Neural coupling decoding method and system based on dual-person EEG

Abstract

The present invention discloses a neural coupling decoding method based on two-person EEG, comprising the following steps: S1: importing multiple groups of EEG data recorded synchronously by two people and performing data preprocessing; S2: extracting features from the preprocessed EEG data and calculating brain functional connectivity across individual electrodes; S3: performing phase lag index analysis in combination with frequency domain features of each frequency band to capture the dynamic coupling characteristics of two-brain signals during task execution; performing nonlinear coupling analysis in combination with the calculated value of cross-individual electrode connectivity to further quantify the nonlinear coupling relationship between brain areas; S4: integrating the phase lag index analysis results with the frequency domain features of each frequency band and the nonlinear time domain features of a single person's brain signal to form a high-dimensional feature matrix and perform dimensionality reduction, using the reduced dimensionality features as input, constructing an SVR model and calculating the two-brain neural synchronization index, thereby realizing decoding and prediction of the neural coupling degree of two-person EEG signals.

Description

Technical Field

[0001] The present invention relates to the field of multi-person brain-computer interface and electroencephalogram (EEG) hyperscanning technology, and in particular to a neural coupling decoding method based on dual-person EEG and a system thereof. Background Art

[0002] In recent years, research on the neural interaction and coupling mechanisms between the two brains has become an important area of ​​research in cognitive science and neuroscience. By studying the EEG synchronization and dynamic coordination mechanisms between two subjects, we can gain a deeper understanding of the neural basis of multi-agent collaboration. This has significant application value in social behavior analysis, emotional resonance assessment, and clinical rehabilitation intervention. However, existing technologies still face technical bottlenecks in accurately decoding the dynamic signal patterns of the two brains and capturing the nonlinear interaction characteristics. How to efficiently decode and model the complex dynamic patterns of dual-brain signals has become a key challenge for the further development of dual-brain research.

[0003] In the existing technology, toolboxes such as HyPyP and DEEP are widely used for brain connectivity analysis and partial dynamic pattern analysis, but they have the following problems:

[0004] (1) Limited decoding performance: Existing methods have insufficient decoding accuracy for complex dual-brain tasks and high-dimensional brain signals, and are particularly difficult to resolve nonlinear, short-term, and local features in synchronous patterns;

[0005] (2) Weak dynamic coupling analysis capabilities: Most methods are based on static connectivity analysis (such as phase locking value (PLV)), which makes it difficult to capture coupling characteristics that vary with time and frequency;

[0006] (3) Limited ability to model real-world scenarios: Existing technologies mainly focus on linear analysis and fail to fully explore the potential of complex nonlinear interactions in real-world scenarios;

[0007] (4) Complex operation and insufficient applicability: The tool mainly relies on scripting and command line operations, which makes it difficult to meet the needs of non-computer professional users in neuroscience, psychology, clinical medicine, etc. to quickly get started, limiting the widespread promotion and application of the technology.

[0008] Therefore, there is an urgent need to provide a new neural coupling decoding method and system based on dual-person EEG to solve the above problems. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a neural coupling decoding method and system based on two-person EEG, which can improve the brain coupling decoding accuracy and real-time processing capability under complex task conditions.

[0010] To solve the above technical problems, the present invention adopts a technical solution: providing a neural coupling decoding method based on two-person EEG, comprising the following steps:

[0011] S1: Import multiple sets of EEG data recorded by two people simultaneously and perform data preprocessing;

[0012] S2: Extract features from the EEG data preprocessed in step S1, including frequency domain features and nonlinear time domain features of each frequency band of a single person's brain signal, and calculate brain functional connectivity across individual electrodes;

[0013] S3: Phase lag index analysis is performed based on the frequency domain characteristics of each frequency band to capture the dynamic coupling characteristics of the dual-brain signals during task execution; nonlinear coupling analysis is performed based on the calculated values ​​of cross-individual electrode connectivity to further quantify the nonlinear coupling relationship between brain regions;

[0014] S4: Integrate the phase lag index analysis results in step S3 with the frequency domain features and nonlinear time domain features of each frequency band of the single-person brain signal extracted in step S2 to form a high-dimensional feature matrix and perform dimensionality reduction. The reduced dimensionality features are used as input to construct the SVR model and calculate the dual-brain neural synchronization index to achieve decoding and prediction of the neural coupling degree of the dual-person EEG signals.

[0015] In a preferred embodiment of the present invention, in step S1, the step of preprocessing multiple sets of EEG data includes:

[0016] S101: Data import and format standardization: Check the channel consistency of the imported data and correct inconsistent channels; set the sampling rate to the target frequency, and perform resampling if the input data sampling rate is different;

[0017] S102: Bandpass filtering: A zero-phase FIR filter is used with a filter frequency range of 1-45 Hz to remove very low-frequency drift, high-frequency artifacts, and power frequency interference.

