A method and system for identifying key nodes in dynamic epileptogenic brain networks based on SEEG

Abstract

This invention discloses a dynamic epileptogenic brain network key node identification system based on SEEG, belonging to the field of brain network technology. The key technical points are: dynamically segmenting SEEG signals using a sliding time window to obtain signal data across multiple time windows; determining the neural vulnerability index of each node based on the signal data; constructing dynamic functional connectivity networks and causal networks based on the multi-band time-frequency characteristics of the signal data, and determining the global efficiency index and node centrality index of each network; and identifying cross-level key nodes that play a crucial role in the epileptic seizure process. This invention can clearly reveal the temporal evolution of the connection strength, topology, and key node roles of the pathogenic network in different stages such as the pre-seizure, onset, duration, and termination phases during an epileptic seizure, thereby accurately locating the origin and propagation path of epileptic activity.

Description

Technical Field

[0001] This invention relates to the field of brain network technology, and more specifically, to a method and system for identifying key nodes in dynamic epileptogenic brain networks based on SEEG. Background Technology

[0002] For patients with focal epilepsy, surgical resection of the epileptogenic zone is an effective means of achieving a complete cure, but its success depends heavily on the accuracy of the epileptogenic zone's localization. Stereotactic electroencephalography (SEEG), a high-precision neurophysiological signal obtained by directly recording neuronal electrical activity through the implantation of deep electrodes into specific areas of the brain via minimally invasive surgery, has become the "gold standard" for preoperative evaluation of refractory epilepsy.

[0003] Existing technologies document various automated analysis methods based on signal characteristics. For example, methods based on static global brain networks generate a standardized static global brain network, providing a unified template for cross-subject analysis and addressing the issue of electrode implantation differences among patients. However, these methods generate a "static snapshot," failing to capture the millisecond-level dynamic evolution of the pathogenic network's connection strength and topology during epileptic seizures, and thus failing to reveal the origin, propagation path, and temporal role changes of key nodes in epilepsy. Furthermore, existing technologies also document dynamic network models based on node vulnerability. These models calculate the minimum perturbation required for each node to become unstable, placing electrodes within the network interactions to assess their epileptogenicity. While this method considers the dynamic attributes of nodes within the network, its analysis remains limited to the node's own "vulnerability," completely ignoring the dynamic changes in the overall network topology (such as community stability and global efficiency) and their interaction with node attributes. It cannot identify which nodes are bridge nodes leading to the disintegration of the network's modular structure, nor can it distinguish between the driver and follower roles of nodes, resulting in an incomplete characterization of the pathogenic network's organizational patterns.

[0004] The accuracy, interpretability, and direct guidance value for clinical surgery of the aforementioned automated analysis methods are severely limited. Therefore, how to research and design a SEEG-based method and system for identifying key nodes in dynamic epileptogenic brain networks that can overcome these shortcomings is an urgent problem we need to solve. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for identifying key nodes in a dynamic epileptogenic brain network based on SEEG. By employing an adaptive sliding time window technique to dynamically segment SEEG signals throughout the entire time sequence, this method overcomes the inherent limitation of static network analysis methods, which can only provide a "static snapshot." This method can clearly reveal the temporal evolution of the connection strength, topology, and key node roles of the pathogenic network during different stages of an epileptic seizure, including the pre-ictal, onset, duration, and termination phases, thereby accurately locating the origin and propagation path of epileptic activity.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: Firstly, a method for identifying key nodes in a dynamic epileptogenic brain network based on SEEG is provided, including the following steps: The SEEG signal is dynamically segmented using a sliding time window to obtain signal data with multiple time windows; Based on the signal data, the neural vulnerability index of each node is determined; Based on the multi-band time-frequency characteristics in the signal data, dynamic functional connectivity networks and causal networks are constructed respectively, and the global efficiency index and node centrality index of each network are determined. Based on the aforementioned neural vulnerability index, global efficiency index, and node centrality index, key nodes across layers are extracted by fusing node attributes with network topology features.

[0007] Furthermore, the dynamic segmentation includes: An adaptive sliding time window is used to segment the SEEG signal, and the window parameters are dynamically adjusted according to the signal sampling frequency and / or the characteristics of epileptic seizures. The signal data is obtained by standardizing the signal within each time window to eliminate baseline drift and noise interference.

[0008] Furthermore, the multi-band time-frequency features are extracted using wavelet transform, including: The signal data is decomposed into multiple frequency bands using discrete wavelet transform; Calculate the relative power, power spectral entropy, and band energy ratio characteristics for each frequency band; Principal component analysis is used for feature dimensionality reduction, and principal components with a contribution rate greater than or equal to a first threshold are retained to obtain the multi-band time-frequency features.

[0009] Furthermore, the determination of the neural vulnerability index includes: Constructing a dynamic network model based on the state transition matrix; The Lyapunov exponent was used to analyze the system stability of the dynamic network model at different time windows, and the effective window signals in the critical state were screened out. The neural vulnerability index of each node in the effective window signal is calculated by solving the minimum perturbation problem.

[0010] Furthermore, the construction of the dynamic functional connectivity network includes: The functional connection strength between nodes is calculated using the phase-locked value, resulting in the functional connection matrix; Based on the aforementioned functional connectivity matrix, a weighted undirected network is constructed using graph theory, and the global efficiency index at each time point is determined, thus forming a dynamic functional connectivity network.

[0011] Furthermore, the construction of the causal network includes: A directed connection network was established using time-varying Granger causality analysis; Based on the directed connection network, calculate the causal outflow intensity and inflow intensity of the nodes; Based on the causal outflow and inflow intensities, a causal network is constructed by analyzing the direction of network information flow using directed graph theory indices.

[0012] Furthermore, based on the neural vulnerability index, global efficiency index, and node centrality index, key nodes across layers are extracted by fusing node attributes and network topology features, including: Based on the neural vulnerability index and the global efficiency index of the dynamic functional connection network, the functional coupling index is determined by quantifying the synergistic evolution relationship between node vulnerability and the overall synchronous diffusion efficiency of the functional network. Based on the neural vulnerability index and the global efficiency index of the causal network, the causal driving index is determined by quantifying the co-evolutionary relationship between node vulnerability and the information flow driving efficiency of the causal network. The functional coupling index and the causal driving index are fused to obtain the comprehensive dynamic coupling strength; The node centrality indices of the dynamic functional connection network and the causal network are introduced as independent structural evidence, and the cross-level key nodes are identified in combination with the comprehensive dynamic coupling strength.

