CN-122004753-A - Dynamic function network self-adaptive construction method for physiological signal analysis and related equipment

Abstract

The embodiment of the application provides a dynamic functional network self-adaptive construction method and related equipment for physiological signal analysis, belonging to the technical field of biomedical signal processing and computer-aided diagnosis. The method comprises the steps of inputting multichannel physiological signals, extracting preset frequency bands, carrying out time sequence segmentation on the signals by adopting sliding windows, calculating phase locking value matrixes among all channels for each time window, carrying out personalized sparse processing on each connection matrix by adopting a self-adaptive threshold method, and constructing a dynamic graph sequence based on sparse results, wherein node characteristics adopt weight centrality. According to the application, time-varying information of the brain function network is completely reserved through the dynamic graph sequence, noise robustness and individuation processing are realized through the self-adaptive threshold, the learning efficiency of the subsequent graph neural network is enhanced through the node characteristics related to the structure, the accuracy of classification of consciousness disturbance is remarkably improved, and the method has the universality of expanding to other physiological signal analysis and industrial monitoring fields.

Inventors

• CHEN HE
• WANG YONG
• XIAO JIAQING
• GUAN LONGZHOU
• LI XIAOLI

Assignees

• 华南理工大学

Dates

Publication Date

20260512

Application Date

20251225

Claims (10)

1. 1. A method for adaptively constructing a dynamic functional network for physiological signal analysis, the method comprising the steps of: inputting multichannel physiological signals and extracting signals of a preset frequency band; performing time sequence segmentation on the signal of the preset frequency band by adopting a sliding window to obtain a plurality of time windows; for each time window, calculating phase locking values PLV between every two channels to obtain a functional connection matrix of the time window; Performing self-adaptive threshold sparsification processing on the functional connection matrix of each time window, and reserving connection with PLV larger than the self-adaptive threshold to obtain a sparsified connection matrix; constructing a dynamic graph sequence based on the sparse connection matrix of each time window; Outputting the dynamic graph sequence for subsequent classification or regression tasks; wherein the physiological signal comprises at least one of electroencephalogram EEG, electrocardiographic ECG, myoelectric EMG, and electrooculogram EOG.
2. 2. The method of claim 1, wherein when the physiological signal is an electroencephalogram, EEG, each dynamic map is constructed Includes node set Edge set Sum node feature matrix ; Node set Corresponding EEG channels; Edge set The method comprises the steps of connecting the two connection blocks after sparsification, wherein the edge weight is a PLV value of the corresponding connection; Node characteristic matrix The feature of each node in the (3) is the weight centrality of the node, and the node is calculated based on the sparse connection matrix.
3. 3. The method of claim 2, wherein the preset frequency band comprises at least one of a Theta frequency band and an Alpha frequency band.
4. 4. The method of claim 1, wherein the adaptive threshold sparsification process determines the threshold using an OTSU algorithm or a genie-purity algorithm based on a PLV matrix.
5. 5. The method of claim 4, wherein the PLV matrix-based OTSU algorithm comprises: for candidate threshold values Dividing the connection into foreground class and background class, and calculating the variance between classes Select to make Maximum of As an adaptive threshold.
6. 6. The method of claim 2, wherein the node Weight centrality of (2) The calculation formula of (2) is as follows: Wherein, the Is a node A set of neighbor nodes in a sparse connection matrix; Is a node With neighbor nodes Weights of the edges in between.
7. 7. The method of claim 2, wherein the edge set Comprising the following steps: An edge index matrix with dimensions of Representing a source node and a target node of each edge; Edge weight vector with dimensions of Representing the PLV value for each edge.
8. 8. An electronic device comprising a memory storing a computer program and a processor implementing the method of any of claims 1 to 7 when the computer program is executed by the processor.
9. 9. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the method of any one of claims 1 to 7.
10. 10. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the method of any one of claims 1 to 7.

Description

Dynamic function network self-adaptive construction method for physiological signal analysis and related equipment Technical Field The application relates to the technical field of biomedical signal processing and computer-aided diagnosis, in particular to a dynamic function network self-adaptive construction method for physiological signal analysis and related equipment. Background Consciousness disorders (DoC), such as Vegetative State (VS) and Micro Consciousness State (MCS), are common consequences after severe brain injury. Accurate discrimination of these states is critical to prognosis and treatment decisions. Resting electroencephalogram (EEG) is widely used for objective assessment of DoC due to its non-invasive, low cost and high temporal resolution characteristics. In recent years, brain function network analysis methods based on Graph Neural Networks (GNNs) have shown great potential in EEG decoding. The core of the method is to model the brain as a complex network (graph) where nodes represent brain regions (EEG electrodes) and edges represent the strength of the functional connection between brain regions (typically measured using indices such as phase lock values PLV). However, the prior art has significant drawbacks in converting the original EEG signal into high quality map data suitable for GNN: 1) The limitation of static network assumptions is that existing methods typically compute the entire EEG recording (e.g., minutes) as a global average functional connection matrix and treat it as a static map. This approach ignores entirely the dynamic time-varying nature inherent to brain function connections, and a number of neuroscience studies have shown that the neural basis of consciousness is closely related to the dynamic interactions between large-scale brain networks. Static averaging methods wipe out transient network pattern information that may be critical to distinguishing between different recognition states. 2) Network dense and noise problem-directly calculated full connection PLV matrices are dense, containing a large number of weak connections caused by noise or random fluctuations. These redundant connections not only greatly increase the computational burden, but also interfere with the GNN's learning of critical connection patterns, reducing the robustness and generalization ability of the model. In the prior art, fixed threshold values or fixed proportions are generally adopted for sparsification, and the strategy of one-cut cannot adapt to individual differences of connection strength of different patients and different time periods, so that information loss or noise residue is easy to cause. 3) Node feature definition is disjoint from the graph structure in that an initial feature vector needs to be defined for each node when constructing the graph data. The prior art often uses features that are independent of the network topology, such as the band power of the electrodes or the raw signal. The node characteristics and the functional connection relation represented by the edges are independent, the synergetic enhancement representation cannot be formed, and the learning difficulty of the GNN model is increased. Disclosure of Invention The main purpose of the embodiment of the application is to provide a dynamic functional network self-adaptive construction method, electronic equipment, storage medium and program product for physiological signal analysis (such as obstacle-of-consciousness analysis), which can extract and retain dynamic time-varying information of functional connection from original physiological (such as EEG) signals, optimize a network structure in a data-driven self-adaptive manner and generate node characteristics closely related to network topology, thereby providing high-quality input data for a subsequent dynamic graph neural network. To achieve the above object, an aspect of an embodiment of the present application provides a dynamic functional network adaptive construction method for physiological signal analysis, the method including: inputting multichannel physiological signals and extracting signals of a preset frequency band; performing time sequence segmentation on the signal of the preset frequency band by adopting a sliding window to obtain a plurality of time windows; for each time window, calculating phase locking values PLV between every two channels to obtain a functional connection matrix of the time window; Performing self-adaptive threshold sparsification processing on the functional connection matrix of each time window, and reserving connection with PLV larger than the self-adaptive threshold to obtain a sparsified connection matrix; constructing a dynamic graph sequence based on the sparse connection matrix of each time window; Outputting the dynamic graph sequence for subsequent classification or regression tasks; wherein the physiological signal comprises at least one of electroencephalogram EEG, electrocardiographic ECG, myoelec