An acute stress state evaluation method and system based on heart-brain interaction analysis

Abstract

The application provides an acute stress state evaluation method and system based on heart-brain interaction analysis, comprising the following steps: constructing a stress task scene and synchronously collecting electroencephalogram (EEG) and electrocardiogram (ECG) signals; through constructing a segmented polynomial function meeting a shape keeping constraint, adaptively reconstructing an ECG R-R interval sequence, realizing time alignment of the ECG and EEG signals; using wavelet transform to extract a multi-band power sequence; using a multi-channel time delay maximum information coefficient algorithm to quantize bidirectional and direct connection strength between the ECG and each EEG channel, and constructing a multi-band directed weighted network; finally, extracting network features to input a classification model, and outputting an acute stress state evaluation result. The application solves the problems of distortion of heart-brain signal alignment and inaccuracy of interaction quantization in the traditional method, and significantly improves the accuracy and objectivity of acute stress state evaluation in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of signal analysis and psychophysiological state assessment technology, specifically to an acute stress state assessment method and system based on heart-brain interaction analysis. Background Technology

[0002] Acute stress refers to a short-term, rapidly initiating psychophysiological response state that occurs when an individual faces a sudden threat, high-risk decision-making, or intense emotional stimulation. Acute stress often occurs in critical situations, such as law enforcement, emergency response, or high-risk operations, and the quality of the response directly affects an individual's decision-making accuracy, behavioral execution ability, and safety level. Numerous studies have shown that improper regulation of acute stress can not only lead to immediate operational errors but also increase the long-term risk of anxiety disorders, post-traumatic stress disorder, and cardiovascular events. Therefore, establishing objective and reliable methods for assessing acute stress is of great significance for understanding stress response mechanisms and for providing mental health assessment services for high-risk populations.

[0003] Currently, the assessment of acute stress primarily relies on psychometric scales or single-modal physiological signal analysis (such as electroencephalogram (EEG) or electrocardiogram (ECG) signals. However, traditional scale assessments are highly subjective; furthermore, the stress response is essentially the result of the synergistic action of the central nervous system (CNS) and the autonomic nervous system (ANS), and single-modal EEG or ECG signals cannot fully reveal the intrinsic mechanisms of acute stress. Under stress, the cerebral cortex, limbic system, and brainstem structures regulate cardiovascular activity through the central autonomic network (CAN); simultaneously, sensory signals from the heart continuously feed back to the brain through interoceptive pathways, participating in emotional experience, maintaining alertness, and allocating cognitive resources. This brain-heart bidirectional regulatory mechanism is considered an important neurophysiological basis for the generation and regulation of mental stress and emotional responses, and can be characterized by quantifying the interaction between ECG and EEG signals; single-modal analysis cannot fully reveal this bidirectional regulatory mechanism.

[0004] Furthermore, electrocardiogram (ECG) and electroencephalogram (EEG) signals are typical nonlinear neurophysiological signals. Methods for quantifying nonlinear interactions (such as transfer entropy) often rely on long-term steady-state data or struggle to handle direct causal relationships between multi-channel signals. In the joint analysis of ECG and EEG signals, time axis alignment issues also exist. Traditional interpolation methods (such as cubic splines) easily introduce non-physiological overshoot and waveform distortion in critical regions such as the RR interval, affecting the accuracy of subsequent analyses.

[0005] Therefore, there is an urgent need to establish a quantitative analysis method that can effectively capture the bidirectional interaction between the heart and brain, so as to provide a reliable indicator for the assessment of acute stress. Summary of the Invention

[0006] This invention aims to solve the following technical problems: how to accurately align electrocardiogram (ECG) signals and electroencephalogram (EEG) signals in the time dimension while avoiding physiological distortion of waveform morphology; how to quantitatively characterize the nonlinear, direct, and directional information interaction relationship between multi-channel EEG and ECG signals in specific brain frequency bands; and how to construct an assessment model that can comprehensively reflect the collaborative working mode of the central and autonomic nervous systems based on the quantitative relationship of heart-brain interaction, so as to objectively and accurately assess acute stress state.

