Search

CN-121527019-B - Method for positioning epilepsy focus of bilateral temporal lobe epilepsy and processing device

CN121527019BCN 121527019 BCN121527019 BCN 121527019BCN-121527019-B

Abstract

The application provides a method and a processing device for positioning bilateral temporal lobe epilepsy, wherein the method comprises the steps of obtaining multi-mode data of bilateral temporal lobes of a patient, comprising first data, second data and third data, constructing a multi-order neural topology evolution tensor based on anatomical constraint of the first data, functional connection of the second data and electrophysiological 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 epilepsy-inducing activity score of the bilateral temporal lobes, and identifying a core initial node in an epilepsy-inducing network based on the central characteristics of the epilepsy-inducing activity score, the functional connection medium and the electrophysiological abnormal discharge characteristics through a topological dynamics core metric algorithm.

Inventors

  • GUO XIAODONG

Assignees

  • 中国人民解放军联勤保障部队第九八八医院

Dates

Publication Date
20260508
Application Date
20251111

Claims (9)

  1. 1. A method for locating a bilateral temporal lobe epileptogenic focus, comprising: Acquiring multi-mode data of bilateral temporal lobes of a patient, wherein the multi-mode data comprises first data of a high-resolution structural magnetic resonance image, second data of a resting state functional magnetic resonance image and third data of a high-density scalp electroencephalogram; The method comprises the steps of based on anatomical constraint of first data, functional connection of second data and electrophysiological abnormal discharge characteristics of third data, constructing a multi-order neural topological evolution tensor, extracting a BOLD signal time sequence from each region of interest based on bilateral temporal lobes of first data segmentation as the region of interest, calculating a pearson correlation coefficient to obtain a static functional connection matrix of the second data, calculating the static functional connection matrix changing with time through a sliding time window to obtain a dynamic functional connection matrix, converting the dynamic functional connection matrix into an instantaneous network graph, calculating the medium centrality on each instantaneous network graph through a shortest path algorithm, carrying out source positioning on the third data by taking a brain cortex geometric model in the first data as priori knowledge, dynamically projecting abnormal electrophysiological characteristics onto a brain cortex geometric model of a patient, extracting an average value or a peak value of continuous dynamic electrophysiological abnormal discharge indexes based on a preset time window, generating a discrete dynamic characteristic graph, projecting the dynamic characteristic graph onto the brain cortex geometric model through source positioning, generating a dynamic characteristic graph sequence at any time window, aligning the dynamic functional connection matrix, a structural characteristic vector and the dynamic characteristic graph sequence, and fusing the multi-order characteristic graph sequence into a neural evolution state vector, wherein the first data is the first-order evolutionary state feature; inputting the multilevel neural topology evolution tensor into a symmetrical-asymmetrical decoupling graph convolution network model to obtain an epileptogenic activity score of bilateral temporal lobes; Based on the epilepsy activity scoring graph, the functional connection betweenness centrality and the electrophysiological abnormal discharge characteristics, a core starting node in an epilepsy network is identified as an epilepsy induction range through a topological dynamics core measurement algorithm.
  2. 2. The method of locating a bilateral temporal lobe epileptic focus according to claim 1, wherein the bilateral temporal lobe segmented based on the first data is used as the region of interest, comprising: The method comprises the steps of extracting boundaries of gray matter, white matter, cerebrospinal fluid, exposed bones and scalp through a preset segmentation algorithm based on first data, generating a cerebral cortex geometric model, endowing conductivity parameters of each layer of medium to the cerebral cortex geometric model based on a boundary element method, generating a lead matrix through forward model calculation, and taking morphological features extracted from the first data as structural feature vectors.
  3. 3. The method of locating bilateral temporal lobe seizures foci of claim 1, wherein calculating the functional connection matrix and the functional connection betweenness centrality of the second data in the region of interest based on the region of interest comprises: Extracting denoised BOLD signal time sequences from N regions of interest defined by each first data constraint, calculating pearson correlation coefficients between all N x N node pairs, and obtaining N x N static functional connection matrixes; calculating a static function connection matrix which changes along with time through a sliding time window, and obtaining a dynamic function connection matrix; converting the dynamic function connection matrix into instantaneous network diagrams, and calculating the betweenness centrality on each instantaneous network diagram through a shortest path algorithm of graph theory.
