A brain function network training effect evaluation method based on a two-level threshold strategy

Abstract

The application discloses a brain function network training effect evaluation method based on a two-level threshold strategy and belongs to the technical field of brain science and artificial intelligence auxiliary diagnosis. In view of the problem that a single threshold method in traditional task state brain function network analysis is easy to make core connection be submerged by noise and low in analysis precision, the application proposes a two-level threshold strategy: firstly, a key sub-network is extracted through a fixed threshold to eliminate weak connection noise; then a series of sparseness is set on the basis of the key sub-network to perform multi-scale binary processing, so that a binary brain network under different sparseness levels is obtained. On this basis, clustering coefficients, characteristic path lengths and other graph theory topological properties are calculated, and significant differences between an untrained group and a trained group are analyzed by statistics comparison, so that neural electrophysiological indexes related to visual motion training are identified. The application can effectively improve the precision and stability of topological property calculation, accurately capture a dynamic change process of a training-induced brain function network from high local clustering to a sparse and efficient architecture, and provide reliable neurophysiological evaluation basis for the fields of motor skill acquisition, virtual reality training and neural rehabilitation.

Description

Technical Field

[0001] This invention belongs to the fields of brain science and artificial intelligence-assisted diagnosis technology, specifically involving a method for evaluating the training effect of brain functional networks based on a two-level threshold strategy, and in particular a multi-scale analysis framework that integrates multi-band static functional connectivity, key sub-network extraction and local-global graph theory indicators, for analyzing training-induced changes in brain functional connectivity and extracting neuroplasticity markers in human-computer interaction visual-motor tasks. Background Technology

[0002] In recent years, task-based electroencephalography (EEG) has been widely applied to research on visual-motor learning and neuroplasticity due to its ability to capture neurophysiological changes during cognitive and motor processing. By dividing the brain into several electrode nodes and calculating their phase synchronization strength, researchers can construct functional connectivity networks that reflect the synchronization relationships of neural activity, and then analyze topological properties such as clustering coefficients and characteristic path lengths to reveal the network reorganization characteristics induced by motor learning. However, visual-motor tasks have complex attributes such as high real-time dynamics, nonlinear response characteristics, and human-computer closed-loop feedback. Neural activity changes rapidly during task execution, resulting in a large number of weak connections in functional connectivity caused by background noise or transient perturbations. If the original network matrix is ​​directly binarized using a fixed threshold or a sparsity threshold, the core connection information is easily submerged, thereby reducing the accuracy and stability of topological property calculation. In complex task-oriented EEG scenarios, traditional methods struggle to accurately capture the dynamic transformation of brain networks from highly localized clusters to sparse and efficient architectures before and after training. They also cannot optimize the structural analysis path in real time based on the dynamic changes in network state. Overall, they suffer from low analysis efficiency, poor adaptability, and insufficient robustness, which limits their ability to provide reliable neurophysiological evidence in fields such as motor skill acquisition, virtual reality training, and neurorehabilitation. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for evaluating the training effect of brain functional networks based on a two-level threshold strategy. By employing phase-locked value (PLV) calculation, fixed threshold extraction of key subnetworks, multi-sparseness binarization processing, topological index calculation, and statistical comparison, the invention uncovers significant changes in the topology of brain networks between the untrained group (UG) and the trained group (TG). Specifically, this invention first introduces PLV phase synchronization calculation and fixed threshold key subnetwork extraction to eliminate weak connection noise and highlight core visuomotor pathways. Second, a series of sparsity settings are used to obtain multi-scale binary networks, improving the model's ability to capture local-to-global integration transitions. Furthermore, this invention proposes a topological index system based on clustering coefficients, feature path lengths, node degree, and local efficiency, and combines statistical tests to achieve interpretability analysis. This method not only uncovers training-induced plasticity changes in brain functional networks but also possesses high biological interpretability, supporting the localization of key brain regions related to visuomotor learning, and has significant application value in rehabilitation, virtual reality training, and human-computer interaction.

