An epileptic seizure zone positioning system, device and medium

Abstract

The present invention provides an epilepsy attack zone positioning system, equipment and medium. The system includes a data acquisition unit: used to acquire spontaneous epilepsy data, epilepsy induced data and epilepsy imaging data; a feature extraction unit: used to extract features from spontaneous epilepsy data to acquire graph feature information; a structure construction unit: used to acquire electroencephalogram signals in epilepsy induced data, and construct graph structure information through significance test; an attack zone acquisition unit: used to cluster graph structure information and graph feature information based on an unsupervised clustering process, acquire electrode sites in the attack zone, and acquire the epilepsy attack zone according to the electrode sites in the attack zone and epilepsy imaging data. The present invention organically combines epilepsy induced data, spontaneous epilepsy data and epilepsy imaging data, and accurately and quickly obtains the patient's epilepsy attack zone electrode or brain region position through signal processing and machine learning methods, helping doctors locate the epilepsy attack zone before surgery and assisting doctors in making judgments.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to an epileptic seizure zone positioning system, equipment and medium. Background Art

[0002] Epilepsy is a chronic disease in which sudden abnormal discharges of brain neurons lead to temporary brain dysfunction. It is a neurological disease. The current method of treating epilepsy is generally to locate the epileptic seizure area and remove it through surgery.

[0003] There are few studies on the detection of epileptic seizure zone. Some researchers have used methods such as phase-locked values ​​to classify electrode sites, but this requires electrode site labels and does not consider the whole-brain connection relationship.

[0004] The relevant data of epilepsy disease include spontaneous epilepsy data and epilepsy induced data. Doctors can obtain the following data before surgery:

[0005] 1. Using stereoelectroencephalography Invasive electrodes are placed inside the patient's skull to record intracranial stereoelectroencephalogram (SEEG).

[0006] 2. Spontaneous epilepsy data: The patient's spontaneous data can be recorded based on the stereo electroencephalogram to obtain electrode electroencephalogram (EEG) signals during epileptic seizures, inter-epileptic periods (no epileptic seizures), sleep and wakefulness.

[0007] 3. Epilepsy-induced data: Cortico-cortical evoked potintial data can be obtained by actively stimulating the electrodes. By stimulating a pair of intracranial electrode sites to record the responses of the whole-brain electrodes, the effective connection between electrodes or brain regions can be measured.

[0008] Before surgery, doctors often determine the location of the epilepsy onset zone based on spontaneous epilepsy data combined with observational imaging data such as MRI and CT images, and then find the brain area that needs to be removed or coagulated by determining the electrode position in the onset zone, and then perform surgery. However, epilepsy-induced data often accounts for a small proportion of the location of the epilepsy onset zone. Therefore, the existing technology generally has the following defects:

[0009] 1. Most existing technologies only make judgments based on spontaneous epilepsy data without considering induced epilepsy data, ignoring the impact of validity connection relationship information on epilepsy.

[0010] 2. The existing technology is relatively simple in processing spontaneous epileptic EEG data, and most of them only use simple signal processing methods to process it, which often misses potential feature information.

[0011] 3. Existing technologies basically use self-supervision methods, which require manual labeling of electrode sites or manual judgment, which is time-consuming and labor-intensive, and has low processing efficiency. Due to differences in experience among clinical professionals, diagnosis becomes difficult to assess.

[0012] 4. The existing technology has weak generalization ability and requires a large number of existing samples for testing, and cannot perform targeted testing based on the actual condition of each patient.

[0013] Therefore, there is a need for an efficient and accurate epileptic onset zone positioning system that can solve the above problems. Summary of the invention

[0014] Based on the problems existing in the prior art, the present invention provides an epileptic seizure zone positioning system, device and medium. The specific scheme is as follows:

[0015] An epileptic seizure zone positioning system, comprising:

[0016] Data acquisition unit: used to acquire spontaneous epilepsy data, epilepsy induced data and epilepsy image data, wherein the epilepsy induced data includes EEG signals of electrode sites under induced stimulation;

[0017] Feature extraction unit: used for extracting features from the spontaneous epilepsy data to obtain graph feature information;

[0018] A structure construction unit: used for obtaining EEG signals of multiple electrode sites under the same induced stimulation in the epilepsy induced data, and constructing graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under the induced stimulation;

[0019] An onset zone acquisition unit is used to cluster the graph structure information and the graph feature information based on an unsupervised clustering process, acquire the onset zone electrode site, and acquire the epileptic onset zone according to the onset zone electrode site and the epilepsy image data.

[0020] In a specific embodiment, the feature extraction unit specifically includes:

[0021] Preprocessing module: used for obtaining spontaneous EEG signals from the spontaneous epilepsy data and preprocessing the spontaneous EEG signals;

[0022] Feature extraction module: used to extract features from the preprocessed spontaneous EEG signals and obtain time-frequency features;

[0023] Information acquisition module: used to process the time-frequency features based on a deep learning network to obtain the time-frequency features after dimensionality reduction, and use the time-frequency features after dimensionality reduction as the graph feature information.