[0007] To achieve the above objectives, this invention provides a method for assessing acute stress based on heart-brain interaction analysis, comprising the following steps: A virtual reality interactive scenario incorporating stress-induced tasks was constructed to induce acute stress in the subjects, and multi-channel EEG and ECG signals were simultaneously collected while the subjects were in the virtual reality interactive scenario.

[0008] In this invention, subjects perform stress-induced tasks in a constructed high-pressure virtual reality scenario, and simultaneously collect multi-channel EEG signals and single-channel ECG signals, saving the data to a computer.

[0009] The virtual reality scenario combines the traditional Stroop color word test and mental arithmetic stress task. Test subjects complete the above test tasks in a highly realistic virtual scenario, thereby effectively inducing acute stress.

[0010] Preprocessing was performed on the EEG and ECG signals respectively. Based on the R-wave peak position and RR interval information of the ECG signal, the ECG signal was divided into several continuous RR interval sequences.

[0011] In this embodiment, electrocardiogram (ECG) and electroencephalogram (EEG) signals of the subjects are acquired simultaneously, wherein the sampling frequency f of the EEG signals is... EEG The sampling frequency f of the electrocardiogram signal ECG To maintain consistency, bandpass filtering and baseline drift correction were sequentially applied to the electrocardiogram (ECG) signal, and bandpass filtering and power line interference removal were applied to the electroencephalogram (EEG) signal. Principal component analysis was used to remove signals such as eye movement and electromyography (EMG).

[0012] In the preprocessed electrocardiogram (ECG) signal, a peak detection algorithm is used to identify the R-wave peak position corresponding to each cardiac cycle, resulting in an R-wave peak time series: (1) Based on the time interval between adjacent R wave peaks, an electrocardiogram RR interval sequence is constructed: (2) in, (3) Using adjacent R-wave peak values ​​as boundaries, the electrocardiogram signal is divided into several consecutive RR intervals: (4) Each RR interval contains a complete ECG waveform structure, which is used for subsequent time reconstruction processing.

[0013] Within each RR interval sequence shown, a piecewise polynomial reconstruction function satisfying morphology preservation constraints is constructed to adaptively reconstruct the ECG signal in time, thereby generating a reconstructed ECG signal aligned with the time dimension of the EEG signal.

[0014] In this embodiment, based on the sampling time sequence t of the EEG signal EEG Determine the set of target time sampling points corresponding to the current RR interval: (5) Using the original sampling points of the ECG signal within the interval as control points, a piecewise polynomial reconstruction function that satisfies the morphology preservation constraint is constructed, so that it can achieve continuous mapping on the time axis while maintaining the monotonicity and local extremum structure of the ECG waveform.

[0015] Within each subinterval, define a piecewise third-order polynomial function: (6) When adjacent sampling points of the ECG RR interval signal show a monotonically increasing or decreasing trend, the sign of the first derivative of the polynomial within the corresponding interval is kept consistent to avoid non-physiological oscillations or overshoot. To satisfy the morphology preservation constraint, the first derivative at the endpoints of the interval is estimated using a weighted difference method: (7) in, (8) When the waveform changes trend, the endpoint slope is automatically suppressed to prevent unreasonable oscillations in the waveform.

[0016] Once the endpoint function values ​​and first derivatives are determined, the coefficients of the third-order polynomial within each subinterval can be uniquely determined. This constitutes a complete piecewise polynomial reconstruction function.

[0017] Finally, the reconstructed ECG signal segments obtained during each RR interval are spliced ​​together in chronological order to generate a complete reconstructed ECG signal. .

[0018] Using the above method, adaptive time reconstruction of the ECG RR interval sequence was achieved without changing the original waveform morphology of the ECG signal. This allows the ECG signal to be accurately aligned with the sampling time axis of the EEG signal, providing a reliable synchronous data foundation for subsequent heart-brain interaction analysis.

[0019] Wavelet transform was used to process the aligned EEG signal and ECG RR interval sequence, respectively, and the average wavelet power in different frequency bands was extracted as a function of time to generate the average wavelet power time series in different frequency bands of ECG and EEG.