[0013] Furthermore, the determination of the functional coupling index includes: Obtain time-series data of the neural vulnerability index for each node within each time window; Based on the dynamic functional connection network, global efficiency time series data corresponding to each time window are extracted; Calculate the dynamic Pearson correlation coefficient between the time series of the neural vulnerability index of each node and the time series of the global efficiency; For each node, calculate the absolute value of the maximum correlation coefficient between its neural vulnerability index and global efficiency within a preset time window before the epileptic seizure. After normalizing the absolute value of the maximum correlation coefficient, it is defined as the functional coupling index of the node.

[0014] Furthermore, the determination of the causal driving index includes: Obtain time-series data of the neural vulnerability index for each node within each time window; Based on the causal network, extract the causal global efficiency time series data corresponding to each time window; For each node, calculate the time-delay cross-correlation function between its neural vulnerability index time series and the causal global efficiency time series; Determine the time delay parameter that maximizes the time delay cross-correlation function; When the time delay parameter is less than or equal to zero, the maximum time delay cross-correlation function value is multiplied by a preset driving direction weighting coefficient to obtain a preliminary causal driving value. The preliminary causal driving value is normalized and multiplied by the average change gradient within a preset time window before the epileptic seizure to obtain the final causal driving index.

[0015] Secondly, a SEEG-based dynamic epileptogenic brain network key node identification system is provided. This system is used to implement the SEEG-based dynamic epileptogenic brain network key node identification method as described in any one of the first aspects, including: The signal segmentation module is used to dynamically segment the SEEG signal using a sliding time window to obtain signal data for multiple time windows. The node analysis module is used to determine the neural vulnerability index of each node based on the signal data. The network analysis module is used to construct dynamic functional connectivity networks and causal networks based on the multi-band time-frequency characteristics in the signal data, and to determine the global efficiency index and node centrality index of each network. The node identification module is used to extract key nodes across layers by fusing node attributes and network topology features based on the neural vulnerability index, global efficiency index and node centrality index.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention overcomes the inherent limitation of static network analysis methods, which can only provide "static snapshots," by employing an adaptive sliding time window technique to perform full-time dynamic segmentation of SEEG signals. This method can clearly reveal the temporal evolution of the connection strength, topology, and key node roles of the pathogenic network during different stages of epileptic seizures, such as the pre-seizure, onset, duration, and termination phases, thereby accurately locating the origin and propagation path of epileptic activity; 2. This invention, by quantifying the synergistic evolution relationship between the node's neural vulnerability index and the global efficiency indicators of functional and causal networks, for the first time dynamically correlates the node's own dynamic attributes with the network topology. This enables the invention to effectively distinguish between the "driver" nodes that play a dominant role in epileptic networks and the "bridge nodes" that cause the abnormal synchronization to spread, thus solving the difficulty of existing methods in distinguishing node roles. 3. This invention constructs a multi-criteria decision-making model, which weights and integrates dynamic coupling strength with multiple node centrality indices for comprehensive judgment. Cross-validation from multiple complementary dimensions of dynamics and topology can reduce the risk of misjudgment, make the identification results more reliable, and provide more sufficient evidence to support clinical surgical planning. 4. The key parameters such as the time window and feature dimensionality reduction threshold in this invention can be dynamically adjusted according to the signal characteristics and epileptic seizure patterns. It adopts a data-driven strategy and can adapt to SEEG data from patients with different sampling frequencies and different types of epilepsy, thus having good versatility and clinical application prospects. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is the overall flowchart of Embodiment 1 of the present invention; Figure 2 This is a flowchart of signal segmentation in Embodiment 1 of the present invention; Figure 3 This is a flowchart illustrating the calculation of the neural vulnerability index in Embodiment 1 of the present invention; Figure 4 This is a flowchart of the construction of two networks in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the results of the neural vulnerability index in Embodiment 1 of the present invention; Figure 6 This is a system block diagram in Embodiment 3 of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are only for explaining this invention and are not intended to limit this invention.

[0019] Example 1: A key node identification system for dynamic epileptogenic brain networks based on SEEG, such as... Figure 1 As shown, it includes the following steps: S1: The SEEG signal is dynamically segmented using a sliding time window to obtain signal data with multiple time windows; S2: Determine the neural vulnerability index of each node based on signal data; S3: Based on the multi-band time-frequency characteristics in the signal data, construct dynamic functional connectivity networks and causal networks respectively, and determine the global efficiency index and node centrality index of each network. S4: Based on the neural vulnerability index, global efficiency index and node centrality index, this study integrates node attributes with network topology features to identify cross-level key nodes that play a crucial role in the epileptic seizure process.

[0020] This invention utilizes a sliding time window to segment the SEEG signal, with adaptive adjustment capabilities for the window parameters. The segmented processing allows for analysis not across the entire signal segment, but rather across dozens or even hundreds of consecutive time segments. For each time window, a set of indicators is calculated, such as neural vulnerability index, global efficiency, and node centrality. Arranging these indicators in chronological order yields the time-series data for each indicator. Analyzing this time-series data provides a clear understanding of the evolution of the attributes of each node and the entire network over time during the pre-seizure, seizure onset, seizure duration, and seizure termination phases. By identifying which node's vulnerability increases first, which network's information flow efficiency changes first, and the propagation patterns of these changes in time and space, the origin and propagation paths of epilepsy can be inferred, thus completely overcoming the limitations of static snapshots.

[0021] Furthermore, this invention integrates node and network processing, not considering nodes in isolation, but evaluating them within their respective network environments. It constructs an undirected dynamic functional connection network that reflects synchronization relationships, and a directed causal network that reflects information flow, characterizing the network topology from the two dimensions of "cooperation" and "driving force," respectively.

