Intelligent sleep disorder monitoring system and method based on multi-mode physiological signals

Abstract

The invention relates to the technical field of intelligent medical monitoring, and discloses a sleep disorder intelligent monitoring system and method based on a multi-modal physiological signal, and the method comprises the steps: collecting electrocardio, respiration and body movement signals, calculating a finite time Lyapunov index by using a phase-space reconstruction technology, taking the finite time Lyapunov index as a feedback control variable, and calculating the sleep disorder intelligent monitoring system based on the multi-modal physiological signal; noise intensity in a Langevin equation is dynamically adjusted, adaptive stochastic resonance enhancement is performed on the preprocessed signal, weak characteristics are improved by using noise energy, and causal coupling intensity between cross-modal symbol transfer entropy and net information flow index quantification systems is further calculated through phase-space symbolization mapping. And in combination with Lempel-Ziv complexity, a multi-dimensional judgment logic is constructed, and sleep staging and obstacle event results are output. According to the method, the problems that nonlinear physiological signal features are difficult to extract and a system coupling mechanism is unknown are solved, and the robustness and accuracy of sleep monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical monitoring technology, specifically to an intelligent monitoring system and method for sleep disorders based on multimodal physiological signals. Background Technology

[0002] Sleep disorders affect human physical and mental health and quality of life. Accurate sleep monitoring is a prerequisite for the diagnosis and treatment of sleep-related diseases. With the development of biomedical engineering technology, sleep monitoring technology based on physiological signals has gradually evolved from traditional laboratory polysomnography monitoring to portable and home-based intelligent monitoring. In order to improve the comprehensiveness and accuracy of monitoring, modern monitoring systems collect multimodal physiological data such as electrocardiogram, respiration, and body movement, and capture physiological state changes during sleep by fusing multi-source information.

[0003] Existing sleep monitoring signal processing techniques rely on traditional time-domain analysis, frequency-domain analysis, and time-frequency joint analysis methods. For example, power spectrum features of the signal are extracted by fast Fourier transform or wavelet transform, or heart rate variability index is used to assess the regulatory function of the autonomic nervous system. In the signal preprocessing stage, existing technologies use linear signal processing methods such as low-pass filtering, notch filters, or adaptive filtering to eliminate baseline drift, power frequency interference, and electromyographic noise.

[0004] However, the human physiological system is inherently a highly complex nonlinear dynamic system, and the physiological signals it generates exhibit non-stationarity and chaotic characteristics. Existing linear analysis methods struggle to characterize the nonlinear dynamic evolution hidden within signal fluctuations. Furthermore, traditional denoising techniques treat all background noise as harmful interference and filter it out, ignoring the fact that in nonlinear systems, a suitable amount of noise can enhance weak characteristic signals through stochastic resonance mechanisms, leading to the loss of crucial microscopic dynamic information during the preprocessing stage. Therefore, this invention proposes an intelligent sleep disorder monitoring system and method based on multimodal physiological signals to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent sleep disorder monitoring system and method based on multimodal physiological signals. It solves the problems of existing sleep monitoring technologies, such as difficulty in feature extraction when processing non-stationary physiological signals, inability to use noise to enhance weak signals, and lack of quantitative analysis of dynamic coupling mechanisms between multiple systems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides an intelligent sleep disorder monitoring system based on multimodal physiological signals, comprising: The system includes a data acquisition unit, a processor unit, and a result output unit. The data acquisition unit is configured to acquire multimodal physiological signals of the subject, including electrocardiogram signals, respiratory signals, and body movement signals. The processor unit is configured to perform cascaded processing on multimodal physiological signals, and it integrates a signal preprocessing module, a phase space reconstruction module, a local dynamic instability assessment module, an adaptive stochastic resonance enhancement module, a phase space symbolization module, a cross-modal causal analysis module, and a sleep state determination module. The local dynamic instability assessment module is configured to quantify the local divergence characteristics of the system state and feed the calculation results back to the adaptive stochastic resonance enhancement module. The adaptive stochastic resonance enhancement module uses the finite-time Lyapunov exponent output by the local dynamic instability assessment module as a feedback control variable to dynamically adjust the noise intensity of the injected system, thereby achieving nonlinear stochastic resonance enhancement of the preprocessed time series. The sleep state determination module outputs the final monitoring results based on the dynamic indicators and causal coupling indicators.

[0007] Preferably, the signal preprocessing module is configured to perform baseline drift removal and standard fractional normalization to eliminate low-frequency interference in the signal and unify data dimensions. The phase space reconstruction module uses the mutual information function minimum method to determine the time delay parameter, uses the spurious neighbor ratio method to determine the embedding dimension parameter, and maps the one-dimensional time series to a high-dimensional topological space based on the determined parameters to restore the dynamic attractor structure of the system.

[0008] Preferably, the local dynamic instability assessment module characterizes the local instability of the system by calculating the finite-time Lyapunov exponent of the phase space trajectory. The calculation of the finite-time Lyapunov exponent is based on the Euclidean distance between the reference trajectory point and the nearest neighbor trajectory point in the phase space. By quantifying the natural logarithm of the rate of change of the Euclidean distance over a preset evolution time, a quantitative index reflecting the local trajectory divergence rate of the system at the current moment is obtained.

[0009] Preferably, the adaptive stochastic resonance enhancement module constructs a bistable system model based on the Langevin equation and uses the stochastic resonance mechanism to enhance weak feature signals using noise energy. The Langevin equation describes that the rate of change of the enhanced signal over time is determined by the resultant force of three parts: the nonlinear restoring force originating from the negative gradient of the bistable state function, the external driving force corresponding to the input time series, and the random noise term containing random fluctuations.

[0010] Preferably, the adaptive stochastic resonance enhancement module introduces an adaptive feedback control mechanism based on system instability. The system pre-sets a dynamic critical response interval. When the finite-time Lyapunov exponent is detected to fall into this interval, it indicates that the system is in a critical state sensitive to noise. At this time, the controlled noise intensity function in the random noise term is controlled to be exponentially positively correlated with the finite-time Lyapunov exponent, thereby injecting an appropriate amount of noise at the moment when the signal is most unstable to induce stochastic resonance. When the finite-time Lyapunov exponent exceeds this interval, the noise injection is turned off to avoid excessive noise overwhelming the effective signal.

[0011] Preferably, the phase space symbolization module uses symbolic dynamics to discretize the continuous signal. The phase space symbolization module divides the phase space into several non-overlapping topological regions by setting a set of hyperplanes in the reconstructed phase space, and assigns a unique integer symbol according to the region position of the enhanced signal state vector at the current time, thereby transforming the complex dynamic trajectory into a symbol sequence, reducing the computational complexity while retaining the nonlinear dynamic characteristics.

