Closed-loop regulation and control method and system based on autonomic nerve function elasticity dynamic evaluation

Abstract

The invention provides a closed-loop regulation and control method and system based on elasticity dynamic evaluation of autonomic nerve function, and relates to the technical field of biomedical engineering and digital therapy, and the method comprises the following steps: step 1, collecting electrocardiosignals of a user in real time, and preprocessing the electrocardiosignals to obtain a clean RR interval sequence; 2, extracting multi-dimensional heart rate variability characteristics based on the RR interval sequence, and calculating an elasticity index of the autonomic nerve function according to the multi-dimensional heart rate variability characteristics; 3, comparing the elasticity index with a dynamic threshold value preset according to the current state of the user, judging an instability risk level, and generating a control instruction containing a stimulation type and parameters; and 4, triggering at least one non-invasive nerve regulation stimulation based on the control instruction. According to the invention, a multi-modal nerve regulation and control closed loop is driven based on real-time quantification of a core physiological index of autonomic nerve function elasticity, so that the mode conversion from passive evaluation to active intervention is realized.

Description

Technical Field

[0001] This invention relates to the fields of biomedical engineering and digital therapy, and in particular to a closed-loop regulation method and system based on dynamic assessment of the elasticity of autonomic nervous function. Background Technology

[0002] Micro-arousals and attentional distraction are two common state disorders affecting core human quality of life. Micro-arousals are brief cortical arousals lasting 3-15 seconds during sleep, disrupting sleep continuity and being a major factor contributing to daytime fatigue and insomnia. Attentional distraction, on the other hand, is characterized by the involuntary detachment of attention from the current focus during waking tasks and is closely related to weakened cognitive control.

[0003] Existing technologies for addressing these issues are limited and fragmented in their approaches. In managing micro-arousal, the clinical gold standard, polysomnography (PSG), can only perform post-event analysis and cannot achieve predictive intervention. In managing attention loss, mainstream behavioral cognitive training and neurofeedback technologies require long-term active user participation, resulting in low compliance, while drug intervention has issues with side effects and tolerability.

[0004] More importantly, existing technological solutions treat micro-awakening and attention deficit as independent issues belonging to different clinical departments, failing to recognize the underlying shared physiological and pathological core—an imbalance in the functional elasticity of the autonomic nervous system. This system is the center for maintaining homeostasis, and a decline in its functional elasticity (i.e., the system's ability to recover homeostasis after being subjected to stress) is a common physiological bridge connecting sleep disorders and cognitive deficits. When elasticity decreases, the nervous system is more susceptible to micro-awakening triggered by minor disturbances during sleep, and also more prone to attentional regulation failures due to cognitive load when awake.

[0005] While heart rate variability (HRV) is the gold standard non-invasive indicator for assessing autonomic nervous system function, its application has long been limited to macroscopic stress assessment or long-term health trend analysis. Current technologies have failed to utilize the dynamic characteristics of HRV as a real-time control signal to predict and simultaneously suppress microscopic physiological events (micro-arousals) and cognitive events (attention deficit).

[0006] Therefore, there is an urgent need for a closed-loop regulation method and system based on dynamic assessment of the elasticity of autonomic nervous function to achieve micro-arousal and inhibition of attentional distraction, so as to improve the core quality of human life. Summary of the Invention

[0007] One of the objectives of this invention is to provide a closed-loop regulation method and system based on dynamic assessment of autonomic nervous system functional elasticity. Based on the scientific concept that "autonomic nervous system functional elasticity is a common physiological bridge connecting the two," this invention drives a multimodal neural regulation closed loop by quantifying the core physiological indicator of "autonomic nervous system functional elasticity" in real time, thereby realizing a shift from "passive assessment" to "active intervention."

[0008] The present invention provides a closed-loop regulation method based on dynamic assessment of autonomic nerve function elasticity. Step 1: Real-time acquisition of the user's electrocardiogram signal and preprocessing it to obtain a clean RR interval sequence. Step 2: Based on the RR interval sequence, extract multidimensional heart rate variability features and calculate the elasticity index of autonomic nervous function accordingly. Step 3: Compare the elasticity index with a preset dynamic threshold based on the user's current state to determine the level of instability risk and generate control instructions that include stimulus type and parameters. Step 4: Based on the control command, trigger at least one non-invasive neuromodulation stimulus; Among them, neuromodulation stimulation includes transcutaneous vagal nerve stimulation and phase-locked acoustic stimulation; the elasticity index is a composite index that integrates at least three physiological dimensions: basic vagal tone, stress recovery speed, and system complexity.

[0009] Preferably, the calculation steps for the elasticity index of autonomic nervous system function specifically include: Calculate the time-domain characteristics, frequency-domain characteristics, and nonlinear characteristics from the RR interval sequence; Based on time-domain features, frequency-domain features, and nonlinear features, physiological indicators of three dimensions—basic vagal tone, stress recovery speed, and system complexity—are synthesized. Physiological indicators of three dimensions—basic vagal tone, stress recovery speed, and system complexity—are used as input features. These indicators are then nonlinearly integrated using a machine learning fusion model pre-trained on a clinical dataset to output the elasticity index of autonomic nerve function. Among them, the basic vagal tone is synthesized by weighting the normalized root mean square difference of the RR interval with the high-frequency power; the stress recovery speed is obtained by quantifying the recovery dynamics of heart rate variability after a preset stress event; and the system complexity is characterized by the entropy value of the RR interval sequence at a specific scale in multi-scale entropy analysis.

[0010] Preferably, after step one and before step two, a security screening step is also included: Calculate the full width at half maximum (FWHM) of the peak interval histogram of the RR intervals. If FWHM < 10 ms, the corresponding risk sub-score for cardiac interference is 0.4. Calculate the power value in the ultra-low frequency band. If the power is greater than 5ms...2 / Hz, corresponding to a cardiac interference risk sub-score of 0.3; York knife test was used to analyze the determinism of RR interval sequences. If the p-value was >0.05, the corresponding risk sub-score for cardiac interference was 0.3. The above sub-scores are summed to obtain a comprehensive risk score. If the comprehensive risk score is ≥0.85, the identification of cardiac disease interference is considered successful. After successful identification, the subsequent intervention process is immediately terminated, a security alarm is sent to the user terminal, and the RR interval sequence, characteristic indicators and risk score data of this screening are automatically recorded.

[0011] Preferably, the parameters of the neural modulation stimulation are dynamically adapted, as follows: During sleep, the type, frequency, and intensity of transcutaneous vagus nerve stimulation and acoustic stimulation are adapted according to sleep stage, risk of micro-arousals, and core pathological dimensions of the disease: During deep sleep, transcutaneous vagus nerve stimulation uses an ultra-low frequency of 0.1-1Hz to maintain stimulation, with an intensity of 40%-65% of the user's comfort threshold. For Parkinson's disease and obstructive sleep apnea, a sub-range within the frequency range and a pulse width of 200-500μs can be adapted to enhance the regulation of the pathological dimensions corresponding to the disease. During the sleep onset period, 10-15Hz periodic pulse transcutaneous vagus nerve stimulation is used, which can be superimposed with 40-45dB SPL low-intensity white noise. A single continuous use is ≤15 minutes, and the transcutaneous vagus nerve stimulation intensity is 40%-50% of the comfort threshold, which is suitable for the physiological characteristics of the sleep onset period. During sleep, transcutaneous vagus nerve stimulation uses 0.5-1Hz ultra-low frequency stimulation at an intensity of 30%-40% of the comfort threshold. Acoustic stimulation can be selectively activated according to disease needs, with pink noise being the preferred choice and low-intensity white noise being superimposed for PTSD patients. During the pre-arousal period, trigger 10-15Hz short-pulse interventional transcutaneous vagal nerve stimulation for 5-8 seconds at an intensity of 60%-70% of the comfort threshold. For obstructive sleep apnea, acoustic stimulation can be added simultaneously to help maintain physiological homeostasis. In a conscious state, the stimulus type and parameters are adapted based on the cognitive, emotional state, and disease type indicated by the elasticity index: When predicting attention loss, the phase-locked acoustic stimulation that is locked to the user's ECG R wave is triggered first, with an R wave delay of 200-250ms and an intensity of 55-65dB SPL. When modulating diseases specifically, for attention deficit hyperactivity disorder, depression, mild cognitive impairment, and post-traumatic stress disorder, suitable acoustic stimulation types include: binaural melodic, pure tone alternation, or transcutaneous vagus nerve stimulation frequency and intensity range, which can be used to target and improve disease-related autonomic dysfunction. Configure unified security constraints, specifically including: When the elasticity index is <0.3, the intensity of transcutaneous vagus nerve stimulation is uniformly limited to 0.5-1.0 mA; When the LF / HF ratio is greater than 3, high-frequency stimulation above 10Hz is paused and switched to the 0.1-1Hz ultra-low frequency maintenance mode. The acoustic stimulation intensity should not exceed 65 dB SPL, and the transcutaneous vagus nerve stimulation current intensity should not exceed 5.0 mA.

