Dynamic function network adaptive construction method for physiological signal analysis and related equipment

Abstract

The embodiment of the invention provides a dynamic function network adaptive construction method for physiological signal analysis and related equipment, and belongs to the technical field of biomedical signal processing and computer-aided diagnosis. The method comprises the following steps: inputting a multi-channel physiological signal and extracting a preset frequency band; performing time sequence segmentation on the signal by adopting a sliding window; calculating a phase locking value matrix among all channels for each time window; carrying out personalized sparsification processing on each connection matrix by adopting a self-adaptive threshold value method; and constructing a dynamic graph sequence based on a sparse result, wherein node features adopt weighting centrality. According to the method, the time-varying information of the brain function network is completely reserved through the dynamic graph sequence, the noise robustness and personalized processing are realized through the adaptive threshold, the learning efficiency of the subsequent graph neural network is enhanced through the structure-related node features, the accuracy of disturbance of consciousness classification is remarkably improved, and the classification efficiency of the disturbance of consciousness is improved. And the method has the universality of expanding to other physiological signal analysis and industrial monitoring fields.

Description

Technical Field

[0001] This application relates to the fields of biomedical signal processing and computer-aided diagnostic technology, and in particular to a dynamic functional network adaptive construction method and related equipment for physiological signal analysis. Background Technology

[0002] Disorders of consciousness (DoC), such as vegetative state (VS) and minimally conscious state (MCS), are common consequences of severe brain injury. Accurate differentiation of these states is crucial for prognostic assessment and treatment decisions. Resting-state electroencephalography (EEG) is widely used for the objective assessment of DoC due to its non-invasive, low-cost, and high temporal resolution characteristics.

[0003] In recent years, graph neural network (GNN)-based brain functional network analysis methods have shown great potential in EEG decoding. The core of this method is to model the brain as a complex network (graph), where nodes represent brain regions (EEG electrodes) and edges represent the strength of functional connections between brain regions (typically measured using metrics such as phase-locked value (PLV)). However, existing techniques have significant limitations in converting raw EEG signals into high-quality graph data suitable for GNNs: 1) Limitations of the static network assumption: Existing methods typically calculate an entire EEG recording (e.g., several minutes) as a globally averaged functional connectivity matrix, treating it as a static graph. This approach completely ignores the inherent dynamic and time-varying nature of brain functional connectivity, while numerous neuroscience studies have shown that the neural basis of consciousness is closely related to the dynamic interactions between large-scale brain networks. The static averaging method obliterates transient network pattern information that may be crucial for distinguishing different states of consciousness.

[0004] 2) Network density and noise issues: Directly computed fully connected PLV matrices are dense, containing numerous weak connections caused by noise or random fluctuations. These redundant connections not only significantly increase the computational burden but also interfere with the GNN's learning of key connection patterns, reducing the model's robustness and generalization ability. Existing techniques typically use fixed thresholds or fixed ratios for sparsity reduction. This "one-size-fits-all" strategy cannot adapt to the individual differences in connection strength among different patients and at different times, easily leading to information loss or residual noise.

[0005] 3) Node feature definition is disconnected from graph structure: When constructing graph data, an initial feature vector needs to be defined for each node. Existing techniques often use features unrelated to the network topology, such as the band power of electrodes or the original signal. This results in node features being independent of the functional connections represented by edges, failing to form a synergistically reinforcing representation and increasing the learning difficulty of the GNN model. Summary of the Invention

[0006] The main objective of this application is to propose a dynamic functional network adaptive construction method, electronic device, storage medium, and program product for physiological signal analysis (such as consciousness disorder analysis). This method can extract and retain dynamic time-varying information of functional connections from raw physiological (such as EEG) signals, optimize the network structure through a data-driven adaptive approach, and generate node features closely related to the network topology, thereby providing high-quality input data for subsequent dynamic graph neural networks.

