critical point prediction in an embodiment of the present invention; Figure 2 This is a flowchart of a closed-loop deep brain stimulation method based on brain state critical point prediction in an embodiment of the present invention. Figure 3 This is a schematic diagram of the critical point prediction principle for the transition from slow-wave sleep to wakefulness in an embodiment of the present invention (including wakefulness distance state space, time series fitting, bifurcation diagram and critical slowing warning signal). Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The term "comprising" and its variations as used herein are open-ended inclusion, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0020] It should be noted that the concepts of "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0021] It should be noted that the terms "a" and "a plurality of" used in this application disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0022] like Figure 1-3 As shown, the present invention provides a slow-wave sleep closed-loop regulation system based on brain state critical point prediction, including: a neural signal acquisition module, a critical point prediction module, and a closed-loop control module.

[0023] The neural signal acquisition module is used to continuously acquire electroencephalogram (EEG) and electromyogram (EMG) signals from the subjects and to apply electrical stimulation at the parafacial nucleus of the facial nerve.

[0024] The critical point prediction module is used to process the EEG and EMG signals in real time, extract multidimensional features representing the brain state, construct the Euclidean state space of the brain state, calculate the awake distance from the current brain state to the awake state, and fit the temporal evolution of the awake distance based on a dynamic model with fold / saddle-node bifurcation characteristics. Combined with the critical slowing warning index, it determines the tipping point of the transition from slow wave sleep to wakefulness and the timing of intervention. A closed-loop control module, connected to the critical point prediction module and the electrical stimulation module respectively, is used to trigger predictive closed-loop electrical stimulation applied to the parafacial nucleus in response to the intervention timing, before the brain state evolves to the critical point of transition from slow-wave sleep to wakefulness, so as to maintain the slow-wave sleep state.

[0025] This embodiment provides a slow-wave sleep closed-loop regulation system based on brain state critical point prediction, also known as a predictive onset intervention via neural tipping detection (POINT). The system includes a neural signal acquisition module, a critical point prediction module, and a closed-loop control module. During the transition from slow-wave sleep to wakefulness, the system continuously records electroencephalogram (EEG) and electromyographic (EMG) signals, extracts multidimensional features in real time for each time window, constructs an Euclidean state space of the brain state, and calculates the "wake distance" from the current state to the wakeful state. Then, it fits a bifurcation dynamic model of the node sellae to the temporal evolution of the wake distance and, combined with a critical slowing-down warning index, predicts in real time the critical point of the slow-wave sleep-wake transition and its arrival time, which is used to determine whether and when to trigger electrical stimulation of the parafacial nucleus. Through the coordinated work of the above modules, electrical stimulation of the parafacial nucleus can be triggered in advance before the critical point of the transition from slow-wave sleep to wakefulness arrives, achieving predictive rather than reactive intervention. Specifically, by constructing a multidimensional Euclidean state space and quantifying the "wakefulness distance," combined with a bifurcation dynamics model and a critical slowing warning index, the system can objectively determine the transition trend and intervene before wakefulness occurs, blocking or delaying the transition process. Compared to passive closed-loop stimulation that can only be remedied after the fact and slow-acting open-loop regulation schemes, this embodiment significantly improves the effect of maintaining the continuity of deep sleep, and predicts the critical point in real time and intervenes in advance to maintain deep sleep without relying on genetic manipulation.

[0026] In one embodiment, the neural signal acquisition module includes: The cranial nail electrode, the EEG acquisition electrode fixed to the anterior and posterior fontanelle regions of the skull of the subject, and the electromyography acquisition electrode fixed to the neck muscles, are used to acquire the EEG signals and electromyography signals. Deep electrodes, stereotactically implanted into the parafacial nucleus of the facial nerve in the subject, are used to apply electrical stimulation under the control of the closed-loop control module. The deep electrodes consist of two strands of nickel-titanium alloy twisted into a single conductive wire, approximately 4 mm in length, with an insulating coating to ensure that the electrode discharges only at the tail end, achieving targeted electrical stimulation. Specifically, the experimenter implanted deep electrodes into the parafacial nucleus of the facial nerve in mice using stereotactic surgery (target coordinates: AP=-5.3mm, ML=±0.7mm, DV=-3.4mm), and fixed cranial screw electrodes near the anterior and posterior fontanelles and in the neck muscles. This implantation method and electrode design effectively ensured the accuracy of targeted stimulation and the stability of signal acquisition. Before the formal experiment, response surface methodology was used to characterize the parameter-response relationship of neuronal activity induced by deep brain stimulation in the parabrachial nucleus (PB)-parafacial nucleus (PZ) circuit, thereby determining the optimal stimulation parameters (such as pulse intensity, width, and frequency) for subsequent experiments.