[0012] The present invention also provides a closed-loop control system based on dynamic assessment of autonomic nervous function elasticity, comprising: a signal acquisition module, an elasticity calculation module, an intelligent decision-making module, and a multimodal stimulation module; The system includes a signal acquisition module that acquires the user's electrocardiogram (ECG) signal in real time and preprocesses it to obtain a clean RR interval sequence; an elasticity calculation module that extracts multi-dimensional heart rate variability features based on the RR interval sequence and calculates the elasticity index of autonomic nervous function accordingly; an intelligent decision-making module that compares the elasticity index with a preset dynamic threshold based on the user's current state to determine the level of instability risk and generates control instructions containing stimulation type and parameters; and a multimodal stimulation module that triggers at least one non-invasive neuromodulation stimulation based on the control instructions. Among them, neuromodulation stimulation includes transcutaneous vagal nerve stimulation and phase-locked acoustic stimulation; the elasticity index is a composite index that integrates at least three physiological dimensions: basic vagal tone, stress recovery speed, and system complexity.

[0013] Preferably, the closed-loop control system based on dynamic assessment of autonomic neural function elasticity further includes: a personalized adaptive engine; the personalized adaptive engine is configured as follows: Record users' historical physiological responses to stimuli; Based on recorded historical physiological response data, the model parameters of the elasticity calculation module, the dynamic threshold of the intelligent decision-making module, and / or the stimulation parameters of the multimodal stimulation module are dynamically optimized using a Bayesian optimization algorithm.

[0014] Preferably, the multimodal stimulation module is connected to a wearable stimulation terminal, which integrates a transcutaneous vagus nerve stimulation unit and an acoustic stimulation unit. Wearable stimulation terminals include, but are not limited to, ear-worn, wrist-worn, and chest-worn devices.

[0015] Preferably, the acoustic stimulation unit includes: a non-air-conducting loudspeaker, an air-conducting loudspeaker, or a combination thereof.

[0016] Preferably, it is implemented on a wearable device platform, which includes a physiological signal sensing subject and a wearable stimulation terminal, and the two interact with each other via wireless communication.

[0017] Preferably, the stimulation parameters of the percutaneous vagus nerve stimulation unit include: a current intensity range of 0.1-5.0 mA, a stimulation frequency range of 0.1-30 Hz, and a pulse width of 200-500 μs.

[0018] Preferably, the closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity further includes an assessment module. The evaluation module performs the following operations: When the elasticity index increases by ≥0.2, the baseline vagal tone increases by ≥0.15, the stress recovery speed increases by ≥0.1, and the system complexity increases by ≥0.1, it is considered that the physiological dimensions are effective. If the score on the clinical symptom scoring scale decreases by a predetermined amount, it is considered effective in the clinical dimension. If both the physiological and clinical dimensions are effective, output the conclusion that the intervention is effective; otherwise, output optimization suggestions that require adjustment of the intervention parameters.

[0019] The present invention has the following beneficial effects: Fundamental: For the first time, it starts from the unified physiological core of "autonomic nervous system functional elasticity" and addresses sleep and cognitive problems simultaneously, achieving a "root cause treatment" rather than a "symptom treatment".

[0020] Foresight and Synergy: It achieves "pre-emptive" inhibition of micro-awakening (triggered 0.5-1 seconds before the event) and "instant" intervention for inattention, breaking down traditional disciplinary barriers.

[0021] Precise and personalized: HRV is used as a real-time biofeedback signal and continuously optimized through machine learning to ensure that the timing, mode and intensity of intervention are optimal for each individual.

[0022] Closed-loop automation: It constructs a complete "perception-decision-intervention" closed loop, requiring no active user participation throughout the process, resulting in a high user experience and compliance.

[0023] Intrinsic safety: By screening for rhythmic source imbalances, the risk of providing incorrect interventions to unsuitable populations is fundamentally avoided.

[0024] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0025] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0026] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a closed-loop regulation method based on dynamic assessment of autonomic nervous function elasticity in an embodiment of the present invention. Figure 2 This is a schematic diagram of a closed-loop regulation system based on dynamic assessment of autonomic nervous function elasticity in an embodiment of the present invention. Figure 3 This is a schematic diagram of the hardware architecture of a device that can be implemented in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the quantitative calculation of the elasticity index of autonomic nervous function in an embodiment of the present invention. Figure 5 This is a schematic diagram illustrating the differences in deployment forms in embodiments of the present invention; Figure 6 This is a schematic diagram of the triple security analysis in an embodiment of the present invention; Figure 7 This is a schematic diagram illustrating the principle of the optimized triggering mechanism in an embodiment of the present invention; Figure 8 This is a schematic diagram of another closed-loop regulation method based on dynamic assessment of autonomic nervous function elasticity in an embodiment of the present invention. Detailed Implementation

[0027] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0028] This invention provides a closed-loop regulation method and system based on dynamic assessment of autonomic nervous system functional elasticity, such as... Figure 1 As shown, it includes: Step 1: Acquire the user's electrocardiogram (ECG) signal in real time and preprocess it to obtain a clean RR interval sequence; Step 2: Based on the RR interval sequence, extract multidimensional heart rate variability features and calculate the elasticity index of autonomic nervous function accordingly. Step 3: Compare the elasticity index with a preset dynamic threshold based on the user's current state to determine the level of instability risk and generate control instructions that include stimulus type and parameters. Step 4: Based on the control command, trigger at least one non-invasive neuromodulation stimulus; Among them, neuromodulation stimulation includes transcutaneous vagus nerve stimulation (tVNS) and phase-locked acoustic stimulation; the Harmonia Index (HI index or Hi) is a composite index that integrates at least three physiological dimensions: baseline vagal tone (D1), stress recovery speed (D2), and system complexity (D3).

[0029] The calculation steps for the elasticity index of autonomic nervous system function specifically include: Calculate the time-domain characteristics, frequency-domain characteristics, and nonlinear characteristics from the RR interval sequence; Based on time-domain features, frequency-domain features, and nonlinear features, physiological indicators of three dimensions—basic vagal tone, stress recovery speed, and system complexity—are synthesized. Physiological indicators of three dimensions—basic vagal tone, stress recovery speed, and system complexity—are used as input features. These indicators are then nonlinearly integrated using a machine learning fusion model pre-trained on a clinical dataset to output the elasticity index of autonomic nerve function. Among them, the basic vagal tone is synthesized by weighting RMSSD and high-frequency power according to fixed weight coefficients; the stress recovery speed is obtained by quantifying the recovery kinetic slope of heart rate variability after a preset stress event; and the system complexity is characterized by the entropy value of the RR interval sequence at a specific scale in multi-scale entropy analysis.