[0007] To achieve the above objectives, one aspect of this application proposes a dynamic functional network adaptive construction method for physiological signal analysis, the method comprising: Input multi-channel physiological signals and extract signals from preset frequency bands; The signal in the preset frequency band is time-series segmented using a sliding window to obtain multiple time windows; For each time window, calculate the phase lock value (PLV) between all pairs of channels to obtain the functional connection matrix for that time window; Adaptive threshold sparsification is applied to the functional connection matrix of each time window, retaining connections with PLV greater than the adaptive threshold, to obtain a sparse connection matrix. A dynamic graph sequence is constructed based on the sparsed connection matrix of each time window; Output the dynamic graph sequence for subsequent classification or regression tasks; The physiological signals include at least one of electroencephalography (EEG), electrocardiography (ECG), electromyography (EMG), and electrooculography (EOG).

[0008] In some embodiments, when the physiological signal is an electroencephalogram (EEG), each constructed dynamic graph Includes node set Edge set and node feature matrix ; Node set Corresponding EEG channel; Edge set It consists of connections retained after sparsification, and the edge weight is the PLV value of the corresponding connection. Node feature matrix Each node in the matrix is ​​characterized by its weighted degree centrality, which is calculated based on the sparsed connection matrix.

[0009] In some embodiments, the preset frequency band includes at least one of the Theta band (4-8 Hz) and the Alpha band (8-13 Hz).

[0010] In some embodiments, the sliding window has a window length of 2 seconds, a step size of 2 seconds, and no overlap between windows.

[0011] In some embodiments, the adaptive threshold sparsity processing uses the OTSU algorithm based on the PLV matrix or the Gini impurity algorithm to determine the threshold.

[0012] In some embodiments, the OTSU algorithm based on the PLV matrix includes: For candidate thresholds The connections are divided into foreground and background classes, and the variance between classes is calculated. , choose to The largest As an adaptive threshold.

[0013] In some embodiments, node Weighted degree centrality The calculation formula is:

[0014] in, For nodes The set of neighboring nodes in the sparse connectivity matrix; For nodes with neighboring nodes The weight of the edges between them.

[0015] In some embodiments, the edge set include: Edge index matrix, dimension 1 , representing the source node and target node of each edge; Edge weight vector, dimension , representing the PLV value corresponding to each edge.

[0016] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0017] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0018] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0019] Compared with the prior art, this application has the following beneficial effects: 1) Complete preservation of dynamic information: By generating a sequence of dynamic graphs through a sliding window, rather than a single static graph, key information about the fluctuation of brain functional connections over time is fully preserved, providing a richer data foundation for distinguishing rapidly changing states of consciousness.

[0020] 2) Personalized adaptive denoising: Adaptive thresholding methods based on the distribution of the data itself (such as OTSU) are used instead of fixed global thresholds. This allows for personalized sparsification of connection strength features for different patients and different time windows, which can more effectively filter noise and highlight discriminative strong connection patterns.

[0021] 3) Synergistic Enhancement of Node Features and Graph Structure: An innovative approach uses weighted degree centrality as a node feature, directly calculated from the network's connection weights, making the node features highly correlated with the graph's topology. This provides powerful prior information for graph neural networks, significantly improving the model's message passing efficiency and feature learning ability.

[0022] 4) High compatibility and complete information retention: The constructed graph is a weighted graph, which retains the PLV weight information. It can be directly used in graph neural networks that support edge weights to improve performance, or it can be used in unweighted graph neural networks through simple transformation (such as binarization), which has high compatibility.

[0023] 5) Provide high-quality input for high-order models: The dynamic graph sequences produced by this method are a prerequisite for running advanced models such as Dynamic Graph Neural Networks (DGNN), thus enabling full exploitation of temporal dependencies. Experiments show that the classification accuracy can be improved from about 62.9% for the static graph-based GCN model to about 76.7% for the DGCN model based on the dynamic graph sequences constructed by this method. Attached Figure Description

[0024] Figure 1 This is a flowchart of a dynamic functional network adaptive construction method for consciousness disorder analysis in the embodiments of this application.