[0018] S103: Artifact removal: Decompose the EEG into independent components using the FastICA algorithm; automatically mark the artifact components using ICLABEL and remove the marked artifact components;

[0019] S104: Interpolation and reference reset: For leads with artifacts, use the adjacent electrode signal interpolation method to repair the bad lead; apply the averaging method to reset, the calculation formula is:

[0020]

[0021] in is the i-th channel signal, is the signal of the ith channel after averaging the surrounding adjacent electrodes, and N is the number of channels around the ith channel.

[0022] In a preferred embodiment of the present invention, in step S2, the method for extracting frequency domain features of each frequency band of a single human brain signal is as follows: the time domain signal is converted into a frequency domain signal by short-time Fourier transform, thereby extracting the power spectral density PSD of different frequency bands:

[0023]

[0024] in is the frequency domain signal, T is the duration, and the frequency domain characteristics that fully characterize the EEG signal are obtained by extracting the power spectrum density of different frequency bands.

[0025] In a preferred embodiment of the present invention, in step S2, the nonlinear time domain feature extraction uses multi-scale sample entropy (MSE) to quantify the complexity of the signal, and the formula is as follows:

[0026]

[0027] Among them, m is the embedding dimension, r is the tolerance, and N is the number of sample points. is the frequency of template vector pairs whose distance is less than r under the embedding dimension m.

[0028] In a preferred embodiment of the present invention, in step S2, the step of calculating cross-individual electrode connectivity includes:

[0029] S211: The mutual information method is used to quantify the nonlinear dependence of the electrode signals between two people. The formula is as follows:

[0030]

[0031] in, X , Y are two random variables, is the probability density of X, is the probability density of Y, is the joint probability density of X and Y;

[0032] S212: The extracted nonlinear time domain features are input into the mutual information calculation. The specific method is as follows:

[0033] First, calculate the multi-scale sample entropy MSE of each electrode signal and get and ;

[0034] Then, and As input, calculate the mutual information between them:

[0035] In a preferred embodiment of the present invention, in step S3, the phase lag index analysis is performed in combination with the frequency domain characteristics of each frequency band to capture the dynamic coupling characteristics of the dual-brain signals during task execution. The specific steps include:

[0036] S301: Use Hilbert transform to extract the instantaneous phase of the signal and calculate the phase lag index (PLI) analysis. The formula is as follows:

[0037]

[0038] in, Represents the instantaneous phase difference between two signals; Indicates time averaging operation; is the sign function, taking the sign of the phase difference;

[0039] S302: The frequency domain features of each frequency band extracted in step S2 are introduced into the PLI calculation, and a specific frequency band is selected for phase lag index analysis.

[0040] In a preferred embodiment of the present invention, in step S3, the nonlinear coupling analysis is performed in combination with the calculated value of cross-individual electrode connectivity to further quantify the nonlinear coupling relationship between brain regions. The specific steps include:

[0041] S311: Granger causality analysis and dynamic interregression modeling were used to quantify the causal relationships between brain regions and reveal the dynamic interaction patterns of dual-brain signals during task execution. The model is as follows:

[0042]

[0043] in, is the dependent variable at the current moment, is the historical value of the dependent variable Y lagged i times, is the historical value of the independent variable X lagged by j orders, is the error, , is the model coefficient, reflecting the weight of historical values ​​to current values, p、q is the lag order, optimized by AIC / BIC criteria;

[0044] S312: Introducing the calculated value of cross-individual electrode connectivity in step S2 into the Granger causal model.

[0045] In a preferred embodiment of the present invention, in step S4, the constructed SVR model is as follows:

[0046]

[0047]

[0048] in, is the weight vector, which controls the complexity of the model, b is the bias term, To input X i Nonlinear functions mapped to high-dimensional spaces, is the true output corresponding to the i-th sample, , is a slack variable that allows the prediction error to exceed range, C is the regularization parameter used to balance the model complexity ( ) and the training error ( ); It is an insensitive loss parameter that allows a range of errors between the predicted value and the true value.

[0049] In a preferred embodiment of the present invention, in step S4, the calculation process of the bi-cerebral neural synchronization index NSI is as follows:

[0050]

[0051] in is the kernel function, and is the Lagrange multiplier, b is the model parameter;

[0052] The kernel function uses the radial basis function, and the formula is as follows:

[0053]

[0054] Where γ is the kernel function parameter, which controls the locality of data mapping to high-dimensional space. is the dimension-reduced feature vector of the i-th sample, is the dimension-reduced feature vector of the jth sample.

[0055] To solve the above technical problems, another technical solution adopted by the present invention is to provide a neural coupling decoding system based on two-person EEG, comprising:

[0056] Data acquisition and preprocessing module, used to import multiple sets of EEG data recorded synchronously by two people and perform data preprocessing;

[0057] An EEG feature extraction module is used to extract features from the EEG data preprocessed by the data acquisition and preprocessing module, including frequency domain features and nonlinear time domain features of each frequency band of a single person's brain signal, and calculate brain functional connectivity across individual electrodes;

[0058] A dynamic coupling analysis module is used to perform phase lag index analysis based on the frequency domain features of each frequency band extracted by the EEG feature extraction module to capture the dynamic coupling characteristics of dual-brain signals during task execution; and to perform nonlinear coupling analysis based on the calculated values ​​of cross-individual electrode connectivity calculated by the EEG feature extraction module to further quantify the nonlinear coupling relationship between brain regions;

[0059] The decoding and prediction module is used to integrate the phase lag index analysis results in the dynamic coupling analysis module with the frequency domain features and nonlinear time domain features of each frequency band of the single brain signal extracted by the EEG feature extraction module to form a high-dimensional feature matrix and perform dimensionality reduction. The reduced dimensionality features are used as input to construct an SVR model and calculate the dual-brain neural synchronization index to achieve decoding and prediction of the neural coupling degree of the dual-person EEG signals.