[0022] Based on two key indicators, the functional coupling index and the causal driving index, node attributes and network topology attributes are dynamically correlated. The functional coupling index quantifies the synergistic relationship between node vulnerability and the overall network synchronization efficiency. If the increase in the vulnerability of a node shows a high correlation (high correlation coefficient) with the sharp change in the synchronization efficiency of the entire network, it indicates that the node is a hub or "bridge node" connecting various parts of the network. Its instability will cause a drastic change in the overall network communication efficiency, i.e., the disintegration of the modular structure.

[0023] Subsequently, the time-delay cross-correlation between node vulnerability and the global efficiency of the causal network is calculated, and the direction of the time delay is determined. If the change in vulnerability of a node precedes the change in the efficiency of the entire causal network (time delay parameter ≤ 0), it indicates that the instability of that node has triggered a change in the subsequent information flow of the entire network, and this node is the "driver". Conversely, if the change in vulnerability of a node lags behind the network, it is more likely to be a "follower" or a passively affected node.

[0024] Finally, by integrating the functional coupling index, the causal driving index, and the node centrality index (the node centrality index is a static indicator that measures the structural importance of a node in the network), the key nodes identified are truly "cross-level key nodes" that have a core position in both dynamics and topology.

[0025] In step S1, as Figure 2 As shown, the sliding time window technique divides a continuous SEEG signal into multiple small time segments according to a certain time length. Each time segment is an independent time window, and the signal data within the window can be regarded as a relatively stable segment for analysis.

[0026] In this embodiment, the window slides sequentially on the SEEG signal according to a set step size. The step size can also be dynamically adjusted according to the actual situation, generally 10%-25% of the window length, to ensure that there is a certain overlap between windows, thereby reducing information omission and more comprehensively analyzing signal characteristics.

[0027] Because SEEG signals at different sampling frequencies contain different frequency components and have varying time resolutions, high sampling frequencies of 2000Hz and above can capture faster and more detailed changes in neural electrical activity. Shorter windows, such as 50-100ms, can be used to ensure sufficient time resolution for analyzing transient features like high-frequency oscillations. Conversely, lower sampling frequencies below 500Hz are better suited to longer windows, such as 300-500ms, to ensure sufficient frequency resolution for low-frequency components. Therefore, window parameters can be dynamically adjusted based on the signal sampling frequency. For sampling frequencies between 500Hz and 2000Hz, a window of 100-300ms can be used.

[0028] Within each time window, the segmented signals are standardized to eliminate baseline drift and noise interference. Standardization typically employs the Z-score standardization method, converting the signal data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Standardization effectively eliminates baseline drift and noise interference, making the signal data more stable and reliable. For example, baseline drift can easily lead to biases in the calculation of features such as the neural vulnerability index, and noise interference can easily mask the true node attributes and network topology features, thus affecting the identification results of key nodes.

[0029] In step S2, as Figure 3 As shown, the determination of the neural vulnerability index includes: constructing a dynamic network model based on the state transition matrix; using the Lyapunov index to analyze the system stability of the dynamic network model in different time windows and screening out the effective window signals in the critical state; and calculating the neural vulnerability index of each node in the effective window signals by solving the minimum perturbation problem.

[0030] Specifically, each node corresponds to a brain region, and each node receives signals through corresponding electrodes. The neurovulnerability index treats a brain region as a dynamic system and calculates the magnitude of the disturbance required to transition it from a stable state (normal discharge) to an unstable state (epilepsy-like discharge). The lower the vulnerability index, the more easily the node becomes out of control, and the stronger its epileptogenic potential may be.

[0031] The SEEG signal is a voltage time series and cannot be used directly. It needs to be reconstructed from a one-dimensional time series into a multi-dimensional state space. Through standard processing of nonlinear dynamics, the system dynamics hidden behind the time series can be revealed.

[0032] First, extract the time delay from the SEEG signal. and embedding dimension For a SEEG channel signal A point in its state space can be represented as: At each point in time The dynamic state of a system is a vector defined by the signal values ​​at the current time and several subsequent time points. This vector constitutes a point in the state space. Connecting the state points at all time points forms the trajectory of the system in the state space.

[0033] Next, each state point is found in the state space. nearest neighbor The state transition can be approximated using a locally linear approach: , For a description in time The local state transition matrix of how the nearby system evolves, matrix It includes the local dynamics of the node within the current time window.

[0034] The Lyapunov exponent (LE) is a classic indicator for measuring the sensitivity of a dynamical system to initial conditions, essentially a quantification of the "butterfly effect." The sign of the Lyapunov exponent is crucial: when... The system is stable and the trajectory converges; when The system is in a critical state (edge ​​state); when The system is chaotic, with divergent trajectories. Since brain networks in a critical state are most susceptible to small perturbations, potentially transitioning to a seizure state, calculating vulnerability is most meaningful at this point; therefore, [the system is chosen]. The analysis is performed within a time window close to zero. The calculation is based on the state space trajectory and the state transition matrix, and is estimated by analyzing the divergence or convergence rate of adjacent points on the trajectory over time.

[0035] Within the selected critical window, the nonlinear dynamic system is linearized and approximated near the equilibrium point. A perturbation input is defined. Add it to the system dynamics equations. Find the disturbance input. The minimum energy, such as minimizing This also satisfies a constraint: the perturbation is sufficient to push the system state from its current stable equilibrium point to the boundary of the stable region. The objective optimization problem is transformed into a linear quadratic regulator problem, or it can be solved by solving the relevant algebraic Riccati equations to find the minimum perturbation energy. The minimum perturbation energy value is the neural vulnerability index of that node in the current time window. The smaller the value, the more vulnerable the node.

[0036] In step S3, as Figure 4 As shown, to capture the neural information carried by oscillations at different frequencies, it is necessary to first perform time-frequency analysis on the SEEG signal to obtain multi-band time-frequency features for constructing subsequent directed and undirected networks. For example, multi-band time-frequency features can be extracted using wavelet transform, including: decomposing the signal data into multiple frequency bands using discrete wavelet transform; calculating the relative power, power spectral entropy, and band energy ratio characteristics of each frequency band; and performing feature dimensionality reduction using principal component analysis, retaining principal components with a contribution rate greater than or equal to a first threshold to obtain the multi-band time-frequency features. In some optional examples, the first threshold can be adaptively determined based on the inflection point of the cumulative contribution rate curve.