[0027] In one embodiment, the critical point prediction module includes: a feature extraction unit, a state space mapping unit, a dynamic analysis unit, and an early warning determination unit.

[0028] The feature extraction unit is used to filter and preprocess the real-time acquired EEG and EMG signals, and segment them according to a preset time window to extract multidimensional features representing brain states. Specifically, it is configured as follows: The real-time acquired EEG and EMG signals are filtered using a first bandpass filter and a second bandpass filter corresponding to the physiological frequency bands, respectively. The first and second bandpass filters are zero-phase Butterworth filters of a preset order. Specifically, the first bandpass filter is a zero-phase fourth-order Butterworth bandpass filter for EEG signals, with a frequency range of 0.1-100Hz, and the second bandpass filter is a zero-phase fourth-order Butterworth bandpass filter for EMG signals, with a frequency range of 10-300Hz. The filtered EEG and EMG signals are segmented according to preset time windows. There is a preset overlap rate between adjacent preset time windows. The duration of the preset time window is 4 seconds and the preset overlap rate between adjacent preset time windows is 50%. The root mean square amplitude of EEG and EMG signals within each preset time window is calculated. Preset time windows whose root mean square amplitude exceeds a preset multiple, such as 2.5 times, of the median root mean square amplitude of all preset time windows are identified as artifact periods and removed, thereby improving the purity of subsequent feature calculations. The EEG and EMG signals within each preset time window after removing the artifact time periods are used to perform feature calculations to obtain a preset dimension, such as a 35-dimensional feature vector, used to characterize the brain state.

[0029] The feature calculations include: 1) EEG frequency domain features: Differential filtering is used to remove 1 / f non-periodic components, and the power spectral density of the EEG signal is estimated using the Thomson multi-taper method (four-taper). The power is normalized to relative frequency band power and the logarithm is taken to obtain the power and frequency band power ratio of the δ (0.5-4Hz), θ (4-8Hz), σ (10-14Hz), β (14-30Hz), and γ (30-80Hz) frequency bands, as well as the peak oscillation frequency and temporal coherence of each frequency band; 2) EMG features: Time domain analysis (root mean square amplitude, zero-crossing rate) and frequency domain analysis (high-frequency power) are performed on the EMG signal; 3) Nonlinear dynamic features: Sample entropy, Hjorth parameter, and spectral slope analysis are performed on the EEG and EMG signals; 4) Features inspired by the CATCH-22 standard time series feature set. The above feature calculations were completed in MATLAB R2023a.

[0030] The state space mapping unit is used to normalize the multidimensional features, construct a Euclidean state space of the brain state with the feature centroid of the awake state of the subject as the reference point, and calculate the Euclidean distance from the brain state feature vector of the current time period to the feature centroid of the awake state as the awake distance. Specifically, it is configured as follows: Obtain the feature vectors for each time period output by the feature extraction unit; Identify at least two consecutive wakefulness periods that first occur after a slow-wave sleep state, and determine the start time of the first wakefulness period among the at least two consecutive wakefulness periods as the wakefulness onset time; Using the start time of wakefulness as the time reference, the feature vectors of each time period during the transition from slow-wave sleep to wakefulness are obtained, and the feature vectors of each time period are subjected to global z-score normalization, that is, normalization is performed based on the mean and standard deviation of the feature vectors of all time periods during the transition period to eliminate the differences in the dimensions of each feature. Construct a Euclidean state space with the characteristic centroid of the conscious state as the origin; The current brain state feature vector is mapped to the Euclidean state space, and the Euclidean distance from the current brain state feature vector to the origin is calculated. This Euclidean distance is defined as the awakening distance s(t), i.e., s(t) = ||x(t)||, where x(t) is the current brain state feature vector, quantifying the degree to which the current brain state deviates from the awakening state. The awakening initiation centroid is the median of the feature vectors in the first 5 minutes after the onset of awakening, and the transition period extraction range is from 10 minutes before the onset of awakening to 5 minutes after.

