A method and processing device for locating epileptogenic foci of bilateral temporal lobe epilepsy

Abstract

The application provides a method and a processing device for positioning an epileptogenic focus of bilateral temporal lobe epilepsy, the method comprising: acquiring multi-modal data of bilateral temporal lobes of a patient, including first data, second data and third data; constructing a multi-order neural topology evolution tensor based on anatomical constraints of the first data, functional connections of the second data and electro-physiological abnormal discharge characteristics of the third data; inputting the multi-order neural topology evolution tensor into a symmetric-asymmetric decoupling graph convolution network model to obtain an epileptogenic activity score map of the bilateral temporal lobes; and identifying a core starting node in an epileptogenic network as an epileptogenic focus based on the epileptogenic activity score map, functional connection betweenness centrality and electro-physiological abnormal discharge characteristics through a topology dynamics core degree measurement algorithm. The application can decouple weak asymmetric characteristics in bilateral lesions, solves the problem of extracting weak difference characteristics of bilateral temporal lobes, and improves the ability to identify the core starting area of the entire epileptogenic network.

Description

Technical Field

[0001] This application relates to the field of intelligent biomedical signal processing technology, specifically a method and processing device for locating epileptogenic foci in bilateral temporal lobe epilepsy. Background Technology

[0002] The temporal lobe is located in the area behind the temple. Bilateral temporal lobe epilepsy typically begins in either the left or right temporal lobe. Due to pathological activity in both temporal lobes, the functional and electrophysiological asymmetry between the epileptogenic and healthy sides is significantly reduced, rendering traditional lateralization indicators ineffective. This poses a significant challenge to identifying and locating the epileptogenic focus in both temporal lobes. Therefore, overcoming these technical problems and deficiencies is a key issue that needs to be addressed. Summary of the Invention

[0003] To overcome the aforementioned problems in the prior art, this application provides a method and processing apparatus, which adopts the following technical solution:

[0004] In a first aspect, this application provides a method for locating epileptogenic foci in bilateral temporal lobe epilepsy, including:

[0005] Acquire multimodal data of the patient's bilateral temporal lobes, including first data, second data, and third data;

[0006] Based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data, a multi-level neural topological evolution tensor is constructed.

[0007] By inputting the multi-order neural topology evolution tensor into the symmetric-asymmetric decoupled graph convolutional network model, the epileptogenic activity score map of the bilateral temporal lobes is obtained.

[0008] Using a topological dynamics core metric algorithm, based on epileptogenic activity scoring maps, functional connectivity betweenness centrality, and electrophysiological abnormal discharge characteristics, core initiation nodes in the epileptogenic network are identified as epileptogenic foci.

[0009] Furthermore, based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data, a multi-order neural topological evolution tensor is constructed, including:

[0010] The bilateral temporal lobes were selected as regions of interest based on the first data segmentation.

[0011] Calculate the functional connectivity matrix and functional connectivity betweenness centrality of the second data in the region of interest based on the region of interest.

[0012] Using the cerebral cortex geometric model in the first data as prior knowledge, the source of the third data is located, and the abnormal electrophysiological features are dynamically projected onto the patient's cerebral cortex geometric model.

[0013] The dynamic functional connection matrix, structural feature vector, and dynamic feature map sequence are aligned and fused into a multi-order neural topological evolution tensor.

[0014] Furthermore, the bilateral temporal lobes, segmented based on the first data, are selected as regions of interest, including:

[0015] Based on the first data, the boundaries of gray matter, white matter, cerebrospinal fluid, exposed bone, and scalp are extracted using a preset segmentation algorithm to generate a cerebral cortex geometric model. The conductivity parameters of each layer of the cerebral cortex geometric model are assigned based on the boundary element method. The lead matrix is ​​generated through forward model calculation, and the morphological features extracted from the first data are used to construct a structural feature vector.

[0016] Furthermore, based on the region of interest, the functional connectivity matrix and functional connectivity betweenness centrality of the second data in the region of interest are calculated, including:

[0017] From the N regions of interest defined by each first data constraint, extract the time series of the denoised BOLD signal, calculate the Pearson correlation coefficient between all N×N node pairs, and obtain the N×N static functional connectivity matrix.

[0018] The dynamic functional connection matrix is ​​obtained by calculating the static functional connection matrix that changes over time using a sliding time window.

[0019] The dynamic function connection matrix is ​​transformed into an instantaneous network graph. On each instantaneous network graph, betweenness centrality is calculated using the shortest path algorithm in graph theory.

[0020] Furthermore, using the cerebral cortex geometric model in the first data as prior knowledge, source localization is performed on the third data, and abnormal electrophysiological features are dynamically projected onto the patient's cerebral cortex geometric model, including:

[0021] Based on the constraints of the cerebral cortex geometric model and the lead matrix, the denoised EEG signal is inverted into cortical source signal;

[0022] Frequency domain analysis of cortical source signals is performed to separate epileptogenic core features from background activity. The separated epileptogenic core features are quantified to generate a dynamic electrophysiological abnormal discharge index.