[0025] Figure 2 This is a schematic diagram of the hardware structure of the electronic device in the embodiments of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0028] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0029] 1) Disorder of consciousness: refers to a state of decreased level of consciousness or altered content of consciousness due to severe brain injury. In this application, it specifically refers to the minimally conscious state and the vegetative state that need to be distinguished.

[0030] 2) Electroencephalography (EEG): A physiological monitoring technique that records the electrical activity of neuronal groups in the cerebral cortex using electrodes placed on the scalp. In this application, it specifically refers to resting-state electroencephalography, i.e., the brain electrical signals collected when the patient is quiet, with eyes closed, and without a specific task.

[0031] 3) Phase Lock Value (PLV): A metric used to quantify the phase synchronization between two neural oscillatory signals. Its value ranges from [0, 1], with a larger value indicating a more stable phase relationship and stronger neural synchronization. PLV is a core metric for constructing brain functional networks.

[0032] 4) Functional connectivity: refers to the degree of functional association between different brain regions, which is usually measured by calculating the statistical dependence (PLV) of neural signals (such as EEG) between two brain regions.

[0033] 5) Dynamic functional connectivity: This specifically refers to the phenomenon that the strength of brain functional connectivity changes over time. This application captures this dynamic characteristic by analyzing the changes in PLV over a continuous time window.

[0034] 6) Sliding window: A time series analysis method that uses a fixed-length window to slide along the time axis to divide a continuous signal into a series of continuous, possibly overlapping, short time periods for analyzing the local characteristics of the signal.

[0035] 7) Window length: The time span of the sliding window. In this application, it specifically refers to a duration of 2 seconds, which has been verified to be sufficient to calculate a stable PLV and capture meaningful dynamic changes in brain networks.

[0036] 8) Step size: The time interval between each movement of the sliding window. In this application, it specifically refers to a 2-second interval, which means that there is no overlap between windows, so as to ensure the independence of signals within each time window and balance computational efficiency.

[0037] 9) Functional connectivity matrix: A symmetric matrix whose element (i, j) represents the functional connectivity strength (e.g., PLV value) between the i-th channel and the j-th channel. In this application, an independent functional connectivity matrix is ​​calculated for each time window.

[0038] 10) Threshold sparsity: A graph data processing technique that converts a dense graph into a sparse graph by setting a threshold value, retaining only connections with weights greater than the threshold value, thereby achieving the purpose of noise reduction and dimensionality reduction.

[0039] 11) Global absolute threshold: A fixed threshold applied to the entire connectivity matrix. In this application, 0.65 was experimentally determined as the optimal global absolute threshold for filtering significant functional connections from the PLV matrix.

[0040] 12) Graph: A data structure used to represent entities and their relationships, consisting of nodes (or vertices) and edges. In this application, a graph represents a snapshot of the brain's functional network within a time window.

[0041] 1) Node: In a graph structure, a node represents an entity. In this application, each node corresponds to an EEG acquisition electrode channel, representing a specific brain region.

[0042] 13) Edge: In a graph structure, an edge is a line segment connecting two nodes, representing the relationship between the nodes. In this application, an edge represents a significant functional connection (PLV > threshold) between two brain regions, and the edge weight is the PLV value of that connection.

[0043] 14) Node feature: The attribute vector associated with each node. In this application, node feature specifically refers to the degree centrality of each node (brain region).

[0044] 15) Degree centrality: One of the metrics for measuring the centrality of a node in graph theory, representing the number of edges connected to that node. In this application, it specifically refers to weighted degree centrality, which is the sum of the weights of all adjacent edges of a node, used to characterize the local importance of that node in the current network.

[0045] 16) Dynamic graph sequence: A collection of graph structures arranged in chronological order. In this application, dynamic graph sequence... This is the final output of this method, which fully characterizes the evolution of brain functional networks over time. The graph represents the t-th time window.

[0046] In computer-aided diagnosis of disorders of consciousness (DoC) based on electroencephalography (EEG), the cornerstone of the entire technical process lies in how to construct input data from raw, continuous, multi-channel EEG signals that can effectively characterize brain state and is suitable for modern deep learning models (especially graph neural networks). The quality of this data construction process directly determines the upper limit of subsequent model performance.