[0031] The dynamics analysis unit is used to construct and identify dynamic models with saddle-node bifurcation characteristics based on the time series of historical wakefulness distances. Through steady-state analysis of the dynamic models, it determines the critical point for the transition from slow-wave sleep to wakefulness and the corresponding critical distance threshold. Specifically, it is configured as follows: Obtain the wakefulness distance time series output by the state space mapping unit; A dynamic model with saddle-node bifurcation characteristics is constructed, which takes the form ds / dt=r+c(t)-h*s-(s / K). 2 Where s is the wakefulness distance, r is the rate of change from the start of wakefulness (corresponding to sleep maintenance drive), K is the maximum wakefulness distance at the slow-wave sleep baseline, c(t) is the control parameter characterizing the wakefulness drive and increases linearly with time at a rate m, i.e., c(t) = c0 + m*t, and h is the scaling factor for adjusting the wakefulness effect, to describe the temporal evolution of the wakefulness distance during the transition from slow-wave sleep to wakefulness, and to use the wakefulness distance time series to perform parameter identification on the dynamic model to obtain model parameters; Steady-state analysis is performed on the dynamic model based on the model parameters to solve for the bifurcation point where the system transitions from bistable to monostable: the equilibrium equation is obtained by solving ds / dt=0, and the stability of the equilibrium point (such as the eigenvalues ​​of the Jacobian matrix) is analyzed. When the upper stable attractor disappears (i.e., the system transitions from bistable to monostable), the critical value of the corresponding control parameter c is the critical control parameter for the transition from slow-wave sleep to wakefulness. The target equilibrium point, i.e. the equilibrium point when ds / dt=0, is calculated based on the critical control parameters. The sobriety distance value corresponding to the target equilibrium point is determined as the critical distance threshold and output to the early warning judgment unit.

[0032] The parameter identification process employs a two-stage optimization strategy: a coarse-grid search is performed within a preset parameter space to obtain candidate parameter combinations; then, a local fine-tuning search is performed based on these candidate parameter combinations to maximize the goodness of fit (Rfit). 2 The joint objective function (R) of minimizing the root mean square error (RMSE) and the objective function of minimizing the root mean square error (RMSE) 2 -RMSE) is used to refine the parameters (r, K, c, h, m) through a local search to determine the model parameters of the dynamic model; the dynamic model is solved using numerical integration methods such as ode45, and the results are based on the coefficient of determination R. 2 The root mean square error (RMSE) was used to evaluate the fitting accuracy of the dynamic model to the sobriety distance time series. The differential equations were solved numerically using MATLAB's ode45 function, and the results were evaluated with the coefficient of determination R0. 2 The fitting accuracy was evaluated. The saddle-node bifurcation dynamic model captured the bistable to monostable dynamic characteristics of the transition from slow-wave sleep to wakefulness. Parameter identification ensured the fitting accuracy between the model and the actual wakefulness distance time series, while steady-state analysis accurately located the transition critical point, providing an objective critical distance threshold for subsequent early warning determination.

[0033] The early warning determination unit is used to compare the real-time awakening distance with the critical distance threshold corresponding to the critical point. When the real-time awakening distance is less than the critical distance threshold and the duration reaches a preset duration, it determines that the intervention opportunity is met and outputs a trigger signal to the closed-loop control module. Specifically, it is configured as follows: Obtain the current sobriety distance output by the state space mapping unit and the critical distance threshold output by the dynamic analysis unit; The current lucidity distance is subjected to causal median filtering with a window length of 10 time periods to suppress temporal fluctuation noise and ensure the accuracy of the lucidity distance; The filtered sobriety distance is compared with the critical distance threshold. When the filtered sobriety distance is less than the critical distance threshold and the state continues for a predetermined time (e.g., 1 minute), it is determined that the intervention opportunity is met, and a trigger signal is output to the closed-loop control module. If the filtered sobriety distance is not less than the critical distance threshold, or if the duration of the filtered sobriety distance being less than the critical distance threshold does not reach the predetermined time, no trigger signal is output, and monitoring continues. In this way, combined with the critical slowing warning signal (increased autocorrelation and variance), the warning determination unit can detect the conversion trend in advance. When the sobriety distance remains below the critical threshold, it determines it is time for intervention and provides a trigger signal to the closed-loop control module. The critical slowing warning signal is calculated as follows: using a sliding window of 50 time periods (approximately 3.3 minutes), after linear detrending within the window, the lag-1 autocorrelation and variance of the residuals are calculated as the early warning signal.