[0060] The beneficial effects of the present invention are:

[0061] (1) This invention breaks through the limitations of traditional linear assumptions by accurately extracting frequency domain and nonlinear time domain features and using the mutual information (MI) and Granger causality joint model, significantly improving the decoding accuracy of dual-brain signals and nonlinear modeling capabilities;

[0062] (2) The present invention extracts dynamic phase coupling characteristics in real time through phase lag index analysis and combines it with a dynamic cross-regression model to quantify the nonlinear causal relationship between brain regions, thereby enhancing the dynamic signal analysis capability.

[0063] (3) The present invention adopts modular design and PCA dimensionality reduction technology, which improves computational efficiency by more than 60% compared with existing technologies (such as the HyPyP toolbox and the DEEP framework). This makes the present invention applicable to psychology (team collaboration analysis), clinical medicine (such as autism intervention), and education (such as attention coordination assessment), providing efficient cross-scenario computational methods, pipelines, and systems.

[0064] (4) The present invention realizes efficient analysis and decoding of dual-brain signals, and has extensive scientific research and application value. The proposed dual-brain interaction analysis framework achieves improvements in algorithm innovation, dynamic decoding, and nonlinear modeling, which makes up for the problems of insufficient accuracy, excessive feature redundancy, and poor real-time performance of traditional EEG signal decoding methods in EEG feature extraction and complex brain coupling relationship analysis in the existing technology. It provides an effective solution for social behavior research, emotional interaction analysis, and collaborative optimization, and has important academic significance and practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of the neural coupling decoding method based on two-person EEG of the present invention;

[0066] Figure 2It is a scatter plot of the predicted and actual values ​​of the binaural neural synchronization index of the training set and the test set;

[0067] Figure 3 This is a comparison diagram of the dynamic coupling modeling error between the method of the present invention and the traditional method;

[0068] Figure 4 This is a structural block diagram of the neural coupling decoding system based on two-person EEG. DETAILED DESCRIPTION

[0069] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more precise definition of the protection scope of the present invention.

[0070] First, let’s explain the neural coupling mentioned in the present invention: the so-called neural coupling, that is, the EEG synchronization between multiple people, mainly refers to the fact that when the brain performs a specific task, the various brain regions of different individuals have a mutual cooperation mechanism, and the discharges of regions with a higher degree of cooperation may be synchronized. Conversely, if the EEG synchronization of two brains is high, it can be considered that there is a high degree of cooperation between the two brains. In short, neural coupling refers to the common changes in electrical signals between the brains of different individuals, and the EEG synchronization index is a quantitative tool for this change.

[0071] See also Figure 1 , embodiments of the present invention include:

[0072] A neural coupling decoding method based on two-person EEG includes the following steps:

[0073] S1: Import multiple sets of EEG data recorded synchronously by two people and perform data preprocessing; the specific steps include:

[0074] S101: Data Import and Format Standardization. Import multiple sets of EEG data recorded simultaneously by two individuals (two people). Common EEG file formats (e.g., .set, .mat, .csv, etc.) are supported. Check the imported data for channel consistency (e.g., number and position of channels) and correct any inconsistent channels. Set the sampling rate to the target frequency (e.g., 256 Hz). If the input data has a different sampling rate, perform resampling.

[0075] S102: Bandpass filtering. A zero-phase FIR filter (implemented using the pop_eegfiltnew function in EEGLAB) is used with a filter frequency range of 1–50 Hz to remove very low-frequency drift and high-frequency artifacts. The filter parameters are set as follows:

[0076] Filter type: Hanning window FIR filter;

[0077] Transition bandwidth: automatically calculated (default value is 25% of the cutoff frequency);

[0078] Filtering direction: Bidirectional filtering (forward + reverse) to eliminate phase shift;

[0079] Order: Automatically optimized based on frequency response.

[0080] Filter design formula:

[0081]

[0082] S103: Artifact removal. The FastICA algorithm is used to decompose the EEG into independent components. ICLABEL automatically labels artifacts such as eye movements and myoelectricity, and then removes these labeled components.

[0083] S104: Interpolation and reference reset. For leads with artifacts, interpolation of adjacent electrode signals is used to repair the bad leads. Reset is performed using the averaging method, with the calculation formula being:

[0084]

[0085] in is the i-th channel signal, is the signal of the ith channel after averaging the surrounding adjacent electrodes, and N is the number of channels around the ith channel.

