Electroencephalogram high-order connection analysis method based on time-spectrum hypernetwork

Abstract

The invention relates to an electroencephalogram high-order connection analysis method based on a time-spectrum super network, which is technically characterized by comprising the following steps of: constructing an electroencephalogram stimulation experiment group, collecting multichannel electroencephalogram data in a stimulation process, preprocessing multichannel electroencephalogram signals, constructing a brain time-spectrum super network, performing feature extraction and quantization on the brain time-spectrum super network, and analyzing the electroencephalogram high-order connection of the brain. Time domain connection strength features and spectral domain graph topological features are obtained and analyzed, a classification model is constructed to distinguish subjects with effective and ineffective spinal cord stimulation, and correlation between the features and clinical behavior scores is revealed through statistical analysis. According to the method, a multi-channel electroencephalogram signal acquisition and graph signal processing method is combined, and the brain network response process in a real application scene can be simulated, so that the objectivity and the interpretability of neural regulation effect evaluation are improved, the quantitative characterization function of brain time-spectrum collaborative dynamics under the neural regulation condition is realized, and the neural regulation effect evaluation accuracy is improved. And a new technical means is provided for quantitative analysis of complex brain function state changes.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology and relates to dynamic EEG analysis, particularly a method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks. Background Technology

[0002] With the continuous development of neuroimaging technology, multichannel electroencephalography (EEG) and related high-density recording methods have been widely used for the dynamic observation of brain functional networks. These techniques can capture neural electrical activity with high temporal resolution, providing an important approach to understanding the network mechanisms of the brain under complex states. In recent years, research has gradually recognized that the realization of higher brain functions depends not only on the independent activation of specific brain regions, but also on the dynamic and coordinated integration of higher-order information among multiple brain regions.

[0003] In brain network analysis, functional connectivity is a commonly used approach, typically constructed by calculating bivariate indicators such as linear correlation and phase synchronization between signals from different brain regions. This method can reveal pairwise connection patterns between brain regions and has been applied to the evaluation of numerous neuromodulation effects. However, traditional functional connectivity has inherent limitations in characterizing complex brain states such as disorders of consciousness. The brain is a typical complex system; one brain region often interacts simultaneously with multiple other brain regions, forming higher-order interaction patterns that transcend pairwise relationships. This higher-order interaction is crucial for achieving hierarchical integration of consciousness, but traditional analytical frameworks based on pairwise connectivity struggle to fully capture the dynamic nature of this multi-regional collaboration.

[0004] In recent years, hypernetwork analysis, as an emerging branch of graph theory, has provided a powerful mathematical tool for characterizing high-order relationships between multiple nodes. Unlike ordinary networks that only describe connections between nodes, hypernetworks connect multiple nodes simultaneously through "hyperedges," enabling a more natural modeling of the collaborative activities of brain region clusters. Combining hypernetwork theory with neural time series analysis to construct brain time-spectral hypernetworks holds promise for simultaneously capturing the high-order dynamic features of brain interaction patterns in both the temporal and spectral domains, thereby providing a more comprehensive characterization of the integration mechanisms of neural information processing.

[0005] However, existing research still has significant limitations in assessing the effects of neuromodulation using features of higher-order brain networks, specifically in the following aspects: 1. Limitations of network modeling Most existing EEG analysis studies are still limited to channel-level time-frequency features or functional networks based on pairwise connections, failing to effectively incorporate higher-order information about the simultaneous interaction of multiple brain regions, thus limiting their ability to describe complex brain network states.

[0006] 2. Lack of dynamic and spectral domain fusion analysis Current methods often examine temporal connectivity separately from network topology attributes, lacking a unified framework to extract joint features that can reflect both transient changes in connectivity strength (temporal domain) and characterize the overall integration and robustness of the network (spectral domain).

[0007] 3. Lack of quantization and verification frameworks for high-order features. Although higher-order interactions are theoretically significant, the key challenge in the field remains how to construct stable supernetworks from neural electrical signals, extract discriminative time-spectral features, and establish a quantifiable and verifiable analysis process to objectively distinguish different neural regulatory response states. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for analyzing higher-order connectivity of brainwaves based on time-spectrum supernetworks. By constructing and quantifying the time-spectrum supernetwork characteristics of the brain under specific states, this method achieves accurate characterization and objective evaluation of higher-order network responses induced by neuromodulation. It provides new and more sensitive biomarkers and analytical tools for understanding regulatory mechanisms, predicting individual therapeutic effects, and optimizing parameters, thereby solving the limitations of existing technologies that rely too heavily on traditional second-order functional connectivity analysis and ignore higher-order dynamic interactions in multiple brain regions when evaluating the effects of neuromodulation.