[0020] In this embodiment, in order to simultaneously capture the changes in the time-frequency space of the EEG signal and the reconstructed ECG RR interval sequence, the above data is subjected to Morlet wavelet transform to obtain the time-frequency energy map of the ECG and EEG signals.

[0021] Subsequently, new time series of average wavelet power changes over time for predefined frequency bands (δ: 0.2-3 Hz, θ: 3-8 Hz, α: 8-13 Hz, β: 13-30 Hz, γ: 30-50 Hz) of ECG and EEG signals were calculated.

[0022] The bidirectional connectivity strength between the average power time series of ECG and EEG in a specific frequency band is calculated using the multi-channel time-delay maximum information coefficient algorithm.

[0023] In this embodiment, the time-delay maximum information coefficient algorithm is only applicable to the calculation of correlations between two variables and cannot meet the needs of multivariate correlation studies. In particular, due to the transmission of information flow, there are often both direct and indirect causal relationships between multi-channel neurophysiological signals. Only direct causal relationships can reflect the essential interaction mechanism between phenomena; therefore, research often focuses more on direct directed connections. To this end, the multi-channel time-delay maximum information coefficient algorithm (M-TDMIC) is proposed to quantify the direct directed connections between multi-channel neurophysiological signals.

[0024] Considering the intermediate variable z, the M-TDMIC definition between x and y is: (9) Where v is a random vector, and k, l, and m are the embedding dimensions. and The time-delay prediction range satisfies .like ,but ;like ,but Embedded vector Representing past values ​​of z can be used to predict time delays. The future y provides useful information. Embedded vector Representing past values ​​of x can be used to predict time lag moments. This provides useful information for the future.

[0025] like If the causal relationship is positive, it indicates a direct causal relationship between x and y; otherwise, it indicates an indirect causal relationship. In this embodiment, x and y represent the time series of average power of cardiac and cerebral electrocardiograms.

[0026] A multi-band bidirectional connection network for the heart and brain was constructed based on the bidirectional connection strength value.

[0027] In this embodiment, the connectivity between heart and brain signals was quantified using the multi-channel time-delay maximum information coefficient method, resulting in 65×65 directed weighted adjacency matrices for predefined frequency bands (64-channel EEG signals and single-channel reconstructed ECG RR interval signal sequences). Then, a multi-band bidirectional heart-brain interaction network was constructed using electrodes as network nodes and the connection strength between electrodes as edges.

[0028] Features of the multi-band brain-heart bidirectional interaction network are extracted, input into a pre-trained deep learning classification model, and output the acute stress state assessment results.

[0029] In the embodiments, one can either directly input the mind-brain bidirectional interaction network into a pre-trained deep learning classification model, or input the network features of the mind-brain bidirectional interaction network into a classification learning model and output the acute stress state assessment results.

[0030] The present invention also provides an acute stress state assessment system based on heart-brain interaction analysis; An acute stress state assessment system based on heart-brain interaction analysis, comprising: The scene generation module is configured to: construct highly realistic virtual reality scenes, combined with traditional stress test tasks; The acquisition module is configured to acquire multi-channel EEG and ECG signals to be evaluated. The sequence construction module is configured to: extract the RR interval sequence from the ECG signal and construct the ECG interval sequence; and perform adaptive temporal reconstruction of the ECG RR interval sequence by constructing a piecewise polynomial reconstruction function that satisfies the morphology preservation constraint, so as to generate a reconstructed ECG RR interval sequence aligned with the time dimension of the EEG signal. The module for constructing a bidirectional interaction network between the heart and brain is configured to: calculate the maximum information coefficient of the multivariate time delay between the multi-band reconstructed ECG RR interval sequence and the EEG signal, and construct a multi-band bidirectional interaction network between the heart and brain. The stress state assessment module is configured to process the heart-brain bidirectional interaction network through a trained acute stress state classification model to generate and output acute stress state assessment results.