[0024] In a specific embodiment, the seizure zone acquisition unit is provided with a clustering module;

[0025] Clustering module: used to construct a graph filter according to the graph structure information, and to construct a graph representation feature according to the graph feature information;

[0026] Performing feature fusion through graph convolution operation to obtain fusion features, each graph convolution includes matrix multiplication of the graph filter and the graph representation feature, and the number of graph convolutions is adaptively determined according to the intra-cluster distance;

[0027] The fusion features are clustered to obtain a plurality of clusters, and the seizure area cluster is screened to obtain the seizure area cluster, wherein the seizure area cluster only includes the seizure area electrode sites.

[0028] In a specific embodiment, the epilepsy image data includes CT images and MRI images;

[0029] The seizure area acquisition unit is also provided with a positioning module;

[0030] Positioning module: used to perform image registration on the CT image and the MRI image to obtain the first image data; determine the electrode position of the electrode site of the seizure area in the image according to the first image data; match the electrode position and the preset image template to obtain the epileptic seizure area of ​​the brain.

[0031] In a specific embodiment, the fusion features are clustered into two types to obtain a seizure area cluster and a non-seizure area cluster, wherein the seizure area cluster only includes the seizure area electrode sites, and the non-seizure area cluster only includes the non-seizure area electrode sites;

[0032] A cluster with fewer electrode sites is selected as the seizure zone cluster.

[0033] In a specific embodiment, the spontaneous data includes spontaneous EEG signals of an epileptic patient during an epileptic seizure;

[0034] The induced data include the electroencephalographic signals of all the electrode sites in the skull under the stimulation when a pair of electrode sites are stimulated.

[0035] In a specific embodiment, the preprocessing includes removing artifacts, removing power frequency interference, filtering, downsampling, and de-emphasis.

[0036] In a specific embodiment, the feature extraction module extracts features from the preprocessed spontaneous EEG signal through Fourier transform or wavelet transform to obtain time-frequency features.

[0037] In a specific embodiment, the process of constructing the graph filter includes:

[0038] Obtaining an adjacency matrix from the graph structure information;

[0039] Performing Laplace transformation on the adjacency matrix to construct a normalized Laplace matrix;

[0040] A graph filter is constructed according to the Laplacian matrix.

[0041] In a specific embodiment, the expression of the Laplacian matrix is:

[0042]

[0043] in, represents the Laplace matrix, D represents the degree matrix, and A represents the adjacency matrix;

[0044] The expression of the graph filter is:

[0045]

[0046] Among them, G represents the graph filter, I represents the unit matrix, represents the Laplacian matrix.

[0047] In a specific embodiment, it is characterized in that the deep learning network includes an autoencoder model, and the information acquisition module processes the time-frequency features based on the autoencoder model.

[0048] In a specific embodiment, the process of obtaining the graph feature information specifically includes:

[0049] Step 1: construct a feature vector according to the time-frequency feature;

[0050] Step 2, performing dimensionality reduction and feature learning through the encoding part in the autoencoder model to obtain a low-dimensional hidden feature representation of an intermediate state;

[0051] Step 3, converting the low-dimensional hidden feature representation into the feature vector through a decoding part in the autoencoder model;

[0052] Step 4: Adjust network parameters based on the loss function;

[0053] Iterate the above steps 1 to 4 until the iteration stop condition is met, and use the low-dimensional feature representation that meets the iteration stop condition as the graph feature information.

[0054] A computer device, comprising:

[0055] one or more processors;

[0056] A memory for storing one or more programs;

[0057] When the one or more programs are executed by the one or more processors, the one or more processors implement the following processing:

[0058] The data acquisition unit acquires spontaneous epilepsy data, epilepsy induced data and epilepsy imaging data;

[0059] The feature extraction unit extracts features from the spontaneous epilepsy data to obtain graph feature information;

[0060] The structure construction unit obtains electroencephalogram signals of multiple electrode sites under the same induced stimulation in the epilepsy induced data, and constructs graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under the induced stimulation;

[0061] The seizure zone acquisition unit clusters the graph structure information and the graph feature information based on an unsupervised clustering process to acquire the seizure zone electrode sites, and acquires the epileptic seizure zone according to the seizure zone electrode sites and the epilepsy image data.

[0062] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following processing:

[0063] The data acquisition unit acquires spontaneous epilepsy data, epilepsy induced data and epilepsy imaging data;

[0064] The feature extraction unit extracts features from the spontaneous epilepsy data to obtain graph feature information;

[0065] The structure construction unit obtains electroencephalogram signals of multiple electrode sites under the same induced stimulation in the epilepsy induced data, and constructs graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under the induced stimulation;

[0066] The seizure zone acquisition unit clusters the graph structure information and the graph feature information based on an unsupervised clustering process to acquire the seizure zone electrode sites, and acquires the epileptic seizure zone according to the seizure zone electrode sites and the epilepsy image data.

[0067] The present invention has the following beneficial effects:

[0068] In response to the existing technology, the present invention provides an epilepsy onset zone positioning system, equipment and medium, which organically combine epilepsy induced data, spontaneous epilepsy data and epilepsy imaging data, and accurately and quickly obtain the epilepsy onset zone electrode or brain area position for different patients through signal processing and machine learning methods, helping doctors locate the epilepsy onset zone before surgery and assisting doctors in making judgments.

[0069] The characteristic relationship is represented by spontaneous epilepsy data, and the structural connection relationship is represented by epilepsy-induced data. The epileptic onset area is located and judged by combining the characteristic relationship and the structural connection relationship. Compared with the existing technology, the impact of the whole brain connection relationship on epilepsy is taken into account, and it is more comprehensive and accurate.