[0012] Preferably, the cross-modal causal analysis module uses symbolic transfer entropy to quantify the direction and intensity of information flow between different physiological systems. By calculating the reduction in uncertainty of the future state prediction of the target signal under the joint condition of the known historical state of the target signal and the historical state of the source signal, the bidirectional symbolic transfer entropy from respiratory signal to electrocardiogram signal and from electrocardiogram signal to respiratory signal is obtained.

[0013] Preferably, the cross-modal causal analysis module further calculates the net information flow index, which is defined as the difference between the bidirectional symbol transfer entropy and is used to analyze the master-slave driving relationship between physiological systems.

[0014] Preferably, the sleep state determination module comprehensively utilizes dynamic characteristics and information theory indicators for multidimensional state identification. The sleep state determination module calculates the Lempel-Ziv complexity (disorder scale) of the enhanced signal as a measure of signal complexity, and combines it with the net information flow index to execute the determination logic: when the net information flow index shows significant positive coupling, it is identified as a respiratory-driven mode; when it shows significant negative coupling, it is identified as a cardiovascular-driven mode; when the coupling strength is weak, it further distinguishes between the awake phase and the REM phase based on the level of Lempel-Ziv complexity, thereby achieving accurate monitoring of sleep stages and sleep disorder events.

[0015] A second aspect of this invention provides an intelligent monitoring method for sleep disorders based on multimodal physiological signals, comprising the following steps: S1. Acquire the subject's electrocardiogram data sequence, respiratory motion data sequence, and triaxial acceleration data sequence, and perform baseline drift removal and standard score standardization on the data sequences; S2. Perform phase space reconstruction on the standardized data sequence, map the one-dimensional time series to a reconstruction vector in a high-dimensional topological space, and calculate the finite-time Lyapunov exponent of the phase space trajectory. S3. Construct an adaptive stochastic resonance feedback loop based on the finite-time Lyapunov exponent: When the finite-time Lyapunov exponent is within a preset dynamic critical response range, Gaussian white noise is injected into the bistable nonlinear system based on the Langevin equation, and the noise intensity is set to be exponentially positively correlated with the finite-time Lyapunov exponent. The enhanced signal is output using the stochastic resonance effect, and the enhanced signal is then mapped to phase space symbolization. By dividing the phase space region, the continuous signal is converted into a discrete symbol sequence. S4. Calculate the bidirectional symbol transfer entropy between the symbol sequence corresponding to the respiratory signal and the symbol sequence corresponding to the electrocardiogram signal, and calculate the net information flow index based on the difference in bidirectional symbol transfer entropy, and calculate the disorder scale of the enhanced signal. S5. Determine the sleep stage status and sleep disorder events of the subject based on the finite-time Lyapunov index, the symbolic transfer entropy, the net information flow index, and the disorder scale.

[0016] This invention provides an intelligent monitoring system and method for sleep disorders based on multimodal physiological signals. It has the following beneficial effects: 1. This invention utilizes the finite-time Lyapunov exponent as a feedback variable to dynamically adjust the noise intensity in the Langevin equation, thereby achieving adaptive stochastic resonance enhancement. It can capture moments of local dynamic instability in the system and enhance weak physiological characteristic signals using noise energy. This overcomes the problem that traditional linear filtering easily loses key dynamic details when processing non-stationary and nonlinear sleep physiological signals, and improves the signal-to-noise ratio and feature discernibility of the signal.

[0017] 2. This invention, through phase space symbolization and cross-modal causal analysis modules, calculates the net information flow index based on symbolic transfer entropy. Under a unified information theory framework, it can quantify the bidirectional coupling strength and driving response relationship between the respiratory and cardiovascular systems, accurately analyze the physiological network regulation mechanism during sleep, and thus provide in-depth physiological and pathological basis for the determination of sleep stages and sleep apnea and other disorder events by distinguishing between respiratory driving modes and cardiovascular driving modes.

[0018] 3. This invention integrates the Lempel-Ziv complexity algorithm with multidimensional dynamic indicators to construct sleep state determination logic. When the net information flow index indicates a weakly coupled state, it further introduces the Lempel-Ziv complexity to evaluate the nonlinear complexity of the signal sequence, solving the technical problem that the rapid eye movement phase and the awake phase are easily confused in terms of conventional time-frequency characteristics. It constructs a multidimensional feature space that includes dynamic stability, causal coupling and sequence complexity, thereby improving the robustness and accuracy of the sleep disorder monitoring system. Attached Figure Description

[0019] Figure 1 This is a structural block diagram of the system of the present invention; Figure 2 This is a flowchart illustrating the overall process of the method of the present invention. Figure 3 This is a schematic diagram of the signal processing logic for the adaptive stochastic resonance enhancement and cross-modal causal analysis of the present invention. Detailed Implementation

[0020] The technical solutions in 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, and 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.

[0021] See attached document Figure 1 The present invention provides an intelligent sleep disorder monitoring system based on multimodal physiological signals. The intelligent sleep disorder monitoring system includes a data acquisition unit, a processor unit, and a result output unit.

[0022] The data acquisition unit is configured to acquire multimodal physiological signals of the subject. The data acquisition unit includes an electrocardiogram (ECG) signal acquisition module, a respiratory signal acquisition module, and a body motion signal acquisition module. The ECG signal acquisition module is configured to acquire the subject's ECG data sequence or photoplethysmography (PPG) pulse wave data sequence. The respiratory signal acquisition module is configured to acquire the subject's respiratory motion data sequence. The body motion signal acquisition module is configured to acquire the subject's triaxial acceleration data sequence. The data acquisition unit is communicatively connected to the processor unit, and the data acquisition unit transmits the acquired multimodal physiological signals to the processor unit.

[0023] The processor unit is configured to store, compute, and logically process the received multimodal physiological signals. The processor unit includes a signal preprocessing module, a phase space reconstruction module, a local dynamic instability assessment module, an adaptive stochastic resonance enhancement module, a phase space symbolization module, a cross-modal causal analysis module, and a sleep state determination module.

[0024] The signal preprocessing module is configured to receive multimodal physiological signals from the data acquisition unit. The signal preprocessing module performs baseline drift removal and Z-score (standard score) standardization on each time series of the multimodal physiological signals. The signal preprocessing module outputs the standardized time series. During the processing, the signal preprocessing module retains the high-frequency non-stationary components and background noise components in the time series and does not perform low-pass filtering operation with a cutoff frequency lower than the preset value.