[0031] Compared with the prior art, the present invention has the following beneficial effects: By constructing virtual reality interactive scenarios and simultaneously collecting multi-channel EEG and ECG signals, stress responses were induced in subjects under high ecological validity in a near-real environment. By introducing key feature points of ECG signals as time constraints, a piecewise polynomial reconstruction function that satisfies the morphology preservation constraint is constructed to perform adaptive time reconstruction of ECG signals, thereby generating reconstructed ECG signals that are aligned with the time dimension of EEG signals. This effectively avoids the waveform overshoot and distortion problems caused by traditional cubic spline interpolation, and provides a reliable data foundation for assessing acute stress states. The average wavelet power obtained by wavelet transform can comprehensively capture the interactive information contained in the time and frequency domains of ECG and EEG signals. By using the multi-channel time-delay maximum information coefficient algorithm to process the average wavelet power time series of specific frequency bands of the heart and brain, it is possible to effectively capture the linear and nonlinear bidirectional connectivity between the heart and brain. This overcomes the technical shortcomings of traditional linear analysis methods (such as Pearson correlation coefficient, coherence method, etc.) or nonlinear analysis methods that rely on long steady-state data (such as transfer entropy, fuzzy entropy, etc.) which are difficult to accurately quantify the bidirectional interaction between the heart and brain. It can comprehensively analyze the collaborative interaction patterns of the central nervous system and the autonomic nervous system under acute stress from the perspective of network topology, avoiding the problem that single-modal signals cannot fully reflect the characteristics of stress.

[0032] This application achieves the goal of improving the accuracy, robustness, and objectivity of acute stress state assessment in complex environments by integrating heart-brain interaction analysis. Attached Figure Description

[0033] Figure 1 A flowchart of an acute stress state assessment method based on heart-brain interaction analysis provided in an embodiment of the present invention; Figure 2 This is a schematic diagram comparing the performance of the adaptive time reconstruction algorithm based on shape preservation constraints with traditional methods in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the principle of the Multi-channel Time Delay Maximum Information Coefficient (M-TDMIC) algorithm used to quantify the direct causal relationship from x to y in an embodiment of the present invention. Figure 4 This is a schematic diagram of a pre-constructed directed weighted network for bidirectional brain-heart interaction under a single frequency band, as described in an embodiment of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0035] Example 1 In this embodiment, step S1 constructs a high-stress interactive scenario based on virtual reality (VR), which integrates the classic Stroop color word test and mental arithmetic task. Subjects wear a 64-lead EEG cap and single-lead ECG electrodes. During the task execution within the VR scenario, EEG and ECG signals are simultaneously acquired at a sampling rate of 1000Hz, and the data is saved.

[0036] In this embodiment, the parameter values ​​are strictly based on physiological principles and signal processing standards: ECG filtering from 0.5 to 40 Hz effectively removes baseline drift and high-frequency noise while preserving key R-wave features; EEG filtering from 0.5 to 50 Hz fully covers the core frequency bands required for stress assessment, and, combined with a 50 Hz notch filter and PCA algorithm, accurately removes power frequency and EMG artifacts. In the feature extraction stage, considering the physiological characteristics of "short-term rapid response" in acute stress, a 1-second time window is used to effectively balance the stability of frequency domain feature extraction with temporal domain resolution; simultaneously, a 0.1-second sliding step (90% data overlap) ensures the smoothness of the feature sequence, meeting the high temporal resolution requirement for accurately capturing sub-second-level dynamic information interaction between the heart and brain.

[0037] Step S2 specifically includes: ① The electrocardiogram signal was bandpass filtered from 0.5 to 40 Hz, the electroencephalogram signal was bandpass filtered from 0.5 to 50 Hz, and a 50 Hz notch filter was used to remove power frequency interference. Then, principal component analysis (PCA) was used to remove artifacts such as electrooculogram.

[0038] ② The peak position of the R wave in the electrocardiogram signal is identified using peak detection algorithms such as Pan-Tompkins, and a sequence is obtained. Calculate the RR interval An RR interval sequence was constructed, and the electrocardiogram signal was divided into several consecutive intervals using adjacent R waves as boundaries. .

[0039] ③ For each RR interval Ω i Based on the EEG signal sampling time sequence t EEG Determine the set of target time sampling points within this interval. .