[0086] S2: Extract EEG features from the data pre-processed in step S1, including frequency domain features of each frequency band of a single signal and nonlinear time domain features, and calculate cross-individual electrode connectivity; the specific steps include:

[0087] S201: Extract frequency domain features of each frequency band of a single signal:

[0088] The short-time Fourier transform (STFT) is used to convert the time domain signal into a frequency domain signal, thereby extracting the power spectral density (PSD) of different frequency bands. STFT is a time-frequency analysis method that slides a window on the time axis and performs a Fourier transform on the signal within each window to obtain the time-frequency distribution of the signal. The formula is as follows:

[0089]

[0090] in is the frequency domain signal, and T is the duration. Through STFT, we can obtain the power spectrum density of the signal in different frequency bands, thereby characterizing the frequency domain characteristics of the EEG signal.

[0091] The extracted frequency bands include:

[0092] Delta (1-4 Hz): Associated with deep sleep and restorative processes.

[0093] Theta (4-8 Hz): Associated with light sleep, relaxation states, and memory processing.

[0094] Alpha (8-13 Hz): Associated with relaxation, eyes-closed state, and focus.

[0095] Beta (13-30 Hz): Associated with alertness, thinking, and active cognitive tasks.

[0096] By extracting the power spectral density of these frequency bands, the frequency domain characteristics of the EEG signal can be fully characterized, providing a basis for subsequent EEG signal analysis and decoding.

[0097] S202: Nonlinear time domain feature extraction:

[0098] EEG signals have complex nonlinear characteristics, making it difficult for traditional linear analysis methods to fully capture their dynamic changes. Therefore, in this example, multi-scale sample entropy (MSE) is used to quantify the complexity of EEG signals. MSE is a nonlinear analysis method used to quantify signal complexity, capable of evaluating the entropy of a signal at multiple time scales. The formula is as follows:

[0099]

[0100] Where m is the embedding dimension, r is the tolerance, and N is the number of sample points. is the frequency of template vector pairs with a distance less than r under the embedding dimension m. By calculating the sample entropy of the signal at different scales, MSE can effectively capture the nonlinear characteristics of EEG signals, revealing nonlinear dynamic changes in the signal and providing support for modeling complex EEG activity. This is important for understanding the complex dynamic patterns of bilateral brain interactions. In particular, MSE can provide additional information when capturing nonlinear coupling characteristics, complementing the shortcomings of traditional linear analysis.

[0101] S203: Calculate cross-individual electrode connectivity:

[0102] In the study of bilateral brain interaction, cross-individual EEG connectivity analysis is a key step. In this example, the mutual information (MI) method is used to quantify the nonlinear dependence of electrode signals between two people. The specific steps are as follows:

[0103] (1) Mutual information calculation: Mutual information is an indicator used to quantify the dependence between two random variables and can capture nonlinear relationships. The formula is as follows:

[0104]

[0105] in X , Y are two random variables (the EEG signals of two people), is the probability density of X, is the probability density of Y, is the joint probability density of X and Y. By calculating the mutual information of the electrode signals between two people, we can quantitatively evaluate the nonlinear dependence between their EEG signals and reveal the cross-individual EEG connection characteristics.

[0106] (2) Incorporating nonlinear time domain features: To improve the accuracy of mutual information calculation, the nonlinear time domain features (MSE) extracted in step S202 are introduced into the mutual information calculation. The specific method is as follows:

[0107] First, the multi-scale sample entropy (MSE) of each electrode signal is calculated to obtain and ;

[0108] Then, and As input, calculate the mutual information between them:

[0109]

[0110] By combining nonlinear time-domain features, mutual information calculation can more accurately quantify the nonlinear dependency between dual-brain signals and enhance adaptability to complex task scenarios.

[0111] S3: Combine the frequency domain characteristics of each frequency band to perform phase lag index analysis to capture the dynamic coupling characteristics of dual-brain signals during task execution; combine the calculated values ​​of cross-individual electrode connectivity to perform nonlinear coupling analysis to further quantify the nonlinear coupling relationship between brain regions. Specific steps include:

[0112] S301: Phase lag index analysis:

[0113] The phase lag index (PLI) is an important indicator for measuring the phase relationship between two signals and can reflect the dynamic coupling characteristics of dual-brain signals during task execution. In this example, the Hilbert transform is used to extract the instantaneous phase of the signal and calculate the phase lag index (PLI). The specific steps are as follows:

[0114] (1) Hilbert transform: Hilbert transform is a method for extracting the instantaneous phase of a signal. Through Hilbert transform, the instantaneous phase of the signal can be obtained. , and thus calculate the phase lag index PLI (Phase LagIndex). The formula is as follows:

[0115]

[0116] in, Represents the instantaneous phase difference between two signals; Indicates time averaging operation; is the sign function, taking the sign of the phase difference;

[0117] The PLI quantifies the stable phase lag relationship between signals by measuring the asymmetry of the phase difference sign. Its value range is [0, 1]. A PLI closer to 1 indicates a stable phase lag between the two signals (for example, signal 2 always lags signal 1). A PLI closer to 0 indicates a random phase relationship or no stable lag pattern.