[0070] By integrating signal processing, deep learning, and machine learning through feature extraction and characterization, we can discover potential features in the data and obtain more accurate detection results.

[0071] Compared with supervised and semi-supervised learning, the unsupervised clustering method does not require time-consuming training, nor does it require the use of a large number of manually labeled samples for pre-training or supervised learning, which shortens the detection time and improves the detection efficiency.

[0072] Since the model is an adaptive graph convolution, it can adaptively determine the number of convolution iterations, and thus can perform targeted detection based on the actual conditions of different patients, with strong generalization ability and high detection accuracy.

[0073] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0075] Figure 1 is a structural diagram of an epilepsy onset zone positioning system according to Embodiment 1 of the present invention;

[0076] Figure 2 Schematic diagram of the principle of the epilepsy onset zone positioning system according to Embodiment 1 of the present invention;

[0077] Figure 3 is a schematic diagram of epilepsy inducing data electrodes according to Embodiment 1 of the present invention;

[0078] Figure 4 is a schematic diagram of epilepsy induction data electrode induction according to embodiment 1 of the present invention;

[0079] Figure 5 is a feature extraction flow chart of Embodiment 1 of the present invention;

[0080] Figure 6 is a schematic diagram of an automatic encoder model according to Embodiment 1 of the present invention;

[0081] Figure 7is an electrode site adjacency matrix diagram of Example 1 of the present invention;

[0082] Figure 8 is a clustering flow chart of Embodiment 1 of the present invention;

[0083] Fig. 9 is a flowchart of the adaptive graph convolution of embodiment 1 of the present invention;

[0084] Fig.10 is a diagram of experimental results of the epilepsy onset zone positioning system according to Example 1 of the present invention;

[0085] Fig.11 It is a schematic diagram of the structure of a computer device according to Embodiment 2 of the present invention.

[0086] Reference numerals:

[0087] 1-data acquisition unit; 2-feature extraction unit; 3-structure construction unit; 4-onset area acquisition unit; 21-preprocessing module; 22-feature extraction module; 23-information acquisition module; 41-clustering module; 42-positioning module; 12-computer device; 14-external device; 16-processing unit; 18-bus; 28-system memory. DETAILED DESCRIPTION

[0088] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0089] The present invention studies the connection relationship of the epilepsy network through CCEP induction points and constructs graph structure information. Graph feature information is obtained by extracting EEG feature information from spontaneous epilepsy data. The graph feature information and graph structure information are combined, and unsupervised clustering is performed using an adaptive graph convolution method based on machine learning to obtain two categories: epilepsy onset area and epilepsy non-onset area, which helps doctors locate the epilepsy onset area before surgery and assists doctors in making judgments.

[0090] Compared with traditional clustering algorithms (such as Kmeans and spectral clustering) that only use node features for clustering, the adaptive unsupervised clustering provided by the present invention is a new type of clustering algorithm that is suitable for attribute graphs and uses the edge connection relationship in the graph structure as a structural feature to aggregate the features of each node and its k-order neighbors through graph convolution operations to obtain better feature representation.

[0091] The present invention provides an epileptic seizure zone positioning system, device and medium. The epileptic seizure zone is positioned by organically combining spontaneous epilepsy data and epilepsy induced data, using spontaneous data to characterize feature relationships, and using induced data to characterize structural connection relationships, fully considering the impact of brain connection relationships on epilepsy. By extracting features through signal processing methods and deep learning methods, potential features can be discovered to achieve better detection results. Using the adaptive graph convolution method, there is no need for pre-training or supervised learning with manually labeled samples, nor is there any need for training and learning on a large number of existing samples, which greatly reduces labor costs, shortens detection time, and improves detection efficiency. Since the model is an adaptive graph convolution, the number of convolution iterations can be adaptively determined, so that targeted detection can be performed according to different patients, and the generalization ability is strong.

[0092] Example 1

[0093] This embodiment proposes a system for locating epileptic seizure areas. The module diagram of the system is shown in the attached manual. Figure 1 As shown in the schematic diagram, Figure 2 The specific scheme is as follows:

[0094] An epileptic seizure zone positioning system comprises a data acquisition unit 1, a feature extraction unit 2, a structure construction unit 3 and an epileptic seizure zone acquisition unit 4. The data acquisition unit 1 is connected to the feature extraction unit 2 and the structure construction unit 3 respectively, and the epileptic seizure zone acquisition unit 4 is connected to the feature extraction unit 2 and the structure construction unit 3 respectively.

[0095] Data acquisition unit 1: used to acquire spontaneous epilepsy data, epilepsy induced data and epilepsy imaging data, wherein the epilepsy induced data includes the EEG signal of the electrode site under induced stimulation;

[0096] Feature extraction unit 2: used for extracting features from spontaneous epileptic data to obtain graph feature information;

[0097] Structure construction unit 3: used to obtain EEG signals of multiple electrode sites under the same induced stimulation in epilepsy induced data, and construct graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under induced stimulation;

[0098] The seizure zone acquisition unit 4 is used to cluster the graph structure information and graph feature information based on an unsupervised clustering process, obtain the electrode site of the seizure zone, and obtain the epileptic seizure zone according to the electrode site of the seizure zone and the epilepsy imaging data.