[0030] The process, following step one and preceding step two, also includes a safety screening step. Rhythm source imbalance analysis was performed on the RR interval sequence. Cardiac disease interference was identified by its temporal regularity, frequency domain formant characteristics, and nonlinear deterministic features. If the identification was successful, the subsequent intervention process was terminated.

[0031] Among them, the parameters of the neural modulation stimulation are dynamically adapted; During sleep, the frequency and intensity of transcutaneous vagus nerve stimulation are adapted according to sleep stage and the risk of micro-arousal. During deep sleep, ultra-low frequency maintenance stimulation of 0.1-1Hz is used, and short pulse intervention stimulation of 10-15Hz is used during the pre-arousal period. In a conscious state, when the elasticity index indicates a predicted loss of attention, phase-locked acoustic stimulation that is locked to the user's ECG R wave is preferentially triggered.

[0032] The closed-loop regulation method based on dynamic assessment of autonomic nervous system functional elasticity of the present invention is based on the scientific concept that "autonomic nervous system functional elasticity is a common physiological bridge connecting the two." By quantifying the core physiological indicator of "autonomic nervous system functional elasticity" in real time, it drives a multimodal neural regulation closed loop, realizing the transformation from "passive assessment" to "active intervention".

[0033] The acquisition equipment and parameters for ECG data acquisition and preprocessing are as follows: Core components: Adaptable flexible patch single / dual-lead ECG sensor (dry electrode / wet electrode / flexible electrode), supporting I2C / SPI / UART universal communication protocols, electrode material is medical-grade Ag / AgCl, skin contact impedance ≤5kΩ; Key parameters: Consumer-grade device sampling rate ≥128Hz (quantization error ≤±1LSB), medical-grade ≥256Hz (quantization error ≤±0.5LSB), compliant with IEC 60601-2-27 medical electrical equipment standard and ISO 10993 biocompatibility standard; Expansion module: Integrated triaxial IMU (range ±2g / ±8g, sampling rate 100Hz) for body motion sensing, optional PPG blood oxygen sensor (sampling rate 50Hz, blood oxygen saturation measurement range 70%-100%) to assist multimodal calibration. Preprocessing Flow (Reproducible Engineering Steps): Filtering: A 4th-order Butterworth bandpass filter (0.5–40Hz) is used, superimposed with an adaptive notch filter (50 / 60Hz power frequency interference suppression, attenuation ≥40dB); R-wave detection: For consumer-grade applications, an improved Pan-Tompkins algorithm is used (dynamic threshold adjustment is introduced, false detection rate ≤3%), while for medical-grade applications, a lightweight CNN model is used (input is a 32×32 RR interval sequence slice, inference accuracy ≥98.5%); RR interval screening: Extreme values ​​are removed based on the 3σ criterion (300–2000ms are retained), and missing values ​​are filled using linear interpolation (when the missing value rate is ≤5%). Effective sinus RR interval percentage: Consumer-grade ≥85%, Medical-grade ≥95%; Signal quality assessment: Classified by a dual index of "signal-to-noise ratio (SNR≥15dB) + R-wave detection rate (≥95%)", Grade A (SNR≥25dB), Grade B... Level 1 (15-25dB), Level C (<15dB). Level C signals are for reference only and do not trigger intervention.

[0034] Safety screening steps (triple-check closed loop): Rhythm source imbalance analysis is performed on the RR interval sequence to identify cardiac disease interference; specifically: Time domain analysis: The peak interval histogram full width at half maximum (FWHM) is calculated using a sliding window (30 RR intervals / window). Risk is marked when FWHM < 10ms for 3 consecutive windows. Frequency domain analysis: The power spectrum was fitted using an AR model (order p=16), and the power in the ultra-low frequency band (VLF: 0–0.0033Hz) was >5ms. 2 / Hz and resonance peak width > 0.001Hz are flagged as risky; Nonlinear analysis: Poincaré plot SD1 / SD2 ratio < 0.5 + Yorkshire knife test p-value > 0.05, jointly labeled with cardiac interference; Decision-making and emergency response logic: Risk scoring: A comprehensive score (time domain weight 0.4, frequency domain 0.3, nonlinearity 0.3) is calculated using a random forest binary classification model (n_estimators=100, maximum depth 8); Intervention lockout: When the score is ≥0.85, the stimulus output is cut off within 10μs, with dual warnings via device indicator light (solid red) + APP push (SMS backup); Recovery mechanism: The intervention function can only be restored after the user restarts the device and completes a secondary safety screening (score ≤0.6). The specific process is as follows... Figure 6 As shown, the sub-score is calculated by triple verification of time domain, frequency domain, and power nonlinearity, and the sum is used to obtain the comprehensive risk score; when the score is ≥0.85, the intervention is terminated immediately and an alarm is sent.

[0035] For multi-dimensional HRV feature extraction and HI index calculation, multimodal HRV feature extraction (30 seconds / analysis window); temporal features: SDNN, RMSSD, pNN50, SDSD, Mean RR (calculation accuracy ≤ ±0.1ms); frequency domain features: LF power (0.04–0.15Hz), HF power (0.15–0.4Hz), LF / HF ratio (FFT points 256, frequency resolution ≤ 0.5Hz); nonlinear features (innovative dual-entropy fusion model, different from traditional single entropy value evaluation, tested on 10-20 healthy volunteers, entropy stability ICC=0.87): multi-scale entropy (MSE): 2–5 scales are calculated in normal state, and 2–3 scales are calculated in low power mode, using "sliding window binning + probability density fitting". Dynamic adjustment of analysis window length: When the target number of RR intervals is >100, the window length = 100 x (60 / real-time heart rate) seconds. When real-time requirements are high, it can be shortened to >50 RR intervals, and the MSE scale is adjusted to 2-3 simultaneously. Synthesis of three-dimensional physiological indicators (Min-Max standardized to the 0–1 range): D1 (Basic Vagal Tone): A weighted composite of two classical HRV indices highly dependent on the vagus nerve. Wherein, RMSSD: root mean square of the difference between adjacent RR intervals; HF power: power spectral density in the high-frequency band (0.15-0.40 Hz). Composite formula: Basic tone index = α; In the formula, and For pre-configured fixed weight coefficients, , α represents the root mean square of the difference between adjacent RR intervals and the power spectral density after standardization, respectively; the weights can be verified by multiple linear regression of 200 clinical data (aged 18-65 years), R²=0.82, p<0.001, reflecting the basic activity of the vagus nerve; D2 (Stress Recovery Rate): Focuses on the dynamic changes in high-frequency power after a disturbance event; identifies... The instantaneous drop in power serves as a marker of a disturbance event. It can be categorized into real-time calculation and long-term analysis. Real-time calculation involves calculating... The recovery slope of power within a specific time window after the event. Synthesis formula: Stress Recovery Index = ; : Indicates that after a disturbance event occurs, Linear trend of power change over time. For calculations over long periods (more than 300 RR intervals); define the baseline period ( ): Data from the period before death should be used to define the recovery period ( ): A period of time after the stress ends For example, data within 60 seconds; calculate the baseline mean heartbeat interval. Recovery time modeling (τ): By fitting recovery period data to baseline mean heart rate interval The process involves solving for the recovery time constant τ, and the relevant formula is: In the formula, This is the offset after stress. The time constant is D2. The formula for calculating D2 is as follows: Or; where, Maximum recovery time constant; This indicates the normalization process for the data; if 60 seconds is used as the maximum reference, the formula can also be... ; D3 (System Complexity): Implemented by weighted fusion using the dual entropy algorithm (MSE+SE). The specific steps are as follows: Step 1: Preprocess the RR interval sequence (length ≥ 100), remove outliers with RR interval difference > 50ms to ensure sequence continuity; Step 2: Multi-scale entropy (MSE) calculation (1) Signal quality judgment: Calculate the RMSSD value of the sequence. If RMSSD < 20ms (low signal quality), select the scale range s = 2-3; if RMSSD ≥ 20ms (normal signal quality), select the scale range s = 2-5; 2) Coarse-grained processing: For each scale s, according to the formula Generate coarse-grained sequences (where Xi is the original RR interval value, (3) Sample entropy calculation: For each coarse-grained sequence, the sample entropy is calculated using embedding dimension m=2 and similarity tolerance r=0.2× sequence standard deviation as parameters; (4) MSE value determination: The mean of sample entropy at all scales is taken as the final MSE value; Step 3: Sample entropy (SE) calculation For the preprocessed original RR interval sequence, the sample entropy at a single scale is calculated directly using embedding dimension m=2 and similarity tolerance r=0.2× sequence standard deviation as parameters to obtain the SE value; Step 4: Standardization processing To eliminate dimensional differences, the MSE and SE values ​​are mapped to the 0-1 interval: ; ; Step 5: Weighted fusion is performed by summing the values ​​according to preset weights to obtain the D3 value: D3 = 0.6 × MSEnorm + 0.4 × SEnorm. The final D3 value is limited to the range of 0-1.