[0025] The phase space reconstruction module is connected to the signal preprocessing module. The phase space reconstruction module is configured to receive the normalized time series and map the one-dimensional time series to a high-dimensional topological space. The phase space reconstruction module targets specific time points... signal value Constructing Reconstructed Vectors Reconstructing vectors The mathematical expression is: ; in, Indicates at time The signal value; Indicates the time delay parameter; Indicates the embedding dimension parameter; This represents the matrix transpose operation.

[0026] The phase space reconstruction module determines the time delay parameter by calculating the minimum value of the mutual information function. The embedding dimension parameter is determined by the spurious neighbor method. The phase space reconstruction module outputs a sequence of reconstructed vectors.

[0027] The local dynamic instability assessment module is connected to the phase space reconstruction module. The local dynamic instability assessment module is configured to calculate the finite-time Lyapunov exponent of the phase space trajectory. Limited time Lyapunov index Used to quantify the system at time... The local orbital divergence rate and the local dynamic instability assessment module calculate the finite-time Lyapunov exponent. The formula is: ; in, Indicates time The Euclidean distance between a reference trajectory point and its nearest neighbor trajectory point in phase space; Indicates the time elapsed since the beginning of evolution The Euclidean distance between the back reference trajectory point and the nearest neighbor trajectory point; Represents the natural logarithm operation.

[0028] The local dynamic instability assessment module outputs a finite-time Lyapunov exponential sequence that varies with time.

[0029] The adaptive stochastic resonance enhancement module is connected to both the signal preprocessing module and the local dynamic instability assessment module. The module is configured to enhance weak signals using a nonlinear stochastic resonance mechanism. It receives the original, standardized time series as system input and also receives the finite-time Lyapunov exponent. As a feedback control variable, the adaptive stochastic resonance enhancement module runs a bistable system model based on the Langevin equation, which is expressed as follows: ; in, This represents the enhanced signal output by the adaptive stochastic resonance enhancement module; This represents the standardized input signal value. Represents system state variables; Represents system state variables bistable state function; This represents Gaussian white noise with zero mean and one variance. This represents the controlled noise intensity function.

[0030] Bistable state function set in the adaptive stochastic resonance enhancement module The expression is: ; in, and These are the preset system structure parameters.

[0031] The adaptive stochastic resonance enhancement module is based on the finite-time Lyapunov exponent. Adjusting the controlled noise intensity function Controlled noise intensity function The expression is: ; in, This represents the preset lower threshold for instability. The Lyapunov index represents a finite-time condition. This indicates the preset upper limit threshold for instability; Indicates the basic noise intensity factor; This represents the gain adjustment coefficient; Represented by natural constant An exponential function with base 0.

[0032] The adaptive stochastic resonance enhancement module outputs the enhanced signal. .

[0033] The phase space symbolization module is connected to the adaptive stochastic resonance enhancement module. The phase space symbolization module is configured to continuously enhance the signal. The signal is converted into a discrete symbol sequence, and the phase space symbolization module enhances the signal. The phase space is divided into Non-overlapping regions The phase space symbolization module is based on time. The region containing the state vector is assigned a sign value. The phase space symbolization module outputs a symbol sequence. .

[0034] The cross-modal causality analysis module is connected to the phase space symbolization module. The cross-modal causality analysis module is configured to receive symbol sequences from different physiological modalities and calculate the symbol transfer entropy. Let the symbol sequence corresponding to the respiratory signal be... The symbol sequence corresponding to the electrocardiogram signal is The cross-modal causal analysis module calculates the symbol transfer entropy from respiratory signals to electrocardiogram signals. Symbolic propagation entropy The calculation formula is: ; in, This represents the summation over all possible combinations of states; Indicates the electrocardiogram signal at The symbolic state at any given moment; Indicates the ECG signal at time The historical state vector; Indicates respiratory signals in The historical state vector at any given time; Represents the probability distribution function; It represents a logarithmic operation with base 2.

[0035] The cross-modal causal analysis module is also configured to calculate the net information flow index. Net Information Flow Index The calculation formula is: ; in, The symbolic transfer entropy represents the signal from the electrocardiogram signal to the respiratory signal. The cross-modal causal analysis module outputs the symbolic transfer entropy value and the net information flow index.

[0036] The sleep state determination module is connected to both the local dynamic instability assessment module and the cross-modal causality analysis module. The sleep state determination module is configured to receive finite-time Lyapunov exponential sequences, symbol transfer entropy values, and net information flow exponents. The sleep state determination module is also configured to calculate the enhanced signal. The Lempel-Ziv complexity quantifies the randomness of the sequence. The sleep state determination module determines the sleep stage and sleep disorder events of the subject based on the received indicators. The result output unit is connected to the sleep state determination module and is configured to display or transmit the determination results.

[0037] See attached document Figure 2 This invention provides an intelligent monitoring method for sleep disorders based on multimodal physiological signals, which includes the following steps: The subject's electrocardiogram (ECG), respiratory motion, and triaxial acceleration data sequences were acquired, and preprocessing was performed on these data sequences. Specifically, the preprocessing included baseline drift removal processing of the ECG, respiratory motion, and triaxial acceleration data sequences to eliminate low-frequency interference caused by sensor drift or slow body movement. Z-score normalization is performed on the baseline-drift-free data sequence to convert it into a normalized sequence with a mean of zero and a standard deviation of one. In order to preserve the nonlinear dynamic characteristics implied in the physiological signal, the preprocessing operation does not perform low-pass filtering with a cutoff frequency lower than a preset value, thereby fully preserving the high-frequency non-stationary components and background noise components in the signal.

[0038] The phase space of the preprocessed normalized sequence is reconstructed, and the local dynamic instability index is calculated for each time step in the normalized sequence. Construct high-dimensional reconstruction vectors In constructing the reconstructed vector First, the mutual information function is calculated, and the lag time corresponding to the mutual information function reaching its first minimum value is determined as the time delay parameter. Simultaneously, the spurious neighbor method is used to calculate the dimension when the proportion of spurious neighbors is lower than a preset threshold, and this value is determined as the embedding dimension parameter. Based on a determined time delay parameter and embedding dimension parameters Mapping a one-dimensional time series to Trajectory in Vie Euclidean space; Subsequently, the finite-time Lyapunov exponent for the phase space trajectory is calculated. The calculation process is as follows: finding the time in phase space. The nearest neighbor of the reference trajectory point is calculated, and the evolution time between the reference trajectory point and its nearest neighbor is calculated. The distance divergence rate after the distance. Finite-time Lyapunov exponent. The higher the value, the more active the sleep physiology system is at that moment. The higher the degree of chaos, the more sensitive the system is to initial conditions.