[0004] The technical solution of this invention is: a method for evaluating the training effect of brain functional networks based on a two-level threshold strategy, the method comprising the following steps: Step 1: EEG signal acquisition. Task-state EEG data were acquired from the subjects during the visual-motor task. The subjects included the untrained group (UG) and the trained group (TG). Step 2: Perform standard preprocessing on the acquired task-oriented EEG data to remove noise and artifacts. This includes electrode localization, removal of useless electrodes, removal of task-irrelevant segments, bandpass filtering (0.5-45Hz), bad channel identification and interpolation, independent component analysis (ICA), and finally decompose it into EEG signals in multiple frequency bands. Step 3: Brain Functional Network Construction. Based on the preprocessed signals from each frequency band, a brain functional network of the subjects is constructed. Specifically, each EEG channel is used as a node in the network, and the phase synchronization between different channels is calculated as the connection weight between nodes, thereby obtaining the whole-brain functional connectivity matrix; Step 4: A two-level thresholding strategy is used to process the functional connectivity matrix to obtain a binary brain network; Step 5: Based on graph theory, calculate the topological properties of the binary brain network at each sparsity level obtained in Step 4; Step 6: Perform statistical analysis on the topological attributes of the untrained and trained groups to identify the topological attributes with significant differences and their corresponding brain regions, which will serve as neuroelectrophysiological indicators for evaluating the effectiveness of visual-motor training.

[0005] Furthermore, the pre-set visual-motor task in step 1 is a control task for a dynamically unstable system in a loop, involving manipulating a joystick to maintain the balance of a virtual inverted pendulum. This task requires the subject to continuously adjust the control input based on real-time visual feedback to keep the virtual inverted pendulum upright.

[0006] Furthermore, the specific steps of EEG data preprocessing in step 2 are as follows: Step A1: Perform electrode localization, remove useless electrodes, and remove task-irrelevant segments from the data; Step A2: Remove 50Hz power frequency interference from the original EEG signal; Step A3: Remove electrooculography (EOG) and electromyography (EMG) artifacts from EEG signals using independent component analysis; Step A4: Use a bandpass filter to filter the signal to Delta (0.5-4 Hz), Theta (4-8 Hz), Alpha (8-13 Hz), Beta (13-30 Hz), and Gamma (30-45 Hz).

[0007] Furthermore, the method for constructing the brain functional network in step 3 is as follows: Step B1: Analyze the EEG signals from any two channels of the preprocessed EEG signal. and Extract each electrode at each time point instantaneous phase and ; Step B2: Calculate the signal and Phase-locked value between, in, This represents the total number of time points. PLV values ​​range from 0 (out of sync) to 1 (fully synchronized). Step B3: Combine the PLV values ​​of all electrode pairs to obtain a The connection matrix, where This represents the number of electrodes.

[0008] Furthermore, the two-level threshold strategy in step 4 is as follows: Step C1: First-level thresholding – Key subnetwork extraction. A hard threshold is applied to the connection matrix constructed in Step 3 to initially remove weak connections and retain strong connections to form the key subnetwork. The threshold for this step can be set empirically or adaptively determined based on data characteristics. Using this threshold, connections with PLV values ​​below the threshold are considered noisy weak connections and set to zero, while connections with PLV values ​​above the threshold retain their original values, thus obtaining the weighted matrix of the key subnetwork. Step C2: Second-level threshold processing – multi-sparseness fine analysis. Based on the key sub-networks obtained in Step C1, a series of sparsity thresholds are set ranging from 22% to 50%, with a step size of 1%. The key sub-networks are binarized to obtain binary brain networks at different sparsity levels. This processing method ensures the integrity of core connections and highlights the transformation from training-induced local clustering to global integration through multi-scale sparsity, thus providing robust, multi-level input for subsequent topology analysis.

[0009] Furthermore, the topological attributes mentioned in step 5 include clustering coefficients, feature path lengths, local efficiency, and node degree. Clustering coefficient The formula used to measure the tightness of connections between nodes in a network is: Among them, is the node The node degree is the maximum number of edges that can exist between neighboring nodes. The network clustering coefficient is the average of the clustering coefficients of all nodes. Characteristic path length measures the efficiency of global information transmission in a network. It is the average of the shortest path lengths between all pairs of nodes, and is usually defined as the average of the shortest path lengths between all pairs of nodes in the network. The calculation formula is: in, Represents a node and nodes The shortest path length, The number of nodes; Node degree D measures the degree of connectivity of a node in a network. The degree of a node is defined as the number of edges directly connected to that node. The formula is:

[0010] Furthermore, the statistical analysis process of the topological attributes in step 6 is as follows: Step D1: Use the Shapiro-Wilk test to check the normality of the data; Step D2: For data that conforms to a normal distribution, use an independent samples t-test; for data that does not conform to a normal distribution, use a Mann-Whitney U test. Step D3: Bonferroni multiple comparison correction was performed to determine the significant difference. Step D4: By comparing the differences between the two groups, identify the frequency bands, network topology attributes, and key brain regions that caused significant changes during training.