[0040] ④ Construct a piecewise third-order polynomial function on the original ECG sampling points. Based on the trend of adjacent sampling points, the endpoint slope is estimated by weighted difference. When the waveform changes trend, the slope at the endpoints is automatically suppressed, thereby forcing the polynomial to remain monotonic within that interval and avoiding overshoot and oscillation.

[0041] ⑤ Based on the determined endpoint values ​​and slope, solve for the polynomial coefficients to construct a complete piecewise polynomial reconstruction function. Calculate the function value at the target time point as the reconstructed signal value.

[0042] ⑥ All reconstructed ECG signal fragments obtained within the RR interval are spliced ​​together in chronological order to form a reconstructed ECG signal that is completely aligned with the EEG signal.

[0043] Step S3 specifically includes: using Morlet wavelets with a center frequency of 1Hz to perform continuous wavelet transform on the aligned EEG and reconstructed ECG signals. Within five frequency bands—δ (0.2-3Hz), θ (3-8Hz), α (8-13Hz), β (13-30Hz), and γ (30-50Hz)—a time window of 1 second and a step size of 0.1 seconds are set, and the average wavelet power within each time window is calculated to generate a time series of the average wavelet power for each frequency band.

[0044] Step S4 specifically includes: For each frequency band, using the average power time series of ECG as variable y, the average power time series of a certain EEG channel as variable x, and the sequences of the other 63 EEG channels as conditional variables z, calculating the M-TDMIC values ​​from x to y and from y to x. The embedding dimension is... k , l , m Time Delay By traversing all EEG channels, a 65x65 directed weighted adjacency matrix is ​​ultimately generated for each frequency band.

[0045] Furthermore, in the Multi-channel Time Delay Maximum Information Coefficient (M-TDMIC) algorithm, the embedding dimension... k , l , m and time delay , The specific basis for determination and the method for obtaining the value are as follows: 1. Embedding Dimension k , l , m Basis for determination and value: Basis for determination: According to the phase space reconstruction theory (Takens' theorem) in nonlinear dynamics, the embedding dimension determines the extent to which the system state is fully unfolded in phase space. If the dimension is too low, it will lead to phase space trajectory folding (i.e., an increase in false nearest neighbors); if the dimension is too high, it will introduce noise interference and exponentially increase computational complexity.

[0046] Calculation Method: This invention employs the False Nearest Neighbors (FNN) method to adaptively determine the embedding dimension of each variable. Specifically, for each frequency band of ECG and EEG power time series, the proportion of false nearest neighbors under different embedding dimensions is calculated. When this proportion first drops to a preset threshold (e.g., below 5%), the corresponding dimension is the optimal embedding dimension.

[0047] Actual values: In the actual implementation of assessing acute stress, considering the data length and signal-to-noise ratio of the average wavelet power sequences of ECG and EEG, the variables... x , y , z Embedding dimension k , l , m The preferred value range is a positive integer within [2, 5]. To reduce computational complexity while ensuring sufficient information capture capability, a default value is usually used. k = l = m =3.

[0048] 2. Time delay , Basis for determination and value: Physiological basis: Time delay Representative signal x (e.g., EEG of a specific brain region) response to signals y (Such as electrocardiogram) The information transmission lag time that has an impact. Under acute stress, the neurophysiological transmission time between the brain and heart (central nervous system to autonomic nervous system) or between the heart and brain (intersensory feedback) objectively exists, usually between hundreds of milliseconds and several seconds.

[0049] Mathematical basis: The selection of delay vectors must ensure that the correlation between delay vectors is neither completely redundant nor completely uncorrelated. The time prediction range of a condition variable is usually related to... Keep synchronized or set to unit step size.

[0050] Calculation Method: Preferably, this invention uses the Mutual Information (MI) method or combines it with physiological prior knowledge to determine the optimal time delay. Calculation Sequence x ( t )and The self-mutual information function between them, when the mutual information function first reaches a local minimum, the corresponding This is the optimal time delay basis for phase space reconstruction.