[0099] Specifically, data acquisition unit 1: is used to acquire epilepsy imaging data, spontaneous epilepsy data and epilepsy induced data. Among them, epilepsy imaging data include CT images and MRI images, and spontaneous epilepsy data are spontaneous EEG signal data collected by the hospital from patients during epileptic seizures and during epileptic seizures (including awake and asleep states), among which the data during epileptic seizures are the most important. CCEP (Cortico-cortical evoked potential) is called the cortical-cortical evoked potential. By stimulating a pair of electrode sites, the EEG signals of all electrode sites in the skull under this induced stimulation are obtained. After stimulating all electrode pairs, an induced brain connection relationship can be obtained, and effective and functional connections can be explored. Brain electrodes are arranged in pairs, and the EEG responses of all electrode sites are obtained by stimulating a pair of electrodes. As shown in the attached manual Figure 3 As shown, when the stimulus is Figure 3 When a pair of electrode sites are Figure 4 The EEG responses of all electrode sites.

[0100] Feature extraction unit 2: used to extract features from spontaneous epileptic data to obtain graph feature information. The feature extraction unit 2 specifically includes a preprocessing module 21, a feature extraction module 22 and an information acquisition module 23.

[0101] Preprocessing module 21: used to obtain spontaneous EEG signals from spontaneous epilepsy data and preprocess the spontaneous EEG signals;

[0102] Feature extraction module 22: used to extract features from the preprocessed spontaneous EEG signals to obtain time-frequency features;

[0103] Information acquisition module 23: used to process the time-frequency features based on the deep learning network to obtain the time-frequency features after dimensionality reduction, and use the time-frequency features after dimensionality reduction as graph feature information.

[0104] The specific feature extraction process is as shown in the attached manual. Figure 5 As shown. First, the preprocessing module is required to preprocess the original spontaneous EEG signal. The preprocessing includes signal processing methods such as removing artifacts, removing power frequency interference, filtering, downsampling, and debulking. After obtaining the preprocessed signal, the feature extraction module extracts signal features through Fourier transform or wavelet transform to obtain time-frequency features. For example, the wavelet power spectrum can be used to obtain various power features in different frequency bands. The information acquisition module uses the deep learning-based autoencoder model and other methods to obtain the time-frequency features after dimensionality reduction through an unsupervised method, and uses the time-frequency features as the graph feature information.

[0105] The acquisition process of graph feature information specifically includes: constructing a feature vector based on time-frequency features; performing dimensionality reduction and feature learning through the encoding part of the autoencoder model to obtain a low-dimensional hidden feature representation of an intermediate state; converting the low-dimensional hidden feature representation into a feature vector through the decoding part of the autoencoder model; adjusting the network parameters based on the loss function; iterating the above steps until the iteration stop condition is met, and using the low-dimensional feature representation that meets the iteration stop condition as the graph feature information. Specifically, based on the original recorded spontaneous signal, the time domain features can be obtained after preprocessing operations, or the time-frequency features can be obtained by using Fourier transform and wavelet transform. Based on these features, a feature vector that can represent the original waveform is constructed. After that, the encoding part of the autoencoder model is used to reduce the dimension and learn features. After the encoding process, a low-dimensional hidden feature representation of an intermediate state will be obtained. Then, this low-dimensional feature representation is converted back to the original feature vector through the decoding process, and the network parameters are adjusted through the loss function. Finally, after multiple rounds of iterations, the input features and output features obtained are close, and the low-dimensional feature representation of the intermediate state can be used as the graph feature information.

[0106] The autoencoder model based on deep learning is used at the end of the feature extraction process. This model is an unsupervised feature extraction method, which is consistent with the unsupervised clustering process, making the entire system an unsupervised learning process. Taking the autoencoder model as an example, the unsupervised feature extraction process is introduced: the autoencoder model (Autoencoder) is a neural network model whose goal is to copy input features to output. By compressing the input features to a low-dimensional latent space feature representation, and then reconstructing and increasing the dimensionality of this low-dimensional feature representation, so that the reconstructed output features are closer to the input features, it is believed that the low-dimensional latent space feature representation can represent the original high-dimensional features.

[0107] The network structure is as shown in the instruction manual. Figure 6 As shown in the figure, each layer can be connected with a fully connected layer, and activation functions such as ReLu and Sigmoid are added in the middle to realize nonlinear transformation. The loss function can be selected as mean square error, etc. The training process is realized by setting the loss function to calculate the distance between the input feature and the output feature. Generally, it can be realized using common deep learning frameworks such as TensorFlow or Pytorch.

[0108] The autoencoder model or its variants, such as the variable autoencoder model, can be used as a feature extraction method.

[0109] Structural construction unit 3: used to obtain EEG signals of multiple electrode sites under the same induced stimulation in epilepsy induced data, and construct graph structure information through significance test, the graph structure information includes an adjacency matrix graph of brain connection relationship under induced stimulation.

[0110] Based on the principle of obtaining epilepsy-induced data, the recorded induced EEG signals are generally recorded repeatedly. Epilepsy-induced data include the EEG signals of all intracranial electrode sites under the induced stimulation obtained when a pair of electrode sites are stimulated. Among the repeatedly recorded EEG signals, the EEG signals under the same induced stimulation are screened out, that is, the EEG signals under the stimulation of the same electrode pair. The structure construction unit 3 selects the baseline and EEG response for significance test for the screened EEG signals, and selects the part whose significance level meets the preset conditions to obtain the adjacency matrix diagram of the brain connection relationship under the induced stimulation. The matrix diagram is the graph structure information. The electrode site adjacency matrix diagram is as attached to the instruction manual. Figure 7 As shown, the vertical axis represents the induced stimulation, that is, the electrode pair that gives stimulation, the horizontal axis represents all the response electrodes, and different grayscales represent different significance levels.