[0034] In one embodiment, the closed-loop control module is configured to: Receive the trigger signal output by the critical point prediction module; In response to the trigger signal, before the brain state evolves to the critical point (i.e. before the transition from slow-wave sleep to wakefulness occurs), an electrical stimulation pulse sequence, such as including 1000 pulses with an intensity of 20uA and a pulse width of 5ms, is applied to the parafacial nucleus of the facial nerve. This promotes the maintenance of slow-wave sleep and blocks the transition process by electrically activating the γ-aminobutyric acid-producing neurons of the parafacial nucleus of the facial nerve. During the application of the electrical stimulation pulse sequence, EEG and EMG signals are continuously acquired, and brain state monitoring results generated by the critical point prediction module based on the continuously acquired EEG and EMG signals are obtained. When the brain state monitoring results indicate that the current wakefulness distance is greater than the critical distance threshold and this state persists for a preset duration, the application of the electrical stimulation pulse sequence is terminated to complete a single predictive intervention. Based on the signal from the warning judgment unit, this module applies electrical stimulation before the transition threshold, activating γ-aminobutyric acid (GABAergic) neurons in the parafacial nucleus of the facial nerve to inhibit arousal drive, thereby maintaining slow-wave sleep. During stimulation, the wakefulness distance is continuously monitored, and stimulation is terminated when the wakefulness distance exceeds the critical threshold, achieving a closed-loop regulation of "prediction-intervention-feedback" to avoid overstimulation.

[0035] Based on the same concept, this embodiment also provides a slow-wave sleep closed-loop regulation method based on brain state critical point prediction, such as... Figure 2 As shown, it includes: Continuous acquisition of electroencephalogram (EEG) and electromyogram (EMG) signals from the subjects; The electroencephalogram (EEG) and electromyogram (EMG) signals are processed in real time to extract multidimensional features representing brain states, construct a Euclidean state space for brain states, and calculate the awakening distance from the current brain state to the awake state. The temporal evolution of the wakefulness distance is fitted based on a dynamic model with saddle-node bifurcation characteristics. Combined with critical slowing warning indicators, the critical point of transition from slow-wave sleep to wakefulness and the timing of intervention are determined. In response to the timing of the intervention, a predictive closed-loop electrical stimulation of the parafacial nucleus is triggered before the brain state evolves to the critical point of transition from slow-wave sleep to wakefulness, in order to maintain the slow-wave sleep state.

[0036] Those skilled in the art will understand that the regulation method of this embodiment includes an offline training phase and a real-time closed-loop phase. The offline training phase includes: acquiring EEG and EMG signals from the subject's historical records, extracting multidimensional features representing brain states; constructing a Euclidean state space for brain states, calculating the historical wakefulness distance time series; fitting the temporal evolution of the historical wakefulness distance based on a dynamic model with saddle node bifurcation characteristics, and determining the critical point and corresponding critical distance threshold for the transition from slow-wave sleep to wakefulness through steady-state analysis of the dynamic model. The above-mentioned offline training phase constructs a dynamic model and determines the critical point using historical data, providing model parameters and critical distance thresholds for the real-time closed-loop phase, ensuring the accuracy of real-time prediction.

[0037] The real-time closed-loop phase includes: continuously collecting EEG and EMG signals from the subjects, extracting multidimensional features representing brain state, and calculating the current awake distance; comparing the current awake distance with the critical distance threshold in conjunction with a critical slowing warning index; determining that intervention is timely when the current awake distance is less than the critical distance threshold and the duration reaches a preset duration; responding to the intervention timing, applying predictive closed-loop electrical stimulation to the parafacial nucleus of the facial nerve before the brain state evolves to the critical point; continuously monitoring the current awake distance during the application of electrical stimulation; terminating the application of electrical stimulation when the current awake distance is greater than the critical distance threshold and the duration reaches a preset duration to maintain slow-wave sleep. The above real-time closed-loop phase achieves a closed-loop process of "real-time prediction - early intervention - feedback termination" by continuously collecting signals, calculating the awake distance, and comparing the critical threshold, ensuring that awakening is blocked before the transition critical point and maintaining the continuity of slow-wave sleep.

[0038] Based on the same concept, the present invention provides an electronic device, comprising: a memory for storing a processing program; and a processor, wherein the processor executes the processing program to implement the steps of the slow-wave sleep closed-loop regulation method based on brain state critical point prediction as described in Embodiment 2.

[0039] In some embodiments of this application, a readable storage medium is also provided, on which a processing program is stored, which, when executed by a processor, implements the steps of the slow-wave sleep closed-loop regulation method based on brain state critical point prediction as described in Embodiment 2.