[0009] The present invention solves the existing technical problems by adopting the following technical solution: A method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks includes the following steps: Step 1: Construct the EEG stimulation experimental group and collect multi-channel EEG data during the stimulation process; Step 2: Preprocess the multi-channel EEG signals acquired in Step 1; Step 3: Construct a brain time-spectrum supernetwork based on the preprocessed multi-channel EEG signals; Step 4: Extract and quantize features from the brain temporal-spectral hypernetwork constructed in Step 3 to obtain temporal connectivity strength features and spectral topology features; Step 5: Analyze the temporal connectivity strength features and spectral topology features extracted and quantified in Step 4, construct a classification model to distinguish between subjects whose spinal cord stimulation is effective and ineffective, and reveal the correlation between features and clinical behavior scores through statistical analysis.

[0010] Furthermore, the specific implementation method of step 1 is as follows: select subjects with impaired consciousness who meet the criteria as the experimental group, and simultaneously collect multi-channel EEG signals during the stimulation test phase after spinal cord stimulation electrode implantation; the signal acquisition covers three phases: the baseline period before stimulation, the stimulation period, and the resting period after stimulation.

[0011] Furthermore, the spinal cord stimulation test in step 1 includes stimulation sequences with two frequency parameters: 5Hz and 70Hz. The acquisition process at each frequency includes: 15 minutes of baseline EEG recording, 15 minutes of EEG recording under continuous stimulation, and 20 minutes of resting-state EEG recording after stimulation. Finally, one of the frequencies is selected as the formal stimulation parameter based on the degree of improvement in background EEG activity.

[0012] Furthermore, step 2 includes the following steps: Step 2.1: Downsample the acquired multi-channel EEG data to 250Hz, and apply a 1-40Hz bandpass filter and a 50Hz notch filter to remove high-frequency noise and power frequency interference. Step 2.2: Rereference the signal using an average reference; Step 2.3: Based on the spherical spline interpolation method, the missing channel signals are reconstructed using the existing channel location information to form a unified lead distribution; Step 2.4: Apply independent component analysis to identify and remove artifacts related to eye movement and electrocardiogram.

[0013] Furthermore, step 2.4 is followed by step 2.5: performing a manual quality check on the preprocessed signal to ensure that the signal quality meets the requirements of subsequent analysis.

[0014] Furthermore, step 3 includes the following steps: Step 3.1: Based on the preprocessed EEG signals during the stimulation period, a high-order connectivity model between multiple brain regions is constructed using the correlation-weighted sparse representation method to generate a set of hyperedges; Step 3.2: Through the hyperedge weight learning optimization process, combined with L1 and L2 regularization constraints, a data-driven hyperedge weight allocation is obtained. L1 is used to introduce sparsity constraints, prompting the model to select the most representative few connections from many possible higher-order connections, thereby enhancing the interpretability of the hypernetwork and the ability to extract key features. L2 is used to introduce smoothness constraints, stabilize the weight learning process, improve the model's generalization ability, and make the constructed hypernetwork have better stability and consistency in different subjects or different time periods. Step 3.3: Based on the hyperedge set and its weights, construct the hypernetwork association matrix H and weight matrix W, and then calculate the symmetric hypernetwork similarity matrix S to fully represent the brain's high-order dynamic interaction patterns.

[0015] Furthermore, step 4 includes the following steps: Step 4.1: Temporal connectivity strength feature extraction. Extract the upper triangular elements of the hypernetwork similarity matrix S and concatenate them to form a temporal feature vector Ft. This vector is used to directly reflect the high-order connectivity strength information between different brain region combinations. Step 4.2: Spectral domain graph topological feature extraction. The hypernetwork similarity matrix S is considered as the adjacency matrix of graph G. Its graph Laplacian matrix L is calculated, and eigenvalue decomposition is performed on L to obtain its eigenvalue sequence as the spectral domain feature vector Fs. The second smallest eigenvalue is separately identified as the key topological feature Fs. 2 , used to characterize the overall connectivity robustness of the network; Step 4.3: Feature selection. A prior selection algorithm is used to reduce the dimensionality and filter the extracted time-domain feature vector Ft and spectral-domain feature vector Fs. For Ft, a recursive feature elimination method based on linear support vector machines is used to select the top 20 features. For Fs, except for the key feature Fs... 2 In addition, the recursive feature elimination method was used to select the top 6 features from the remaining features; finally, these were merged to form a feature subset containing 27 of the most discriminative features, which were used for subsequent model construction and efficacy evaluation.