[0023] Based on a preset time window, the average or peak value of the continuous dynamic electrophysiological abnormal discharge index is extracted to generate discrete dynamic feature maps. These feature maps are then projected onto the geometric model of the cerebral cortex through source localization and generate a sequence of dynamic feature maps along with the time window.

[0024] Furthermore, the symmetric-asymmetric decoupled graph convolutional network model includes two branches, one of which is a symmetric branch and the other is an asymmetric branch.

[0025] Furthermore, the multi-order neural topological evolution tensor is input into a symmetric-asymmetric decoupled graph convolutional network model to obtain epileptogenic activity scoring maps of the bilateral temporal lobes, including:

[0026] The symmetric-asymmetric decoupled graph convolutional network model iterates along the time dimension from t=1 to t=T, extracting an instantaneous graph at each time step t;

[0027] The symmetric part of the adjacency matrix is ​​averaged by symmetric branching, and the average value is aggregated by standard GCN to obtain the background activity feature representation;

[0028] Asymmetric differences are calculated on the asymmetric parts of the adjacency matrix through asymmetric branches, and then aggregated using a differentiated GCN to obtain epileptogenic driving feature representations.

[0029] Based on advanced distance metrics such as Wasserstein distance, the similarity between background activity features and epileptogenic driving features is penalized, and symmetric and asymmetric features are forced to learn independent and non-overlapping feature spaces to obtain symmetric and asymmetric branch features.

[0030] After feature aggregation at all time steps T, the symmetric and asymmetric branch features are fused to form the final node features. The probability that the final node features belong to the epileptogenic focus is calculated through the output layer, and the output vector is then... The epileptogenic activity score vector is obtained and mapped onto the geometric model of the cerebral cortex to obtain the epileptogenic activity score map.

[0031] Secondly, this application also provides a device for locating and treating epileptogenic foci in bilateral temporal lobe epilepsy, comprising:

[0032] The data acquisition module is used to acquire multimodal data of the patient's bilateral temporal lobes, including first data, second data, and third data;

[0033] A multi-level neural topology evolution tensor construction module is used to construct a multi-level neural topology evolution tensor based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data.

[0034] The module for obtaining the epileptogenic activity score map of the bilateral temporal lobes is used to input the multi-order neural topology evolution tensor into the symmetric-asymmetric decoupled graph convolutional network model to obtain the epileptogenic activity score map of the bilateral temporal lobes.

[0035] The epileptogenic focus identification module is used to identify the core initiation node in the epileptogenic network as the epileptogenic focus by using the topological dynamics core metric algorithm based on the epileptogenic activity score map, functional connectivity betweenness centrality and electrophysiological abnormal discharge characteristics.

[0036] Thirdly, this application provides an electronic device, comprising:

[0037] One or more processors; a memory; and one or more computer programs, wherein the one or more computer programs are stored in the memory, and the one or more computer programs include instructions that, when executed by the device, cause the device to perform the method as described in the first aspect.

[0038] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to perform the method described in the first aspect.

[0039] Fifthly, this application provides a computer program that, when executed by a computer, performs the method described in the first aspect.

[0040] In one possible design, the program in the fifth aspect can be stored wholly or partially on a storage medium packaged with the processor, or it can be stored wholly or partially on a memory not packaged with the processor.

[0041] This application has the following beneficial effects:

[0042] 1. This application acquires multimodal data from the bilateral temporal lobes of a patient, including first data, second data, and third data; based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data, a multi-level neural topological evolution tensor is constructed. This allows for focusing on the most probable lesion region based on the region of interest, and the multi-level neural topological evolution tensor integrates data from different modalities, capturing the spatiotemporal dynamics and heterogeneous relationships between these different modalities.

[0043] 2. This application obtains epileptogenic activity scores for both temporal lobes by inputting multi-order neural topological evolution tensors into a symmetric-asymmetric decoupled graph convolutional network model. Using a topological dynamics core metric algorithm, based on the epileptogenic activity scores, functional connectivity betweenness centrality, and electrophysiological abnormal discharge characteristics, the core initiation nodes in the epileptogenic network are identified as epileptogenic foci. The symmetric-asymmetric decoupled graph convolutional network model identifies highly active regions, and the topological dynamics core metric algorithm filters out the core initiation epileptogenic foci from these regions. This symmetric-asymmetric decoupled graph convolutional network model solves the challenge of extracting subtle differential features from both temporal lobes. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the method for locating epileptogenic foci in bilateral temporal lobe epilepsy according to an embodiment of this application.

[0045] Figure 2 This is a flowchart illustrating the construction of a multi-order neural topology evolution tensor according to an embodiment of this application.