[0111] A significance test is to make an assumption about the parameters or distribution of the population (random variable) in advance, and then use the sample information to determine whether the assumption (alternative hypothesis) is reasonable, that is, to determine whether there is a significant difference between the true situation of the population and the original hypothesis. In other words, the significance test is to determine whether the difference between the sample and the hypothesis is purely due to chance variation, or caused by the inconsistency between the hypothesis and the true situation of the population. The significance test is to test the assumptions made for the population, and its principle is to accept or deny the hypothesis based on the "principle of actual impossibility of small probability events."

[0112] Seizure area acquisition unit 4: clusters the graph structure information and graph feature information based on the unsupervised clustering process, obtains the electrode site of the seizure area, and obtains the epileptic seizure area according to the electrode site of the seizure area and the preset image data. In the unsupervised clustering process, an adaptive graph convolution method is used to combine the graph structure signal and the graph feature information for processing. Compared with self-supervised learning and semi-supervised learning, it does not require a large number of samples annotated by doctors for pre-training or supervised learning. Seizure area acquisition unit 4 includes a clustering module 41 and a positioning module 42.

[0113] The clustering module 41 specifically includes: constructing a graph filter through Laplace transform of graph structure information, and constructing graph representation features according to graph feature information; performing feature fusion through graph convolution operation to obtain fusion features, each graph convolution includes matrix multiplication of the graph filter and the graph representation features, and the number of graph convolutions is adaptively determined according to the distance within the cluster; clustering is performed according to the fusion features to obtain multiple clusters, and the electrode sites of the seizure area are screened.

[0114] The graph structure information is obtained from epilepsy induced data, which contains a large amount of information about electrode sites. Each electrode site is regarded as a node in the graph structure. Each node has its own feature, which is associated with the graph feature information. The feature information of all nodes together constitutes the complete graph feature information. The adjacency matrix obtained from epilepsy induced data reflects the connection relationship of all nodes.

[0115] The adaptive unsupervised clustering provided in this embodiment is a new type of clustering algorithm, which is applicable to attribute graphs and uses the edge connection relationship in the graph structure as a structural feature to aggregate the features of each node and its k-order neighbors through graph convolution operations to obtain better feature representation.

[0116] Given a property graph ,in is a set of nodes, E is a set of edges, forming an adjacency matrix ,represent and There are undirected edges between them, and the attribute characteristics of all nodes in the graph are , represents the attributes of each node. The goal is to cluster all nodes into m clusters, and the set of clusters is C={C1,...,Cm}.

[0117] First, we need to construct the adjacency matrix A into a normalized Laplace matrix , and then construct the graph filter .in, represents the Laplace matrix, D represents the degree matrix, A represents the adjacency matrix, I represents the identity matrix, represents the Laplacian matrix. The process of a graph convolution is the process of a matrix multiplication: X̂=GX, and the process of k-order graph convolution is , the process is as attached in the instruction manual Figure 8 shown.

[0118] In the adaptive graph convolution model, the graph structure information is used to construct a graph filter in the form of Laplace transform, and the graph representation feature is constructed according to the graph feature information. The multiplication of the two matrices of the graph filter and the graph representation feature is regarded as a graph convolution process, and an adaptive graph convolution operation is performed to blend the two features to obtain a fusion feature, where the number of convolutions is adaptively determined according to the distance within the cluster. Clustering is performed based on the fusion features to obtain multiple clusters, each of which contains multiple nodes. The seizure area cluster is screened out from the multiple clusters, and the seizure area cluster only includes the electrode sites of the seizure area. The corresponding epileptic seizure area is obtained according to the electrode sites of the seizure area. It should be noted that due to different clustering models, multiple clusters may be generated, and the seizure area cluster also includes multiple sub-clusters. Since the model is an adaptive graph convolution, the number of convolution iterations can be adaptively determined, so it can perform targeted detection according to the conditions of different patients, and has strong generalization.

[0119] The process of adaptive graph convolution is as shown in the attached manual. Fig. 9 As shown, define the maximum number of iterations max_iter, define t as the number of loops, That is, the sum of the intra-cluster distances of all clusters. The process of adaptive convolution is to decide whether to jump out of the loop by comparing the intra-cluster distances in each iteration. When the loop is jumped out, the current result is considered to be optimal. At the same time, the value of k obtained is the number of convolutions of the graph convolution. Finally, the k-order graph signal is obtained. , and then obtain the clustering result through spectral clustering. Iterate according to the change of distance within the cluster, and perform a convolution in each iteration to obtain the characteristics of the node's k-order neighbors.

[0120] In the two-category clustering mode, two clusters are obtained by clustering, namely the seizure area cluster and the non-seizure area cluster. Clustering based on fusion features can only obtain two clusters, avoiding the screening process. Among them, the seizure area cluster only includes the seizure area electrode sites, and the non-seizure area cluster only includes the non-seizure area electrode sites, and the position of the seizure area electrode sites in the image corresponds to the epileptic seizure area. Since each cluster contains multiple nodes, each node corresponds to an electrode site, therefore, the seizure area cluster and the non-seizure area cluster here are both electrode site sets, and the set contains multiple electrode sites. The seizure area cluster contains one or more seizure area electrode sites, and the non-seizure area contains one or more non-seizure area electrode sites.