[0037] Specifically, the SEEG signal (voltage time series) of each channel is decomposed into a series of predefined frequency bands, such as δ, θ, α, β, and γ bands. Then, wavelet coefficients representing the intensity of different frequency components are extracted from each frequency band.

[0038] To further reduce noise and condense information, statistical characteristics of each frequency band are calculated based on wavelet coefficients. These characteristics include, but are not limited to, relative power, power spectral entropy, and band energy ratio. Relative power refers to the percentage of energy in that frequency band relative to the total energy of the entire band, reflecting the dominance of oscillations in that band. Power spectral entropy measures the degree of disorder in the power spectrum; a low entropy value indicates that energy is concentrated in a few frequency bands, indicating more ordered activity; a high entropy value indicates that energy is evenly distributed, indicating more chaotic activity. Band energy ratio characterizes the energy ratio between different frequency bands, reflecting the balance between frequency bands, such as γ / δ.

[0039] After calculating the statistical features across multiple frequency bands, the resulting statistical features have high dimensionality and contain redundant information. Therefore, principal component analysis (PCA) is used to select the top principal components with the highest contribution rates from the high-dimensional statistical feature set. These principal components are linear combinations of the original multi-frequency band features, representing the core dynamic change patterns of the SEEG signal in the multi-frequency domain. It should be noted that the number of principal components selected can be flexibly designed according to requirements and is not limited here.

[0040] The core of functional connectivity is measuring synchronization, especially phase synchronization. To meet dynamic requirements, this invention constructs a dynamic functional connectivity network based on instantaneous phase. The construction of the dynamic functional connectivity network includes: calculating the functional connectivity strength between nodes using phase-locked values ​​to obtain a functional connectivity matrix; based on the functional connectivity matrix, constructing a weighted undirected network using graph theory, and determining the global efficiency index at each time point to form the dynamic functional connectivity network.

[0041] Specifically, based on the relative power of a single frequency band in each time window, a relative power time-series curve can be obtained for each frequency band. For each frequency band, a significant difference is compared between its relative power during the seizure and before the seizure. Considering that the data may not follow a normal distribution, a paired t-test or a non-parametric Wilcoxon signed-rank test is used. Furthermore, if the p-value is insufficient to represent significance, the magnitude of the difference can be considered, and the effect size can be calculated, such as Cohen's d. A larger effect size indicates a more dramatic increase in power during the seizure and a stronger association with epileptic activity.

[0042] The Hilbert transform is performed on the frequency band with the greatest significant difference to calculate the instantaneous phase value of the signal at each time point. For a pair of nodes i and j, the phase-locked value (PLV) within the time window is calculated using the following formula: .

[0043] A symmetric matrix can be obtained by calculating the phase lock-in value of the signals between nodes. Matrix elements This represents the degree of synchronization between brain regions i and j in a specific frequency band of neural oscillations. It is an undirected, weighted matrix. Synchronization level is also functional connectivity strength, which is not distance itself. To calculate the shortest path, it needs to be converted to distance. Generally, the reciprocal of the functional connectivity strength or the result of subtracting the functional connectivity strength from 1 can be used as the distance. For example, if the functional connectivity strength between nodes i and j is 0.9, then the distance between them is short, such as 1 / 0.9 ≈ 1.11; if the functional connectivity strength is 0.1, then the distance is long, such as 1 / 0.1 = 10.

[0044] Global efficiency is a metric that measures the overall efficiency of network information transmission. It is calculated as the average transmission efficiency between all pairs of nodes in the network, and the transmission efficiency between a pair of nodes is defined as the reciprocal of the shortest path length between them. A symmetric matrix (functional connectivity matrix) can be transformed into a distance matrix. Then, a graph theory algorithm is used to calculate the shortest path length between any two nodes on the distance matrix, constructing a dynamic functional connectivity network. Dijkstra's algorithm can be used as the graph theory algorithm. Finally, the global efficiency metric of the dynamic functional connectivity network is calculated by calculating the efficiency between all pairs of nodes.

[0045] The core of causal networks is measuring directional influence, that is, whether the past value of one signal can predict the future value of another signal. The construction of a causal network includes: establishing a directed network using time-varying Granger causality analysis; calculating the causal outflow and inflow strengths of nodes based on the directed network; and analyzing the direction of network information flow using directed graph theory indicators based on the causal outflow and inflow strengths to construct the causal network.

[0046] Specifically, the dimensionality-reduced principal components are directly used as the input signal for Granger causality analysis. Through dimensionality reduction and noise reduction, the focus is on the frequency bands most relevant to epilepsy. The principal component time series are low-dimensional feature sequences obtained by multi-band time-frequency feature extraction and principal component analysis of the SEEG signal, with each principal component representing a neurophysiological activity pattern. Stationarity tests are performed on the principal component time series within each sliding time window. The AFD test is used to verify the stationarity of each principal component sequence. For sequences that do not meet the stationarity requirement, first-order differencing is performed until the stationarity test is passed, ensuring the effectiveness of subsequent vector autoregressive modeling.

[0047] For each time window, the optimal lag order is determined using the information criterion method after stationarization. Specifically, Bayesian information criterion values ​​from 1 to 15 are calculated, and the lag order that minimizes the Bayesian information criterion value is selected as the optimal order of the vector autoregressive model for that time window. This step ensures that the model can fully capture the dynamic relationships between the principal components.

[0048] Within each given time window, the optimal lag order is first determined based on information criteria (such as AIC or BIC), and then a multivariate vector autoregressive (VAR) model is constructed. The model construction process includes the following steps: First, multiple principal components extracted through principal component analysis are used as endogenous variables in the model; next, the ADF unit root test is used to ensure that all variables meet the stationarity requirement, and differencing is performed if non-stationary sequences exist; then, the VAR model parameters are estimated using the least squares (OLS) method to obtain the coefficient matrix of the lagged terms of each variable; finally, the model fit is verified using residual autocorrelation tests (such as the LM test) and normality tests. Next, for each pair of principal components, Granger causality tests are performed sequentially. Specifically, for any two principal components, such as principal component a and principal component b, it is tested whether the lagged terms of principal component a have a significant predictive power for the future values ​​of principal component b. This test is implemented using the F-test method, calculating the corresponding F-statistic, and determining the significance level of Granger causality based on its value, while recording all test results for subsequent analysis.