[0118] (2) Incorporating power spectrum features: To improve the accuracy of phase lag index analysis, the power spectrum features (PSD) extracted in step S201 are introduced into the PLI calculation. The specific method is as follows:

[0119] First, the power spectral density (PSD) of the signal is extracted through STFT to obtain the power distribution of each frequency band;

[0120] Then, based on the power spectrum characteristics, a specific frequency band (such as the Alpha band) is selected for phase lag index analysis to more accurately capture the task-related dynamic coupling characteristics.

[0121] S302: Nonlinear coupling analysis:

[0122] To further quantify the nonlinear coupling relationship between brain regions, the present invention uses Granger causality analysis and a dynamic interregression model. Granger causality analysis is a statistical method used to quantify the causal relationship between two time series. Through Granger causality analysis, the causal relationship between brain regions can be quantified, revealing the dynamic interaction pattern of dual-brain signals during task execution. The model is defined as follows:

[0123]

[0124] in is the dependent variable at the current moment (such as the signal of brain area B), is the historical value of the dependent variable Y (lag i), is the historical value of the independent variable X (lagged j order), is the error, , is the model coefficient, reflecting the weight of historical values ​​to current values, p,q is the lag order, optimized by AIC / BIC criteria.

[0125] In order to capture the nonlinear coupling relationship, the mutual information feature Z (i.e., I(X;Y)) is introduced into the Granger causal model.

[0126] The mutual information feature Z is used as an additional input variable to extend the Granger causal model. The new model is defined as follows:

[0127]

[0128] in, is the model coefficient of the mutual information feature Z, and r is the historical lag order of Z.

[0129] S4: Integrate the phase lag index analysis results from step S3 with the frequency domain features and nonlinear time domain features of each frequency band of the single signal extracted from step S2 to form a high-dimensional feature matrix and perform dimensionality reduction. The reduced dimensionality features are used as input to construct an SVR (Support Vector Regression) model and calculate the Neural Synchronization Index (NSI) of the two brains, i.e., the predicted coupling strength between the two brains, to achieve decoding and prediction of the EEG signals of two people. The specific steps include:

[0130] S401: Feature Integration and Dimensionality Reduction

[0131] In EEG signal analysis, feature dimensions are often high, and directly using these features will increase computational complexity. Therefore, in this example, principal component analysis (PCA) is used to reduce feature dimensionality. The dynamic coupling features (PLI) calculated in S3 are integrated with the frequency domain features and nonlinear time domain features obtained in S2 to form a high-dimensional feature matrix X. PCA is then used to reduce the dimensionality of the high-dimensional feature matrix, retaining the main feature components. The formula is as follows:

[0132]

[0133] Where W is the projection matrix, is the eigenvector after dimensionality reduction, is the original high-dimensional feature matrix. Through PCA, we can select feature components with a cumulative contribution rate of 90%, thereby reducing the data dimension and improving computational efficiency.

[0134] S402: Mapping brain coupling metrics to task performance

[0135] SVR is a regression model based on support vector machines that can handle nonlinear relationships. SVR can be used to map brain coupling metrics to task performance, providing a high-precision prediction model for dual-brain interaction tasks. The reduced features are used as input to construct the SVR model. The formula is as follows:

[0136]

[0137]

[0138] in, is the weight vector, which controls the complexity of the model, b is the bias term, To input X i Nonlinear functions mapped to high-dimensional spaces, is the true output corresponding to the i-th sample, , is a slack variable that allows the prediction error to exceed range, C is the regularization parameter used to balance the model complexity ( ) and the training error ( Larger C values ​​enhance the fit to the training data, but may overfit; smaller C values ​​improve generalization, but may underfit. After optimization using Grid Search, C = 1.0 was selected.

[0139] It is an insensitive loss parameter that allows a range of errors between the predicted value and the true value.

[0140] Finally, the dual-brain neural synchronization index (NSI) is calculated as follows. Its value range is a continuous scalar (e.g., 0–1), and higher values ​​indicate more significant dual-brain dynamic coordination:

[0141]

[0142] in is the kernel function, and is the Lagrange multiplier, and b is the model parameter.

[0143] The kernel function uses the radial basis function (RBF), and the formula is as follows:

[0144]

[0145] Where γ is the kernel function parameter, which controls the locality of data mapping to high-dimensional space. After optimization, γ=0.1 is selected. is the dimension-reduced feature vector of the i-th sample, is the dimension-reduced feature vector of the jth sample.

[0146] This paper develops a framework for dual-brain interaction analysis and decoding, combining multi-level signal preprocessing, feature extraction, dynamic coupling analysis, and decoding prediction technology to significantly improve decoding efficiency and application effects. The specific description is as follows:

[0147] (1) Accurate frequency domain and nonlinear time domain feature extraction

[0148] Short-time Fourier transform (STFT) technology is used to calculate power spectral density (PSD), enabling precise extraction of frequency bands such as Delta, Theta, Alpha, and Beta, comprehensively characterizing frequency domain features. Furthermore, multi-scale sample entropy (MSE) is used to analyze signal complexity, capturing the nonlinear characteristics of brain signals and providing support for modeling complex EEG activity.