[0121] In the multi-class clustering mode, clustering results in more than two clusters. For example, the categories are divided into multiple areas such as epilepsy onset area, rapid spread area, irritation area, non-epileptic area, etc. The category selection can be specific according to the actual situation.

[0122] After obtaining the electrode sites in the onset area, it is necessary to perform electrode positioning. The positioning module 42 combines the epilepsy image data and the electrode sites in the onset area to obtain the positions of the electrode sites in the onset area, and then obtains the epilepsy onset area of ​​the brain in combination with the preset image template. The positioning module specifically includes: obtaining CT images and MRI images of epilepsy patients, and performing image registration on the CT images and MRI images to obtain first image data. The electrode position of the electrode sites in the onset area in the image is determined according to the first image data; the electrode position is matched with the preset image template to obtain the epilepsy onset area of ​​the brain.

[0123] The results of epilepsy onset zone detection are as attached in the instruction manual. Fig.10As shown. Among them, the two circular areas of different depths correspond to the two electrode sites. The depth of the electrode is used to determine whether it is in the seizure area. The shallower electrode corresponds to the non-seizure area, and the deeper electrode corresponds to the seizure area. After experimental verification, the system clustering results were compared with the annotations given by the hospital, and better results were obtained, which proved the applicability and reliability of the system provided in this embodiment. According to the electrode brain area correspondences marked by the doctor, the posterior parahippocampal gyrus, temporal lobe, temporal polar plane and other brain areas are brain areas related to the seizure area. According to the results obtained by the system, the electrodes and brain areas are matched, and the final result is the posterior parahippocampal gyrus, temporal lobe, temporal polar plane, inferior frontal gyrus and other areas, which basically include the seizure area to be removed by the actual surgery, proving the practicality and usability of this system.

[0124] The present embodiment provides an epilepsy onset area positioning system, which organically combines epilepsy-induced data and spontaneous epilepsy data, characterizes feature relationships through spontaneous epilepsy data, characterizes structural connection relationships through epilepsy-induced data, and locates and determines the epilepsy onset area by integrating feature relationships and structural connection relationships, which is highly comprehensive and accurate. By integrating signal processing, deep learning, and machine learning through feature extraction and characterization methods, it is possible to discover potential features in the data and obtain more accurate detection results. Compared with supervised and semi-supervised learning, the use of unsupervised clustering methods does not require time-consuming training, nor does it require the use of a large number of manually labeled samples for pre-training or supervised learning, which shortens the detection time and improves the detection efficiency. Since the model is an adaptive graph convolution, the number of convolution iterations can be adaptively determined, so that targeted detection can be performed according to the actual conditions of different patients, with strong generalization ability and high detection accuracy.

[0125] Example 2

[0126] Instructions attached Fig.11 This is a schematic diagram of the structure of a computer device provided in Example 2 of the present invention. Fig.11 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0127] As the instruction manual Fig.11 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 connecting different system components (including the system memory 28 and the processing unit 16). The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the device computer 12, including volatile and non-volatile media, removable and non-removable media. The system memory 28 may include computer system readable media in the form of volatile memory.

[0128] The computer device 12 may also communicate with one or more external devices 14 (e.g., a keyboard, a pointing device, a display, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices.

[0129] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing a control method of an epileptic seizure zone positioning system, the method comprising:

[0130] 101. The data acquisition unit 1 acquires spontaneous epilepsy data, epilepsy induced data and epilepsy image data; the epilepsy induced data includes the electroencephalogram signal under induced stimulation.

[0131] 102. The feature extraction unit 2 extracts features from spontaneous epileptic data to obtain graph feature information.

[0132] 103. The structure construction unit 3 obtains EEG signals of multiple electrode sites under the same induced stimulation in the epilepsy induced data, and constructs graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under the induced stimulation.

[0133] 104. The seizure zone acquisition unit 4 clusters the graph structure information and the graph feature information based on an unsupervised clustering process, acquires the seizure zone electrode sites, and acquires the epileptic seizure zone according to the seizure zone electrode sites and the epilepsy imaging data.

[0134] Among them, step 102 and step 103 can be performed simultaneously or sequentially.

[0135] This embodiment applies a control method for an epilepsy onset zone locating system to a specific computer device, and stores the method in a memory. When an executor executes the memory, the method is run to locate the epilepsy onset zone. The method is quick and convenient to use and has a wide range of applications.

[0136] Of course, those skilled in the art can understand that the processor can also implement the technical solution provided by any embodiment of the present invention.

[0137] Example 3

[0138] This embodiment 3 provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, a control method for an epilepsy onset zone locating system is implemented. The method includes:

[0139] 101. The data acquisition unit 1 acquires spontaneous epilepsy data, epilepsy induced data and epilepsy image data; the epilepsy induced data includes the electroencephalogram signal under induced stimulation.

[0140] 102. The feature extraction unit 2 extracts features from spontaneous epileptic data to obtain graph feature information.

[0141] 103. The structure construction unit 3 obtains EEG signals of multiple electrode sites under the same induced stimulation in the epilepsy induced data, and constructs graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under the induced stimulation.