[0036] In addition, during the preprocessing in step one, for sequence lengths of 50-99, the MSE scale is fixed at 2-3; for sequence lengths <50, only scale 2 is retained. In this embodiment, the beneficial effects of the dual-entropy algorithm are as follows: by covering multi-scale global features with MSE and supplementing single-scale local features with SE, the fused D3 value improves the accuracy of identifying the complexity of the autonomic nervous system and enhances the robustness to low-quality RR interval sequences, effectively solving the problem that the single entropy index is easily affected by noise.

[0037] HI index nonlinear integration (fully aligned with PSG-HI1.0.PY); Model parameters: GBR model (n_estimators=120, max_depth=6, learning_rate=0.05, minimum number of sample splits=10); Training data: The model is jointly trained based on "public standard database + clinical labeled data". The labeling work is completed by more than 2 associate chief physicians, and the labeling accuracy is ≥97%; Performance indicators: consumer-grade hardware inference latency ≤150ms, medical-grade ≤50ms; HI index value is 0.0–1.0. If it exceeds the range, it is corrected by the average of the adjacent 10 seconds (correction error ≤±0.025); Validation results: In the test set (n=300), the HI index distinguishes between "stable / unstable" states with AUC=0.93, sensitivity=89%, and specificity=91%. Risk grading (personalized threshold adjustment); High risk (instability): HI < 0.3 (initial threshold, the adaptive engine can dynamically adjust it in the range of 0.25–0.35); Medium risk (borderline): 0.3 ≤ HI < 0.7; Low risk (stable): HI ≥ 0.7; Threshold calibration: The initial threshold is based on the baseline threshold of the population. After 10 interventions, the adaptive engine updates the personalized threshold based on user response data. The quantitative calculation process for the elasticity index of autonomic nervous function is as follows: Figure 4 As shown; For risk assessment and control instruction generation (logic closed-loop optimization), the decision module executes the full-link logic of "state perception → risk linkage → intervention matching → feedback verification", as shown in Figure 5: State perception (multi-source data fusion, false judgment rate ≤2%): Awake state: LF / HF < 1.8 + IMU body movement > 0.1g (lasting ≥ 3 seconds) + time dimension (6:00-23:00); Light sleep state: 1.8 ≤ LF / HF < 3.0 + IMU body movement ≤ 0.1g (lasting ≥ 10 seconds) + time dimension (23:00-6:00); Deep sleep state: LF / HF ≥ 3.0 + IMU body movement ≤ 0.05g (lasting ≥ 20 seconds) + stable dual entropy value (fluctuation ≤ ±0.05). Risk linkage determination: High-risk trigger: HI index meets target + secondary verification of physiological characteristics (e.g., during deep sleep, HF power needs to decrease by ≥30% within 1 second; during wakefulness, RMSSD needs to decrease by ≥25%); Delayed trigger: High-risk state lasts for ≥0.5 seconds (to avoid accidental triggering due to instantaneous fluctuations), intervention response delay ≤100ms. Intervention matching rules are as follows: Feedback verification: Calculate the HI index recovery rate within 10 seconds after intervention (recovery rate = (HI after intervention - HI before intervention) / HI before intervention). If the recovery rate is <10%, automatically adjust the parameters (e.g., increase tVNS intensity by 10%) and perform a second intervention.

[0038] For neuromodulation stimulation execution (safety + comfort dual guarantee); VNS stimulation unit; hardware parameters: constant current source output (0.1–5.0mA, 0.1mA step, accuracy ±0.05mA), pulse width 200–500μs (continuously adjustable); safety design: impedance monitoring: real-time monitoring of electrode-skin impedance (normal range 1–10kΩ), intensity decreases by 50% when impedance >15kΩ, stimulation is paused and "electrode poor contact" is indicated when impedance >20kΩ; overcurrent protection: maximum current threshold 5.0mA, output is cut off within 10μs when threshold is exceeded; temperature protection: the stimulation terminal has a built-in NTC sensor, stimulation is paused when surface temperature >40℃, and resumed when it drops below 37℃; comfort optimization: adopts "gradually increasing and decreasing" pulse waveform (rising edge / falling edge 10μs) to reduce electrical stimulation discomfort (user comfort feedback ratio ≥90%, n=100). Acoustic stimulation unit: Hardware parameters: bone conduction speaker (frequency response range 200–2000Hz), stimulation frequency 400–800Hz (speech band, low interference), intensity ≤65dBSPL (equivalent to whisper); Phase locking: triggered by R-wave detection signal, delay 200–250ms (verified in 30 clinical cases, this delay precisely matches the "vagal nerve excitation window" of the heartbeat cycle, improving intervention effectiveness by 35%, p<0.01); Comfort optimization: adopts "narrow pulse + gradual intensity" waveform (rising edge 20μs, falling edge 20μs) to avoid sudden noise interference, and the proportion of users' subjective "no obvious interference" feedback is ≥92% (n=100); Safety constraints: single continuous stimulation during wakefulness ≤3 times (8-second interval), acoustic stimulation during sleep is only used in high-risk scenarios of light sleep, intensity ≤55dB SPL, to avoid waking deep sleep.

[0039] The specific implementation of closed-loop control for stimulation is as follows: Post-intervention data acquisition: Immediately after the first intervention, 30 seconds of ECG signals are acquired. The preprocessing procedure of "4th-order Butterworth bandpass filtering (0.5–40Hz) + adaptive notch filtering (50 / 60Hz) + improved Pan-Tompkins algorithm R-wave detection + 3σ criterion screening" is followed to ensure that the effective RR interval ratio is ≥85%. The HI index recovery rate is calculated (formula: (mean HI value 30 seconds after intervention - HI value before intervention) / HI value before intervention × 100%). If the quality of some prognostic signals is grade C (SNR < 15dB), it is judged as "recovery rate cannot be calculated", and a second intervention is directly triggered. Secondary intervention logic: If the recovery rate is <10% (or the signal quality is substandard), a secondary intervention will be automatically initiated after 5 seconds, with parameter adjustments ≤15% (e.g., tVNS intensity increased by 10%, acoustic frequency shifted by 50Hz, pulse width adjusted by 10%). After the secondary intervention, ECG signals will be collected again for 30 seconds for subsequent parameter optimization. Overload protection: The number of high-risk interventions in the same state within 24 hours is ≤8, with no limit across states. The cumulative stimulation duration is ≤60 minutes. If the limit is exceeded, the system will automatically switch to "low-intensity maintenance mode" (tVNS uses 0.1-0.5Hz ultra-low frequency stimulation, intensity is 30%-40% of the comfort threshold, acoustic stimulation ≤50dB SPL), and a dual notification will be sent via device indicator light (flashing yellow) and APP push notification. Data-driven optimization: All "stimulation parameters - physiological responses (HI recovery rate, D1 / D2 / D3 changes)" data are synchronized to the personalized adaptive engine in real time. Only when the signal quality after intervention is ≥ Grade B (SNR ≥ 15dB) is it included in the "effective intervention record," providing data support for Bayesian optimization. Emergency interruption mechanism: Users can pause stimulation with one click via the device's physical button (press and hold for 2 seconds) or the APP, with a response delay ≤ 50ms. After interruption, a safety screening (comprehensive risk score < 0.85) must be completed before intervention can be resumed. The first intervention after resumption requires a 10-second baseline ECG signal to recalibrate the HI index baseline value.