[0049] After performing Granger causality tests on all principal component pairs, the p-values ​​were corrected using a false discovery rate (FCR) correction method. A significance level threshold of 0.05 was set, retaining only significant causal relationships corrected for the FCR. The directed connection strengths of significant Granger causal relationships were quantified into corresponding F-statistics, and a directed weighted connection matrix was constructed.

[0050] Based on the directed weighted connection matrix for each time window, a corresponding directed connection network is constructed. In this network, network nodes correspond to principal components, directed edges represent significant Granger causal relationships between principal components, edge direction indicates the causal flow, and edge weight represents the strength of the causal effect. By traversing the entire SEEG recording period through a sliding time window, a series of directed connection networks arranged in chronological order are obtained, forming a dynamic directed connection network sequence. It should be noted that while the network construction is based on principal components to reduce dimensionality, the time series data corresponding to each principal component must establish a clear and accurate correlation with the original electrode channels. The final key node indicators need to be calculated and ranked for each original electrode channel. The specific correlation is as follows: First, using the loading matrix obtained from principal component analysis, the weight coefficients of the original electrode channels corresponding to each principal component are extracted to establish a linear combination relationship between the principal components and the electrode channels. Next, the variance contribution rate of each original electrode channel to the principal components is calculated, and the channel contribution distribution is visualized through heatmaps or bar charts. Then, the dynamic changes of the principal component time series are analyzed synchronously with the spatiotemporal discharge modes of the original electrode channels, and multimodal data fusion is achieved by combining the intracranial spatial coordinates of the electrodes. Finally, the principal component signals are back-mapped to the original electrode space using a deconvolution algorithm, and the accuracy of the correlation is verified through residual analysis.

[0051] For each time window, a directed connection network is constructed, and topological metrics of the network nodes are calculated. These include calculating the causal outflow strength of each node (the sum of connection weights from that node to all other nodes), the causal inflow strength of each node (the sum of connection weights from all other nodes to that node), and the net causal flow strength of each node (the difference between outflow and inflow strengths). These metrics collectively characterize the causal driving role of each principal component node in the epilepsy network.

[0052] The calculation principle of the global efficiency index for causal networks is similar to that for dynamic functional connectivity networks. However, in causal networks, the communication efficiency is zero for node pairs that do not have a directed path.

[0053] The centrality of network nodes is comprehensively evaluated from multiple complementary dimensions. For dynamic functional connectivity networks, the following combination of indicators—weighted degree centrality, betweenness centrality, and eigenvector centrality—is preferred. Among them, weighted degree centrality reflects the local connectivity strength of nodes in functional synchronization; betweenness centrality can identify information transmission hub nodes in the functional network; and eigenvector centrality can assess the global influence of nodes in the functional network.

[0054] Specifically, based on the constructed dynamic functional connectivity matrix, the weighted degree centrality index of each node is calculated. For the functional connectivity network within each time window, the weighted adjacency matrix of that window is first extracted. The element values ​​of this matrix represent the phase lock values ​​between corresponding node pairs. The weighted degree centrality of a node is defined as the sum of the weights of all edges connected to that node, specifically calculated as the cumulative sum of the element values ​​of all elements in the corresponding row or column of that node.

[0055] Furthermore, this invention employs a weighted shortest path algorithm to calculate the betweenness centrality of nodes. First, the connection weights in the functional connectivity matrix are converted into distance metrics using the reciprocal of the weights or a negative logarithm transformation. Then, the Floyd-Warshall algorithm or Dijkstra's algorithm is used to calculate the shortest paths between all pairs of nodes in the network. The betweenness centrality of a node is defined as the proportion of all shortest paths passing through that node to the total number of paths, specifically calculated by counting the frequency of shortest paths between all pairs of nodes passing through that node. Betweenness centrality reflects the bridging role of a node in network information transmission.

[0056] Furthermore, this invention employs a power-law iteration method to calculate the eigenvector centrality of functionally connected networks. First, a weighted adjacency matrix of the network is constructed. Then, the principal eigenvectors of the matrix are calculated iteratively; each component of this eigenvector represents the eigenvector centrality value of the corresponding node. Normalization is performed during the iteration process to ensure numerical stability. Iteration stops when the difference between two consecutive iterations is less than a preset threshold. Eigenvector centrality reflects the importance of a node's adjacency to highly connected nodes.

[0057] For causal networks, the following combination of metrics is preferred: out-degree centrality, in-degree centrality, directed betweenness centrality, and hub value centrality. Out-degree centrality characterizes the causal driving strength of a node; in-degree centrality characterizes the degree to which a node is influenced by the network; directed betweenness centrality can identify key hubs in the transmission of causal flows; and authority value centrality and hub value centrality can jointly assess the bidirectional importance of a node in a causal network.

[0058] Specifically, the in-degree centrality and out-degree centrality of nodes are calculated based on a directed weighted causal connection matrix. The out-degree centrality of a node is defined as the sum of the weights of all directed edges from that node to all other nodes, i.e., the sum of all element values ​​in the corresponding row of that node. The in-degree centrality of a node is defined as the sum of the weights of all directed edges from all other nodes to that node, i.e., the sum of all element values ​​in the corresponding column of that node. Out-degree centrality reflects a node's causal driving ability, while in-degree centrality reflects the degree to which a node is influenced by the network.

[0059] Furthermore, considering the directional nature of the network, the directed betweenness centrality of nodes is calculated. Based on directed weighted causal networks, a shortest path algorithm suitable for directed graphs is used to calculate the directed shortest paths between all pairs of nodes. The directed betweenness centrality of a node is defined as the proportion of paths passing through that node in all directed shortest paths; only pairs of nodes with directed connected paths are considered in the statistics. Directed betweenness centrality reflects the pivotal role of a node in the propagation of directed causal flows.

[0060] Furthermore, the HITS algorithm is used to calculate the authority center and hub center of a node. Through an iterative process, the authority value (representing the node's importance as an information receiver) and hub value (representing the node's importance as an information source) of a node are calculated simultaneously. In each iteration, the node's authority value is updated to the sum of the hub values ​​of all its incoming source nodes, and the node's hub value is updated to the sum of the authority values ​​of all its outgoing target nodes. Normalization is performed during the iteration until convergence. These two metrics together characterize the bidirectional importance of a node in the causal network.