[0046] Figure 3 This is a schematic diagram of the dynamic projection process of abnormal electrophysiological characteristics according to an embodiment of this application;

[0047] Figure 4 This is a flowchart illustrating the process of obtaining epileptogenic activity scores of the bilateral temporal lobes according to an embodiment of this application.

[0048] Figure 5 This is a schematic diagram of a device for locating and treating epileptogenic foci in bilateral temporal lobe epilepsy according to an embodiment of this application. Detailed Implementation

[0049] 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 pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0050] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0051] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0052] Example 1:

[0053] Please refer to Figure 1 The flowchart below illustrates a method for locating epileptogenic foci in bilateral temporal lobe epilepsy, as provided in this application embodiment. The specific implementation process is as follows:

[0054] Step S1: Acquire multimodal data of the patient's bilateral temporal lobes, including high-resolution structural magnetic resonance imaging (SMRI), resting-state functional magnetic resonance imaging (fMRI), and high-density scalp electroencephalography (EEG). The high-resolution SMRI is used as the first data, the resting-state fMRI as the second data, and the high-density scalp EEG as the third data.

[0055] It should be noted that high-resolution structural magnetic resonance imaging (MRI) is a non-invasive, non-ionizing radiation imaging technique that reveals details of different parts of the patient's body and techniques, clearly displaying the anatomical structure of the human body. This application establishes a geometric model of the patient's cerebral cortex using T1-weighted images and segments the bilateral temporal lobes and hippocampus. Using the cortical geometric model provided by the high-resolution structural MRI images as prior anatomical knowledge, subsequent steps restrict the search for current sources to the gray matter of the cerebral cortex, narrowing the source space and making the calculated origin locations of spikes and slow waves more accurate and reliable.

[0056] It should be noted that the resting-state functional magnetic resonance imaging (fMRI) images in this application are BOLD (blood oxygen level dependent) signals acquired while the patient is awake or without a specific task. These signals record spontaneous low-frequency fluctuations in the brain in default mode, revealing the collaborative working relationship between different brain regions. This application constructs a whole-brain and local functional connectivity matrix using time-series data of BOLD signals from resting-state fMRI images, with a particular focus on the connectivity within and between the bilateral temporal lobes, specifically the connectivity strength between different subregions within the left temporal lobe, the connectivity strength between different regions within the right temporal lobe, and the connectivity strength between different subregions of the left and right temporal lobes (spanning hemispheres).

[0057] It should be noted that high-density scalp EEG, through a uniformly distributed mesh electrode cap, can achieve dense coverage of the entire head, face, and back of the neck, capturing brain electrical activity. This application uses 128 / 256-lead high-density scalp EEG to obtain transient epileptic abnormal discharges in the patient's brain, extracting electrophysiological characteristics, especially the dynamic spatial distribution characteristics of sharp and slow wave activities.

[0058] Step S2: Based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data, construct a multi-level neural topological evolution tensor.

[0059] In this embodiment, a multi-level neural topological evolution tensor is constructed based on the first data anatomical constraints, the second data functional connectivity, and the third data dynamic electrophysiological characteristics. Please refer to [reference needed]. Figure 2 ,include:

[0060] Step 21: The bilateral temporal lobes are selected as regions of interest based on the first data segmentation.

[0061] It should be noted that before the bilateral temporal lobes segmented based on the first data are used as the region of interest, the process includes: extracting the boundaries of gray matter, white matter, cerebrospinal fluid, exposed bone, and scalp based on the first data using a preset segmentation algorithm to generate a cerebral cortex geometric model; assigning conductivity parameters of each layer of the cerebral cortex geometric model based on the boundary element method; generating a lead matrix through forward model calculation; determining the mapping relationship from the current source to the scalp potential; and using the morphological features extracted from the first data as structural feature vectors.

[0062] It should be noted that the bilateral temporal lobes of the first data segmentation are divided into N predefined regions. These N predefined regions serve as nodes for functional connectivity and electrophysiological characteristics, ensuring that the data are aligned within the same anatomical space.

[0063] Step 22: Calculate the functional connectivity matrix and functional connectivity betweenness centrality of the second data in the region of interest based on the region of interest.

[0064] In this embodiment of the application, the calculation of the functional connectivity matrix of the second data in the region of interest based on the region of interest includes: extracting the denoised BOLD signal time series from N regions of interest defined by each first data constraint, calculating the Pearson correlation coefficient between all N×N node pairs, obtaining an N×N static functional connectivity matrix, calculating the static functional connectivity matrix that changes over time through a sliding time window, and obtaining a dynamic functional connectivity matrix.

[0065] In this embodiment of the application, the computation function connects betweenness centrality, including:

[0066] The dynamic function connection matrix is ​​transformed into a forward model into an instantaneous network graph. On each instantaneous network graph, betweenness centrality is calculated using the shortest path algorithm in graph theory.