[0039] Based on the Lyapunov index of finite time An adaptive stochastic resonance feedback loop is constructed to nonlinearly enhance the characteristics of weak signals. A bistable nonlinear system based on the overdamped Langevin equation is established. The bistable nonlinear system receives a standardized sequence as the input signal and receives the finite-time Lyapunov exponent. The potential function of a bistable nonlinear system serves as the feedback control variable for adjusting noise intensity. Set as a system state variable The fourth-degree polynomial, i.e. ,in and The positive real parameter determines the height of the bistable potential barrier and the location of the potential well. Furthermore, a controlled noise injection strategy is implemented: real-time monitoring of the Lyapunov exponent over a finite time period. The value of the Lyapunov exponent when the time is finite At the preset instability lower threshold Compared with the preset instability upper limit threshold When the system is in a metastable state, the controlled noise intensity function is calculated. Controlled noise intensity function The value of the Lyapunov exponent with finite time There is an exponential positive correlation, with the strength being Injecting Gaussian white noise into a bistable nonlinear system utilizes the stochastic resonance effect to transfer noise energy to signal energy, thereby making the weak periodic components or transient abrupt components that were originally submerged in the background noise appear in the system's state variables. The Lyapunov exponent is enhanced in the output when the time is finite. When not within the above interval, the controlled noise intensity function When the value is set to zero, the system only performs nonlinear following of the input signal, and the final output is the enhanced signal. .

[0040] For the enhanced signal Perform phase space symbol mapping to generate discrete symbol sequences for the enhanced signal. The phase space is reconstructed again, and the reconstructed high-dimensional phase space is divided into... Each sampling point in the sequence is divided into non-overlapping hypercube regions or sectors based on the Poincaré section. For each sampling point, the region number to which the phase space vector corresponding to that sampling point falls is mapped to a unique integer symbol. This step converts a continuous analog signal into a symbol sequence consisting of a finite number of integers. This symbolization process can filter out minor redundant fluctuations in signal amplitude, focusing on the topological evolution of the signal dynamics.

[0041] Calculate the symbol transfer entropy and net information flow index between symbol sequences of different physiological modalities to quantify cross-system causal coupling relationships, and select electrocardiogram symbol sequences. and respiratory symbol sequence Calculate the symbolic transfer entropy from the respiratory system to the cardiovascular system. The net information flow index quantifies the information flow in a known ECG symbol sequence. Its own historical state and respiratory symbol sequence Under historical conditions, respiratory symbol sequences ECG symbol sequence The reduction in uncertainty in the current state prediction; similarly, the symbolic transfer entropy from the cardiovascular system to the respiratory system is calculated. Subsequently, the net information flow index was calculated. Net Information Flow Index equal minus .

[0042] Sleep stages and sleep disorder events are determined based on the finite-time Lyapunov index, symbolic transfer entropy, and net information flow index. The determination logic includes: when the finite-time Lyapunov index is detected... It is in the low value range, and the symbol propagation entropy Higher than At that time, the subject was determined to be in non-rapid eye movement (NREM) deep sleep, which is characterized by a stable physiological coupling of breathing and heart rate. When the Lyapunov index was monitored for a limited time... It is in the high value range, and the net information flow index When the value approaches zero, the subject is determined to be in the rapid eye movement (REM) phase or the awake phase; Furthermore, if in the net information flow index While approaching zero, the signal is enhanced. If the Lempel-Ziv complexity (disorder scale) is higher than the preset lucidity threshold, it is determined to be a lucid period; otherwise, it is determined to be a REM period. When the net information flow index is detected... The Lyapunov exponent can reverse from positive to negative within a short period of time, or within a finite time. When a transient spike occurs, it is determined that a micro-awakening event or a sleep apnea event has occurred, and the start and end times of the event are marked.

[0043] See attached document Figure 3 This embodiment details how to convert a one-dimensional physiological signal time series into a high-dimensional topological structure that can reflect the dynamic characteristics of the system, and quantifies the local stability of the topological structure.

[0044] The phase space reconstruction and local dynamic instability assessment process first includes determining the optimal embedding parameters. For the normalized time series output from the signal preprocessing module, the phase space reconstruction module calculates the time delay parameters. and embedding dimension parameters .

[0045] For time delay parameters To determine the time delay, the phase space reconstruction module employs the average mutual information method. It calculates the average mutual information function of the standardized time series at different lag times. The average mutual information function measures the statistical dependence between signal values ​​at two moments, including both linear and nonlinear correlations. The phase space reconstruction module iterates through the lag times, searching for the first minimum point of the average mutual information function as the lag time increases. The lag time corresponding to this first minimum point is then determined as the time delay parameter. Choosing the first minimum point ensures that the components of the reconstructed vector maintain dynamic correlation while eliminating excessive information redundancy, thereby maximizing the unfolding of the system's phase space trajectory.

[0046] For embedding dimension parameters To determine the dimensionality, the phase space reconstruction module employs the false neighbor method. This method is based on the premise that two seemingly adjacent trajectory points in a low-dimensional space may be superimposed due to projection effects, while in a higher-dimensional space, these two trajectory points may be separated. If the distance between two points no longer increases with the increase in dimension, they are considered true neighbors. The phase space reconstruction module gradually increases the dimension from the minimum dimension, calculating the nearest neighbor of all points in the phase space at the current dimension, and calculating the proportion of previously adjacent points whose distances change abruptly when the dimension increases by one (i.e., the false neighbor ratio). When the false neighbor ratio drops to zero or below a preset minimum threshold, the phase space reconstruction module stops increasing the dimension and determines the current dimension as the embedded dimension parameter. .

[0047] After determining the time delay parameter and embedding dimension parameters Next, the phase space reconstruction module performs phase space trajectory reconstruction. The phase space reconstruction module uses the sliding window technique to reconstruct the time in the standardized time series. , build A column vector of dimension, i.e., a reconstructed vector. .

[0048] Subsequently, the local dynamic instability assessment module performs local trajectory divergence rate calculation, which is achieved by calculating the finite-time Lyapunov exponent of the phase space trajectory. To quantify the system at time Local instability, finite-time Lyapunov exponent It describes the average exponential rate at which two adjacent trajectories in phase space separate or converge in a short period of time.