[0061] After calculating multiple centrality indicators, standardization and fusion can be performed using methods such as indicator standardization, weight allocation, and comprehensive centrality score calculation to obtain the final node centrality indicator.

[0062] In step S4, based on the neural vulnerability index, global efficiency index, and node centrality index, key cross-level nodes that play a crucial role in the epileptic seizure process are identified by fusing node attributes and network topology features. This includes: determining the functional coupling index by quantifying the co-evolutionary relationship between node vulnerability and the overall synchronous diffusion efficiency of the functional network based on the neural vulnerability index and the global efficiency index of the dynamic functional connectivity network; determining the causal driving index by quantifying the co-evolutionary relationship between node vulnerability and the information flow driving efficiency of the causal network based on the neural vulnerability index and the global efficiency index of the causal network; fusing the functional coupling index and the causal driving index to obtain the comprehensive dynamic coupling strength; and introducing the node centrality index of the dynamic functional connectivity network and the causal network as independent structural evidence, and combining it with the comprehensive dynamic coupling strength to identify key cross-level nodes.

[0063] Specifically, for the calculation of the functional coupling index, the time-series data of the neural vulnerability index of each node under each time window are first obtained to form a node vulnerability time series. Simultaneously, the global efficiency index of each time window calculated based on the dynamic functional connectivity network is extracted to form a global efficiency time series of the functional network. The time series all cover the preset key time windows before the onset of epileptic seizures to ensure that characteristic changes before the onset of seizures can be captured.

[0064] For each node, the dynamic Pearson correlation coefficient is calculated between its neural vulnerability index over time and the global efficiency of the functional network over time to assess the strength and trend of their correlation over time. Neural vulnerability index (a micro-attribute of the node) and global efficiency (a macro-topological characteristic of the network) belong to different conceptual levels. Therefore, both indices are Z-score standardized to map their values ​​to the same dimensional interval (mean 0, standard deviation 1). A sliding window is then used to simultaneously extract time segments of both indices to ensure consistent temporal granularity of the analysis window. The neural vulnerability index reflects the stability margin of a node in the local network, while global efficiency characterizes the overall network's information transmission capability. The two are intrinsically linked through node-network interaction mechanisms (e.g., abnormal discharge of a vulnerable node can lead to a decrease in global efficiency). This invention aims to capture the dynamic coupling pattern between node attributes and network topology at different stages of a seizure by calculating the dynamic Pearson correlation coefficient. When the absolute value of the correlation coefficient increases significantly, it suggests that the node may affect the global network function through its own stability changes, providing a quantitative basis for identifying cross-level key nodes.

[0065] Within a pre-defined time window prior to an epileptic seizure, the absolute maximum value of the correlation coefficient between the neural vulnerability index and global efficiency of each node is identified. This maximum value is used as the original feature value and standardized using a min-maximum normalization method, ensuring that the functional coupling index values ​​of all nodes are distributed between zero and one, thus obtaining the standardized functional coupling index.

[0066] Specifically, for the calculation of the causal driving index, the time series data of the neural vulnerability index and the global efficiency of the causal network for each node are first obtained. For each node, the time-delay cross-correlation function between its neural vulnerability index time series and the causal global efficiency time series is calculated. By systematically introducing positive and negative time-delay parameters, the correlation strength between the two time series at different time offsets is analyzed.

[0067] Next, determine the time delay parameter value that maximizes the time delay cross-correlation function. When the time delay parameter is less than or equal to zero, it indicates that the node's vulnerability changes precede or are synchronized with changes in the network's global efficiency, and the node has a potential driving role; when the time delay parameter is greater than zero, it indicates that the node's vulnerability changes lag behind changes in network efficiency, and the node is more likely to be a follower node.

[0068] For nodes with time-delay parameters less than or equal to zero, the maximum time-delay cross-correlation function value is multiplied by a preset driving direction weighting coefficient to obtain a preliminary causal driving value. This weighting coefficient is set based on the magnitude of the time delay; the more negative the time delay, the stronger the leading effect, and the larger the weighting coefficient is assigned. Subsequently, the preliminary causal driving value is normalized and multiplied by the average gradient of change within a preset time window before the epileptic seizure to finally obtain the causal driving index.

[0069] Next, the functional coupling index and the causal driving index of each node are weighted and fused to obtain the comprehensive dynamic coupling strength. The weight allocation is optimized and determined based on the discriminative efficacy of the two indices in the preliminary experiments to ensure that their contributions to the final result are balanced.

[0070] Finally, a multi-criteria decision-making model was established, using comprehensive dynamic coupling strength as the primary criterion and node centrality as structural evidence for comprehensive judgment. An adaptive threshold mechanism was set to identify nodes that exhibit outstanding performance in both dynamic characteristics and topological structure, ultimately determining the cross-level key nodes that play a crucial role in the epileptic seizure process.

[0071] Specifically, a multi-level decision feature system is established, encompassing both dynamic characteristics and network topology characteristics. The first-level feature is the comprehensive dynamic coupling strength, which integrates the functional coupling index and the causal driving index to characterize the dynamic driving ability of nodes in the epilepsy network. The second-level feature is a set of node centrality indicators, including weighted degree centrality and betweenness centrality extracted from dynamic functional connectivity networks, as well as structural importance indicators such as out-degree centrality, in-degree centrality, and directed betweenness centrality extracted from causal networks.

[0072] All decision features are standardized and preprocessed, and a min-maximum normalization method is used to map each feature value to the range of zero to one. Based on the feature importance assessment results, appropriate weight coefficients are assigned to different features. The comprehensive dynamic coupling strength is given a high weight as the core dynamic feature, and the node centrality index is given a corresponding weight as structural evidence. The weight allocation ratio is determined through pre-experiment verification to ensure the dominant role of dynamic features while taking into account the auxiliary judgment value of topological structure.