[0040] It is understood that, for the aforementioned slow-wave sleep closed-loop regulation system based on brain state critical point prediction, if all of them are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0041] Computer-readable storage media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0042] The program code for executing the technical solutions disclosed in this application can be written in any combination of one or more programming languages. These programming languages ​​include object-oriented programming languages—such as Java and C++—and conventional procedural programming languages—such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A slow-wave sleep closed-loop regulation system based on brain state critical point prediction, characterized in that, include: The neural signal acquisition module is used to continuously acquire electroencephalogram (EEG) and electromyogram (EMG) signals from the subjects and to apply electrical stimulation at the parafacial nucleus of the facial nerve. The critical point prediction module is used to process the EEG and EMG signals in real time, extract multidimensional features representing the brain state, construct the Euclidean state space of the brain state, calculate the awake distance from the current brain state to the awake state, and fit the time evolution of the awake distance based on the dynamic model with the bifurcation characteristics of the saddle node. Combined with the critical slowing warning index, the critical point of the transition from slow wave sleep to wakefulness and the timing of intervention are determined. A closed-loop control module is used to respond to the intervention timing by triggering predictive closed-loop electrical stimulation applied to the parafacial nucleus of the face before the brain state evolves to the critical point of transition from slow-wave sleep to wakefulness, in order to maintain the slow-wave sleep state.

2. The slow-wave sleep closed-loop regulation system based on brain state critical point prediction according to claim 1, characterized in that, The neural signal acquisition module includes: The cranial nail electrode, the EEG acquisition electrode fixed to the anterior and posterior fontanelle regions of the skull of the subject, and the electromyography acquisition electrode fixed to the neck muscles, are used to acquire the EEG signals and electromyography signals. A deep electrode is implanted into the parafacial nucleus of the facial nerve in the subject via stereotactic positioning, and is used to apply the electrical stimulation under the control of the closed-loop control module.

3. The slow-wave sleep closed-loop regulation system based on brain state critical point prediction according to claim 1, characterized in that, The critical point prediction module includes: The feature extraction unit is used to filter and preprocess the real-time acquired EEG and EMG signals, and segment them according to a preset time window to extract multidimensional features that characterize brain state. The state space mapping unit is used to normalize the multidimensional features, construct the Euclidean state space of the brain state with the feature centroid of the awake state of the subject as the reference point, and calculate the Euclidean distance from the brain state feature vector of the current time period to the feature centroid of the awake state as the awake distance. The dynamic analysis unit is used to construct and identify dynamic models with saddle node bifurcation characteristics based on the time series of historical wakefulness distances. By performing steady-state analysis on the dynamic model, the critical point for the transition from slow-wave sleep to wakefulness and the corresponding critical distance threshold are determined. The early warning determination unit is used to compare the real-time awakening distance with the critical distance threshold corresponding to the critical point. When the real-time awakening distance is less than the critical distance threshold and the duration reaches the preset duration, it determines that the intervention opportunity is met and outputs a trigger signal to the closed-loop control module.

4. The slow-wave sleep closed-loop regulation system based on brain state critical point prediction according to claim 3, characterized in that, The feature extraction unit is configured as follows: The real-time acquired EEG and EMG signals are filtered by a first bandpass filter and a second bandpass filter corresponding to the physiological frequency band, respectively. The first bandpass filter and the second bandpass filter are zero-phase preset order Butterworth filters. The filtered EEG and EMG signals are segmented according to a preset time window, wherein adjacent preset time windows have a preset overlap rate. Calculate the root mean square amplitude of EEG and EMG signals within each preset time window, and identify preset time windows whose root mean square amplitude exceeds a preset multiple of the median root mean square amplitude of all preset time windows as artifact periods and remove them. The EEG and EMG signals within each preset time window after removing the artifact time periods are used to perform feature calculations to obtain a preset dimension feature vector for characterizing the brain state. The feature calculation includes performing spectral analysis on the EEG signal to obtain EEG frequency domain features, performing time domain analysis and frequency domain analysis on the electromyography signal to obtain electromyography features, and performing nonlinear dynamic analysis on at least one of the EEG signal and electromyography signal to obtain nonlinear dynamic features.