  4. 4. The method of locating bilateral temporal lobe epilepsy as defined in claim 1 wherein the source locating the third data using the cortical geometry model in the first data as a priori knowledge, dynamically projecting the abnormal electrophysiological characteristics onto the patient cortical geometry model comprises: Inverting the denoised electroencephalogram signals into cortex source signals based on constraints of the brain cortex geometric model and a lead matrix; frequency domain analysis is carried out on the cortex source signals, the core characteristics of the epilepsy in the background activities are separated, quantification is carried out on the separated core characteristics of the epilepsy, and a dynamic electrophysiological abnormal discharge index is generated; Based on a preset time window, extracting an average value or a peak value of continuous dynamic electrophysiological abnormal discharge indexes, generating a discrete dynamic feature map, projecting the discrete dynamic feature map onto a cerebral cortex geometric model through source positioning, and generating a dynamic feature map sequence along with the time window.
  5. 5. The method of claim 1, wherein the symmetric-asymmetric decoupled graph convolution network model comprises two branches, wherein one branch is a symmetric branch and the other branch is an asymmetric branch.
  6. 6. The method for locating bilateral temporal lobe epilepsy as in claim 5 wherein inputting the multilevel neural topology evolution tensor into a symmetric-asymmetric decoupling graph convolution network model to obtain an epileptic activity score of bilateral temporal lobe comprises: The symmetric-asymmetric decoupling graph convolution network model performs iterative loop along the time dimension from t=1 to t=t, and extracts an instantaneous graph at each time step T; averaging symmetrical parts in the adjacent matrix through symmetrical branches, and aggregating the average value through a standard GCN to obtain background activity characteristic representation; Calculating asymmetric difference calculation on an asymmetric part in the adjacent matrix through asymmetric branches, and polymerizing by using a differential GCN to obtain an epilepsy driving characteristic representation; punishment of the similarity of the background active feature and the epilepsy driving feature based on Wasserstein distance and other advanced distance measures, forcing the symmetrical feature and the asymmetrical feature to learn feature spaces which are independent and non-overlapping, and obtaining symmetrical branch features and asymmetrical branch features; After feature aggregation of all time steps T is completed, the symmetrical branch features and the asymmetrical branch features are fused to form final node features, the probability that the final node features belong to an epileptogenic focus is calculated through an output layer, then an epileptogenic activity scoring vector with an output vector of N multiplied by 1 is obtained, and the scoring vector is mapped onto a cerebral cortex geometric model to obtain an epileptogenic activity scoring graph.
  7. 7. A device for positioning a bilateral temporal lobe epileptogenic focus for implementing the method of any of claims 1-6, comprising: The data acquisition module is used for acquiring multi-mode data of bilateral temporal lobes of a patient, and comprises first data of a high-resolution structural magnetic resonance image, second data of a resting state functional magnetic resonance image and third data of a high-density scalp electroencephalogram; The multi-order neural topology evolution tensor construction module is used for constructing multi-order neural topology evolution tensor based on anatomical constraint of first data, functional connection of second data and electrophysiological abnormal discharge characteristics of third data, and comprises extracting BOLD signal time sequences from all the interested regions based on bilateral temporal lobes of first data segmentation, calculating pearson correlation coefficients to obtain a static functional connection matrix of the second data, calculating the static functional connection matrix changing with time through a sliding time window to obtain a dynamic functional connection matrix, converting the dynamic functional connection matrix into an instantaneous network map, calculating the medium centrality on each instantaneous network map through a shortest path algorithm, carrying out source positioning on the third data by taking a cerebral cortex geometric model in the first data as priori knowledge, dynamically projecting abnormal electrophysiological characteristics onto a cerebral cortex geometric model of a patient, extracting average value or peak value of continuous dynamic electrophysiological abnormal discharge indexes based on a preset time window, generating a discrete dynamic characteristic map by source positioning projection onto the cerebral cortex geometric model, generating a dynamic characteristic map sequence by the source positioning projection, and aligning the dynamic characteristic vector and the dynamic characteristic sequence to form an evolution topological feature vector; The epilepsy-inducing activity scoring graph acquisition module is used for inputting the multi-order neural topology evolution tensor into a symmetric-asymmetric decoupling graph convolution network model to acquire an epilepsy-inducing activity scoring graph of the bilateral temporal lobes; the epileptogenic focus identification module is used for identifying a core starting node in an epileptogenic network as an epileptogenic focus based on an epileptogenic activity score, a functional connection betweenness centrality and electrophysiological abnormal discharge characteristic through a topological dynamics core measurement algorithm.
  8. 8. An electronic device, comprising: One or more processors; a memory for storing one or more programs; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-6.
  9. 9. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method according to any of claims 1-6.