[0016] Furthermore, step 5 includes the following steps: Step 5.1: Using the selected feature subset, train a linear discriminant analysis classifier to distinguish between the effective and ineffective spinal cord stimulation groups; Step 5.2: Use five-fold cross-validation to evaluate the classifier performance. The performance metrics include accuracy, sensitivity, specificity, and F1 score. Step 5.3: Use the nonparametric Wilcoxon rank-sum test to analyze the statistical differences in time-spectral hypernetwork characteristics between the effective group and the invalid group, and calculate the effect size; Step 5.4: Spearman rank correlation analysis was used to explore the correlation between the extracted features and the changes in the scores of the revised Coma Recovery Scale. All statistical tests were corrected for error discovery rate.

[0017] The advantages and positive effects of this invention are: 1. This invention breaks through the dimensional limitations of traditional brain network analysis, achieving dynamic quantification of high-order brain interactions. Traditional EEG-based functional connectivity analysis often focuses on linear or phase relationships between brain regions, failing to characterize collaborative or competitive interactions occurring simultaneously across multiple brain regions. This invention innovatively introduces hypernetwork theory, constructing a brain time-spectral hypernetwork through "correlation-weighted sparse representation" and "data-driven hyperedge learning," elevating the analytical dimension from "edge" to "hyperedge," thereby enabling natural and effective modeling of collaborative activity patterns across clusters of multiple brain regions. This not only theoretically aligns more closely with the nature of the brain as a complex system but also, in practice, captures the dynamic integration of high-order neural information that may be crucial for transitions in consciousness, often overlooked by traditional methods.

[0018] 2. This invention integrates both temporal and spectral domain perspectives, providing a more comprehensive and robust feature description of brain network states. Beyond network construction, this invention further extracts two complementary features from the constructed supernetwork: temporal connectivity strength features directly reflect the instantaneous changes in the strength of interactions between different brain region combinations; spectral topological features characterize the overall network's connectivity robustness and information integration efficiency. This joint feature extraction strategy, combining "microscopic connectivity strength" and "macroscopic topological properties," enables the analysis method to sensitively capture stimulus-induced transient network reorganization and assess overall network stability changes, thus providing a multi-dimensional and more informative quantitative profile of neuromodulation effects.

[0019] 3. This invention establishes a closed-loop, quantifiable, and verifiable analysis process from data to decision-making, significantly improving the objectivity and accuracy of the assessment. This invention not only proposes new features but also provides a complete analysis process, including rigorous signal preprocessing and channel standardization, construction of a stable supernetwork based on regularization optimization, recursive feature selection combining prior knowledge and data-driven approaches, and classification prediction using a linear discriminant analysis model. This process ensures the reliability and reproducibility of the analysis results. Through five-fold cross-validation, this method demonstrates excellent classification performance in distinguishing between subjects with and without spinal cord stimulation, significantly outperforming methods based solely on time-frequency features or traditional functional connectivity networks. This proves that the proposed supernetwork features possess powerful clinical discriminative efficacy.

[0020] 4. This invention reveals a direct correlation between higher-order brain network features and clinical behavioral improvement, providing biomarkers with clear physiological significance for predicting treatment efficacy. The significant value of this invention lies not only in achieving high-precision classification but also in rigorous statistical analysis confirming a significant correlation between the extracted time-spectral hypernetwork features and changes in the revised Coma Recovery Scale (CRS) score—the clinical gold standard. This means that these calculated features are not "black box" indicators but are intrinsically linked to behavioral improvements in the subject's level of consciousness. This provides clinicians with a potential, objective, and quantifiable "reading," which can assist in early assessment of the subject's possible response trends to specific stimulus parameters, thus laying the foundation for individualized and precise neuromodulation programs. Attached Figure Description

[0021] Figure 1 This is a flowchart of the electroencephalogram (EEG) higher-order connectivity analysis method of the time-spectral supernetwork of the present invention. Detailed Implementation

[0022] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0023] A method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks, such as Figure 1As shown, it includes the following steps: Step 1: Construct an EEG stimulation experimental group and collect multi-channel EEG data during the stimulation process.