[0067] It should be noted that transforming the dynamic functional connectivity matrix into a transient network graph includes: applying a threshold to the dynamic functional connectivity matrix, generating a binarized adjacency matrix, and obtaining a graph sequence that varies over time. The binarized adjacency matrix transforms connection strength into discrete connectivity information. The graph sequence that varies over time is a structure composed of nodes and edges. The set of nodes represents N regions of interest, and the set of edges is determined by the non-zero elements in the binarized adjacency matrix.

[0068] It should be noted that the betweenness centrality of each instantaneous network graph, calculated using the shortest path algorithm in graph theory, can be represented as: ,in Indicates betweenness centrality, This represents the number of times the target node v is passed through in the shortest path from source node s to target node t. This represents the total number of shortest paths from source node s to target node t. Indicates a time window.

[0069] Step 23: Using the cerebral cortex geometric model from the first dataset as prior knowledge, source localization is performed on the third dataset. Abnormal electrophysiological features are then dynamically projected onto the patient's cerebral cortex geometric model. Please refer to [link / reference needed]. Figure 3 ,include:

[0070] Step 231: Based on the constraints of the cerebral cortex geometric model and the lead matrix, the denoised EEG signal is inverted into cortical source signal.

[0071] It should be noted that the process of converting denoised EEG signals into cortical source signals includes:

[0072] The positive problem is to calculate the scalp voltage V(t) based on the cortical current source J(t) and the lead matrix L, that is: Where V(t) represents the voltage signal measured on the M electrodes, This represents the current density at N cortical source points. , This indicates measurement noise. Scalp voltage is the voltage signal measured by scalp electrodes over time, which is also the electroencephalogram (EEG) signal.

[0073] It should be noted that the forward model calculation solves the forward problem, that is, given the distribution of cortical current sources in the cerebral cortex, it calculates the potential distribution generated by these cortical current sources at the electrode locations on the scalp.

[0074] The inverse problem is based on known scalp voltage. and lead matrix Solve for the unknown cortical current source J(t). And because... The inverse problem is underdetermined and ill-determined, meaning it has infinitely many solutions and is sensitive to noise.

[0075] To address underdetermined and ill-determined problems, an objective function is constructed using minimum norm constraints and structural constraints: ,in for The estimate, It is a data fitting term that can measure the estimation source. generated voltage The error between the measured voltage and the actual voltage It is the inverse of the noise covariance matrix, used to load data fitting terms and reduce the impact of high-noise electrodes. It is a regularization term; the most probable solution is selected through the regularization term. It is a regularization parameter that controls the strength of the constraint. It is the weighted identity matrix, and is the inverse of the source covariance matrix. This is the constraint matrix.

[0076] Solve for the inverse matrix W using a closed-form solution based on regularized least squares: For each time point of V(t), matrix multiplication is used. The denoised EEG signal V(t) is inverted point by point into cortical source signal. for The transpose of .

[0077] Step 232 involves performing frequency domain analysis on the cortical source signal to separate the core epileptogenic features from the background activity. The separated core epileptogenic features are then quantified to generate a dynamic electrophysiological abnormal discharge index. Step 232 transforms electrophysiological activity into a quantifiable epileptogenic activity index.

[0078] In the embodiments of this application, the core epileptogenic features include high-frequency features, spike wave index, and abnormal slow wave index.

[0079] In this embodiment of the application, the quantification of the separated epileptogenic core features includes:

[0080] The high-frequency characteristics are calculated by using the Hilbert transform to calculate the instantaneous envelope of the high-frequency oscillation signal, and the average instantaneous power per unit time is calculated at each node.

[0081] It should be noted that high-frequency oscillations are one of the reliable biomarkers for the core region of the epileptogenic focus. The types include ripples, fast oscillations, and ultra-high frequencies. Among them, the frequency band of ripples is 80~250Hz, the frequency band of fast oscillations is 250~500Hz, and the frequency band of ultra-high frequencies is >500Hz.

[0082] The spike index is calculated by measuring the frequency or integral area of ​​spikes at each node per unit time.

[0083] It should be noted that spikes are instantaneous high-frequency components with a duration of 20~70ms. When performing frequency domain analysis, they manifest as a wideband power increase.

[0084] The abnormal slow-wave index preserves the slow-wave frequency band signal by digitally filtering the core epileptogenic features, and then obtains the power spectral density of the slow-wave frequency band signal through Fourier transform. For power spectral density Integrate the signal to obtain the integral within a specific frequency range. Use the integral result as the absolute power of the abnormal slow wave band and the absolute power value as the abnormal slow wave index.

[0085] It should be noted that, in the embodiments of this application, absolute power = Assuming the calculation is for the theta wave, then =4Hz, =8Hz.

[0086] It should be noted that in awake adults, focal or persistent high-amplitude theta waves are a manifestation of cortical dysfunction. Focal high-amplitude theta waves... Theta waves are a strong indicator of localized brain dysfunction or structural lesions. The frequency band of theta waves ranges from 4 to 8 Hz. The frequency band of the wave is 0.5~4Hz.