[0049] Specifically, the local dynamic instability assessment module first searches for the time in phase space. Reconstructed vector The nearest neighbor vector, will As a reference point, after excluding the relationship with After the temporally adjacent subsequent points, search in phase space for... The point with the smallest Euclidean distance is denoted as . Calculate the reconstructed vector of the reference point. with nearest neighbor vector Initial Euclidean distance between Initial Euclidean distance The calculation formula is: ; in, This represents Euclidean norm operations.

[0050] Next, the local dynamic instability assessment module tracks the evolution of these two points over time. After a preset evolution time step... Later, the reference point evolved into The nearest neighbor vector evolves into Calculate the evolved Euclidean distance. .

[0051] Finally, the local dynamic instability assessment module calculates the finite-time Lyapunov exponent based on the rate of change of distance. Limited time Lyapunov index The calculation formula is: ; in, Represents the natural logarithm operation.

[0052] When the calculated finite-time Lyapunov exponent When it is positive, it indicates that the adjacent trajectories in the phase space are within the evolution time step. The exponential separation within the system means that at time... Sensitive to initial conditions, it exists in a chaotic or unstable state; when the Lyapunov exponent is finite-time... When the value is negative or close to zero, it indicates that the adjacent trajectories are within the same evolution time step. The system remains close to or converges within a certain time frame, meaning that at time 10:00, the system... In a stable or periodic state, the local dynamic instability assessment module will calculate the finite-time Lyapunov exponent. The output is sent to the adaptive stochastic resonance enhancement module as the basis for subsequent noise injection decisions.

[0053] See attached document Figure 3 This embodiment details the bistable nonlinear system model constructed within the adaptive stochastic resonance enhancement module. The module constructs a nonlinear system conforming to the overdamped Langevin equations. This nonlinear system simulates the motion of a Brownian particle in a bistable potential field. This motion is influenced by a deterministic potential force, a periodic input signal driving force, and random noise fluctuations. The Langevin equations describing the evolution of this nonlinear system are as follows: ; in, This represents the rate of change of the output response over time. Represents the potential function The determined nonlinear restoring force; This represents the normalized signal value input to the adaptive stochastic resonance enhancement module, which corresponds to the external modulation force on the Brownian particle in the physical model. This represents the random noise term.

[0054] In the Langevin equation, the bistable state function The topology of the nonlinear system is determined by the adaptive stochastic resonance enhancement module, which adopts the classic symmetric double-well potential function.

[0055] The symmetric double-well potential function in the adaptive stochastic resonance enhancement module is configured by the system structure parameters. and Set the barrier height Set the barrier height The principle is to enable the system to operate solely based on the input signal value without external noise. The energy is insufficient to drive the system state variables. Overcoming barriers A transition occurs from one potential well to another; and after injecting auxiliary noise of appropriate strength, the signal value... It works in synergy with noise to make the system state variables Able to match the input signal value A synchronous transition occurs between two potential wells when rhythms of the same frequency transition. This phenomenon is called stochastic resonance. Through this transition process, the output response signal is enhanced. The amplitude will be significantly greater than the input signal value. This amplifies the amplitude of subtle features.

[0056] Random noise term in Langevin equations Defined as controlled noise intensity function With unit Gaussian white noise The product of, i.e. ; in, It satisfies the statistical properties: the mean is zero, and the autocorrelation function is Dirac. function; The value of the controlled noise intensity function directly determines the random energy level of the injected system, which will be dynamically adjusted by the subsequent feedback control mechanism based on local dynamic instability.

[0057] See attached document Figure 3 This embodiment details how the adaptive stochastic resonance enhancement module uses the finite-time Lyapunov exponent as a feedback quantity to dynamically adjust the noise energy injected into the bistable system.

[0058] The core of the adaptive noise intensity control mechanism lies in identifying the metastable region during the state transition process of a biophysiological system and precisely matching the optimal random resonance noise intensity within this region. The adaptive random resonance enhancement module pre-sets an instability lower limit threshold. and the upper limit threshold of instability Instability lower limit threshold The value of is typically set to a positive real number slightly greater than zero, used to indicate that the system begins to deviate from its stable periodic trajectory and exhibits initial divergent characteristics. Instability upper limit threshold. The value of is set to the average Lyapunov exponent when the system is in a fully developed chaotic state, indicating that the system has entered a highly disordered state. The resulting numerical range is... Defined as the dynamic critical response interval. The adaptive stochastic resonance enhancement module monitors the finite-time Lyapunov exponent input from the local dynamic instability assessment module in real time. The adaptive stochastic resonance enhancement module will use the current finite-time Lyapunov exponent. The control is compared with the dynamic critical response range, and a segmented control strategy is executed based on the comparison results.

[0059] See attached document Figure 3 This embodiment details how the adaptive stochastic resonance enhancement module uses the finite-time Lyapunov exponent as feedback to dynamically adjust the noise energy injected into the bistable system. The core of the noise intensity adaptive control mechanism lies in identifying the metastable region during the state transition process of the biophysiological system and matching the optimal stochastic resonance noise intensity within the metastable region. The adaptive stochastic resonance enhancement module pre-sets an instability lower limit threshold. and the upper limit threshold of instability Instability lower limit threshold The value is set to a positive real number slightly greater than zero, used to indicate that the system begins to deviate from its stable periodic trajectory and exhibits initial divergent characteristics; the instability upper limit threshold. The value of is set as the average Lyapunov exponent when the system is in a fully developed chaotic state, used to indicate that the system has entered a highly disordered state, and the numerical range determined therefrom is given. It is defined as the dynamic critical response interval.

[0060] The adaptive stochastic resonance enhancement module monitors in real time the finite-time Lyapunov exponent input by the local dynamic instability assessment module. The adaptive stochastic resonance enhancement module will enhance the current finite-time Lyapunov exponent. The control is compared with the dynamic critical response range, and a segmented control strategy is executed based on the comparison results.

[0061] When the Lyapunov index with a limited time is detected Located within the dynamic critical response range (i.e. When this occurs, it indicates that the sleep physiology system is currently in a critical state that is extremely sensitive to disturbances, such as the transition from non-rapid eye movement (NREM) to rapid eye movement (REM) sleep, or the pre-arousal stage during deep sleep. In this state, weak pathological signals are often masked by background fluctuations, but the dynamic structure of the system has not yet completely disintegrated. At this time, the adaptive stochastic resonance enhancement module activates the noise injection function and calculates the controlled noise intensity function according to the exponential growth law. .

[0062] When the Lyapunov index with a limited time is detected Less than the lower threshold of instability When this occurs, it indicates that the system is in a stable deep sleep state or a regular periodic motion state, with a high signal-to-noise ratio and obvious dynamic characteristics, requiring no additional noise assistance. At this time, the adaptive stochastic resonance enhancement module controls the noise intensity function... The value is set to zero, meaning no auxiliary noise is injected.