[0051] Actual value: In this embodiment, the step size of the wavelet power time series extracted in step S3 is set to 0.1 seconds. This is combined with the physiological delay of neural conduction along the brain-heart axis (approximately 100 to 2000 milliseconds). The physical time constraint is between [0.1 s, 2.0 s]. Therefore, in the algorithm implementation, the time delay parameter... The optimal value range for this is an integer between 1 and 20 (corresponding to the time window step size). In global optimal search, the prediction delay is typically... Values Or set To conduct single-step dynamic correlation evaluation.

[0052] Steps S5 and S6 specifically include: using a 65×65 directed weighted adjacency matrix of five frequency bands as input features, and feeding it into a pre-trained graph convolutional neural network (GCN). This GCN model is trained on a large amount of labeled (high stress / low stress) sample data. The model outputs a probability value between 0 and 1; when the probability is greater than 0.5, it is determined to be a "high acute stress state," and vice versa for a "low acute stress state."

[0053] To ensure the reliability of the evaluation model, this study recruited a total of 80 subjects and collected data using the Neuroscan 64-channel EEG system in a provincial key laboratory. A total of 12,000 effective training samples were extracted using a 3-second sliding window and a 1-second step strategy. The samples were divided into training, validation, and independent test sets in an 8:1:1 ratio. Finally, the model was trained using a 10-fold cross-validation method, achieving an accuracy rate of over 92% on the independent test set.

[0054] Example 2 This embodiment is basically the same as Embodiment 1, except for step S6. After constructing the multi-band brain-heart bidirectional interaction network, instead of directly using the network matrix as the input to the deep learning model, the graph theory features of each frequency band network are calculated, such as the out-degree, in-degree, and betweenness centrality of nodes, as well as the network's global efficiency, clustering coefficient, and feature path length. These cross-band and cross-node graph theory features are straightened and concatenated into a feature vector, which is then input into a pre-trained support vector machine (SVM) classifier to obtain the stress state evaluation result. This method has a smaller computational load and lower hardware requirements.

[0055] Example 3 This embodiment applies the core algorithm to the analysis of event-related potentials (ERPs). In the VR scene, specific stress stimuli (such as color-meaning mismatched Stroop stimuli) are used as event markers. EEG and ECG data are extracted around each event (e.g., 200 milliseconds before to 800 milliseconds after stimulus presentation) to form multiple short-term data segments. For each segment, the method in steps S2-S5 is applied to construct a dynamic, time-evolving heart-brain interaction network sequence. In step S6, this network sequence is the input to the classification model, and the model performs state assessment by learning the dynamic evolution of heart-brain interaction patterns under stress.

[0056] Experimental verification of beneficial effects: To verify the effectiveness of the proposed method based on Mind-Brain Two-Way Interaction Analysis (M-TDMIC), this study conducted a quantitative comparative experiment and performed statistical tests and evaluations on its classification and recognition performance in "acute stress state" and "resting state" compared with existing technologies.

[0057] 1. Experimental Design and Dataset Size This experiment recruited 80 healthy participants. During the experiment, participants were presented with a high-intensity scenario combining the Stroop color word test and mental arithmetic tasks using a high-fidelity virtual reality (VR) device to induce a stress response. The period during which participants performed the high-intensity task was labeled as the "acute stress state," and the period before the task when they wore the device and rested with their eyes closed was labeled as the "resting state."

[0058] All subjects simultaneously acquired 64-channel EEG and single-channel ECG signals using a multi-channel Neuroscan EEG system. A sliding window technique (3-second window length, 1-second step) was used to slice the data, yielding 6,400 valid state data samples (half stress samples and half resting samples). The dataset was randomly divided into training, validation, and independent test sets in an 8:1:1 ratio.

[0059] 2. Comparison of Schemes To highlight the technical advantages of this invention, the following five comparative schemes were set up in the experiment: Option A (Scale and Subjective Self-Rating): State thresholds are defined based on the State Anxiety Inventory (STAI) score and real-time subjective arousal scores between tasks.

[0060] Option B (single-modal EEG analysis): Only the power spectral density (PSD) features of different frequency bands of EEG are extracted and input into a deep convolutional neural network (CNN) for state classification.