[0040] The iterative calculation logic of the dynamic adaptive W-weighted ratio includes a collaborative mechanism of immediate application closed loop and medium-to-long-term qualitative analysis. The weight range and safety constraints during the iteration process comply with the safety limitations of claim 10 for transcutaneous vagus nerve stimulation (tVNS) units and multimodal stimulation: Immediate application closed loop (calculation-output-control): Immediate calculation: For the first time, a general weighted ratio (W1=0.35, W2=0.35, W3=0.30) is used to quickly calculate the HI index and three-dimensional data (latency ≤200ms, medical grade ≤100ms); when abnormal conditions are met, at least one second-order iterative calculation is performed according to the targeted weighting strategy; Immediate output: Outputs "real-time value + risk level" (high / medium / low), simplified to "risk level + stimulation suggestion" for consumer-grade applications, and "value + abnormal dimension marker +..." for medical-grade / independent clinical modules. "Pathological indications"; Real-time control: Based on the output results, control the wearable stimulation terminal to dynamically adjust stimulation parameters (current intensity, frequency, phase-locking timing), and simultaneously link physiological signal monitoring (heart rate, blood oxygen) to trigger safety control (such as immediately reducing intensity in case of abnormality); Mid-to-long-term qualitative analysis (trend-guidance): Data accumulation: The system summarizes the audit logs, output data, and control terminal execution records calculated in real time according to the preset period (7 days for consumer-grade, 30 days for medical-grade, and independent clinical modules can be configured as needed) to form a mid-to-long-term dataset; Qualitative analysis: Analyze the core trends based on the dataset, including: Dimensional trends: Mean changes and fluctuations of D1 / D2 / D3 (e.g., monthly average increase of D1 ≥0.1 → improvement in vagal tone); HI index trends: Overall improvement / decrease trend, percentage of stable intervals (e.g., 20% increase in the percentage of 30-day HI index ≥0.5 → improvement in autonomic nervous function); Abnormal pattern qualitative analysis: Identify core abnormal dimensions (e.g., "D2 consistently low → stress recovery disorder type", "D1+D3"). (Dual anomalies → complex disorder); Guided optimization: Feedback qualitative analysis conclusions to the personalized adaptive engine to optimize real-time application logic: Adjust the general weighting ratio of real-time calculation (e.g., D1 dominant anomaly → general W1 increased to 0.4); optimize anomaly detection threshold (e.g., small long-term fluctuations → anomaly threshold narrowed to ±15%); update the stimulation strategy of the control terminal (e.g., slow improvement of D2 → enhanced tVNS stimulation during stress periods); Safety constraints: The real-time logic parameters after medium- and long-term optimization must still comply with the W weighting constraint and HI index circuit breaker mechanism of claim 10 to ensure that the optimization does not exceed the safety boundary. The specific optimization triggering mechanism is as follows: Figure 7 As shown.

[0041] In addition, a flowchart of a specific implementation case is as follows: Figure 8As shown, a closed-loop regulation method based on dynamic assessment of autonomic nervous system functional elasticity comprises the following steps: Step 1: Signal acquisition (ECG signal); Step 2: Preprocessing and safety screening; Step 3: HI index calculation; Step 4: Intelligent decision-making; Step 5: Multimodal intervention; After multimodal intervention, physiological feedback is fed back to Step 1 to form a closed loop.

[0042] This invention also provides a closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity, such as... Figure 2 As shown, it includes: signal acquisition module 1, elastic computing module 2, intelligent decision-making module 3, and multimodal stimulation module 4; The system includes a signal acquisition module 1 that acquires the user's electrocardiogram (ECG) signal in real time and preprocesses it to obtain a clean RR interval sequence; an elasticity calculation module 2 that extracts multi-dimensional heart rate variability features based on the RR interval sequence and calculates the elasticity index of autonomic nervous function accordingly; an intelligent decision-making module 3 that compares the elasticity index with a preset dynamic threshold based on the user's current state to determine the level of instability risk and generates control instructions containing stimulation type and parameters; and a multimodal stimulation module 4 that triggers at least one non-invasive neuromodulation stimulation based on the control instructions. The closed-loop system of this invention adopts an embedded architecture that combines hardware and software, such as... Figure 3 As shown in the system hardware architecture diagram, it mainly includes the following core modules: The physiological signal acquisition module's core component employs a flexible, patch-type multi-lead ECG sensor (at least dual leads) to improve the signal-to-noise ratio and robustness of R-wave detection. Key parameters: Sampling rate: ≥ 250 Hz to ensure R-wave localization accuracy. Analog front-end: Integrates a low-noise amplifier and a 0.5–40 Hz bandpass filter to suppress baseline drift and power line interference. Communication interface: Communicates with the main control unit via Bluetooth Low Energy (BLE) or a dedicated wireless protocol. Expansion options: Can integrate an inertial measurement unit (IMU) for motion sensing and to assist in sleep staging.

[0043] The elastic computing module and intelligent decision-making module are mounted on a high-performance, low-power embedded microprocessor (such as the ARM Cortex-M7 series), with an optional neural network acceleration unit. Core functions include: running real-time ECG signal processing and R-wave detection algorithms; carrying the autonomic nervous system elasticity index calculation model; embedding a rhythm-source imbalance screening submodule as a pre-intervention safety checkpoint; and executing state judgment (sleep / wake) and intervention decision-making logic.

[0044] The multimodal stimulation module includes a transcutaneous vagus nerve stimulation (TVNS) unit and a phase-locked acoustic stimulation (PLS) unit. The TVNS unit uses a clip-on or over-ear tVNS device with a constant current source output. Parameter range: current 0.1–5 mA (0.1 mA steps), frequency 0.1–30 Hz, pulse width 200–500 μs. Safety design: real-time impedance monitoring and overcurrent protection. The PLS unit uses a bone conduction speaker or miniature air conduction speaker embedded in an ear-worn device. Stimulation type: can emit a "click" sound locked to the heartbeat R-wave (duration 50 ms, intensity <60 dB SPL) or binaural beats. Integration: the stimulation terminal is a wearable device, wirelessly connected to the sensing unit for all-weather, uninterrupted control. Figure 5 The diagram shows the implementation of separate and integrated deployments, that is, the multimodal stimulation module and the physiological signal acquisition module can be deployed in an integrated manner or separately.

[0045] Among them, neuromodulation stimulation includes transcutaneous vagal nerve stimulation and phase-locked acoustic stimulation; the elasticity index is a composite index that integrates at least three physiological dimensions: basic vagal tone, stress recovery speed, and system complexity.

[0046] The system complies with the IEC 60601-1-2 electromagnetic compatibility standard for medical electrical equipment during the design phase and has been verified through the following tests: Radiated emission test: confirming that the spatial radiation of the equipment under operating conditions is below the limit requirements; Conducted emission test: verifying that the conducted interference of the power line meets the requirements of Class B equipment; Electrostatic discharge immunity: passing ±8kV contact discharge and ±15kV air discharge tests; Radio frequency electromagnetic field immunity: the system functions normally without interference under a field strength of 3V / m; Electrical fast transient / burst immunity: verifying the system stability under power grid fluctuations; The test results show that under typical dynamic usage scenarios (including user walking, turning over in sleep, etc.), the signal-to-noise ratio of ECG signal acquisition remains better than 20dB, meeting the accuracy requirements of heart rate variability analysis; The stimulation parameters of the transcutaneous vagus nerve stimulation unit used in this system are as follows: current intensity: 0.1-5.0 mA, strictly controlled below the sensory threshold; stimulation frequency: 0.1-30 Hz, avoiding known nerve excitation sensitive frequency bands; pulse width: 200-500 μs, optimizing charge transfer efficiency.