[0063] When the Lyapunov index with a limited time is detected Greater than the upper limit threshold of instability When this occurs, it indicates that the system is in a completely random or highly complex turbulent state. In this case, injected noise will only further degrade the signal quality. Therefore, the adaptive stochastic resonance enhancement module will also control the noise intensity function. The value is set to zero to implement a silent strategy.

[0064] Through the above mechanism, the adaptive stochastic resonance enhancement module only activates stochastic resonance gating when the system is in a specific dynamically sensitive window period. It uses controlled noise energy to raise the weak signal so that it can overcome the potential barrier of the bistable system, thereby exhibiting periodic transitions at the output end and realizing selective signal enhancement for specific nonlinear states.

[0065] See attached document Figure 3 This embodiment details how the adaptive stochastic resonance enhancement module solves the Langevin equation based on a discretized numerical calculation method to generate the enhanced signal.

[0066] The adaptive stochastic resonance enhancement module employs the Euler-Maria-Maurice numerical iterative algorithm to discretize and solve the continuous Langevin dynamic equations. The module sets the sampling time step. Sampling time step It is equal to the reciprocal of the sampling frequency of the data acquisition unit, for the i-th discrete time series. Each sampling point corresponds to a time. The adaptive stochastic resonance enhancement module is based on time. System status Calculation time System status The discretized iterative calculation formula is as follows: ; in, Indicates the next sampling time The enhanced signal amplitude; Indicates the current sampling time The enhanced signal amplitude; Indicates the sampling time step; Indicates at time The input is the standardized time series signal value; Indicates at time The controlled noise intensity function value determined by the noise intensity adaptive control mechanism; This represents a sequence of Gaussian random numbers that follows a standard normal distribution. The mean is zero and the variance is one; Represents the bistable state function For state variables The negative gradient represents the deterministic restoring force; Represents external driving force; It represents random fluctuation force.

[0067] During the iterative calculation process, the adaptive stochastic resonance enhancement module executes the following dynamic evolution logic: First, system state variables It is initialized to zero or a random, small perturbation value at the bottom of the potential well; then, the adaptive stochastic resonance enhancement module calculates the resultant force when the input signal... The amplitude is small, and the controlled noise intensity function When the net force is zero, the net force is insufficient to overcome the height of the potential barrier. At this time, the system state variables The signal is confined to oscillate slightly near the bottom of a single-sided potential well, and the amplified output signal is then... Manifestation of input signal The linear small amplitude response.

[0068] When input signal Contains weak feature information and a controlled noise intensity function When at a non-zero optimization level, random fluctuation force and external driving force Nonlinear cooperation occurs, in the input signal During the positive half-cycle, external driving forces raise the system state variables. Moving towards the top of the potential barrier, the injected random noise energy provides the additional thrust needed to cross the barrier, thus changing the system's state variables. With a high probability of crossing the barrier point And it transitions to the right potential well; similarly, with the input signal... The negative half-cycle, system state variables With the assistance of reverse driving force and noise, it jumps back to the left potential well.

[0069] Through the above process, the input signal value, which was originally weak and unable to independently drive the system to cross the potential barrier, is transformed. The particle transition rhythm of the bistable system was controlled with noise assistance, due to the distance between the bottoms of the two potential wells. This is a macroscopic physical quantity determined by the system's structural parameters; the distance between the bottoms of the two potential wells is significantly greater than the original amplitude of the weak input signal. Therefore, the system state variable... The large-amplitude synchronous transition between the two potential wells directly constitutes the enhanced signal. Enhanced signal It retains the original input signal value The frequency and phase characteristics, along with the large-amplitude waveform derived from the potential energy of a bistable system, enable the dynamic extraction of subtle physiological characteristics. The adaptive stochastic resonance enhancement module then calculates the enhanced signal... The sequence is transmitted to the phase space symbolization module.

[0070] See attached document Figure 3 This embodiment details how the phase space symbolization module converts the continuously varying enhanced signal into a discrete symbol sequence, extracting the topological features of the system dynamics. The phase space symbolization module first processes the enhanced signal from the adaptive stochastic resonance enhancement module... Perform a second phase space reconstruction; the phase space symbolization module targets the enhanced signal. Determine the reconstruction parameters, including the symbolization time delay parameter. and symbolic embedding dimension parameters The phase space symbolization module uses the sliding window method to construct the enhanced signal state vector. Enhance signal state vector The mathematical expression is: ; in, Indicates the enhanced signal at the 1st The amplitude at each sampling time; Represents discrete points in time; This represents the matrix transpose operation.

[0071] Next, the phase space symbolization module performs a phase space segmentation operation based on the Poincaré section principle. Define a set of hyperplanes in the dimensional phase space, the set of hyperplanes containing A series of non-intersecting hyperplanes Each hyperplane divides the entire phase space into A set of non-overlapping topological regions, denoted as .

[0072] The hyperplane is constructed using a maximum entropy partitioning strategy, and the phase space symbolization module statistically enhances the signal state vector. The projection distribution in each direction of phase space is adjusted. The position and normal direction of each hyperplane make the enhanced signal state vector Fall into each topological region Since the probabilities are equal, the maximum entropy partitioning strategy can ensure that the generated symbol sequence has the maximum information capacity and avoid statistical deviations in subsequent entropy calculations due to the low frequency of some symbols.

[0073] After completing the region partitioning, the phase space symbolization module performs the symbol mapping step, checking each sampling time. The corresponding enhanced signal state vector The spatial coordinates are used to determine the topological region where the enhanced signal state vector is located. The phase space symbolization module defines the symbol mapping function. The symbolic mapping function maps continuous vectors to discrete integer symbols. The expression is: ; in, Indicates the first The symbol value generated at each moment; The range of values ​​for is the set of integers. ; Indicates that it belongs to the symbol; Indicates the first A topological region.

[0074] By performing the above mapping on the entire time series, the phase space symbolization module generates the symbol sequence corresponding to the enhanced signal. symbol sequence The expression is: ; in, Indicates the total length of the symbol sequence.

[0075] Phase space symbolization is essentially a coarse-grained processing method. By converting continuous amplitude variations into discrete region numbers, the phase space symbolization module filters out the enhanced signal. The residual minute random jitters and amplitude noise only preserve the orbital characteristics of the system state transitions between different topological regions. This symbol sequence based on the Poincaré section principle can robustly characterize the switching behavior of the sleep physiological system between different attractor orbits, providing an interference-resistant discretized data foundation for subsequent calculation of cross-modal causal information flow.