[0011] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention proposes for the first time a two-level threshold strategy for task-oriented brain functional network analysis. The first level, a fixed threshold, is used to initially extract key subnetworks, removing a large number of noisy and weak connections. The second level, multi-sparseness fine analysis, can more sensitively capture the dynamic changes in network topology attributes before and after training, based on the key subnetworks. This mechanism effectively solves the problem that existing single-threshold methods easily lead to core connections being overwhelmed by noise and resulting in low analysis accuracy. (2) This invention uses a two-level threshold strategy to make the calculation of topological attributes based on more reliable core connections, which significantly improves the calculation accuracy and stability of indicators such as clustering coefficient and feature path length, and can accurately capture the dynamic transformation process of training-induced brain functional networks from high local clustering to sparse and efficient architecture. Attached Figure Description

[0012] Figure 1 This is a diagram showing the electrode location distribution of the 30 channels implemented in this invention.

[0013] Figure 2 This is a flowchart of the task-state brain functional network analysis method based on a two-level threshold strategy of the present invention.

[0014] Figure 3 This is a heatmap of the brain functional connectivity matrix constructed based on PLV in an embodiment of the present invention.

[0015] Figure 4 This is a statistical chart showing the inter-group differences in clustering coefficients at different sparsities in each frequency band in an embodiment of the present invention.

[0016] Figure 5 This is a statistical chart showing the inter-group differences in path lengths of two groups at different sparsities in each frequency band, as presented in an embodiment of the present invention.

[0017] Figure 6 This is a statistical chart showing the inter-group differences in node degree across each frequency band in an embodiment of the present invention. Detailed Implementation

[0018] To facilitate understanding of this application, a more comprehensive description of this application will be provided below with reference to the accompanying drawings. Figure 1 This is a diagram showing the distribution of the 30-channel electrode positions in an embodiment of the present invention. Figure 2 This is a flowchart of the task-oriented brain functional network analysis method based on a two-level threshold strategy, as described in this invention. Figure 1 and Figure 2 As shown, the method for dynamic performance analysis of brain functional networks based on sparsity adjustment of key sub-networks includes the following steps: Step 1: Recruit participants and collect data. Recruited participants were divided into an untrained group (UG) and a trained group (TG). EEG data acquisition used a 30-channel recording system (International 10-20 system), and the task was inverted pendulum balance control. Step 2: Preprocessing based on Matlab 2018b and the eeglab 2023 toolbox; Step 2.1: Locate the 30 channel electrodes based on their coordinate positions; Step 2.2: Remove useless segment data other than those in the task state; Step 2.3: Use the notch filter in the toolbox to remove the 50Hz power frequency interference from the EEG signal; Step 2.4: Use independent component analysis to decompose the EEG signal into 30 independent components. Then identify and remove artifacts. Step 2.5: Divide the full-band EEG signal into 5 sub-bands; Step 3: Calculate the PLV connection matrix. Calculate the PLV by frequency band using the same formula as above, resulting in a 30*30 functional connection matrix; Step 4: Obtain the key subnetwork using the fixed threshold method. In this implementation, the fixed threshold for sampling the PLV connection matrix is ​​0.3, retaining strong connections and removing weak connections to form a key subnetwork containing only core synchronization relationships, thus eliminating noise interference; Step 5: Based on the key sub-networks, a series of sparsities are set to obtain a binary brain functional network. In this implementation, based on the constructed sub-networks, the sparsity is set to 22%~50%, and the step size is set to 1% to obtain a series of binary networks; Step 6: Calculate network topology attributes. Calculate node degree, clustering coefficient, characteristic path length, etc. The specific calculation formulas for each attribute have been detailed in the invention content section and will not be repeated here; Step 7: Perform statistical analysis. Use the Shapiro-Wilk test to check the normality of the data. For data that are normally distributed, use the independent samples t-test for between groups (UG vs. TG); for data that are not normally distributed, use the Mann-Whitney U test. Perform Bonferroni multiple comparison correction, with a significance level of 100%. ; Step 8: Results Analysis. Within the given sparsity range, the clustering coefficients of both groups monotonically increase with increasing sparsity across all frequency bands. Furthermore, the clustering coefficients of the TG group are lower than those of the UG group at all sparsity levels, and this difference reaches a significant level at some sparsity points. Within the given sparsity range, the feature path lengths of both groups monotonically decrease with increasing sparsity across all frequency bands. Furthermore, the feature path lengths of the TG group are lower than those of the UG group at all sparsity levels, with significant differences only in the lower sparsity range. This further confirms the optimization of the network topology after training: after training, redundant connections in the brain network are reduced, the global information integration ability is enhanced, and the network structure shifts towards a sparse and efficient mode. Node degree results show that significantly different nodes are distributed in the temporal lobe, frontal lobe, and occipital lobe, almost all located in the right side of the brain.