[0047] The signal preprocessing and feature extraction performed by the signal acquisition module specifically include: R-wave detection and RR interval sequence generation: The R-wave is accurately located using algorithms such as Pan-Tompkins on the original electrocardiogram signal, the continuous RR interval (RR) is calculated, and abnormal beats (such as aberrant beats with a difference of more than 20% from the previous or next point) are eliminated by interpolation to form a clean RR sequence.

[0048] Multimodal HRV feature extraction (each 30 seconds is an analysis window); Temporal features include SDNN (overall variability) and RMSSD (root mean square of adjacent RR differences, reflecting vagal tension).

[0049] Frequency domain characteristics: Low-frequency power (LF: 0.04–0.15 Hz), high-frequency power (HF: 0.15–0.4 Hz), and LF / HF ratio are calculated using Fast Fourier Transform (FFT) or Autoregressive (AR) models.

[0050] Nonlinear characteristics: Calculate multi-scale entropy, especially the entropy value at scale 5 (MSE_scale5), to characterize the system complexity; Safety Screening: Rhythm Source Imbalance Analysis specifically includes: Before calculating the elasticity index, the system first performs a safety screening to exclude interference from cardiogenic diseases (such as atrial fibrillation). This analysis is based on a three-dimensional discriminant feature library: Time Domain Analysis: Calculate the peak interval histogram width (FWHM) of the RR interval sequence. If FWHM < 10 ms, it indicates that the rhythm is too regular and may be cardiogenic interference. Frequency Domain Analysis: Detect the presence of abnormally high power (> 5 ms² / Hz) or non-physiological resonance peaks in the ultra-low frequency band (VLF: 0-0.0033 Hz). Nonlinear Analysis: Perform a Yorkshire knife test. If the p-value > 0.05, it indicates that the signal conforms to the linear random hypothesis and is inclined to be cardiogenic. Decision: Calculate the risk score (Risk_score) through a hybrid decision model. If Risk_score > 0.85, the system immediately locks in intervention and prompts the user to seek medical attention. The elasticity calculation module specifically includes the calculation of the elasticity index of autonomic nervous function, which includes: The autonomic nervous system elasticity quantification model adopts a three-dimensional fusion system, which expands the traditional HRV analysis into a comprehensive assessment of "basal tension-stress recovery-system complexity".

[0051] Three-dimensional physiological index synthesis: including: basic tension index, stress recovery index, and system complexity index; the above three-dimensional indexes are used as inputs and nonlinearly integrated through a pre-trained random forest model to output a normalized autonomic nervous system function elasticity index (HI), ranging from 0.0 to 1.0.

[0052] Risk classification: Based on the calculated HI index, the risk levels are classified as follows: High risk (instability): HI < 0.3; Medium risk (critical): 0.3 ≤ HI < 0.7; Low risk (stable): HI ≥ 0.7.

[0053] The elasticity index (HI index) of autonomic nervous system function is calculated through the following steps: Signal input: The preprocessed clean RR interval sequence is used as input. Parallel calculation of multi-dimensional physiological indicators: The system simultaneously calculates three dimensions of physiological indicators: Basal tension index: Characterizing resting vagal tone, it is obtained by fusing normalized RMSSD and high-frequency power (HF) with fixed weights. Stress recovery index: Quantifying the speed of recovery of the autonomic nervous system from disturbances, it is obtained by calculating the slope of change of high-frequency power (HF) within a preset time window (ΔHF slope) and substituting it into the formula. System complexity index: Reflecting the flexibility of autonomic nervous system regulation, it is directly characterized by the entropy value calculated at scale 5 in the multi-scale entropy analysis of the RR interval sequence. Machine learning fusion: The above three dimensions of indicators are used as input features and input into a random forest regression model pre-trained on a large-scale clinical dataset. The model performs nonlinear fusion and finally outputs a normalized autonomic nervous system elasticity index (HI index) in the range of [0.0, 1.0].

[0054] The decision-making module determines whether the user is asleep or awake based on HRV and IMU data and the real-time HI index.

[0055] Intervention during sleep: Deep sleep (N3) maintenance: When the HI index is stable (≥0.6) and the patient is in deep sleep, maintenance stimulation is performed using 0.1–0.5 Hz ultra-low frequency, low intensity (50% of comfort intensity) tVNS.

[0056] Micro-awakening prospective inhibition: When the system detects a HI index < 0.3 and the HF power drops by more than 30% of the baseline within 1 second (this is a micro-awakening precursor), an interventional tVNS is immediately triggered: frequency 10–15 Hz, duration 5 seconds, intensity at a personalized comfort level.

[0057] Intervention in a conscious state: When the HI index indicates a risk of inattention (HI index < 0.4), phase-locked acoustic stimulation is triggered preferentially. After detecting each R wave, the system emits a brief "click" sound after a 250 ms delay, designed to reset the phase of the default mode network and bring attention back to the current task.

[0058] The decision-making module runs a state machine with the following logic: Status assessment: Based on whether the LF / HF ratio in heart rate variability is greater than 1.8, and combined with body motion data from the inertial measurement unit (IMU), the system comprehensively determines whether the user is awake, in a light sleep or a deep sleep.

[0059] Risk classification: Based on the real-time HI index, three levels of risk are classified: HI index ≥ 0.7 is stable (low risk), 0.3 ≤ HI index < 0.7 is critical (medium risk), and HI index < 0.3 is unstable (high risk).

[0060] Intervention selection: Based on the current state and risk level, the optimal stimulation protocol is matched from the intervention pattern library. For example, if an "instability" risk is detected during sleep accompanied by a sharp drop in HF power, interventional tVNS is triggered; if a risk below the "critical" level is continuously detected during wakefulness, phase-locked acoustic stimulation is triggered.

[0061] Security screening logic Before entering the main algorithm process, the system performs rhythmic source imbalance analysis on the RR interval sequence as a preliminary safety checkpoint for intervention: Temporal regularity test: Calculate the peak interval histogram width (FWHM) of the RR interval sequence. If FWHM < 10ms, it is considered that the rhythm is too regular and is marked as risky.

[0062] Frequency domain anomalous resonance detection: Analyze the power in the ultra-low frequency band (VLF, 0-0.0033 Hz). If the VLF power > 5 ms² / Hz, it is considered to have non-physiological resonance and is marked as risk.

[0063] Nonlinear deterministic analysis: The Yule-Walker method is used to test linearity. If the p-value is > 0.05, it indicates that the signal conforms to the linear stochastic hypothesis, tends to be cardiogenic interference, and is marked as risk.

[0064] System Lockdown: Based on the above three-dimensional analysis results, a comprehensive risk score is calculated. If the comprehensive risk score is >0.85, the system immediately locks all intervention functions and prompts the user to seek medical attention.

[0065] The closed-loop control system based on dynamic assessment of autonomic neural function elasticity also includes: a personalized adaptive engine; the personalized adaptive engine is configured as follows: Record users' historical physiological responses to stimuli; Based on recorded historical physiological response data, the model parameters of the elasticity calculation module, the dynamic threshold of the intelligent decision-making module, and / or the stimulation parameters of the multimodal stimulation module are dynamically optimized using a Bayesian optimization algorithm.