[0047] Traditional EEG feature extraction methods (such as calculating band power and event-related potentials) mainly focus on the time / frequency domain characteristics of individual channels or simple relationships between a few channel pairs, failing to characterize the collaborative working patterns between multiple brain regions from a global perspective. In recent years, the concept of "brain functional networks" has been introduced, modeling the brain as a complex network system where nodes represent brain regions and edges represent functional connections between them. Phase-locked value (PLV), due to its insensitivity to amplitude and its ability to effectively quantify neural oscillation synchronicity, has become one of the most commonly used metrics for constructing EEG functional networks.

[0048] However, existing PLV-based brain network construction methods have serious limitations: 1) Static Network Assumption: To simplify analysis, most studies use signals from the entire EEG recording period (usually 2-5 minutes) to calculate a "globally averaged" PLV matrix. This approach implicitly assumes that the brain's functional connectivity in the resting state is fixed. This completely contradicts the general consensus in neuroscience—that the brain is a dynamic and constantly evolving system. Consciousness, as the highest level of brain function, is considered to be the result of dynamic interactions between different large-scale brain networks (such as the default mode network and the prefrontal-parietal network). The static averaging method obliterates all time-related dynamic information and fails to reveal the reorganization patterns of brain functional connectivity on the second or even millisecond scale, and these dynamic characteristics may be the key to distinguishing between minimally conscious states (MCS) and vegetative states (VS) and other states of impaired consciousness.

[0049] 2) Network density and noise issues: The directly computed fully connected PLV matrix is ​​a dense matrix (e.g., 1891 connections between 62 electrodes), containing a large number of weak connections. These weak connections are likely due to noise, artifacts, or random fluctuations and do not have clear neurophysiological significance. Directly inputting such a dense network into a graph neural network (GNN) will bring significant problems: First, the computational burden is enormous; second, noisy connections will interfere with the model's learning of key connection patterns, reducing the model's generalization ability and robustness.

[0050] 3) Arbitrariness in node feature definition: When inputting graph data into a GNN, initial features need to be defined for each node (electrode). Common practices in existing technologies include using the raw signal of the electrode, its bandwidth power, or simply using one-hot encoding. These features are either unrelated to the graph structure itself and fail to reflect the role of the node in the network, or they lack clear physical meaning and cannot effectively guide the model's learning.

[0051] For example, one existing technical solution is a graph construction method based on a static, averaged functional connectivity matrix, which was commonly used in early studies applying GCN to EEG analysis. Its technical solution can be described in detail below: 1) Data Averaging: Acquire resting-state EEG data for a duration of time (e.g., 3 minutes). After preprocessing and band filtering, ignoring the internal temporal structure of the signal, treat the entire 3-minute data as a whole and calculate the average PLV values ​​between all pairs of channels, thus obtaining a unique, global [C,C] functional connectivity matrix (C is the number of electrodes, e.g., 62). This matrix represents an "average snapshot" of brain functional connectivity throughout the recording period.

[0052] 2) Fixed-threshold sparsity: To reduce the density of the network, a fixed, general threshold is applied to the average matrix (e.g., retaining the top 10% of edges with the highest PLV values, or setting an absolute threshold such as 0.5). After thresholding, a relatively sparse adjacency matrix is ​​obtained.

[0053] 3) Graph construction and node feature definition: Node (Vertex / Node, V): Defined as an EEG electrode.

[0054] Edge (E): The connection retained after thresholding. The edge weight is 0 or 1, that is, either there is an edge connection or there is no edge connection. The resulting graph is an unweighted graph.

[0055] Node features (X): These typically employ features independent of the graph structure, such as: a) The average power of each electrode in a certain frequency band (e.g., Alpha) throughout the entire EEG period.

[0056] b) The original signal sampling points of each electrode (after normalization).

[0057] c) Artificially constructed statistical characteristics (such as mean and variance).

[0058] 4) Output: The final output is a static graph structure. It is used as input for training and classification of static GCN models.