[0061] Option C (single-modal HRV analysis): Only the time-frequency domain and nonlinear features (such as SDNN, LF / HF ratio, sample entropy) of the ECG RR interval are extracted, and support vector machine (SVM) is used for state classification.

[0062] Option D (Traditional Heart-Brain Interaction Method): The traditional transfer entropy (TE) is used to quantify the heart-brain coupling relationship and extract network features to input the classifier.

[0063] Scheme E (the method of this invention): Time reconstruction is achieved by using a piecewise polynomial with shape preservation constraints, and a multi-frequency directed weighted network is constructed by combining multi-channel time-delay maximum information coefficient (M-TDMIC), and finally input into a graph convolutional neural network (GCN) for state classification.

[0064] 3. Quantitatively compare the data results On the independent test set, the classification and identification results of each scheme for "acute stress state" and "resting state" (based on the mean of 10-fold cross-validation) are shown in Table 1 below:

[0065] 4. Statistical test results To verify the statistical significance of the performance improvement of the method of this invention, paired samples were used for each evaluation index in the cross-validation. t test: Compared with single-modal methods: the present invention (Scheme E) significantly outperforms single-modal EEG (Scheme B) and single-modal HRV (Scheme C) in terms of accuracy, sensitivity, and specificity. p <0.001). This demonstrates that the bidirectional interaction between the central nervous system and the autonomic nervous system contains more essential physiological information than single-system signals when distinguishing between stress and resting states.

[0066] Compared to traditional interaction methods, the accuracy of this invention (Solution E) is approximately 6.1% higher than that of traditional transfer entropy (TE) interaction analysis (Solution D), and the difference is statistically significant. p <0.05).

[0067] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A method for assessing acute stress based on heart-brain interaction analysis, characterized in that, Includes the following steps: S1: Construct an interactive scenario that includes a stress-inducing task to induce an acute stress state in the subject, and simultaneously collect multi-channel EEG and ECG signals while the subject is in the interactive scenario; S2: Perform preprocessing on the EEG signal and ECG signal respectively, construct an ECG RR interval sequence based on the ECG signal, construct a piecewise polynomial reconstruction function that satisfies the morphology preservation constraint, perform adaptive time reconstruction on the ECG RR interval sequence, and generate a reconstructed ECG signal aligned with the time dimension of the EEG signal; S3: Use wavelet transform to process the aligned EEG signal and the ECG RR interval sequence respectively, extract the data of the average wavelet power change over time in different predefined frequency bands, and generate the average wavelet power time series in different frequency bands of ECG and EEG; S4: Taking into account other channels besides the target channel as conditional variables, the bidirectional connection strength value between the ECG and EEG average power time series in the predefined frequency band is calculated using the multi-channel time delay maximum information coefficient algorithm. S5: Based on the bidirectional connection strength values, construct a multi-band bidirectional directed weighted interaction network for the heart and brain; S6: Extract the features of the multi-band brain-mind bidirectional interaction network, input them into a pre-trained classification model, and generate acute stress state assessment results.

2. The method for assessing acute stress state based on heart-brain interaction analysis according to claim 1, characterized in that, Step S2 specifically includes: Within each interval of the ECG RR interval sequence, a piecewise third-order polynomial function is constructed using the original sampling point as the control point; Based on the signal change trend between adjacent sampling points, constrain the sign of the first derivative of the polynomial in the corresponding interval so that the reconstructed waveform maintains monotonicity and local extremum structure, and avoids non-physiological overshoot or oscillation. The reconstructed electrocardiogram (ECG) signal is obtained by interpolating at the sampling time points of the EEG signal using the constructed piecewise polynomial function.

3. The method for assessing acute stress state based on heart-brain interaction analysis according to claim 2, characterized in that, The step of constraining the sign of the first derivative of the polynomial within the corresponding interval further includes: when the signals of adjacent sampling points within a certain interval are monotonically increasing or decreasing, constraining the sign of the first derivative of the polynomial in that interval to remain consistent; and automatically suppressing the slope of the endpoints when the waveform changes trend.