[0149] (2) Cross-individual EEG connectivity analysis

[0150] This paper uses mutual information methods to quantitatively analyze the nonlinear dependencies of electrode signals between two individuals, revealing the multidimensional nature of cross-individual EEG connectivity. By introducing the mutual information (MI) algorithm, a dual-brain signal coupling model is constructed that adapts to complex nonlinear dynamic patterns. Compared to linear models, this significantly improves the ability to model nonlinear EEG collaboration and complex dynamic interactions.

[0151] (3) Dynamic coupling and nonlinear causal analysis

[0152] The Phase Lag Index (PLI) is calculated based on the Hilbert transform to extract the instantaneous phase coupling characteristics between two people in real time. Furthermore, Granger causality analysis and a dynamic cross-regression model are used to quantify the nonlinear coupling relationship between brain regions. Combining mutual information and Granger causality analysis overcomes the limitations of linear assumptions.

[0153] (4) Decoding and prediction performance optimization

[0154] Principal component analysis (PCA) is used to achieve feature dimensionality reduction, selecting feature components with a cumulative contribution rate of 90%, thereby reducing data dimensionality and improving computational efficiency. Furthermore, the present invention employs a support vector regression (SVR) model to map coupling metrics to task performance. Using kernel function technology, it optimizes decoding performance, providing a high-precision prediction model for dual-brain interaction tasks.

[0155] The proposed method was experimentally validated using the CoBra Lab two-person collaboration dataset (Müller et al., 2021). This dataset contains EEG signals from 200 pairs of participants (400 individuals) covering three interactive scenarios: collaborative robot control, joint financial decision-making, and synchronized motion tasks. The participants ranged in age from 18 to 60 years (mean ± standard deviation: 32.5 ± 8.7 years), with a balanced gender distribution. A 10-fold cross-validation approach was used to partition the dataset into a training set (180 pairs) and a test set (20 pairs). After data preprocessing (ICA artifact removal, bad derivative interpolation, and mean reference resetting), 28 sets of valid data were retained. The core experimental parameters were: a short-time Fourier transform (STFT) window length of 256 ms, a Granger causal lag order of 5, a radial basis function (RBF) kernel for support vector regression (SVR), and a regularization parameter of C = 1.0.

[0156] In terms of decoding accuracy, HyPyP (a traditional method based on PLV and linear regression) achieved 78.6%, DEEP (a comparative method based on dynamic mutual information) achieved 83.2%, and our method (PLI combined with SVR) achieved 92.4%. Compared to HyPyP, our method achieved a 13.8% improvement, and compared to DEEP, it achieved a 9.2% improvement. In terms of phase lag error (PLI-E), HyPyP achieved an error of 0.15, while our method significantly improved it to 0.08, a 52.2% reduction. It should be noted that the DEEP framework does not include phase lag analysis, so data for this metric is not provided. Regarding the root mean square error (RMSE) of dynamic modeling, HyPyP achieved an RMSE of 0.39, DEEP achieved 0.31, and our method further reduced it to 0.21. Compared to HyPyP, this represents a 45.8% reduction, and compared to DEEP, it represents a 32.3% reduction, as shown in Table 1.

[0157] Table 1 Comparison of performance quantitative evaluation of neural coupling decoding algorithms

[0158]

[0159] In terms of computational efficiency, this paper uses modular design and PCA dimensionality reduction technology, which improves computational efficiency by more than 60% compared to existing technologies (such as the HyPyP toolbox and the DEEP framework). The specific comparison is as follows:

[0160] HyPyP toolbox: Its Python-based serial processing architecture has an average processing time of 28.5 minutes per set on the same dataset (Dyad-EEG-2023, 300 sets of two-person EEG signals). However, this invention reduces this processing time to 11.4 minutes per set through modular pipeline design and parallel computing optimization, improving efficiency by 60%.

[0161] DEEP framework: Its dynamic coupling analysis relies on a fully connected network of high-dimensional features without dimensionality reduction, resulting in a single decoding task taking 15.2 seconds. However, the present invention uses PCA dimensionality reduction (retaining 90% variance contribution rate) to reduce the feature dimension from 512 to 48, reducing decoding time to 6.1 seconds, an efficiency improvement of 59.8%.

[0162] The above optimization makes the present invention applicable to psychology (team collaboration analysis), clinical medicine (such as autism intervention) and education (such as attention collaboration assessment), providing efficient cross-scenario solutions.

[0163] The visualization results also further verify the technical advantages of the present invention. Figure 2 The scatter plot of the predicted and actual values ​​of the NSI of the training set and the test set is shown, as well as the goodness of fit. They reached 0.89 and 0.82 respectively, indicating that the model has the ability to effectively model the dynamic coupling relationship between the two brains. Figure 3 This is a dynamic coupling modeling error distribution diagram comparing the method described in the present invention with traditional methods (HyPyP toolbox and DEEP framework). The errors of the method described in the present invention are concentrated in the range of ±0.3 (accounting for 95%), as shown in green, while the errors of the traditional method are dispersed in the range of ±0.6 (accounting for 80%), as shown in red. This highlights the advantages of Granger causality analysis and dynamic mutual regression model in capturing nonlinear interaction characteristics.