[0142] 104. The seizure zone acquisition unit 4 clusters the graph structure information and the graph feature information based on an unsupervised clustering process, acquires the seizure zone electrode sites, and acquires the epileptic seizure zone according to the seizure zone electrode sites and the epilepsy imaging data.

[0143] Among them, step 102 and step 103 can be performed simultaneously or sequentially.

[0144] The computer storage medium of the present embodiment can adopt any combination of one or more computer-readable media. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable storage media can be, for example, but not limited to: electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this document, computer-readable storage media can be any tangible medium containing or storing programs, which can be used by instruction execution systems, devices or devices or used in combination with them.

[0145] This embodiment applies a control method for an epilepsy onset zone locating system to a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the control method for an epilepsy onset zone locating system provided by the present invention are implemented, which is simple, fast, easy to store, and not easy to lose.

[0146] The present invention provides an epilepsy attack area positioning system, equipment and medium for the prior art, which organically combines epilepsy induced data, spontaneous epilepsy data and epilepsy image data, and accurately and quickly obtains the epilepsy attack area electrode or brain area position for different patients through signal processing and machine learning methods, helping doctors to locate the epilepsy attack area before surgery and assisting doctors to make judgments. The epilepsy spontaneous data characterizes the feature relationship, the epilepsy induced data characterizes the structural connection relationship, and the epilepsy attack area is positioned and judged by integrating the feature relationship and the structural connection relationship. Compared with the prior art, the influence of the whole brain connection relationship on epilepsy is considered, which is comprehensive and accurate. By feature extraction and characterization means, signal processing, deep learning and machine learning are integrated to discover potential features in the data, thereby obtaining more accurate detection effects. Using unsupervised clustering methods, compared with supervised and semi-supervised learning, there is no need for time-consuming training, and there is no need to use a large number of manually labeled samples for pre-training or supervised learning, which shortens the detection time and improves the detection efficiency. Since the model is an adaptive graph convolution, the number of convolution iterations can be adaptively judged, so that targeted detection can be performed according to the actual condition of different patients, with strong generalization ability and high detection accuracy.

[0147] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0148] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

[0149] The above disclosure is only a few specific implementation scenarios of the present invention, but the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. An epilepsy onset area positioning system, characterized in that: include, Data acquisition unit: used to acquire spontaneous epilepsy data, epilepsy induced data and epilepsy image data, wherein the epilepsy induced data includes EEG signals of electrode sites under induced stimulation; Feature extraction unit: used for extracting features from the spontaneous epilepsy data to obtain graph feature information; The feature extraction unit specifically includes: Preprocessing module: used for obtaining spontaneous EEG signals from the spontaneous epilepsy data and preprocessing the spontaneous EEG signals; Feature extraction module: used to extract features from the preprocessed spontaneous EEG signals and obtain time-frequency features; Information acquisition module: used for processing the time-frequency features based on a deep learning network to obtain the time-frequency features after dimensionality reduction, and using the time-frequency features after dimensionality reduction as the graph feature information; A structure construction unit: used for obtaining EEG signals of multiple electrode sites under the same induced stimulation in the epilepsy induced data, and constructing graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under the induced stimulation; An onset zone acquisition unit is used to cluster the graph structure information and the graph feature information based on an unsupervised clustering process, acquire the electrode site of the onset zone, and acquire the epileptic onset zone according to the electrode site of the onset zone and the epilepsy image data; The seizure area acquisition unit is provided with a clustering module; Clustering module: used to construct a graph filter according to the graph structure information, and to construct a graph representation feature according to the graph feature information; Performing feature fusion through graph convolution operation to obtain fusion features, each graph convolution includes matrix multiplication of the graph filter and the graph representation feature, and the number of graph convolutions is adaptively determined according to the intra-cluster distance; Clustering the fusion features to obtain multiple clusters, screening to obtain a seizure area cluster, wherein the seizure area cluster only includes the seizure area electrode sites; The epilepsy image data includes CT images and MRI images; The seizure area acquisition unit is also provided with a positioning module; Positioning module: used to perform image registration on the CT image and the MRI image to obtain the first image data; determine the electrode position of the electrode site of the seizure area in the image according to the first image data; match the electrode position and the preset image template to obtain the epileptic seizure area of ​​the brain.

2. The epilepsy onset zone positioning system according to claim 1, characterized in that: Perform two-type clustering on the fusion features to obtain an onset area cluster and a non-onset area cluster, wherein the onset area cluster only includes the onset area electrode sites, and the non-onset area cluster only includes the non-onset area electrode sites; A cluster with fewer electrode sites is selected as the seizure zone cluster.

3. The epilepsy onset zone positioning system according to claim 1, characterized in that: The spontaneous data include spontaneous EEG signals of epileptic patients during and after an epileptic seizure; The induced data include the electroencephalographic signals of all the electrode sites in the skull under the stimulation when a pair of electrode sites are stimulated.

4. The epilepsy onset zone positioning system according to claim 1, characterized in that: The preprocessing includes removing artifacts, removing power frequency interference, filtering, downsampling and castration.

5. The epilepsy onset zone positioning system according to claim 4, characterized in that: The feature extraction module extracts features from the preprocessed spontaneous EEG signal through Fourier transform or wavelet transform to obtain time-frequency features.