[0024] In this embodiment, the specific implementation method of this step is as follows: Subjects clinically diagnosed with impaired consciousness and meeting the inclusion / exclusion criteria were selected as the experimental group. After spinal cord stimulation electrode implantation surgery, subjects entered the stimulation parameter testing phase. During this phase, a Nicolet amplifier equipped with a 32-lead electrode cap was used to simultaneously acquire multi-channel EEG signals. Signal acquisition strictly followed a three-stage paradigm of "baseline-stimulation-resting," covering: a 15-minute baseline period before stimulation to record uninterrupted background EEG activity; a 15-minute stimulation period during which continuous spinal cord stimulation was applied and EEG was recorded simultaneously; and a 20-minute resting period after stimulation to observe the aftereffects of stimulation. To optimize the testing frequency, two frequency parameters, 5 Hz and 70 Hz, were included, with each frequency independently completing the aforementioned three-stage acquisition process.

[0025] Finally, the improvement in background EEG activity under the two stimulation frequencies was assessed offline, and the frequency that induced more significant improvement was selected as the formal stimulation parameter for the subject.

[0026] Step 2: Preprocess the data collected in Step 1.

[0027] In this embodiment, the specific implementation method of this step includes the following steps: Step 2.1: Downsample the original EEG signal to 250Hz, and apply a 1-40Hz bandpass filter and a 50Hz notch filter to remove high-frequency noise and power frequency interference. Step 2.2: Rereference the signal using an average reference; Step 2.3: To address the channel loss issue caused by using different electrode caps, the missing channel signals are reconstructed using the existing channel location information based on the spherical spline interpolation method, forming a unified lead distribution. Step 2.4: Apply independent component analysis to identify and remove artifacts related to eye movement and electrocardiogram. Step 2.5: Perform a manual quality check on the preprocessed signal to ensure that the signal quality meets the requirements of subsequent analysis.

[0028] Step 3: Construct a brain time-spectrum supernetwork based on the preprocessed multi-channel EEG signals.

[0029] In this embodiment, the specific implementation method of this step includes the following steps: Step 3.1: Based on the preprocessed EEG signals, a high-order connectivity model between multiple brain regions is constructed using a correlation-weighted sparse representation method. Let the preprocessed i-th EEG channel signal be... The total number of channels is N. For each channel, using a correlation-weighted sparse representation method, it is represented as a linear combination of the signals from the other channels. The optimization model is as follows: in, This represents the influence coefficient of the j-th channel on the i-th channel. For sparse regularization parameters, The correlation weighting factor is defined as: in, This represents the Pearson correlation coefficient between channel i and channel j. This is the mean of the standard deviation of the correlation coefficient.

[0030] Perform the above sparse representation process on each channel, and select several channels with the largest absolute value among the influence coefficients to form a superedge, thereby generating a supernetwork structure containing N superedges.

[0031] Step 3.2: Obtain data-driven hyperedge weight allocation through the hyperedge weight learning optimization process. Based on the assumption that the signal has smoothness in the hypernetwork structure, the hyperedge weights are optimized by minimizing the total variation of the signal on the hypernetwork's Laplacian operator, combined with L1 and L2 regularization constraints (where L1 is used to introduce sparsity constraints, prompting the model to select the most representative few connections from many possible higher-order connections, thereby enhancing the interpretability and key feature extraction ability of the hypernetwork; L2 is used to introduce smoothness constraints, stabilize the weight learning process, improve the model's generalization ability, and make the constructed hypernetwork have better stability and consistency in different subjects or different time periods). The objective function is defined as: Where w is the hyperedge weight vector, For regularization parameters, For the supernetwork Laplacian matrix.