[0076] See attached document Figure 3 This embodiment details how the cross-modal causal analysis module, based on information theory principles, uses symbolic sequences to quantify the intensity of directional information transmission between different physiological subsystems. The cross-modal causal analysis module first receives two sets of time-synchronized symbolic sequences output by the phase space symbolization module: a respiratory signal symbol sequence representing the source system. and the electrocardiogram signal symbol sequence representing the target system In order to capture the time dependence and memory characteristics of the system, the cross-modal causal analysis module uses state space reconstruction technology to construct historical state vectors.

[0077] The cross-modal causal analysis module defines the target history embedding length. Heyuan history embedding length For each moment The cross-modal causal analysis module constructs historical state vectors of electrocardiogram signals. and respiratory signal history state vector ECG signal historical state vector Includes ECG signal symbol sequences At any moment and the past The sign value at each moment, i.e. Respiratory signal history state vector Contains respiratory signal symbol sequences At the present moment and the past The sign value at each moment, i.e. Meanwhile, the cross-modal causal analysis module identifies electrocardiogram signal symbol sequences. In the next moment future state .

[0078] Based on the constructed state vector, the cross-modal causal analysis module estimates the joint probability distribution and conditional probability distribution by statistically analyzing the frequency of symbol combinations throughout the sequence. Subsequently, the cross-modal causal analysis module calculates the symbol transfer entropy from the respiratory system to the cardiovascular system. Symbolic propagation entropy Capable of measuring the known historical state of the target system itself. Based on this, the historical state of the source system is introduced. Predicting the future state of the target system The degree of reduction in uncertainty.

[0079] In symbolic transitive entropy, the numerator term The denominator reflects the predicted probability after combining historical information from both systems. This reflects the prediction probability based solely on the target system's own historical information. When there is a causal driving effect between the respiratory system and the cardiovascular system, a respiratory signal historical state vector is introduced. This will significantly improve the ability to predict the future state of ECG signals. The prediction accuracy is such that the numerator is greater than the denominator, thus producing a positive sign transfer entropy. Conversely, if the two systems are independent or there is no driving force in that direction, the numerator and denominator are approximately equal, and the sign transfer entropy is... Approaching zero.

[0080] Similarly, the cross-modal causal analysis module calculates the symbolic transfer entropy from the cardiovascular system to the respiratory system by exchanging the roles of the source and target systems. The cross-modal causal analysis module utilizes the nonlinear statistic of symbolic transfer entropy to overcome the shortcomings of traditional linear correlation analysis, which cannot determine directionality and is insensitive to nonlinear dynamics, and achieves accurate quantification of the hidden coupling mechanism between physiological subsystems during sleep.

[0081] See attached document Figure 3 This embodiment details how the cross-modal causal analysis module derives the net interaction direction based on bidirectional symbolic transfer entropy and defines the master-slave driving relationship between physiological systems. The cross-modal causal analysis module calculates the symbolic transfer entropy from the respiratory system to the cardiovascular system. And the symbolic transfer entropy from the cardiovascular system to the respiratory system Next, the net information flow index is calculated. The net information flow index is used to offset the symmetric correlation components between the two systems due to common history or background noise, thereby highlighting the truly directional causal driving force.

[0082] The cross-modal causal analysis module is based on the calculated net information flow index. The numerical sign and magnitude are used to determine the coupling direction. When the cross-modal causality analysis module determines the net information flow index It is a positive value and its absolute value is greater than the preset coupling significance threshold. (Right now When the cross-modal causal analysis module determines that the physiological network at the current moment is in the respiratory-driven mode, the respiratory central rhythm represented by the respiratory motion data sequence plays a dominant modulatory role on the cardiovascular autonomic regulation represented by the electrocardiogram data sequence. Physiologically, this state corresponds to obvious respiratory sinus arrhythmia, which usually indicates that the subject is in a stable non-rapid eye movement sleep stage and that parasympathetic tone is dominant.

[0083] When the cross-modal causality analysis module determines the net information flow index It is a negative value and its absolute value is greater than the preset coupling significance threshold. (Right now When the cross-modal causal analysis module determines that the physiological network is in a cardiovascular-driven mode, changes in the cardiovascular system take precedence over changes in the respiratory system, and the information flow exhibits reverse transmission characteristics. This reverse drive or strong feedback regulation of breathing by heart rate may occur in specific pathological events or during the transition from deep sleep to wakefulness.

[0084] When the cross-modal causality analysis module determines the net information flow index The absolute value is less than or equal to the preset coupling significance threshold. (Right now When the cross-modal causal analysis module determines that the physiological network at the current moment is in a weakly coupled mode or a decoupled mode, there is no obvious unidirectional driving mechanism between the respiratory system and the cardiovascular system, or the two are independently regulated by the higher-level central nervous system. This decoupled state usually corresponds to the rapid eye movement (REM) sleep period or the wakefulness period. During the REM period, the cerebral cortex is active and issues independent random control commands to the heart rate and breathing, which leads to the breakdown of the originally tight RSA coupling relationship between the two, which is manifested as the net information flow index approaching zero.

[0085] To improve the statistical reliability of the determination, the cross-modal causal analysis module can also perform alternative data testing. The cross-modal causal analysis module generates multiple sets of alternative data sequences by randomly shuffling the time order of the original symbol sequence and calculates the net information flow index distribution of these alternative data sequences. Only when the net information flow index of the original data deviates significantly from the mean of the alternative data distribution, the cross-modal causal analysis module confirms the validity of the above driving relationship; otherwise, it is determined to be a random correlation. The cross-modal causal analysis module outputs the finally determined net information flow index and coupling direction flag to the sleep state determination module.

Claims

1. A sleep disorder intelligent monitoring system based on multimodal physiological signals, characterized in that, include: The data acquisition unit is configured to acquire multimodal physiological signals of the subject, and the data acquisition unit includes an electrocardiogram signal acquisition module, a respiratory signal acquisition module, and a body movement signal acquisition module; The processor unit, which is connected to the data acquisition unit, is configured to process the received multimodal physiological signals. The processor unit includes a signal preprocessing module, a phase space reconstruction module, a local dynamic instability assessment module, an adaptive stochastic resonance enhancement module, a phase space symbolization module, a cross-modal causal analysis module, and a sleep state determination module connected in sequence. The local dynamic instability assessment module is connected to the sleep state determination module, and the adaptive stochastic resonance enhancement module is connected to both the signal preprocessing module and the local dynamic instability assessment module. The adaptive stochastic resonance enhancement module is configured to receive the finite-time Lyapunov exponent output by the local dynamic instability assessment module as a feedback control variable, dynamically adjust the noise intensity of the injected system, and perform nonlinear stochastic resonance enhancement on the signal output by the signal preprocessing module. The result output unit is connected to the sleep state determination module and is configured to output the determination result.