4. The method for assessing acute stress state based on heart-brain interaction analysis according to claim 2, characterized in that, Step S2 specifically includes: Bandpass filtering and baseline drift correction were performed sequentially on the electrocardiogram signal; In the preprocessed electrocardiogram (ECG) signal, a peak detection algorithm is used to identify the R-wave peak position corresponding to each cardiac cycle, resulting in an R-wave peak time series: (1) Based on the time interval between adjacent R wave peaks, an electrocardiogram RR interval sequence is constructed: in, (3) Using adjacent R-wave peak values ​​as boundaries, the electrocardiogram signal is divided into several consecutive RR intervals: (4) Each RR interval contains a complete ECG waveform structure, which is used for subsequent time reconstruction processing.

5. The method for assessing acute stress state based on heart-brain interaction analysis according to claim 4, characterized in that, Based on the sampling time sequence t of the EEG signal EEG Determine the set of target time sampling points corresponding to the current RR interval: (5) Using the original sampling points of the ECG signal within the interval as control points, a piecewise polynomial reconstruction function satisfying the morphology preservation constraint is constructed. Within each sub-interval, a piecewise third-order polynomial function is defined: (6) The first derivative at the endpoints of the interval is estimated using a weighted difference method: (7) in, (8) After determining the endpoint function values ​​and the first derivative, the coefficients of the third-order polynomial in each subinterval are then uniquely determined. This constitutes a complete piecewise polynomial reconstruction function; Finally, the reconstructed ECG signal segments obtained during each RR interval are spliced ​​together in chronological order to generate a complete reconstructed ECG signal.

6. The method for assessing acute stress state based on heart-brain interaction analysis according to claim 1, characterized in that, In step S4, the multi-channel time-delay maximum information coefficient algorithm is used to quantify the direct directed connectivity between multi-channel neurophysiological signals. Considering the condition variable z, the M-TDMIC definition between x and y is: (9) Where v is a random vector, and k, l, and m are the embedding dimensions. and The time-delay prediction range satisfies .

7. The method for assessing acute stress state based on heart-brain interaction analysis according to claim 6, characterized in that, The variables x and y represent the average wavelet power time series of an EEG channel and an ECG channel, respectively, and the condition variable z represents an intermediate variable.

8. The method for assessing acute stress state based on heart-brain interaction analysis according to claim 1, characterized in that, The predefined frequency band in step S3 includes: δ: 0.2-3Hz, θ: 3-8Hz, α: 8-13Hz, β: 13-30Hz, γ: 30-50Hz.

9. The method for assessing acute stress state based on heart-brain interaction analysis according to claim 1, characterized in that, The multi-band brain-mind bidirectional interaction network in step S5 is a directed weighted network, in which the EEG electrodes and ECG signal channels serve as network nodes, and the bidirectional connection strength value serves as the weight of the directed edges between nodes.

10. An acute stress state assessment system based on heart-brain interaction analysis, used to implement the method as described in any one of claims 1 to 9, characterized in that, include: The scene generation and acquisition module is configured to: construct an interactive scene containing a stress-induced task to induce an acute stress state in the subject, and simultaneously acquire multi-channel EEG and ECG signals; The signal preprocessing and time alignment module is configured to: preprocess the signal, construct an ECG RR interval sequence based on the ECG signal, construct a piecewise polynomial reconstruction function that satisfies the morphology preservation constraint, perform adaptive time reconstruction on the ECG RR interval sequence, and generate a reconstructed ECG signal aligned with the time dimension of the EEG signal; The time-frequency feature extraction module is configured to: process the aligned signal using wavelet transform and extract the average wavelet power time series within different predefined frequency bands; The module for constructing a bidirectional heart-brain interaction network is configured to: use a multi-channel time-delay maximum information coefficient algorithm to calculate the bidirectional connection strength value between ECG and EEG power sequences in a specific frequency band, and construct a multi-band bidirectional heart-brain interaction network based on this value. The stress state assessment module is configured to: extract the network features of the network, input them into a pre-trained classification model, and generate and output the acute stress state assessment results.

Images