[0032] Step 3.3: Based on the hyperedge set and its weights, construct the hypernetwork association matrix H and weight matrix W, and further calculate the symmetric hypernetwork similarity matrix. This is used to fully represent the high-order dynamic interaction patterns of the brain. The hypernetwork correlation matrix is ​​defined as follows: Supernetwork Laplacian Matrix Defined as: Wherein, vertex degree matrix The diagonal elements are defined as: Hypernetwork similarity matrix Further expressed as: in, Let be a hyperedge degree matrix, and its diagonal elements are defined as: Step 4: Extract and quantify features from the brain temporal-spectral hypernetwork constructed in Step 3.

[0033] In this embodiment, the specific implementation method of this step includes the following steps: Step 4.1: Temporal connectivity strength feature extraction. Extract the upper triangular elements of the hypernetwork similarity matrix S and concatenate them to form a temporal feature vector Ft. This vector directly reflects the high-order connectivity strength information between different brain region combinations. Step 4.2: Extracting topological features from the spectral domain graph. Treating the hypernetwork similarity matrix S as the adjacency matrix of graph G, calculate its graph Laplacian matrix L: Perform eigenvalue decomposition on the graph Laplacian matrix: in, The eigenvalues ​​form a diagonal matrix, and their sequence of eigenvalues ​​constitutes the spectral domain eigenvectors. The second smallest eigenvalue was identified separately as a key topological feature. Algebraic connectivity is used to characterize the robustness of the overall connectivity of a hypernetwork.

[0034] Step 4.3: Feature selection. A prior selection algorithm is used to select the extracted temporal feature vectors. With spectral domain eigenvectors Perform dimensionality reduction and screening; for The top 20 features were selected using a recursive feature elimination method based on linear support vector machines; for While retaining key features Based on this, the recursive feature elimination method was also used to select the top 6 features from the remaining features; finally, they were merged to form a feature subset containing 27 of the most discriminative features, which were used for subsequent model construction and efficacy evaluation.

[0035] Step 5: Analyze the temporal connectivity strength features and spectral topology features extracted and quantified in Step 4, construct a classification model to distinguish between subjects whose spinal cord stimulation is effective and ineffective, and reveal the correlation between features and clinical behavior scores through statistical analysis.

[0036] In this embodiment, the specific implementation method of this step includes the following steps: Step 5.1: Use the selected feature subset to train a linear discriminant analysis classifier to distinguish between the effective and ineffective spinal cord stimulation groups; Step 5.2: Five-fold cross-validation is used to evaluate the classifier performance. Performance metrics include accuracy, sensitivity, specificity, and F1 score. Step 5.3: Perform a nonparametric Wilcoxon rank-sum test to analyze the statistical differences between the effective and ineffective groups in terms of time-spectral hypernetwork characteristics, and calculate the effect size; Step 5.4: Spearman rank correlation analysis was performed to explore the correlation between the proposed features and the changes in the scores of the revised Coma Recovery Scale. All statistical tests were corrected for error discovery rate.

[0037] Through the above steps, the brainwave higher-order connectivity analysis function based on time-spectral supernetwork proposed in this invention has been realized.

[0038] It should be emphasized that the embodiments described in this invention are illustrative rather than limiting. Therefore, this invention includes, but is not limited to, the embodiments described in the specific implementation. Any other implementations derived by those skilled in the art based on the technical solutions of this invention are also within the scope of protection of this invention.

Claims

1. A method for analyzing higher-order connectivity in electroencephalography (EEG) based on time-spectral hypernetworks, characterized in that: Includes the following steps: Step 1: Construct an EEG stimulation experimental group and collect multi-channel EEG data during the stimulation process; Step 2: Preprocess the multi-channel EEG signals acquired in Step 1; Step 3: Construct a brain time-spectrum supernetwork based on the preprocessed multi-channel EEG signals; Step 4: Extract and quantize features from the brain temporal-spectral hypernetwork constructed in Step 3 to obtain temporal connectivity strength features and spectral topology features; Step 5: Analyze the temporal connectivity strength features and spectral topology features extracted and quantified in Step 4, construct a classification model to distinguish between subjects whose spinal cord stimulation is effective and ineffective, and reveal the correlation between features and clinical behavior scores through statistical analysis.

2. The method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks according to claim 1, characterized in that: The specific implementation method of step 1 is as follows: Select subjects with impaired consciousness who meet the criteria as the experimental group. After the spinal cord stimulation electrode is implanted, multi-channel EEG signals are collected simultaneously during the stimulation test phase. The signal acquisition covers three phases: the baseline period before stimulation, the stimulation period, and the resting period after stimulation.