[0073] A multi-criteria decision-making model based on weighted comprehensive scoring is established. For each node, its multi-feature weighted comprehensive score is calculated, which comprehensively reflects the node's performance in two dimensions: dynamic importance and topological importance. Simultaneously, an adaptive threshold mechanism based on percentiles is set to rank nodes according to their weighted comprehensive scores, selecting the top 20% of nodes as candidate key nodes.

[0074] This invention also designs a dynamically adjustable adaptive threshold strategy to adjust the screening criteria for key nodes based on the specific characteristics of the epilepsy network. First, the statistical distribution characteristics of the weighted composite score of all nodes are calculated, including the mean, standard deviation, and skewness coefficient. A basic threshold level is determined based on the distribution characteristics and appropriately adjusted in conjunction with prior clinical knowledge. For candidate nodes whose weighted composite scores exceed the threshold, they are further required to exhibit outstanding performance in both dynamic and topological characteristics; that is, the overall dynamic coupling strength and the centrality index of each node must both reach their respective sub-threshold criteria.

[0075] Cross-level consistency verification is then performed on the initially selected key nodes. This involves checking whether each candidate node maintains a consistent topological position in both the functional connectivity network and the causal network, verifying the match between its dynamic behavior and structural role. Nodes that excel only at a single level and lack cross-level consistency are eliminated. Ultimately, nodes that hold a key position in both dynamic driving capabilities and network topology are retained as the true cross-level key nodes.

[0076] Furthermore, this invention can classify identified cross-level key nodes into different importance levels based on their comprehensive performance in a multi-criteria decision-making model. Level 1 key nodes are those that excel in both dynamic characteristics and all topological indices; Level 2 key nodes are those that stand out in core dynamic characteristics and major topological indices; and Level 3 key nodes are those that meet the standard in some important features. This hierarchical classification provides a tiered reference for the development of clinical treatment strategies.

[0077] Taking the SEEG signal of a patient with drug-resistant epilepsy who has had six electrodes implanted as an example, the sampling frequency is 1000Hz. The AF nodes corresponding to the six electrodes are analyzed using a 200ms sliding window with a step size of 50ms to obtain signal data for eight consecutive time windows (T1-T8).

[0078] Based on state transition matrix and Lyapunov index analysis, the neural vulnerability index (NVI) of each node in eight time windows was calculated, and the results are shown in Table 1: Table 1. Neurovulnerability Index

[0079] The global efficiency time series of the Dynamic Functional Connection Network (DFM) from T1 to T8 is: [0.65, 0.63, 0.60, 0.58, 0.55, 0.52, 0.45, 0.38]; the global efficiency time series of the Causal Network from T1 to T8 is: [0.55, 0.53, 0.52, 0.54, 0.58, 0.62, 0.75, 0.85]; the node centrality indices (normalized values) are: A: 0.42, B: 0.68, C: 0.95, D: 0.88, E: 0.75, F: 0.46.

[0080] Based on the pre-epidemic data from T1 to T6, the Pearson correlation coefficient between node NVI and the global efficiency of the functional network was calculated, and the functional coupling index (FCI) was obtained after normalization: FCI_A=1.00; FCI_B=0.125; FCI_C=0.979; FCI_D=0.906; FCI_E=0.188; FCI_F=0.083.

[0081] By using time-delay cross-correlation analysis, the lead-lag relationship between node NVI and the global efficiency of the causal network is identified, and the causal driving index (CDI) is calculated: CDI_A=1.00 (time-delay parameter k=-2, leading by 2 windows); CDI_C=0.81 (time-delay parameter k=-1, leading by 1 window); and CDI=0 for the remaining nodes.

[0082] The weights of FCI and CDI are each 0.5. The overall dynamic coupling strength is calculated as follows: Node A = 0.5 × 1.00 + 0.5 × 1.00 = 1.00; Node C = 0.5 × 0.979 + 0.5 × 0.81 = 0.895.

[0083] With a dynamic coupling strength weight of 0.6 and a centrality weight of 0.4, the final key node scores are calculated as follows: Node A = 0.6 × 1.00 + 0.4 × 0.42 = 0.768; Node C: 0.6 × 0.895 + 0.4 × 0.95 = 0.917. Both scores are greater than the set threshold of 0.7, and the key nodes are identified as C and A.

[0084] like Figure 5 As shown, node A's NVI decreased by 28% during T1-T3, identifying it as the epilepsy origin point; its propagation path is A(T1-T3)→C(T3-T5)→D(T5-T7)→the entire network; in addition, node C has both high FCI (0.979) and high centrality (0.95), identifying it as a critical bridge node; node A's time delay parameter k=-2, confirming it as the main driving node; analysis shows that node A is the driver, node C is the bridge node, and node D is the follower.

[0085] Example 2: The difference between Example 2 and Example 1 is that the epileptic seizure process has distinct stages, including the preictal phase, seizure onset, seizure duration, and seizure termination. The seizure onset phase typically manifests as rapid, high-frequency, low-amplitude discharges, suitable for a short window of 50-100 ms to accurately capture transient characteristics; the seizure duration may exhibit rhythmic discharges and synchronized activity, suitable for a medium-length window of 200-300 ms to analyze stable oscillation patterns; the signal changes during the preictal and termination phases are relatively slow, and a longer window of 300-500 ms can be used to capture slow trends. Therefore, the window parameters can also be dynamically adjusted according to the characteristics of epileptic seizures.

[0086] Example 3: A SEEG-based dynamic epileptogenic brain network key node identification system, which is used in the SEEG-based dynamic epileptogenic brain network key node identification system described in Example 1, such as... Figure 6 As shown, it includes a signal segmentation module, a node analysis module, a network analysis module, and a node identification module.

[0087] The system comprises the following modules: a signal segmentation module for dynamically segmenting the SEEG signal using a sliding time window to obtain signal data across multiple time windows; a node analysis module for determining the neural vulnerability index of each node based on the signal data; a network analysis module for constructing dynamic functional connectivity networks and causal networks based on the multi-band time-frequency characteristics of the signal data, and determining the global efficiency index and node centrality index of each network; and a node identification module for identifying key cross-level nodes that play a crucial role in the epileptic seizure process by fusing node attributes with network topology features, based on the neural vulnerability index, global efficiency index, and node centrality index.