2. The intelligent sleep disorder monitoring system based on multimodal physiological signals according to claim 1, characterized in that, The signal preprocessing module is configured to perform baseline drift removal and standard score normalization on each time series of the multimodal physiological signal. The phase space reconstruction module is configured to receive the standardized time series, calculate the minimum value of the mutual information function to determine the time delay parameter, calculate the proportion of false neighbors to determine the embedding dimension parameter, and map the one-dimensional time series into a reconstruction vector sequence in a high-dimensional topological space based on the time delay parameter and the embedding dimension parameter.

3. The intelligent sleep disorder monitoring system based on multimodal physiological signals according to claim 2, characterized in that, The local dynamic instability assessment module is configured to calculate the finite-time Lyapunov exponent of the phase space trajectory, which is used to quantify the local trajectory divergence rate of the system at the current moment. The local orbit divergence rate is obtained by calculating the natural logarithm of the rate of change of the Euclidean distance between the reference trajectory point and the nearest neighbor trajectory point in the phase space after a preset evolution time.

4. The intelligent sleep disorder monitoring system based on multimodal physiological signals according to claim 3, characterized in that, The adaptive stochastic resonance enhancement module is configured to run a bistable system model based on the Langevin equation; The Langevin equation defines the rate of change of the enhanced signal over time as equal to the sum of the nonlinear restoring force, the external driving force, and the random noise term. The external driving force corresponds to the standardized time series of the input, the nonlinear restoring force is determined by the negative gradient of the bistable state function with respect to the system state variables, and the random noise term is defined as the product of the controlled noise intensity function and unit Gaussian white noise.

5. The intelligent sleep disorder monitoring system based on multimodal physiological signals according to claim 4, characterized in that, The adaptive stochastic resonance enhancement module is configured to adjust the controlled noise intensity function according to the finite-time Lyapunov exponent, and the adjustment logic is as follows: A lower instability threshold and an upper instability threshold are preset, and the range between the lower instability threshold and the upper instability threshold is defined as the dynamic critical response interval; When the finite-time Lyapunov exponent is detected to be within the dynamic critical response range, the value of the controlled noise intensity function is set to be exponentially positively correlated with the finite-time Lyapunov exponent. When the finite-time Lyapunov exponent is detected to be outside the dynamic critical response range, the value of the controlled noise intensity function is set to zero.

6. The intelligent sleep disorder monitoring system based on multimodal physiological signals according to claim 1, characterized in that, The phase space symbolization module is configured to perform a second phase space reconstruction on the enhanced signal output by the adaptive stochastic resonance enhancement module to generate an enhanced signal state vector; The phase space symbolization module uses a set of hyperplanes to divide the phase space where the enhanced signal state vector is located into multiple non-overlapping topological regions, and assigns a unique integer symbol according to the topological region into which the enhanced signal state vector falls at the current time, generating a discrete symbol sequence.

7. The intelligent sleep disorder monitoring system based on multimodal physiological signals according to claim 6, characterized in that, The cross-modal causal analysis module is configured to receive the symbol sequence corresponding to the respiratory signal and the symbol sequence corresponding to the electrocardiogram signal, and construct the historical state vector of the respiratory signal and the historical state vector of the electrocardiogram signal, respectively. The cross-modal causal analysis module calculates the symbolic transfer entropy from respiratory signal to ECG signal based on joint probability distribution. The symbolic transfer entropy from respiratory signal to ECG signal is used to measure the reduction in uncertainty of predicting the future state of ECG signal under the condition that the historical state vectors of ECG signal and respiratory signal are known. The cross-modal causal analysis module is also configured to calculate the symbolic transfer entropy from electrocardiogram signals to respiratory signals.

8. The intelligent sleep disorder monitoring system based on multimodal physiological signals according to claim 7, characterized in that, The cross-modal causal analysis module is also configured to calculate a net information flow index, which is equal to the symbolic transfer entropy from the respiratory signal to the electrocardiogram signal minus the symbolic transfer entropy from the electrocardiogram signal to the respiratory signal.

9. The intelligent sleep disorder monitoring system based on multimodal physiological signals according to claim 8, characterized in that, The sleep state determination module is configured to receive the finite-time Lyapunov exponent, the symbol transfer entropy, and the net information flow exponent, and to calculate the disorder scale of the enhanced signal. The sleep state determination module executes the following determination logic: When the net information flow index is positive and greater than the preset coupling significance threshold, the subject is determined to be in respiratory drive mode; When the net information flow index is negative and its absolute value is greater than the preset coupling significance threshold, the subject is determined to be in cardiovascular-driven mode. When the absolute value of the net information flow index is less than or equal to the preset coupling significance threshold, the subject is determined to be in a weak coupling mode, and further, the awake period and the REM period are distinguished based on whether the disorder scale is higher than the preset awakeness threshold.

10. A method for intelligent monitoring of sleep disorders based on multimodal physiological signals, characterized in that, Includes the following steps: S1. Acquire the subject's electrocardiogram data sequence, respiratory motion data sequence, and triaxial acceleration data sequence, and perform baseline drift removal and standard score standardization on the data sequences; S2. Perform phase space reconstruction on the standardized data sequence, map the one-dimensional time series to a reconstruction vector in a high-dimensional topological space, and calculate the finite-time Lyapunov exponent of the phase space trajectory. S3. Construct an adaptive stochastic resonance feedback loop based on the finite-time Lyapunov exponent: When the finite-time Lyapunov exponent is within a preset dynamic critical response range, Gaussian white noise is injected into the bistable nonlinear system based on the Langevin equation, and the noise intensity is set to be exponentially positively correlated with the finite-time Lyapunov exponent. The enhanced signal is output using the stochastic resonance effect, and the enhanced signal is then mapped to phase space symbolization. By dividing the phase space region, the continuous signal is converted into a discrete symbol sequence. S4. Calculate the bidirectional symbol transfer entropy between the symbol sequence corresponding to the respiratory signal and the symbol sequence corresponding to the electrocardiogram signal, and calculate the net information flow index based on the difference in bidirectional symbol transfer entropy, and calculate the disorder scale of the enhanced signal. S5. Determine the sleep stage status and sleep disorder events of the subject based on the finite-time Lyapunov index, the symbolic transfer entropy, the net information flow index, and the disorder scale.

Images