6. The epilepsy onset zone positioning system according to claim 1, characterized in that: The construction process of the graph filter includes: Obtaining an adjacency matrix from the graph structure information; Performing Laplace transformation on the adjacency matrix to construct a normalized Laplace matrix; A graph filter is constructed according to the Laplacian matrix.

7. The epilepsy onset zone positioning system according to claim 6, characterized in that: The expression of the Laplacian matrix is in, represents the Laplace matrix, D represents the degree matrix, and A represents the adjacency matrix; The expression of the graph filter is: Among them, G represents the graph filter, I represents the unit matrix, represents the Laplacian matrix.

8. The epilepsy onset zone positioning system according to claim 1, characterized in that: The deep learning network includes an autoencoder model, and the information acquisition module processes the time-frequency features based on the autoencoder model.

9. The epilepsy onset zone positioning system according to claim 8, characterized in that: The process of obtaining the graph feature information specifically includes: Step 1: construct a feature vector according to the time-frequency feature; Step 2, performing dimensionality reduction and feature learning through the encoding part in the autoencoder model to obtain a low-dimensional hidden feature representation of an intermediate state; Step 3, converting the low-dimensional hidden feature representation into the feature vector through a decoding part in the autoencoder model; Step 4: Adjust network parameters based on the loss function; Iterate the above steps 1 to 4 until the iteration stop condition is met, and use the low-dimensional feature representation that meets the iteration stop condition as the graph feature information.

10. A computer device, characterized in that: The computer device comprises: one or more processors; A memory for storing 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 following processing: The data acquisition unit acquires spontaneous epilepsy data, epilepsy induced data and epilepsy imaging data; The feature extraction unit extracts features from the spontaneous epilepsy data to obtain graph feature information; The feature extraction unit specifically includes: Preprocessing module: used for obtaining spontaneous EEG signals from the spontaneous epilepsy data and preprocessing the spontaneous EEG signals; Feature extraction module: used to extract features from the preprocessed spontaneous EEG signals and obtain time-frequency features; Information acquisition module: used for processing the time-frequency features based on a deep learning network to obtain the time-frequency features after dimensionality reduction, and using the time-frequency features after dimensionality reduction as the graph feature information; The structure construction unit obtains electroencephalogram signals of multiple electrode sites under the same induced stimulation in the epilepsy induced data, and constructs graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under the induced stimulation; The seizure area acquisition unit clusters the graph structure information and the graph feature information based on an unsupervised clustering process to acquire the electrode site of the seizure area, and acquires the epilepsy seizure area according to the electrode site of the seizure area and the epilepsy image data; The seizure area acquisition unit is provided with a clustering module; Clustering module: used to construct a graph filter according to the graph structure information, and to construct a graph representation feature according to the graph feature information; Performing feature fusion through graph convolution operation to obtain fusion features, each graph convolution includes matrix multiplication of the graph filter and the graph representation feature, and the number of graph convolutions is adaptively determined according to the intra-cluster distance; Clustering the fusion features to obtain multiple clusters, screening to obtain a seizure area cluster, wherein the seizure area cluster only includes the seizure area electrode sites; The epilepsy image data includes CT images and MRI images; The seizure area acquisition unit is also provided with a positioning module; Positioning module: used to perform image registration on the CT image and the MRI image to obtain the first image data; determine the electrode position of the electrode site of the seizure area in the image according to the first image data; match the electrode position and the preset image template to obtain the epileptic seizure area of ​​the brain.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, the following processing is performed: The data acquisition unit acquires spontaneous epilepsy data, epilepsy induced data and epilepsy imaging data; The feature extraction unit extracts features from the spontaneous epilepsy data to obtain graph feature information; The feature extraction unit specifically includes: Preprocessing module: used for obtaining spontaneous EEG signals from the spontaneous epilepsy data and preprocessing the spontaneous EEG signals; Feature extraction module: used to extract features from the preprocessed spontaneous EEG signals and obtain time-frequency features; Information acquisition module: used for processing the time-frequency features based on a deep learning network to obtain the time-frequency features after dimensionality reduction, and using the time-frequency features after dimensionality reduction as the graph feature information; The structure construction unit obtains electroencephalogram signals of multiple electrode sites under the same induced stimulation in the epilepsy induced data, and constructs graph structure information through significance test, wherein the graph structure information includes an adjacency matrix graph of brain connection relationship under the induced stimulation; The seizure area acquisition unit clusters the graph structure information and the graph feature information based on an unsupervised clustering process to acquire the electrode site of the seizure area, and acquires the epilepsy seizure area according to the electrode site of the seizure area and the epilepsy image data; The seizure area acquisition unit is provided with a clustering module; Clustering module: used to construct a graph filter according to the graph structure information, and to construct a graph representation feature according to the graph feature information; Performing feature fusion through graph convolution operation to obtain fusion features, each graph convolution includes matrix multiplication of the graph filter and the graph representation feature, and the number of graph convolutions is adaptively determined according to the intra-cluster distance; Clustering the fusion features to obtain multiple clusters, screening to obtain a seizure area cluster, wherein the seizure area cluster only includes the seizure area electrode sites; The epilepsy image data includes CT images and MRI images; The seizure area acquisition unit is also provided with a positioning module; Positioning module: used to perform image registration on the CT image and the MRI image to obtain the first image data; determine the electrode position of the electrode site of the seizure area in the image according to the first image data; match the electrode position and the preset image template to obtain the epileptic seizure area of ​​the brain.

Images