[0059] Analysis of the above methods reveals the following shortcomings that urgently need to be addressed: 1) Complete lack of dynamic information: This is the most fundamental flaw. "Global averaging" completely eliminates all information about the evolution of brain functional connections over time. For patients with consciousness disorders whose levels of consciousness may fluctuate, signs of residual consciousness may only appear in certain specific, brief time windows. Static averaging methods cannot capture these transient but potentially crucial network states, resulting in a significant loss of information for classification during the data preprocessing stage.

[0060] 2) Crude Threshold Selection Strategy: Using a fixed threshold or fixed ratio for sparsification is a very crude strategy. It cannot adapt to the inherent variability in connection strength among different patients and different recording periods. For a patient with generally strong connections, a fixed ratio threshold may retain too many redundant connections; while for a patient with generally weak connections, a fixed absolute value threshold may inadvertently delete relatively important weak connections for that patient, resulting in information loss.

[0061] 3) Information loss due to unweighted edges: Binarization is used in graph data construction, where edge weights are either 0 or 1, allowing only their existence or non-existence, thus constructing an unweighted graph. While this reduces computational burden, it undoubtedly results in the loss of a large amount of information, lowers the upper limit of the model, and is only suitable for training unweighted graph neural networks.

[0062] 4) Node features are disconnected from graph structure: Using average power or the original signal as node features completely ignores the network's topological properties during calculation. This makes node features independent of the functional connections represented by edges, failing to form a mutually reinforcing representation. The GNN model has to learn the relationship between features and structure from scratch, increasing the learning difficulty and reducing efficiency.

[0063] 5) Mismatch between data and model capabilities: Static graph data limits the use to static GNN models, making it impossible to fully utilize more powerful models that can handle temporal dynamics (such as dynamic graph neural networks DGCN), thus artificially suppressing the performance ceiling of the overall technical solution.

[0064] In view of this, this application provides a method for adaptive construction of dynamic functional networks, an electronic device, a storage medium, and a program product for analyzing consciousness disorders. This solution can: 1) Extract and preserve the dynamic time-varying characteristics of brain functional connectivity from EEG signals.

[0065] 2) Effectively denoise and highlight strong connections through an adaptive, data-driven, personalized sparsity strategy.

[0066] 3) By preserving edge weights, we construct edge-weighted graph data to retain the original data to the maximum extent, thereby increasing the upper limit of the model while being compatible with both edge-weighted graph neural networks and edgeless graph neural networks.

[0067] 4) Provide a node feature definition method that is closely related to the graph structure to enhance the learning efficiency and performance of subsequent graph neural networks.

[0068] 5) Provide high-quality, directly input dynamic graph sequence data for dynamic graph neural networks.

[0069] This application provides a method for adaptive construction of dynamic functional networks (fFNN) for analyzing consciousness disorders, relating to the fields of biomedical signal processing and computer-aided diagnosis. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited thereto. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the method for adaptive construction of dynamic functional networks for analyzing consciousness disorders, but is not limited to the above forms.

[0070] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0071] like Figure 1 As shown, this embodiment provides a dynamic functional network adaptive construction method for consciousness disorder analysis, including the following steps: S1. Input multi-channel resting-state electroencephalogram (EEG) signals and extract signals from preset frequency bands.

[0072] In one embodiment, a preprocessed, multi-channel resting-state EEG signal is input. A bandpass filter (such as a 4th-order Butterworth filter) is used to extract specific frequency bands related to the level of consciousness, preferably the Theta (4-8 Hz) and Alpha (8-13 Hz) bands.

[0073] S2. The signal of the preset frequency band is time-series segmented using a sliding window to obtain multiple time windows.

[0074] In one embodiment, a sliding window technique is used to segment the extracted frequency band signal. A preferred window length (win_len) is 2 seconds, and a step size (win_step) is 2 seconds (i.e., no overlap), to capture the rapid dynamic changes of the brain network while ensuring signal segment independence and computational efficiency.

[0075] S3. For each time window, calculate the phase lock value (PLV) between all pairs of channels to obtain the functional connection matrix for that time window.

