Search

CN-122004891-A - Electrocerebral high-order connection analysis method based on time-spectrum super network

CN122004891ACN 122004891 ACN122004891 ACN 122004891ACN-122004891-A

Abstract

The invention relates to an electroencephalogram high-order connection analysis method based on a time-spectrum super network, which is technically characterized by constructing an electroencephalogram stimulation experimental group, collecting multichannel electroencephalogram data in the stimulation process, preprocessing multichannel electroencephalogram data, constructing a brain time-spectrum super network, carrying out feature extraction and quantification on the brain time-spectrum super network, acquiring and analyzing time domain connection strength features and spectral domain diagram topological features, constructing a classification model to distinguish subjects with effective and ineffective spinal cord stimulation, and revealing the correlation between the features and clinical behavior scores through statistical analysis. The invention combines the multichannel brain electrical signal acquisition and graph signal processing method, can simulate the brain network response process under the real application scene, thereby improving the objectivity and the interpretability of the evaluation of the nerve regulation effect, realizing the quantitative characterization function of the brain time-spectrum collaborative dynamics under the nerve regulation condition, and providing a new technical means for the quantitative analysis of the state change of the complex brain function.

Inventors

  • LU JIEWEI
  • YU NINGBO
  • HAN JIANDA
  • WANG JIAYI

Assignees

  • 南开大学
  • 南开大学深圳研究院

Dates

Publication Date
20260512
Application Date
20260121

Claims (8)

  1. 1. A time-spectrum super-network-based electroencephalogram high-order connection analysis method is characterized by comprising the following steps of: Step 1, constructing an electroencephalogram stimulation experimental group and collecting multichannel electroencephalogram data in the stimulation process; Step 2, preprocessing the multichannel electroencephalogram signals acquired in the step 1; Step 3, constructing a brain time-spectrum super network based on the preprocessed multichannel brain electrical signals; step 4, extracting and quantifying the characteristics of the brain time-spectrum super network constructed in the step 3 to obtain time domain connection strength characteristics and spectrum domain diagram topological characteristics; And 5, analyzing the time domain connection strength characteristics and the spectral domain diagram topological characteristics extracted and quantified in the step 4, constructing a classification model to distinguish subjects with effective spinal cord stimulation and subjects without effective spinal cord stimulation, and revealing the correlation between the characteristics and clinical behavior scores through statistical analysis.
  2. 2. The method for analyzing the electroencephalogram high-order connection based on the time-spectrum super network is characterized in that the specific implementation method of the step 1 is that conscious disturbance subjects meeting the standard are selected as experimental groups, multichannel electroencephalogram signals are synchronously acquired in a stimulation test stage after spinal cord stimulation electrode implantation, and the signal acquisition covers three stages of a pre-stimulation baseline stage, a pre-stimulation stage and a post-stimulation resting stage.
  3. 3. The brain electrical high-order connection analysis method based on the time-spectrum super network of claim 2, wherein the spinal cord stimulation test in the step 1 comprises a stimulation sequence of two frequency parameters of 5Hz and 70Hz, each frequency acquisition flow comprises 15 minutes baseline brain electrical recording, 15 minutes continuous brain electrical recording under stimulation and 20 minutes resting state brain electrical recording after stimulation, and finally one frequency is selected as a formal stimulation parameter according to the brain electrical background activity improvement degree.
  4. 4. The method for analyzing the high-order connection of the brain waves based on the time-spectrum super network according to claim 1, wherein the step 2 comprises the following steps: Step 2.1, downsampling the acquired multichannel electroencephalogram data to 250Hz, and applying a 1-40Hz band-pass filter and a 50Hz notch filter to remove high-frequency noise and power frequency interference; step 2.2, re-referencing the signal by using the average reference; Step 2.3, reconstructing the missing channel signals by using the existing channel position information based on a spherical spline interpolation method to form uniform lead distribution; And 2.4, identifying and eliminating artifact components related to eye movement and electrocardio by using independent component analysis.
  5. 5. The method for analyzing the electroencephalogram high-order connection based on the time-spectrum super network according to claim 4, wherein the step 2.4 is further characterized by further comprising the step 2.5 of performing manual quality inspection on the preprocessed signals to ensure that the signal quality meets the requirement of subsequent analysis.
  6. 6. The method for analyzing the high-order connection of the brain waves based on the time-spectrum super network according to claim 1, wherein said step 3 comprises the following steps: Step 3.1, constructing a high-order connection relation model between multiple brain areas by adopting a correlation weighted sparse representation method based on the preprocessed stimulation period electroencephalogram signals, and generating a superside set; Step 3.2, obtaining data-driven superlimit weight distribution by combining L1 and L2 regularization constraint through a superlimit weight learning optimization process, wherein L1 is used for introducing sparsity constraint to promote a model to screen out most representative few connections among a plurality of possible high-order connections, so that the interpretation of a supernetwork and the extraction capability of key features are enhanced, L2 is used for introducing smoothness constraint, stabilizing the weight learning process, and improving the generalization capability of the model, so that the constructed supernetwork has better stability and consistency in different subjects or different time periods; And 3.3, constructing a super network association matrix H and a weight matrix W based on the super-edge set and the weight thereof, and further calculating to obtain a symmetrical super network similarity matrix S which is used for completely representing the high-order dynamic interaction mode of the brain.
  7. 7. The method for analyzing the high-order connection of the brain waves based on the time-spectrum super network according to claim 1, wherein said step 4 comprises the following steps: Step 4.1, extracting the time domain connection strength characteristics, extracting the upper triangle elements of the super-network similarity matrix S, and connecting the upper triangle elements in series to form a time domain characteristic vector Ft, wherein the vector is used for directly reflecting the high-order connection strength information between different brain areas; Step 4.2, extracting topological features of the spectral domain graph, regarding a super-network similarity matrix S as an adjacent matrix of the graph G, calculating a graph Laplace matrix L of the super-network similarity matrix S, and carrying out feature decomposition on the L to obtain a feature value sequence of the super-network similarity matrix L as a spectral domain feature vector Fs, wherein a second small feature value is independently identified as a key topological feature Fs 2 and used for representing overall connection robustness of the network; And 4.3, selecting features, namely performing dimension reduction and screening on the extracted time domain feature vector Ft and the spectral domain feature vector Fs by adopting a priori selection algorithm, screening out the 20 highest-ranking features by using a recursive feature elimination method based on a linear support vector machine for Ft, screening out the 6 highest-ranking features from the rest features by using a recursive feature elimination method except for the key features Fs 2 for Fs, and finally merging to form a feature subset containing 27 most discriminative features for subsequent model construction and curative effect evaluation.
  8. 8. The method for analyzing the high-order connection of the brain waves based on the time-spectrum super network according to claim 1, wherein said step 5 comprises the following steps: Step 5.1, training a linear discriminant analysis classifier by using the screened feature subsets to distinguish a spinal cord stimulation effective group from an ineffective group; Step 5.2, evaluating the performance of the classifier by adopting five-fold cross validation, wherein the performance indexes comprise accuracy, sensitivity, specificity and F1 fraction; step 5.3, analyzing the statistical difference of the effective group and the ineffective group on the time-spectrum super network characteristics by using a nonparametric Wilcoxon rank sum test, and calculating the effect quantity; and 5.4, exploring the correlation between the extracted features and the changes of the revisions of the coma recovery scale of the subject by adopting Spearman grade correlation analysis, wherein all statistical tests are corrected by error discovery rate.