[0087] It should be noted that abnormal slow-wave activity is associated with functional impairment surrounding the epileptogenic focus. The dynamic electrophysiological abnormal discharge index is a measure of epileptogenicity. It is obtained by fusing high-frequency features, the spike index, and the abnormal slow-wave index. The high-frequency features reflect abnormal synchronous bursts at the microscale and are a highly specific marker of the epileptogenic focus. The spike index reflects abnormal excitation at the macroscale and is a highly sensitive marker of the epileptogenic focus. The abnormal slow-wave index reflects persistent local cortical dysfunction and is associated with functional impairment in the area surrounding the epileptogenic focus.

[0088] The Dynamic Electrophysiological Abnormal Discharge Index quantifies and temporally dynamizes abnormal characteristics, enabling a measurement of overall epileptogenic capacity over time. It can also capture the enhancement or weakening of epileptogenic network activity before and after a seizure, as well as its evolutionary trajectory. This is crucial for identifying the initiation point of the epileptogenic focus.

[0089] Step 233: Based on a preset time window, extract the average or peak value of the continuous dynamic electrophysiological abnormal discharge index, generate discrete dynamic feature maps, project them onto the geometric model of the cerebral cortex through source localization, and generate a dynamic feature map sequence with the time window.

[0090] In this embodiment of the application, source localization is projected onto a geometric model of the cerebral cortex, and a dynamic feature map sequence is generated over a time window, including:

[0091] Based on the source localization algorithm, the scalp signal of abnormal discharge is mapped to N source current densities of interest. Based on the source current density within a continuous short time window, the electrophysiological connectivity matrix between the N regions of interest is calculated, and network topology indicators are extracted from the electrophysiological connectivity matrix.

[0092] It should be noted that this application uses the physical electrode positions acquired by the second data acquisition to rigidly register with the cerebral cortex geometric model constructed by the first data, forming a forward propagation model of current propagation in the cortex.

[0093] As can be seen from steps 231-233, this application achieves brain region segmentation through the first data, which defines nodes in different segmented regions. The second data provides functional connection matrices for different nodes in different segmented regions. The third data, based on the cortical ensemble model of the first data, restricts the search space of the current source to the gray matter layer, thereby achieving accurate source localization.

[0094] Step 24: Align and fuse the dynamic functional connectivity matrix, structural feature vector, and dynamic feature map sequence into a multi-order neural topological evolution tensor, including:

[0095] Assign the dynamic function connectivity matrix to the first mode of the tensor to obtain the tensor's topological layer, i.e. Then obtain A dynamically connected tensor of dimension , where is For spatial node indexing, To synchronize time window sets, Indicates modal index, Index for time windows. Indicates a connection. Indicates characteristics. This represents the set of valid source points, signifying the defined region of interest. This application forms the tensor's topology layer by assigning the dynamic function connectivity matrix to the first mode of the tensor.

[0096] The feature vectors containing structural and dynamic feature map sequences are assigned to the node attribute dimension of the tensor to obtain the feature layer of the tensor. This involves broadcasting the structural features along the time dimension and directly associating the dynamic electrophysiological indices with time. The tensor is then assigned values. It should be noted that stacking all modes along the M-dimensional plane forms a fully dynamic fourth-order tensor: .in This represents a placeholder; in node attribute modalities, since the feature belongs only to the node. ,therefore Dimensions are treated as placeholders to maintain the dimensional structure of the fourth-order tensor; For structural feature vectors, This is the dynamic electrophysiological discharge index. For dynamic node feature vectors, it is The combined entity. This represents node attributes, indicating The application constructs a feature layer for a tensor by assigning feature vectors containing structural features and dynamic feature map sequences to the node attribute dimension of the tensor.

[0097] It should be noted that this application can encode the functional connectivity matrix at the voxel level of the second data through functional connectivity, encode the static anatomical constraints extracted from the first data through structural features, and encode the dynamic feature map sequence obtained by localization and quantization of the third data source through electrophysiological features. Among them, the structural feature vector provides spatial information for the multi-order neural topological evolution tensor, which can identify the core region of small epileptogenic foci.

[0098] It should be noted that the structural feature vector in this application includes cortical morphological indices and voxel-level morphological indices. The cortical morphological indices quantify the geometric characteristics of the cortical gray matter layer, including cortical thickness, cortical surface area, and cortical curvature. Cortical thickness is the thickness of the gray matter layer along the normal direction at the node; cortical surface area is the local surface area of ​​the region where the node is located; and cortical curvature is the degree of local bending in the region where the node is located. Voxel-level morphological indices include local gray matter volume and gray matter density. Local gray matter volume is the gray matter volume of the voxel where the node is located or its neighborhood; and gray matter density is the proportion of gray matter components within the voxel where the node is located. The structural feature vector can constrain dynamic functional connectivity and electrophysiological abnormalities.