[0066] First, optimization begins with data collection and recording. The system continuously records the contextual data of each intervention event, forming a historical record database. The recorded data dimensions include at least: Pre-intervention state: HI index at the time of intervention triggering, user state (sleep / awake), and basic physiological indicators. Intervention parameters: type, frequency, intensity, and duration of the stimulus performed. Intervention response: recovery slope and degree of improvement of the HI index and key physiological indicators (such as HF power) after intervention. Long-term effects: correlation with external indicators such as user sleep efficiency and attention assessment results. The optimization parameter space is defined as follows: The adaptive engine maintains a multi-dimensional parameter space as the object of optimization. The core parameters include: Decision threshold domain: such as the adjustable range of critical risk threshold and attention distraction threshold. Stimulation parameter domain: such as the allowable range of current intensity multiplier, pulse frequency, and duration of tVNS. Algorithm weight domain: such as the weight coefficients of each dimension indicator in the elasticity index calculation model. Optimization cycle and triggering: The engine automatically initiates an optimization process at a fixed time period (e.g., every 7 calendar days) or after accumulating a sufficient number of effective intervention records (e.g., no less than 20 times). The effective intervention records include compliant interventions in all states, with cross-state interventions included in the optimization data accumulation. The multi-objective optimization function aims to maximize a comprehensive scoring function that balances three objectives: Effectiveness: the increase in the HI index and the recovery speed of physiological indicators after intervention (highest weight, e.g., 60%). Safety: no side effects recorded, and no overstimulation (HI index improvement within a reasonable range). Comfort: stimulation intensity within the user's personalized comfort range. Core optimization algorithm: The system uses a Bayesian optimization algorithm as the core optimizer. This algorithm constructs a surrogate model to fit the relationship between historical parameters and scoring results, and intelligently explores the parameter space to find the closest approximate parameter combination with the fewest iterations. Its advantage lies in its ability to handle nonlinear, computationally expensive black-box functions, making it particularly suitable for scenarios like this system, which is based on physiological responses and has high experimental costs (each attempt is a real intervention). Parameter update and deployment: After the optimization process, the engine seamlessly updates the newly obtained set of optimal parameters into the system's elastic computing module, intelligent decision-making module, and multimodal stimulation module, completing a learning iteration and achieving personalized adaptation that becomes more accurate with use.

[0067] The multimodal stimulation module is connected to a wearable stimulation terminal, which integrates a transcutaneous vagus nerve stimulation unit and an acoustic stimulation unit.

[0068] The acoustic stimulation unit includes a non-air-conducting loudspeaker.

[0069] A closed-loop control system based on dynamic assessment of autonomic nervous function elasticity is implemented on a wearable device platform, which includes a physiological signal sensing subject and a wearable stimulation terminal, and the two interact with each other via wireless communication.

[0070] The stimulation parameters of the percutaneous vagus nerve stimulation unit include: current intensity range of 0.1-5.0 mA, stimulation frequency range of 0.1-30 Hz, and pulse width of 200-500 μs.

[0071] This embodiment aims to illustrate the actual workflow of the method and system of the present invention through a specific application scenario, demonstrating how it synergistically suppresses sleep micro-awakening and daytime attention loss.

[0072] 1. System initialization and personalized calibration process When the system is used for the first time, an initialization and calibration process is performed to establish a personalized physiological baseline and stimulation parameters for the user.

[0073] Resting-state physiological baseline acquisition: The user is in a calm, seated resting state, and the system continuously acquires ECG signals for 5 minutes through the physiological signal acquisition module. The system calculates the heart rate variability characteristics during this period and establishes personalized baseline values ​​for the user, including: RMSSD baseline value (e.g., 28.5 ms), high-frequency power (HF) baseline value (e.g., 680 ms²), and LF / HF ratio baseline value (e.g., 2.1).

[0074] Stimulation intensity comfort threshold calibration: The system controls the transcutaneous vagus nerve stimulation unit to output weak electrical stimulation in a stepped-increase manner (starting from 0.1 mA, in 0.1 mA increments). The user provides feedback through a user interface (such as a mobile app) when they feel a clear but comfortable tingling sensation, and the system records this current value as the comfort threshold (e.g., 2.0 mA). Subsequently, the system sets the stimulation intensity for daily work at 70% of the comfort threshold (i.e., 1.4 mA).

[0075] 2. The prospective inhibition process of micro-arousals during nighttime sleep Take a typical intervention during a user's nighttime sleep as an example: Continuous physiological state awareness: During the user's sleep, the system continuously collects signals through an electrocardiogram sensor and calculates the autonomic nervous system elasticity index (HI index) every 30 seconds. At the same time, the system determines that the user is in a sleep state based on the LF / HF ratio (>1.8) in heart rate variability and the low body movement signal from the inertial measurement unit.

[0076] Micro-awakening precursor risk identification: At a certain moment, the system detects the following physiological characteristics appearing simultaneously within a 1-second window: The real-time HI index dropped rapidly from 0.52 to 0.25, below the critical threshold for sleep (0.3).

[0077] The high-frequency power (HF) dropped instantaneously by 38% compared to its baseline value. The system determined this phenomenon to be a physiological precursor to micro-arousal and indicated a high risk.

[0078] Precise Intervention Execution and Verification: The decision-making module immediately generates control commands, triggering the transcutaneous vagus nerve stimulation unit. The stimulation unit outputs a short pulse train lasting 5 seconds at a high frequency (15 Hz) and personalized working intensity (1.4 mA). After the intervention, continued system monitoring showed that the user's high-frequency power recovered to more than 85% of the baseline level within 5 seconds, and the HI index also rose to above 0.5. A potential micro-arousal event was successfully suppressed, and sleep continuity was maintained.

[0079] 3. Intervention procedure for attention wandering during daytime waking hours Take, for example, a user intervention during an afternoon work meeting: Continuous cognitive status assessment: During the user's conscious task, the system continuously calculates the HI index. The system determines that the user is conscious based on the LF / HF ratio (<1.5).

[0080] Attention lapse risk assessment: The system detected that the user's HI index was below the attention threshold (0.4) during wakefulness for 12 consecutive seconds, and combined with the regularity of the decline in the electrocardiogram signal, the system determined that the user was at risk of attention lapse.

[0081] Non-invasive neuromodulation execution: The decision-making module generates control commands that preferentially trigger phase-locked acoustic stimulation. After detecting the R-wave of the user's ECG signal in real time, the acoustic stimulation unit precisely delays for 250 milliseconds and emits a soft "click" sound at an intensity of 55 dB SPL for 50 milliseconds via a bone conduction speaker. This stimulation is synchronized with the user's own heart rhythm and aims to reset the active phase of their brain's default mode network, thereby unobtrusively drawing attention back to the current meeting task.

[0082] 4. Long-term adaptive optimization process of the system Data accumulation: The system continuously runs a personalized adaptive engine in the background, recording the timing of each intervention, stimulation parameters, HI index before intervention, HI index after intervention, and recovery rate of physiological indicators.

[0083] Parameter Optimization: After every 7 consecutive days of use, the adaptive engine automatically initiates an optimization cycle. Based on the intervention response data accumulated over the past week, the engine dynamically adjusts system parameters using a Bayesian optimization algorithm, aiming to improve the magnitude of improvement and recovery speed of the HI index. For example, the critical threshold for micro-awakening may be fine-tuned from the initial 0.30 to 0.28, making the system more sensitive to the risk of instability.

[0084] Closed-loop learning: After several weeks of operation, the system parameters (including decision thresholds and stimulus parameters) will converge to the optimal solution that best suits the user's current physiological state, achieving personalized adaptation that becomes more accurate with use.

[0085] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A closed-loop regulation method based on dynamic assessment of autonomic nervous system functional elasticity, characterized in that, include: Step 1: Acquire the user's electrocardiogram (ECG) signal in real time and preprocess it to obtain a clean RR interval sequence; Step 2: Based on the RR interval sequence, extract multidimensional heart rate variability features and calculate the elasticity index of autonomic nervous function accordingly. Step 3: Compare the elasticity index with the dynamic threshold preset based on the user's current state to determine the level of instability risk and generate control instructions containing stimulus type and parameters. Step 4: Based on the control command, trigger at least one non-invasive neuromodulation stimulus; Among them, neuromodulation stimulation includes transcutaneous vagal nerve stimulation and phase-locked acoustic stimulation; the elasticity index is a composite index that integrates at least three physiological dimensions: basic vagal tone, stress recovery speed, and system complexity.