[0076] Specifically, for each time window Calculate the phase-locked value (PLV) between all pairs of channels to form a C×C functional connectivity matrix for this window. The calculation of PLV involves performing a Hilbert transform on the signal to obtain the instantaneous phase and calculating the average consistency of the phase difference vector.

[0077] S4. Perform adaptive threshold sparsification on the functional connection matrix of each time window, retaining connections with PLV greater than the adaptive threshold, to obtain a sparse connection matrix.

[0078] In one embodiment, the functional connection matrix for each time window Adaptive thresholding is performed. A threshold determination method based on matrix data distribution (such as the OTSU algorithm or the Gini impurity algorithm) is used to automatically find the optimal threshold Th. Connections with PLV values ​​greater than Th are retained, while connections with PLV values ​​less than or equal to Th are removed, thereby transforming the dense matrix into a sparse matrix, achieving denoising and dimensionality reduction.

[0079] S5. Construct a dynamic graph sequence based on the sparsed connection matrix of each time window.

[0080] In one embodiment, a graph data structure is constructed for each sparsified time window. .

[0081] Node set : Fixed for all EEG electrode channels, representing different brain regions.

[0082] Edge set It consists of all the connections retained in step S4. Each edge contains the two node indices of the connection (edge_index) and the PLV value of the connection as the edge weight (edge_weight).

[0083] Node feature matrix For each node Calculate its weighted degree centrality as the initial feature. The formula is as follows:

[0084] in For nodes The set of neighboring nodes in the sparse connection matrix. This feature is directly derived from the current network topology, has a clear physical meaning, and characterizes the local importance of the node (brain region) in the network at the current time step.

[0085] S6. Output the dynamic graph sequence, which is used as input to the dynamic graph neural network for classifying states of consciousness impairment.

[0086] The final output is a sequence of dynamic graphs arranged in chronological order. This sequence fully characterizes the evolution of brain functional networks over time during the recording period and can be directly used as input for models such as Dynamic Graph Neural Networks (DGCN) for the classification of states of consciousness impairment.

[0087] The solutions of the embodiments of this application will be described in detail and explained below with reference to specific application examples.

[0088] This embodiment provides a dynamic brain functional network adaptive construction method for consciousness disorder analysis, including the following steps: Step 1: Input and frequency band extraction.

[0089] The input is a preprocessed duration. Multi-channel (C-channel, e.g., C=62) resting-state EEG data (e.g., 60 seconds). A 4th-order Butterworth bandpass filter is used to extract the Theta (4-8Hz) and Alpha (8-13Hz) frequency bands, which are closely related to the level of consciousness. Preferably, these two frequency bands can be processed separately or combined into a single band (e.g., 4-13 Hz) for processing.

[0090] Step 2: Sliding window splitting.

[0091] A sliding window technique is used to perform time-series segmentation on the filtered continuous signal. This is key to achieving "dynamic" analysis. Specifically: a) Window length win_len: Set to 2 seconds. This length is sufficient to calculate a stable PLV while also capturing the rapid dynamic changes in the brain network.

[0092] b) Step size win_step: Set to 2 seconds (i.e., no overlap). This ensures that the signal within each time window is independent, balancing dynamic resolution with computational efficiency.

[0093] c) For a 60-second dataset, we can obtain... A time window.

[0094] Step 3: Dynamic link calculation.

[0095] For each time window t (t = 1 to T), perform the following operations: Step 3.1: Extract the timing signals of all C channels within the window.

[0096] Step 3.2: For each pair of channels (i,j), calculate its phase-locked value (PLV) as follows: a) First, perform Hilbert transform on the signals of both channels respectively:

[0097] To obtain their analytic signals and instantaneous phases and .

[0098] b) Then, calculate the instantaneous phase difference: .

[0099] c) Finally, calculate the PLV value: , where N is the number of sampling points in the window, and j is the imaginary unit.

[0100] Step 3.3: Combine the PLV values ​​of all channel pairs to form a C*C symmetric functional connectivity matrix. Thus, we have obtained a PLV matrix sequence. .