[0099] Step S3: Input the multi-order neural topology evolution tensor into the symmetric-asymmetric decoupled graph convolutional network model to obtain the epileptogenic activity score map of the bilateral temporal lobes.

[0100] In this embodiment, the symmetric-asymmetric decoupled graph convolutional network model includes two branches, one of which is a symmetric branch and the other is an asymmetric branch.

[0101] In this embodiment, a multi-order neural topological evolution tensor is input into a symmetric-asymmetric decoupled graph convolutional network model to obtain epileptogenic activity scoring maps of the bilateral temporal lobes. Please refer to [reference needed]. Figure 4 ,include:

[0102] Step 41: The symmetric-asymmetric decoupled graph convolutional network model iterates along the time dimension from t=1 to t=T. At each time step t, an instantaneous graph is extracted. The instantaneous graph structure includes an adjacency matrix and a feature matrix. The adjacency matrix consists of the edges of the instantaneous graph, representing the dynamic functional connectivity features. The feature matrix consists of the nodes of the instantaneous graph, representing the structural and dynamic electrophysiological features.

[0103] Step 42: Average the symmetrical part of the adjacency matrix through symmetric branching, and aggregate the average value through standard GCN to obtain the background activity feature representation.

[0104] It should be noted that averaging the symmetric portion of the adjacency matrix is ​​achieved through symmetric branching, i.e. The background activity feature representation obtained by aggregating the average value using standard GCN can be expressed as: .in It is a symmetric adjacency matrix. This represents the dynamic functional connectivity strength from node i to node j within the time window t. This represents the dynamic functional connectivity strength from node j to node i within the time window t. This represents the symmetric feature representation of the (l+1)th layer. This represents the symmetric feature representation of the l-th layer. yes The angle matrix. It is the normalization factor. It is the learnable weight matrix of the l-th layer. This represents the activation function.

[0105] Step 43: Calculate the asymmetric difference in the asymmetric part of the adjacency matrix through asymmetric branching, and aggregate it using the differential GCN to obtain the epileptogenic driving feature representation.

[0106] It should be noted that the asymmetric difference calculation by using asymmetric branches to calculate the asymmetric difference of the asymmetric part of the adjacency matrix can be expressed as: ,in The asymmetric difference adjacency matrix over time window t is a The matrix, This represents the dynamic functional connectivity strength from node i to node j within the time window t. This represents the dynamic functional connectivity strength from node j to node i within time window t. If the connectivity from node i to node j is significantly stronger than the connectivity from node j to node i, then... The value will be very large, indicating the existence of obvious asymmetry, which is an important characteristic of the epileptogenic focus as a driving source.

[0107] Step 44: Based on advanced distance metrics such as Wasserstein distance, penalize the similarity between background activity features and epileptogenic driving features, force symmetric and asymmetric features to learn independent and non-overlapping feature spaces, and obtain symmetric branch features and asymmetric branch features.

[0108] It should be noted that the similarity between penalized background activity features and epileptogenic driving features based on high-level distance metrics can be expressed as:

[0109]

[0110] in This represents the decoupling loss. Indicates background activity characteristics, Indicates epileptogenic driving characteristics, This represents the Wasserstein distance regularization coefficient. This represents the Wasserstein distance function.

[0111] It should be noted that the Wasserstein distance is a better way to quantify the asymmetry between the epileptogenic focus and the healthy side. It decouples the driving asymmetric features of the epileptogenic focus from the symmetric background of the normal network, ensuring that the asymmetric branches only focus on pathological biases.

[0112] Step 45: After completing feature aggregation for all time steps T, the symmetric and asymmetric branch features are fused to form the final node features. The probability that the final node features belong to the epileptogenic focus is calculated through the output layer, and the output vector is then obtained. The epileptogenic activity score vector is obtained and mapped onto the geometric model of the cerebral cortex to obtain the epileptogenic activity score map.

[0113] It should be noted that the final node feature can be obtained by merging symmetrical branch features and asymmetrical branch features to form the final node feature.

[0114] Step S4: Using the topological dynamics core metric algorithm, based on the epileptogenic activity score map, functional connectivity betweenness centrality, and electrophysiological abnormal discharge characteristics, identify the core initiation nodes in the epileptogenic network as epileptogenic foci, including:

[0115] For each node in both temporal lobes, the final score of each node is obtained through a composite calculation formula.

[0116] All nodes are sorted in descending order based on the final score sequence. The set of nodes with the highest final score is identified, and the set of nodes that meet the preset threshold is selected as the core node set. The core node set is mapped to the patient's anatomical space. The mapped area is the chord start node of the epileptogenic network, i.e., the epileptogenic focus.

[0117] It should be noted that the final score for each node, obtained through a composite calculation formula, can be expressed as follows: ,in For the final score of the i-th node, The probability of epileptogenic activity at node i is represented, and it is a modulating factor in the final score. For time window indexing, The total number of time windows. For dynamic betweenness centrality, For dynamic electrophysiological indices, , These are adjustable hyperparameters. These are nonlinear coefficients.