5. The slow-wave sleep closed-loop regulation system based on brain state critical point prediction according to claim 3, characterized in that, The state space mapping unit is configured as follows: Obtain the feature vectors for each time period output by the feature extraction unit; Identify at least two consecutive wakefulness periods that first occur after a slow-wave sleep state, and determine the start time of the first wakefulness period among the at least two consecutive wakefulness periods as the wakefulness onset time; Using the moment of awakening as the time reference, feature vectors for each time period during the transition from slow-wave sleep to awakening are obtained, and global normalization is performed on the feature vectors for each time period. Construct a Euclidean state space with the characteristic centroid of the conscious state as the origin; The brain state feature vector of the current time period is mapped to the Euclidean state space, and the Euclidean distance from the brain state feature vector of the current time period to the origin is calculated. The Euclidean distance is defined as the awake distance to quantify the degree to which the current brain state deviates from the awake state.

6. The slow-wave sleep closed-loop regulation system based on brain state critical point prediction according to claim 3, characterized in that, The dynamic analysis unit is configured as follows: Obtain the wakefulness distance time series output by the state space mapping unit; A dynamic model with saddle-node bifurcation characteristics is constructed to describe the temporal evolution of the wakefulness distance during the transition from slow-wave sleep to wakefulness, and the wakefulness distance time series is used to identify the parameters of the dynamic model to obtain the model parameters. Steady-state analysis is performed on the dynamic model based on the model parameters to solve for the bifurcation point where the system changes from bistable to monostable. The control parameters corresponding to the bifurcation point are determined as the critical control parameters for the transition from slow-wave sleep to wakefulness. The target equilibrium point is calculated based on the critical control parameters, and the sobriety distance value corresponding to the target equilibrium point is determined as the critical distance threshold.

7. The slow-wave sleep closed-loop regulation system based on brain state critical point prediction according to claim 3, characterized in that, The dynamic analysis unit is configured to perform parameter identification using a two-stage optimization strategy, which includes: A coarse-grid search is performed within a preset parameter space to obtain candidate parameter combinations; Based on the candidate parameter combination, a local refinement search is performed to determine the model parameters of the dynamic model by optimizing the joint objective function of maximizing the goodness of fit and minimizing the root mean square error. The dynamic model is solved using a numerical integration method, and the fitting accuracy of the dynamic model to the sobriety distance time series is evaluated based on the coefficient of determination and root mean square error.

8. The slow-wave sleep closed-loop regulation system based on brain state critical point prediction according to claim 3, characterized in that, The early warning determination unit is configured as follows: Obtain the current sobriety distance output by the state space mapping unit and the critical distance threshold output by the dynamic analysis unit; Causal median filtering is applied to the current lucidity distance to suppress temporal fluctuation noise; The filtered sobriety distance is compared with the critical distance threshold. When the filtered distance to wakefulness is less than the critical distance threshold and the state continues for a predetermined time, it is determined that the intervention opportunity is met, and a trigger signal is output to the closed-loop control module. Monitoring will continue if the filtered sobriety distance is not less than the critical distance threshold, or if the duration for which the filtered sobriety distance is less than the critical distance threshold does not reach the predetermined time.

9. The slow-wave sleep closed-loop regulation system based on brain state critical point prediction according to claim 1, characterized in that, The closed-loop control module is configured as follows: Receive the trigger signal output by the critical point prediction module; In response to the trigger signal, an electrical stimulation pulse sequence is applied to the parafacial nucleus of the facial nerve before the brain state evolves to the critical point; During the application of the electrical stimulation pulse sequence, EEG and EMG signals are continuously acquired, and brain state monitoring results generated by the critical point prediction module based on the continuously acquired EEG and EMG signals are obtained. When the brain state monitoring results indicate that the current awake distance is greater than the critical distance threshold and the state continues for a preset duration, the application of the electrical stimulation pulse sequence is terminated.

10. A method for slow-wave sleep closed-loop regulation based on brain state critical point prediction, characterized in that, include: Continuous acquisition of electroencephalogram (EEG) and electromyogram (EMG) signals from the subjects; The electroencephalogram (EEG) and electromyogram (EMG) signals are processed in real time to extract multidimensional features representing brain states, construct a Euclidean state space for brain states, and calculate the awakening distance from the current brain state to the awake state. The temporal evolution of the wakefulness distance is fitted based on a dynamic model with saddle-node bifurcation characteristics. Combined with critical slowing warning indicators, the critical point of transition from slow-wave sleep to wakefulness and the timing of intervention are determined. In response to the timing of the intervention, a predictive closed-loop electrical stimulation of the parafacial nucleus is triggered before the brain state evolves to the critical point of transition from slow-wave sleep to wakefulness, in order to maintain the slow-wave sleep state.

Images