[0101] Step 4: Network sparsification: Otsu's method or other threshold finding methods.

[0102] Matrix for each time window Adaptive thresholding is performed to achieve noise reduction and capacity reduction. The specific steps are as follows: Step 4.1: Threshold Determination: An optimal threshold for distinguishing foreground from background (i.e., effective connections from noise) in the PLV matrix is ​​determined using an adaptive thresholding method. Preferred adaptive thresholding methods include the OTSU algorithm based on the PLV matrix and the Gini coefficient. Experiments show that the adaptive threshold Th achieves the best balance between preserving significant functional connections and effectively removing noise.

[0103] The specific operations of the OTSU algorithm based on the PLV matrix and the Gini impurity differentiation are as follows: a) OTSU algorithm based on PLV matrix: Let the threshold be k, and the probability that an edge connection is classified as a valid connection is... The average connection strength assigned as an effective connection is Similarly, the probability that an edge connection is classified as noise is... The average connection strength assigned to noise is There is also the cumulative mean m of the threshold k, and the average connection strength of the entire PLV matrix. Then the adaptive threshold That is, the adaptive threshold is the threshold that maximizes the inter-class variance.

[0104] b) Gini impurity algorithm based on PLV matrix: The definition is the same as above, then the adaptive threshold is... That is, the adaptive threshold is the optimal threshold that minimizes Gini impurity.

[0105] Step 4.2: Sparsification operation: Traversal Of all the elements, only those are retained. The connection of the sparse matrix transforms the dense matrix into a sparse representation.

[0106] Step 5: Dynamic graph sequence construction.

[0107] For each sparsified time window t, construct a graph data structure. : 1) Node set V: fixed as all C EEG electrode channels, .

[0108] 2) Edge set It consists of all the connections retained in step 4. Specifically, it is represented as: 2.1) Edge index, dimension is Each column represents the source and target node indexes of an edge.

[0109] 2.2) Edge weights, with dimension 1 The value is the PLV value of the corresponding connection.

[0110] 3) Node characteristics This embodiment innovatively uses each node Calculate its "weighted degree centrality" as the initial feature. The calculation formula is:

[0111] in For nodes The set of neighboring nodes after thresholding.

[0112] Beneficial effects: This feature originates directly from the graph structure of the current time window itself, has a clear physical meaning (characterizing the local importance of the brain region in the network at the current moment), and provides the model with powerful prior information that is directly related to the connection pattern.

[0113] Ultimately, the eigenvalues ​​of each node form the node feature matrix. The dimension is [C,1].

[0114] Step 6: Output sequence.

[0115] A dynamic graph sequence is constructed using 30 time windows. This sequence fully characterizes the sequence in... The dynamic evolution of brain functional networks over time can be directly used as input for the Dynamic Graph Neural Network (DGCN) model.

[0116] In summary, the method of this embodiment has the following advantages and beneficial effects compared with the prior art: 1) It preserves key dynamic information: By generating graph sequences through sliding windows, the time-varying characteristics of brain functional connectivity are fully preserved, providing a much richer source of information for distinguishing states of consciousness than static methods.

[0117] 2) Effective denoising and structural highlighting: Personalized sparsification is applied to different samples using an adaptive sparsification strategy, effectively filtering noisy connections and highlighting strong functional connections that contribute more to classification. Simultaneously, the individualized brain differences among patients with consciousness disorders behind the different samples are taken into account, enabling subsequent models to focus more on learning key patterns and discovering disease commonalities.

[0118] 3) Preserve edge weight information and be compatible with two types of graph neural network models: By constructing graph data with edge weights to preserve the original edge weights to the maximum extent, the upper limit of the model is improved while being compatible with both graph neural networks with edge weights and graph neural networks without edge weights.

[0119] 4) Synergy between node features and graph structure: Using degree centrality as a node feature is a self-reinforcing representation method. It integrates the connection strength information of the graph into the node features, enabling the GNN to obtain a strong signal of the importance of each node at the beginning of message transmission, which greatly improves the model's feature learning efficiency and effect.