[0118] Example 2:

[0119] Please refer to Figure 5 The present application provides a device for locating and treating epileptogenic foci in bilateral temporal lobe epilepsy, which specifically includes:

[0120] The data acquisition module 501 is used to acquire multimodal data of the patient's bilateral temporal lobes, including first data, second data and third data.

[0121] The multi-level neural topology evolution tensor construction module 502 is used to construct a multi-level neural topology evolution tensor based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data.

[0122] The bilateral temporal lobe epileptogenic activity score map acquisition module 503 is used to input the multi-order neural topology evolution tensor into the symmetric-asymmetric decoupled graph convolutional network model to obtain the bilateral temporal lobe epileptogenic activity score map.

[0123] The epileptogenic focus identification module 504 is used to identify the core initiation node in the epileptogenic network as the epileptogenic focus by using the topological dynamics core metric algorithm based on the epileptogenic activity score map, functional connectivity betweenness centrality and electrophysiological abnormal discharge characteristics.

[0124] This application acquires multimodal data of the bilateral temporal lobes of a patient, including first data, second data, and third data; constructs a multi-order neural topological evolution tensor based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data; inputs the multi-order neural topological evolution tensor into a symmetric-asymmetric decoupled graph convolutional network model to obtain the epileptogenic activity score map of the bilateral temporal lobes; and identifies the core initiation node in the epileptogenic network as the epileptogenic focus using a topological dynamics core metric algorithm based on the epileptogenic activity score map, functional connectivity betweenness centrality, and electrophysiological abnormal discharge characteristics. This application can decouple the weak asymmetric features in the bilateral lesions, solve the problem of extracting weak differential features in the bilateral temporal lobes, and improve the ability to identify the core initiation region of the entire epileptogenic network.

[0125] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for locating epileptogenic foci in bilateral temporal lobe epilepsy, characterized in that, include: Acquire multimodal data of the patient's bilateral temporal lobes, including first data from high-resolution structural magnetic resonance imaging, second data from resting-state functional magnetic resonance imaging, and third data from high-density scalp electroencephalography. Based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data, a multi-order neural topological evolution tensor is constructed, including: using the bilateral temporal lobes segmented from the first data as regions of interest; extracting BOLD signal time series from each region of interest, calculating the Pearson correlation coefficient to obtain the static functional connectivity matrix of the second data, calculating the static functional connectivity matrix that changes over time through a sliding time window to obtain the dynamic functional connectivity matrix, converting the dynamic functional connectivity matrix into an instantaneous network graph, and calculating the betweenness centrality on each instantaneous network graph using the shortest path algorithm; using the cortical geometric model in the first data as prior knowledge to perform source localization on the third data, dynamically projecting the abnormal electrophysiological characteristics onto the patient's cortical geometric model, extracting the average or peak value of the continuous dynamic abnormal electrophysiological discharge index based on a preset time window to generate discrete dynamic feature maps, projecting them onto the cortical geometric model through source localization, and generating a sequence of dynamic feature maps with the time window; aligning and fusing the dynamic functional connectivity matrix, structural feature vectors, and dynamic feature map sequences into a multi-order neural topological evolution tensor; wherein the structural feature vectors are the morphological features extracted from the first data. By inputting the multi-order neural topology evolution tensor into the symmetric-asymmetric decoupled graph convolutional network model, the epileptogenic activity score map of the bilateral temporal lobes is obtained. Using a topological dynamics core metric algorithm, based on epileptogenic activity scoring maps, functional connectivity betweenness centrality, and electrophysiological abnormal discharge characteristics, core initiation nodes in the epileptogenic network are identified as epileptogenic foci.

2. The method for locating the epileptogenic focus of bilateral temporal lobe epilepsy according to claim 1, characterized in that, The bilateral temporal lobes, segmented based on the first data, are used as regions of interest, including: Based on the first data, the boundaries of gray matter, white matter, cerebrospinal fluid, exposed bone, and scalp are extracted using a preset segmentation algorithm to generate a cerebral cortex geometric model. The conductivity parameters of each layer of the cerebral cortex geometric model are assigned based on the boundary element method. The lead matrix is ​​generated through forward model calculation, and the morphological features extracted from the first data are used as structural feature vectors.

3. The method for locating the epileptogenic focus of bilateral temporal lobe epilepsy according to claim 1, characterized in that, The functional connectivity matrix and functional connectivity betweenness centrality of the second data in the region of interest are calculated based on the region of interest, including: From the N regions of interest defined by each first data constraint, extract the time series of the denoised BOLD signal, calculate the Pearson correlation coefficient between all N×N node pairs, and obtain the N×N static functional connectivity matrix. The dynamic functional connection matrix is ​​obtained by calculating the static functional connection matrix that changes over time using a sliding time window. The dynamic function connection matrix is ​​transformed into an instantaneous network graph. On each instantaneous network graph, betweenness centrality is calculated using the shortest path algorithm in graph theory.