Description

Electrocerebral high-order connection analysis method based on time-spectrum super network Technical Field The invention belongs to the technical field of artificial intelligence, relates to electroencephalogram dynamic analysis, and in particular relates to an electroencephalogram high-order connection analysis method based on a time-spectrum super network. Background With the development of neuroimaging technology, multichannel Electroencephalogram (EEG) and its related high-density recording method have been widely used for dynamic observation of brain function network. These techniques are capable of capturing neural electrical activity with high temporal resolution, providing an important way to understand the network mechanisms of the brain in complex states. In recent years, research has increasingly recognized that the implementation of advanced brain functions is not only dependent on independent activation of specific brain regions, but rather on dynamic, coordinated high-level information integration between multiple brain regions. In brain network analysis research, functional connection is a common means, and is generally constructed by calculating bivariate indexes such as linear correlation, phase synchronism and the like between different brain area signals. Such methods can reveal the pattern of connectivity between brain regions and have been applied to the evaluation of many neuromodulation effects. However, traditional functional connections have inherent limitations in characterizing complex brain states such as disturbance of consciousness. The brain is a typically complex system, where one brain region often interacts with multiple other brain regions simultaneously, forming a higher order interaction pattern that goes beyond a pairwise relationship. Such high-order interactions are critical to achieving hierarchical integration of consciousness, whereas traditional pairwise connection-based analysis frameworks have difficulty adequately capturing the dynamic nature of such multi-brain region collaboration. In recent years, supernetwork analysis has provided powerful mathematical tools for characterizing the higher-order relationships between such multiple nodes as an emerging branch of graph theory. Unlike common networks which only describe the connections between nodes, super networks connect multiple nodes simultaneously through "superedges", which can more naturally model the collaborative activity of brain clusters. The super-network theory is combined with neural time sequence analysis to construct a brain time-spectrum super-network, which is hopeful to capture the high-order dynamic characteristics of the brain interaction mode in the time and spectrum domain at the same time, so as to more comprehensively represent the integration mechanism of neural information processing. However, the existing research still has obvious defects in assessing the neuromodulation effect by using the higher order brain network characteristics, and the specific expression is as follows: 1. limitations of network modeling Most of the existing EEG analysis researches are still limited to time-frequency characteristics of channel level or functional networks based on paired connection, and cannot effectively incorporate high-order information of simultaneous interaction of a plurality of brain regions, so that the description capability of the EEG analysis researches on the state of a complex brain network is limited. 2. Lack of dynamic and spectral domain fusion analysis The current method often separates and surveys the time domain connection and the network topology attribute, and lacks a unified framework to extract the joint characteristics which can reflect the transient change of the connection strength (time domain) and can also characterize the integral integration and the robustness (spectral domain) of the network. 3. Lack of quantization and verification framework for higher order features Although the higher-order interaction has important significance in theory, how to construct a stable super-network from the nerve electric signal, extract the time-spectrum characteristics with discriminant power from the super-network, and further establish a quantifiable and verifiable analysis flow for objectively distinguishing different nerve regulation response states, which is still a key problem facing the current field. Disclosure of Invention The invention aims to overcome the defects of the prior art and provide an electroencephalogram high-order connection analysis method based on a time-spectrum super network, which realizes the functions of accurately describing and objectively evaluating the high-order network response caused by nerve regulation by constructing and quantifying the time-spectrum super network characteristics of the brain in a specific state, and provides a new and more sensitive biomarker and analysis tool for understanding a regulation mechanism, predicting individual