2. The closed-loop regulation method based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 1, characterized in that, The specific steps for calculating the elasticity index of autonomic nervous system function include: Calculate the time-domain characteristics, frequency-domain characteristics, and nonlinear characteristics from the RR interval sequence; Based on time-domain features, frequency-domain features, and nonlinear features, physiological indicators of three dimensions—basic vagal tone, stress recovery speed, and system complexity—are synthesized. Physiological indicators of three dimensions—basic vagal tone, stress recovery speed, and system complexity—are used as input features. These indicators are then nonlinearly integrated using a machine learning fusion model pre-trained on a clinical dataset to output the elasticity index of autonomic nerve function. Among them, the basic vagal tone is synthesized by weighting the normalized root mean square difference of the RR interval and the high-frequency power; the stress recovery speed is obtained by quantifying the recovery dynamics of heart rate variability after a preset stress event; the system complexity is obtained by weighting and fusing the multi-scale entropy of the RR interval sequence and the sample entropy according to preset weights; among them, the weight corresponding to the multi-scale entropy is greater than the weight corresponding to the sample entropy.

3. The closed-loop regulation method based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 1 or 2, characterized in that, The process following step one and before step two also includes: a security screening step. Calculate the full width at half maximum (FWHM) of the peak interval histogram of the RR intervals. If FWHM < 10 ms, the corresponding risk sub-score for cardiac interference is 0.

4. Calculate the power value in the ultra-low frequency band. If the power is greater than 5ms... 2 / Hz, corresponding to a cardiac interference risk sub-score of 0.3; York knife test was used to analyze the determinism of RR interval sequences. If the p-value was >0.05, the corresponding risk sub-score for cardiac interference was 0.

3. The above sub-scores are summed to obtain a comprehensive risk score. If the comprehensive risk score is ≥0.85, the identification of cardiac disease interference is considered successful. After successful identification, the subsequent intervention process is immediately terminated, a security alarm is sent to the user terminal, and the RR interval sequence, characteristic indicators and risk score data of this screening are automatically recorded.

4. The closed-loop regulation method based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 1, characterized in that, The parameters of the neural modulation stimulation are dynamically adapted, as follows: During sleep, the type, frequency, and intensity of transcutaneous vagus nerve stimulation and acoustic stimulation are adapted according to sleep stage, risk of micro-arousals, and core pathological dimensions of the disease: During deep sleep, transcutaneous vagus nerve stimulation uses an ultra-low frequency of 0.1-1Hz to maintain stimulation, with an intensity of 40%-65% of the user's comfort threshold. For Parkinson's disease and obstructive sleep apnea, a sub-range within the frequency range and a pulse width of 200-500μs can be adapted to enhance the regulation of the pathological dimensions corresponding to the disease. During the sleep onset period, 10-15Hz periodic pulse transcutaneous vagus nerve stimulation is used, which can be superimposed with 40-45dB SPL low-intensity white noise. A single continuous use is ≤15 minutes, and the transcutaneous vagus nerve stimulation intensity is 40%-50% of the comfort threshold, which is suitable for the physiological characteristics of the sleep onset period. During sleep, transcutaneous vagus nerve stimulation uses 0.5-1Hz ultra-low frequency stimulation at an intensity of 30%-40% of the comfort threshold. Acoustic stimulation can be selectively activated according to disease needs, with pink noise being the preferred choice and low-intensity white noise being superimposed for PTSD patients. During the pre-arousal period, trigger 10-15Hz short-pulse interventional transcutaneous vagal nerve stimulation for 5-8 seconds at an intensity of 60%-70% of the comfort threshold. For obstructive sleep apnea, acoustic stimulation can be added simultaneously to help maintain physiological homeostasis. In a conscious state, the stimulus type and parameters are adapted based on the cognitive, emotional state, and disease type indicated by the elasticity index: When predicting attention loss, the phase-locked acoustic stimulation that is locked to the user's ECG R wave is triggered first, with an R wave delay of 200-250ms and an intensity of 55-65dB SPL. When modulating diseases specifically, for attention deficit hyperactivity disorder, depression, mild cognitive impairment, and post-traumatic stress disorder, suitable acoustic stimulation types include: binaural melodic, pure tone alternation, or transcutaneous vagus nerve stimulation frequency and intensity range, which can be used to target and improve disease-related autonomic dysfunction. Configure unified security constraints, specifically including: When the elasticity index is <0.3, the intensity of transcutaneous vagus nerve stimulation is uniformly limited to 0.5-1.0 mA; When the LF / HF ratio is greater than 3, high-frequency stimulation above 10Hz is paused and switched to the 0.1-1Hz ultra-low frequency maintenance mode. The acoustic stimulation intensity should not exceed 65 dB SPL, and the transcutaneous vagus nerve stimulation current intensity should not exceed 5.0 mA.

5. A closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity, characterized in that, include: Signal acquisition module, elastic computing module, intelligent decision-making module, and multimodal stimulation module; The system includes a signal acquisition module that acquires the user's electrocardiogram (ECG) signal in real time and preprocesses it to obtain a clean RR interval sequence; an elasticity calculation module that extracts multi-dimensional heart rate variability features based on the RR interval sequence and calculates the elasticity index of autonomic nervous function accordingly; an intelligent decision-making module that compares the elasticity index with a preset dynamic threshold based on the user's current state to determine the level of instability risk and generates control instructions containing stimulation type and parameters; and a multimodal stimulation module that triggers at least one non-invasive neuromodulation stimulation based on the control instructions. Among them, neuromodulation stimulation includes transcutaneous vagal nerve stimulation and phase-locked acoustic stimulation; the elasticity index is a composite index that integrates at least three physiological dimensions: basic vagal tone, stress recovery speed, and system complexity.

6. The closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 5, characterized in that, Also includes: a personalized adaptive engine; the personalized adaptive engine is configured as follows: Record users' historical physiological responses to stimuli; Based on recorded historical physiological response data, the model parameters of the elasticity calculation module, the dynamic threshold of the intelligent decision-making module, and / or the stimulation parameters of the multimodal stimulation module are dynamically optimized using a Bayesian optimization algorithm.

7. The closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 5, characterized in that, The multimodal stimulation module is connected to a wearable stimulation terminal, which integrates a transcutaneous vagus nerve stimulation unit and an acoustic stimulation unit. Wearable stimulation terminals include, but are not limited to, ear-worn, wrist-worn, and chest-worn devices.

8. The closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 7, characterized in that, Acoustic stimulation units include: non-air-conducting loudspeakers, air-conducting loudspeakers, or combinations thereof.

9. The closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 5, characterized in that, It is implemented on a wearable device platform, which includes a physiological signal sensing subject and a wearable stimulation terminal, and the two interact with each other via wireless communication.

10. The closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 7, characterized in that, The stimulation parameters of the percutaneous vagus nerve stimulation unit include: current intensity range of 0.1-5.0 mA, stimulation frequency range of 0.1-30 Hz, and pulse width of 200-500 μs.

11. The closed-loop regulation system based on dynamic assessment of autonomic nervous system functional elasticity as described in claim 5, characterized in that, It also includes an evaluation module. The evaluation module performs the following operations: When the elasticity index increases by ≥0.2, the baseline vagal tone increases by ≥0.15, the stress recovery speed increases by ≥0.1, and the system complexity increases by ≥0.1, the physiological dimensions are considered effective. If the score on the clinical symptom scoring scale decreases by a predetermined amount, it is considered effective in the clinical dimension. If both the physiological and clinical dimensions are effective, output the conclusion that the intervention is effective; otherwise, output optimization suggestions that require adjustment of the intervention parameters.

Images