3. The method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks according to claim 2, characterized in that: The spinal cord stimulation test in step 1 includes stimulation sequences with two frequency parameters: 5Hz and 70Hz. The acquisition process at each frequency includes: 15 minutes of baseline EEG recording, 15 minutes of EEG recording under continuous stimulation, and 20 minutes of resting-state EEG recording after stimulation. Finally, one of the frequencies is selected as the formal stimulation parameter based on the degree of improvement in background EEG activity.

4. The method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks according to claim 1, characterized in that: Step 2 includes the following steps: Step 2.1: Downsample the acquired multi-channel EEG data to 250Hz, and apply a 1-40Hz bandpass filter and a 50Hz notch filter to remove high-frequency noise and power frequency interference. Step 2.2: Rereference the signal using an average reference; Step 2.3: Based on the spherical spline interpolation method, the missing channel signals are reconstructed using the existing channel location information to form a unified lead distribution; Step 2.4: Apply independent component analysis to identify and remove artifacts related to eye movement and electrocardiogram.

5. The method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks according to claim 4, characterized in that: Step 2.4 is followed by step 2.5: performing a manual quality check on the preprocessed signal to ensure that the signal quality meets the requirements of subsequent analysis.

6. The method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks according to claim 1, characterized in that: Step 3 includes the following steps: Step 3.1: Based on the preprocessed EEG signals during the stimulation period, a high-order connectivity model between multiple brain regions is constructed using the correlation-weighted sparse representation method to generate a set of hyperedges; Step 3.2: Through the hyperedge weight learning optimization process, combined with L1 and L2 regularization constraints, a data-driven hyperedge weight allocation is obtained. L1 is used to introduce sparsity constraints, prompting the model to select the most representative few connections from many possible higher-order connections, thereby enhancing the interpretability of the hypernetwork and the ability to extract key features. L2 is used to introduce smoothness constraints, stabilize the weight learning process, improve the model's generalization ability, and make the constructed hypernetwork have better stability and consistency in different subjects or different time periods. Step 3.3: Based on the hyperedge set and its weights, construct the hypernetwork association matrix H and weight matrix W, and then calculate the symmetric hypernetwork similarity matrix S to fully represent the brain's high-order dynamic interaction patterns.

7. The method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks according to claim 1, characterized in that: Step 4 includes the following steps: Step 4.1: Temporal connectivity strength feature extraction. Extract the upper triangular elements of the hypernetwork similarity matrix S and concatenate them to form a temporal feature vector Ft. This vector is used to directly reflect the high-order connectivity strength information between different brain region combinations. Step 4.2: Spectral domain graph topological feature extraction. The hypernetwork similarity matrix S is considered as the adjacency matrix of graph G. Its graph Laplacian matrix L is calculated, and eigenvalue decomposition is performed on L to obtain its eigenvalue sequence as the spectral domain feature vector Fs. The second smallest eigenvalue is separately identified as the key topological feature Fs. 2 , used to characterize the overall connectivity robustness of the network; Step 4.3: Feature selection. A prior selection algorithm is used to reduce the dimensionality and filter the extracted time-domain feature vector Ft and spectral-domain feature vector Fs. For Ft, a recursive feature elimination method based on linear support vector machines is used to select the top 20 features. For Fs, except for the key feature Fs... 2 In addition, the recursive feature elimination method was used to select the top 6 features from the remaining features; finally, these were merged to form a feature subset containing 27 of the most discriminative features, which were used for subsequent model construction and efficacy evaluation.

8. The method for analyzing higher-order EEG connectivity based on time-spectral hypernetworks according to claim 1, characterized in that: Step 5 includes the following steps: Step 5.1: Using the selected feature subset, train a linear discriminant analysis classifier to distinguish between the effective and ineffective spinal cord stimulation groups; Step 5.2: Use five-fold cross-validation to evaluate the classifier performance. The performance metrics include accuracy, sensitivity, specificity, and F1 score. Step 5.3: Use the nonparametric Wilcoxon rank-sum test to analyze the statistical differences in time-spectral hypernetwork characteristics between the effective group and the invalid group, and calculate the effect size; Step 5.4: Spearman rank correlation analysis was used to explore the correlation between the extracted features and the changes in the scores of the revised Coma Recovery Scale. All statistical tests were corrected for error discovery rate.

Images