Claims

1. A method for evaluating the training effect of brain functional networks based on a two-level threshold strategy, the method comprising the following steps: Step 1: Collect task-state EEG data of subjects in the visual-motor task. The subjects include untrained group and trained group. Step 2: Preprocess the acquired task-state EEG data to remove noise and artifacts, and decompose it into EEG signals of multiple frequency bands; Step 3: Based on the preprocessed EEG signals of each frequency band, construct the whole-brain functional connectivity matrix; wherein, each EEG channel is used as a node of the network, and the phase synchronization between different channels is calculated as the connection weight between nodes; Step 4: Process the functional connectivity matrix using a two-level thresholding strategy to obtain a binary brain network; Step 5: Based on graph theory, calculate the topological properties of the binary brain network at each sparsity level obtained in Step 4; Step 6: Perform statistical analysis on the topological attributes of the untrained and trained groups to identify the topological attributes with significant differences and their corresponding brain regions, which serve as neuroelectrophysiological indicators for evaluating the effectiveness of visual-motor training.

2. The method according to claim 1, characterized in that, The preset visual-motor task in step 1 is a dynamic unstable system control task for a person in the loop. Specifically, it is a task of controlling a virtual inverted pendulum to maintain balance by manipulating a joystick. The subject is required to continuously adjust the control input based on real-time visual feedback to keep the virtual inverted pendulum in an upright state.

3. The method according to claim 1, characterized in that, The preprocessing in step 2 is as follows: Step A1: Perform electrode localization, remove useless electrodes, and remove task-irrelevant segments from the data; Step A2: Remove 50Hz power frequency interference from the raw EEG signal; Step A3: Remove electrooculography (EOG) and electromyography (EMG) artifacts from EEG signals using independent component analysis; Step A4: Use a bandpass filter to filter the signal to the Delta (0.5-4 Hz), Theta (4-8 Hz), Alpha (8-13 Hz), Beta (13-30 Hz), and Gamma (30-45 Hz) frequency bands.

4. The method according to claim 1, characterized in that, The specific method for constructing the brain functional network in step 3 is as follows: Step B1: Analyze the EEG signals from any two channels of the preprocessed EEG signal. and Extract each electrode at each time point instantaneous phase and ; Step B2: Calculate the signal and Phase-locked value between, ； in, This represents the total number of time points. PLV values ​​range from 0 to 1. Step B3: Combine the PLV values ​​of all electrode pairs to obtain a The connection matrix, where This represents the number of electrodes.

5. The method according to claim 1, characterized in that, The two-level threshold processing in step 4 is as follows: Step C1: First-level threshold processing: Hard threshold processing is performed on the functional connection matrix using a fixed threshold to remove weak connections and retain strong connections to form the weighted matrix of the key sub-networks. Step C2: Second-level threshold processing. Based on the key sub-network, a series of sparsity threshold ranges are set to binarize the key sub-network and obtain binary brain networks under different sparsity levels.

6. The method according to claim 1, characterized in that, The topological attributes in step 5 include clustering coefficient, feature path length, local efficiency, and node degree. The clustering coefficient C is used to measure the tightness of the connections between nodes in the network. The network clustering coefficient is the average of the clustering coefficients of all nodes. The characteristic path length L is used to measure the efficiency of global information transmission in the network, and is defined as the average of the shortest path lengths between all node pairs. The node degree D is used to measure the connectivity of a node in the network, and is defined as the number of edges directly connected to that node.

7. The method according to claim 7, characterized in that, The clustering coefficient The calculation formula is: ； in, It is with nodes The degree of the node, It represents the maximum number of edges that can exist between neighboring nodes. The network clustering coefficient is the average of the clustering coefficients of all nodes. The formula for calculating the feature path length L is: ； in, Represents a node and nodes The shortest path length, The number of nodes; The formula for calculating the node degree D is: ； in, Represents a node With nodes The connection status between them, if a connection exists. .

8. The method according to claim 1, characterized in that, The statistical analysis in step 6 is as follows: Step D1: Use the Shapiro-Wilk test to check the normality of the data; Step D2: For data that conforms to a normal distribution, use an independent samples t-test; for data that does not conform to a normal distribution, use a Mann-Whitney U test. Step D3: Bonferroni multiple comparison correction was performed to determine the significant difference. Step D4: By comparing the differences between the two groups, identify the frequency bands, network topology attributes, and key brain regions that caused significant changes during training.

Images