[0088] Working Principle: This invention overcomes the inherent limitation of static network analysis methods, which can only provide "static snapshots," by employing an adaptive sliding time window technique to perform full-time dynamic segmentation of SEEG signals. This method can clearly reveal the temporal evolution of the connection strength, topology, and key node roles of the pathogenic network during different stages of an epileptic seizure, including the pre-ictal, onset, duration, and termination phases, thereby accurately locating the origin and propagation path of epileptic activity.

[0089] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying key nodes of a dynamic epileptogenic brain network based on SEEG, characterized in that, Includes the following steps: The SEEG signal is dynamically segmented using a sliding time window to obtain signal data with multiple time windows; Based on the signal data, the neural vulnerability index of each node is determined; Based on the multi-band time-frequency characteristics in the signal data, dynamic functional connectivity networks and causal networks are constructed respectively, and the global efficiency index and node centrality index of each network are determined. Based on the aforementioned neural vulnerability index, global efficiency index, and node centrality index, key nodes across layers are extracted by fusing node attributes with network topology features.

2. The method for identifying key nodes of dynamic epileptogenic brain networks based on SEEG according to claim 1, characterized in that, The dynamic segmentation includes: An adaptive sliding time window is used to segment the SEEG signal, and the window parameters are dynamically adjusted according to the signal sampling frequency and / or the characteristics of epileptic seizures. The signal data is obtained by standardizing the signal within each time window to eliminate baseline drift and noise interference.

3. The method for identifying key nodes of dynamic epileptogenic brain networks based on SEEG according to claim 1, characterized in that, The multi-band time-frequency features are extracted using wavelet transform, including: The signal data is decomposed into multiple frequency bands using discrete wavelet transform; Calculate the relative power, power spectral entropy, and band energy ratio characteristics for each frequency band; Principal component analysis is used for feature dimensionality reduction, and principal components with a contribution rate greater than or equal to a first threshold are retained to obtain the multi-band time-frequency features.

4. The method for identifying key nodes of dynamic epileptogenic brain networks based on SEEG according to claim 1, characterized in that, The determination of the neural vulnerability index includes: Constructing a dynamic network model based on the state transition matrix; The Lyapunov exponent was used to analyze the system stability of the dynamic network model at different time windows, and the effective window signals in the critical state were screened out. The neural vulnerability index of each node in the effective window signal is calculated by solving the minimum perturbation problem.

5. The method for identifying key nodes of dynamic epileptogenic brain networks based on SEEG according to claim 1, characterized in that, The construction of the dynamic functional connectivity network includes: The functional connection strength between nodes is calculated using the phase-locked value, resulting in the functional connection matrix; Based on the aforementioned functional connectivity matrix, a weighted undirected network is constructed using graph theory, and the global efficiency index at each time point is determined, thus forming a dynamic functional connectivity network.

6. The method for identifying key nodes of dynamic epileptogenic brain networks based on SEEG according to claim 1, characterized in that, The construction of the causal network includes: A directed connection network was established using time-varying Granger causality analysis; Based on the directed connection network, calculate the causal outflow intensity and inflow intensity of the nodes; Based on the causal outflow and inflow intensities, a causal network is constructed by analyzing the direction of network information flow using directed graph theory indices.

7. The method for identifying key nodes of dynamic epileptogenic brain networks based on SEEG according to claim 1, characterized in that, Based on the neural vulnerability index, global efficiency index, and node centrality index, key nodes across layers are extracted by fusing node attributes and network topology features, including: Based on the neural vulnerability index and the global efficiency index of the dynamic functional connection network, the functional coupling index is determined by quantifying the synergistic evolution relationship between node vulnerability and the overall synchronous diffusion efficiency of the functional network. Based on the neural vulnerability index and the global efficiency index of the causal network, the causal driving index is determined by quantifying the co-evolutionary relationship between node vulnerability and the information flow driving efficiency of the causal network. The functional coupling index and the causal driving index are fused to obtain the comprehensive dynamic coupling strength; The node centrality indices of the dynamic functional connection network and the causal network are introduced as independent structural evidence, and the cross-level key nodes are identified in combination with the comprehensive dynamic coupling strength.

8. The method for identifying key nodes of dynamic epileptogenic brain networks based on SEEG according to claim 1, characterized in that, The determination of the functional coupling index includes: Obtain time-series data of the neural vulnerability index for each node within each time window; Based on the dynamic functional connection network, global efficiency time series data corresponding to each time window are extracted; Calculate the dynamic Pearson correlation coefficient between the time series of the neural vulnerability index of each node and the time series of the global efficiency; For each node, calculate the absolute value of the maximum correlation coefficient between its neural vulnerability index and global efficiency within a preset time window before the epileptic seizure. After normalizing the absolute value of the maximum correlation coefficient, it is defined as the functional coupling index of the node.

9. The method for identifying key nodes of dynamic epileptogenic brain networks based on SEEG according to claim 1, characterized in that, The determination of the causal driving index includes: Obtain time-series data of the neural vulnerability index for each node within each time window; Based on the causal network, extract the causal global efficiency time series data corresponding to each time window; For each node, calculate the time-delay cross-correlation function between its neural vulnerability index time series and the causal global efficiency time series; Determine the time delay parameter that maximizes the time delay cross-correlation function; When the time delay parameter is less than or equal to zero, the maximum time delay cross-correlation function value is multiplied by a preset driving direction weighting coefficient to obtain a preliminary causal driving value. The preliminary causal driving value is normalized and multiplied by the average change gradient within a preset time window before the epileptic seizure to obtain the final causal driving index.

10. A key node identification system for dynamic epileptogenic brain networks based on SEEG, characterized in that, This system is used to implement the SEEG-based method for identifying key nodes in dynamic epileptogenic brain networks as described in any one of claims 1-9, comprising: The signal segmentation module is used to dynamically segment the SEEG signal using a sliding time window to obtain signal data for multiple time windows. The node analysis module is used to determine the neural vulnerability index of each node based on the signal data. The network analysis module is used to construct dynamic functional connectivity networks and causal networks based on the multi-band time-frequency characteristics in the signal data, and to determine the global efficiency index and node centrality index of each network. The node identification module is used to extract key nodes across layers by fusing node attributes and network topology features based on the neural vulnerability index, global efficiency index and node centrality index.

Images