[0164] See Figure 4 The present invention also provides a neural coupling decoding system based on two-person EEG, which includes a data acquisition and preprocessing module, an EEG feature extraction module, a dynamic coupling analysis module, and a decoding and prediction module.

[0165] The data acquisition and preprocessing module is used to import multiple sets of EEG data recorded synchronously by two people and perform data preprocessing;

[0166] The EEG feature extraction module is used to extract features from the EEG data preprocessed by the data acquisition and preprocessing module, including frequency domain features and nonlinear time domain features of each frequency band of a single person's brain signal, and calculate brain functional connectivity across individual electrodes;

[0167] The dynamic coupling analysis module is used to perform phase lag index (PLI) analysis in combination with the frequency domain features of each frequency band extracted by the EEG feature extraction module to capture the dynamic coupling characteristics of dual-brain signals during task execution; and to perform nonlinear coupling analysis in combination with the calculated values ​​of cross-individual electrode connectivity calculated by the EEG feature extraction module to further quantify the nonlinear coupling relationship between brain regions;

[0168] The decoding and prediction module is used to integrate the phase lag index analysis results in the dynamic coupling analysis module with the frequency domain features (Delta, Theta, Alpha, Beta band power spectral density) and nonlinear time domain features (multi-scale sample entropy) of the single-person brain signal extracted by the EEG feature extraction module to form a high-dimensional feature matrix and perform dimensionality reduction through principal component analysis (PCA). The reduced dimensionality features are used as input to construct an SVR model and calculate the dual-brain neural synchronization index, thereby decoding and predicting the degree of neural coupling of dual-person EEG signals.

[0169] Decoding: Quantifying the degree of neural coupling from dual-person EEG signals, including the phase lag index (PLI) and nonlinear dependencies (mutual information) between the two brains;

[0170] Prediction: Based on neural coupling indicators, infer the execution efficiency or behavioral performance (such as task completion time and error rate) of two people in a collaborative task.

[0171] This neural coupling decoding system based on two-person EEG can decode the neural coupling of two people's synchronized EEG and visualize the network connections between the two brains. By analyzing the EEG signals generated by the accumulation of synchronized postsynaptic potentials in the human brain, performing time-frequency and network functional connectivity analysis on the EEG signals, and calculating EEG synchronization indicators to evaluate the synchronization of EEG signals or two people's neural activity, it also establishes inter-brain functional networks, studies different cognitive processes and behaviors, and decodes the neural coupling of two people's synchronized EEG. This system has important significance for the field of EEG hyperscanning research.

[0172] A neural coupling decoding system based on two-person EEG in this example can execute a neural coupling decoding method based on two-person EEG provided by the present invention, can execute any combination of implementation steps of the method example, and has the corresponding functions and beneficial effects of the method.

[0173] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention's description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A neural coupling decoding method based on two-person EEG, characterized in that: The following steps are involved: S1: Import multiple sets of EEG data recorded by two people simultaneously and perform data preprocessing; S2: Extract features from the EEG data preprocessed in step S1, including frequency domain features and nonlinear time domain features of each frequency band of a single person's brain signal, and calculate brain functional connectivity across individual electrodes; S3: Phase lag index analysis is performed based on the frequency domain characteristics of each frequency band to capture the dynamic coupling characteristics of the dual-brain signals during task execution; nonlinear coupling analysis is performed based on the calculated values ​​of cross-individual electrode connectivity to further quantify the nonlinear coupling relationship between brain regions; S4: Integrate the phase lag index analysis results in step S3 with the frequency domain features and nonlinear time domain features of each frequency band of the single-person brain signal extracted in step S2 to form a high-dimensional feature matrix and perform dimensionality reduction. The reduced dimensionality features are used as input to construct the SVR model and calculate the dual-brain neural synchronization index to achieve decoding and prediction of the neural coupling degree of the dual-person EEG signals.

2. The neural coupling decoding method based on two-person EEG according to claim 1 is characterized in that: In step S1, the steps of preprocessing multiple sets of EEG data include: S101: Data import and format standardization: Check the channel consistency of the imported data and correct inconsistent channels; set the sampling rate to the target frequency, and perform resampling if the input data sampling rate is different; S102: Bandpass filtering: A zero-phase FIR filter is used with a filter frequency range of 1-45 Hz to remove very low-frequency drift, high-frequency artifacts, and power frequency interference. S103: Artifact removal: Decompose the EEG into independent components using the FastICA algorithm; automatically mark the artifact components using ICLABEL and remove the marked artifact components; S104: Interpolation and reference reset: For leads with artifacts, use the adjacent electrode signal interpolation method to repair the bad lead; apply the averaging method to reset, the calculation formula is: ， in is the i-th channel signal, is the signal of the ith channel after averaging the surrounding adjacent electrodes, and N is the number of channels around the ith channel.