[0120] 5) Paving the way for high-performance dynamic models: The dynamic graph sequences generated by this method are a prerequisite for running advanced dynamic graph neural networks (such as DGCN), thereby unlocking higher classification performance (such as increasing the accuracy from 62.9% of static GCN to 76.7% of DGCN).

[0121] It should be noted that the above methods are mainly used for the analysis of consciousness disorders by EEG signals, but they are not limited to this. They can also be applied to other physiological signal analysis and disease diagnosis, such as: ECG / EEG multimodal analysis, EMG motor function assessment, EMG neuromuscular disease diagnosis, and EOG sleep staging.

[0122] Alternatively, the change process: a) Data: Replace the input data with multi-lead ECG signals (node ​​= body surface electrode), multi-channel EMG signals (node ​​= muscle recording point), etc.

[0123] b) Connectivity metrics: PLV is suitable for oscillating signals. For other signals, metrics such as Pearson correlation coefficient, mutual information, and Granger causality can be used to calculate the dynamic connectivity between channels.

[0124] c) Tasks: Used for diagnosing arrhythmias (based on dynamic electrocardiogram networks), assessing motor unit synchrony, identifying sleep stages, etc.

[0125] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0126] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0127] Please see Figure 2 , Figure 2 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 201 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 202 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 202 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 202 and is called and executed by the processor 201 using the methods described in the embodiments of this application. Input / output interface 203 is used to implement information input and output; The communication interface 204 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 205 transmits information between various components of the device (e.g., processor 201, memory 202, input / output interface 203, and communication interface 204); The processor 201, memory 202, input / output interface 203 and communication interface 204 are connected to each other within the device via bus 205.

[0128] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0129] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0130] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0131] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0132] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0133] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0134] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0135] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0136] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0137] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0138] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0139] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0140] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0141] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0142] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0143] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for adaptive construction of dynamic functional networks for physiological signal analysis, characterized in that, The method includes the following steps: Input multi-channel physiological signals and extract signals from preset frequency bands; The signal in the preset frequency band is time-series segmented using a sliding window to obtain multiple time windows; For each time window, calculate the phase lock value (PLV) between all pairs of channels to obtain the functional connection matrix for that time window; Adaptive threshold sparsification is applied to the functional connection matrix of each time window, retaining connections with PLV greater than the adaptive threshold, to obtain a sparse connection matrix. A dynamic graph sequence is constructed based on the sparsed connection matrix of each time window; Output the dynamic graph sequence for subsequent classification or regression tasks; The physiological signals include at least one of electroencephalography (EEG), electrocardiography (ECG), electromyography (EMG), and electrooculography (EOG).

2. The method according to claim 1, characterized in that, When the physiological signal is an electroencephalogram (EEG), each dynamic graph constructed... Includes node set Edge set and node feature matrix ; Node set Corresponding EEG channel; Edge set It consists of connections retained after sparsification, and the edge weight is the PLV value of the corresponding connection. Node feature matrix Each node in the matrix is ​​characterized by its weighted degree centrality, which is calculated based on the sparsed connection matrix.

3. The method according to claim 2, characterized in that, The preset frequency band includes at least one of the Theta band and the Alpha band.

4. The method according to claim 1, characterized in that, The adaptive threshold sparsity processing uses the OTSU algorithm based on the PLV matrix or the Gini impurity algorithm to determine the threshold.

5. The method according to claim 4, characterized in that, The OTSU algorithm based on the PLV matrix includes: For candidate thresholds The connections are divided into foreground and background classes, and the variance between classes is calculated. , choose to The largest As an adaptive threshold.

6. The method according to claim 2, characterized in that, node Weighted degree centrality The calculation formula is: in, For nodes The set of neighboring nodes in the sparse connectivity matrix; For nodes with neighboring nodes The weight of the edges between them.

7. The method according to claim 2, characterized in that, The edge set include: Edge index matrix, dimension 1 , representing the source node and target node of each edge; Edge weight vector, dimension , representing the PLV value corresponding to each edge.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images