4. The method for locating the epileptogenic focus of bilateral temporal lobe epilepsy according to claim 1, characterized in that, Using the cerebral cortex geometric model from the first dataset as prior knowledge, source localization is performed on the third dataset, and abnormal electrophysiological features are dynamically projected onto the patient's cerebral cortex geometric model, including: Based on the constraints of the cerebral cortex geometric model and the lead matrix, the denoised EEG signal is inverted into cortical source signal; Frequency domain analysis of cortical source signals is performed to separate epileptogenic core features from background activity. The separated epileptogenic core features are quantified to generate a dynamic electrophysiological abnormal discharge index. Based on a preset time window, the average or peak value of the continuous dynamic electrophysiological abnormal discharge index is extracted to generate discrete dynamic feature maps. These feature maps are then projected onto the geometric model of the cerebral cortex through source localization and generate a sequence of dynamic feature maps along with the time window.

5. The method for locating the epileptogenic focus of bilateral temporal lobe epilepsy according to claim 1, characterized in that, The symmetric-asymmetric decoupled graph convolutional network model contains two branches, one of which is a symmetric branch and the other is an asymmetric branch.

6. The method for locating the epileptogenic focus of bilateral temporal lobe epilepsy according to claim 5, characterized in that, By inputting a multi-order neural topological evolution tensor into a symmetric-asymmetric decoupled graph convolutional network model, epileptogenic activity scores of the bilateral temporal lobes are obtained, including: The symmetric-asymmetric decoupled graph convolutional network model iterates along the time dimension from t=1 to t=T, extracting an instantaneous graph at each time step t; The symmetric part of the adjacency matrix is ​​averaged by symmetric branching, and the average value is aggregated by standard GCN to obtain the background activity feature representation; Asymmetric differences are calculated on the asymmetric parts of the adjacency matrix through asymmetric branches, and then aggregated using a differentiated GCN to obtain epileptogenic driving feature representations. Based on advanced distance metrics such as Wasserstein distance, the similarity between background activity features and epileptogenic driving features is penalized, and symmetric and asymmetric features are forced to learn independent and non-overlapping feature spaces to obtain symmetric and asymmetric branch features. After feature aggregation at all time steps T, the symmetric and asymmetric branch features are fused to form the final node features. The probability that the final node features belong to the epileptogenic focus is calculated through the output layer. The output vector is an N×1 epileptogenic activity score vector. The score vector is mapped onto the cerebral cortex geometric model to obtain the epileptogenic activity score map.

7. A device for locating and treating epileptogenic foci in bilateral temporal lobe epilepsy, used to implement the method described in any one of claims 1-6, characterized in that, include: The data acquisition module is used to acquire multimodal data of the patient's bilateral temporal lobes, including first data of high-resolution structural magnetic resonance imaging, second data of resting-state functional magnetic resonance imaging, and third data of high-density scalp electroencephalography. A multi-order neural topology evolution tensor construction module is used to construct a multi-order neural topology evolution tensor based on the anatomical constraints of the first data, the functional connectivity of the second data, and the electrophysiological abnormal discharge characteristics of the third data. This includes: using the bilateral temporal lobes segmented from the first data as regions of interest; extracting BOLD signal time series from each region of interest; calculating the Pearson correlation coefficient to obtain the static functional connectivity matrix of the second data; calculating the time-varying static functional connectivity matrix using a sliding time window to obtain the dynamic functional connectivity matrix; converting the dynamic functional connectivity matrix into an instantaneous network graph; and calculating each instantaneous network using a shortest path algorithm. The betweenness centrality of the graph; using the cerebral cortex geometric model in the first data as prior knowledge, source localization is performed on the third data, and abnormal electrophysiological features are dynamically projected onto the patient's cerebral cortex geometric model. Based on a preset time window, the average or peak value of the continuous dynamic abnormal electrophysiological discharge index is extracted to generate discrete dynamic feature maps, which are then projected onto the cerebral cortex geometric model through source localization, and a sequence of dynamic feature maps is generated according to the time window; the dynamic functional connectivity matrix, structural feature vector, and dynamic feature map sequence are aligned and fused into a multi-order neural topological evolution tensor; where the structural feature vector is the morphological feature extracted from the first data; The module for obtaining the epileptogenic activity score map of the bilateral temporal lobes is used to input the multi-order neural topology evolution tensor into the symmetric-asymmetric decoupled graph convolutional network model to obtain the epileptogenic activity score map of the bilateral temporal lobes. The epileptogenic focus identification module is used to identify the core initiation node in the epileptogenic network as the epileptogenic focus by using the topological dynamics core metric algorithm based on the epileptogenic activity score map, functional connectivity betweenness centrality and electrophysiological abnormal discharge characteristics.

8. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

9. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-6.

Images