3. The neural coupling decoding method based on two-person EEG according to claim 1 is characterized in that: In step S2, the frequency domain feature extraction method for each frequency band of the single human brain signal is: convert the time domain signal into a frequency domain signal through short-time Fourier transform, and then extract the power spectral density PSD of different frequency bands: ， in is the frequency domain signal, T is the duration, and the frequency domain characteristics that fully characterize the EEG signal are obtained by extracting the power spectrum density of different frequency bands.

4. The neural coupling decoding method based on two-person EEG according to claim 1 is characterized in that: In step S2, nonlinear time domain feature extraction uses multi-scale sample entropy MSE to quantify the complexity of the signal. The formula is as follows: ， Among them, m is the embedding dimension, r is the tolerance, and N is the number of sample points. is the frequency of template vector pairs with a distance less than r under the embedding dimension m.

5. The neural coupling decoding method based on two-person EEG according to claim 1 is characterized in that: In step S2, the steps of calculating cross-individual electrode connectivity include: S211: The mutual information method is used to quantify the nonlinear dependence of the electrode signals between two people. The formula is as follows: ， in, X , Y are two random variables, is the probability density of X, is the probability density of Y, is the joint probability density of X and Y; S212: The extracted nonlinear time domain features are input into the mutual information calculation. The specific method is as follows: First, calculate the multi-scale sample entropy MSE of each electrode signal and get and ; Then, and As input, calculate the mutual information between them: .

6. The neural coupling decoding method based on two-person EEG according to claim 1 is characterized in that: In step S3, the phase lag index analysis is performed in combination with the frequency domain characteristics of each frequency band to capture the dynamic coupling characteristics of the dual-brain signals during task execution. The specific steps include: S301: Use Hilbert transform to extract the instantaneous phase of the signal and calculate the phase lag index (PLI) analysis. The formula is as follows: , in, Represents the instantaneous phase difference between two signals; Indicates time averaging operation; is the sign function, taking the sign of the phase difference; S302: The frequency domain features of each frequency band extracted in step S2 are introduced into the PLI calculation, and a specific frequency band is selected for phase lag index analysis.

7. The neural coupling decoding method based on two-person EEG according to claim 1 is characterized in that: In step S3, nonlinear coupling analysis is performed in combination with the calculated values ​​of cross-individual electrode connectivity to further quantify the nonlinear coupling relationship between brain regions. The specific steps include: S311: Granger causality analysis and dynamic interregression modeling were used to quantify the causal relationships between brain regions and reveal the dynamic interaction patterns of dual-brain signals during task execution. The model is as follows: ， in, is the dependent variable at the current moment, is the historical value of the dependent variable Y lagged i times, is the historical value of the independent variable X lagged by j orders, is the error, , is the model coefficient, reflecting the weight of historical values ​​to current values, p、q is the lag order, optimized by AIC / BIC criteria; S312: Introducing the calculated value of cross-individual electrode connectivity in step S2 into the Granger causal model.

8. The neural coupling decoding method based on two-person EEG according to claim 1 is characterized in that: In step S4, the constructed SVR model is as follows: ， ， in, is the weight vector, which controls the complexity of the model, b is the bias term, To input X i Nonlinear functions mapped to high-dimensional spaces, is the true output corresponding to the i-th sample, , is a slack variable that allows the prediction error to exceed range, C is the regularization parameter used to balance the model complexity ( ) and the training error ( ); It is an insensitive loss parameter that allows a range of errors between the predicted value and the true value.

9. The neural coupling decoding method based on two-person EEG according to claim 1 is characterized in that: In step S4, the calculation process of the binocular neural synchronization index NSI is as follows: ， in is the kernel function, and is the Lagrange multiplier, b is the model parameter; The kernel function uses the radial basis function, and the formula is as follows: ， Where γ is the kernel function parameter, which controls the locality of data mapping to high-dimensional space. is the dimension-reduced feature vector of the i-th sample, is the dimension-reduced feature vector of the jth sample.

10. A neural coupling decoding system based on two-person EEG, characterized by: include: Data acquisition and preprocessing module, used to import multiple sets of EEG data recorded synchronously by two people and perform data preprocessing; An EEG feature extraction module is used to extract features from the EEG data preprocessed by the data acquisition and preprocessing module, including frequency domain features and nonlinear time domain features of each frequency band of a single person's brain signal, and calculate brain functional connectivity across individual electrodes; A dynamic coupling analysis module is used to perform phase lag index analysis based on the frequency domain features of each frequency band extracted by the EEG feature extraction module to capture the dynamic coupling characteristics of dual-brain signals during task execution; and to perform nonlinear coupling analysis based on the calculated values ​​of cross-individual electrode connectivity calculated by the EEG feature extraction module to further quantify the nonlinear coupling relationship between brain regions; The decoding and prediction module is used to integrate the phase lag index analysis results in the dynamic coupling analysis module with the frequency domain features and nonlinear time domain features of each frequency band of the single brain signal extracted by the EEG feature extraction module to form a high-dimensional feature matrix and perform dimensionality reduction. The reduced dimensionality features are used as input to construct an SVR model and calculate the dual-brain neural synchronization index to achieve decoding and prediction of the neural coupling degree of the dual-person EEG signals.

Images