Description

Method for positioning epilepsy focus of bilateral temporal lobe epilepsy and processing device Technical Field The application relates to the technical field of intelligent biomedical signal processing, in particular to a method and a processing device for positioning a bilateral temporal lobe epilepsy induction focus. Background The temporal lobe is located the region behind the temple, and the seizure initiating site of bilateral temporal lobe epilepsy is located left and right both sides temporal lobe, because both sides of bilateral temporal lobe have pathological activity, the function and electrophysiological asymmetry between the bilateral temporal lobe induced epilepsy side and healthy side are obviously reduced, traditional collateral index is invalid, and the identification and positioning of the induced epilepsy kitchen of bilateral temporal lobe brings very big challenges. Therefore, how to overcome the above-mentioned technical problems and drawbacks becomes a major problem to be solved. Disclosure of Invention In order to overcome the problems in the prior art, the application provides a method and a processing device, which adopt the following technical scheme: in a first aspect, the present application provides a method for locating a bilateral temporal lobe epileptogenic focus, comprising: acquiring multi-modal data of bilateral temporal lobes of a patient, including first data, second data, and third data; based on the anatomical constraint of the first data, the functional connection of the second data and the electrophysiological abnormal discharge characteristics of the third data, constructing a multi-order neural topology evolution tensor; inputting the multilevel neural topology evolution tensor into a symmetrical-asymmetrical decoupling graph convolution network model to obtain an epileptogenic activity score of bilateral temporal lobes; Based on the epilepsy activity scoring graph, the functional connection betweenness centrality and the electrophysiological abnormal discharge characteristics, a core starting node in an epilepsy network is identified as an epilepsy induction range through a topological dynamics core measurement algorithm. Further, based on the anatomical constraints of the first data, the functional connections of the second data and the electrophysiological abnormal discharge characteristics of the third data, a multi-order neural topology evolution tensor is constructed, comprising: Bilateral temporal lobes segmented based on the first data as regions of interest; Calculating a functional connection matrix and a functional connection betweenness centrality of second data in the region of interest based on the region of interest; taking the cerebral cortex geometric model in the first data as priori knowledge, carrying out source positioning on the third data, and dynamically projecting abnormal electrophysiological characteristics onto the cerebral cortex geometric model of the patient; and aligning and fusing the dynamic function connection matrix, the structural feature vector and the dynamic feature graph sequence into a multi-order neural topology evolution tensor. Further, a bilateral temporal lobe segmented based on the first data as a region of interest, comprising: The method comprises the steps of extracting boundaries of gray matter, white matter, cerebrospinal fluid, exposed bones and scalp through a preset segmentation algorithm based on first data, generating a cerebral cortex geometric model, endowing conductivity parameters of each layer of medium to the cerebral cortex geometric model based on a boundary element method, generating a lead matrix through forward model calculation, and forming structural feature vectors by morphological features extracted from the first data. Further, calculating the functional connection matrix and the functional connection betweenness centrality of the second data in the region of interest based on the region of interest, comprising: Extracting denoised BOLD signal time sequences from N regions of interest defined by each first data constraint, calculating pearson correlation coefficients between all N x N node pairs, and obtaining N x N static functional connection matrixes; calculating a static function connection matrix which changes along with time through a sliding time window, and obtaining a dynamic function connection matrix; converting the dynamic function connection matrix into instantaneous network diagrams, and calculating the betweenness centrality on each instantaneous network diagram through a shortest path algorithm of graph theory. Further, taking the cortex geometric model in the first data as priori knowledge, performing source positioning on the third data, and dynamically projecting the abnormal electrophysiological characteristics onto the cortex geometric model of the patient, wherein the method comprises the following steps: Inverting the denoised electroencephalogram signals into cortex so