A scalp electroencephalogram-based epilepsy lesion positioning system

Abstract

The application aims to provide a scalp EEG-based epilepsy lesion positioning system, which is based on Granger causality theory, establishes brain function networks of patients in interictal period and ictal period respectively, comprehensively uses effective information in different periods, analyzes the connectivity difference between the initial stage of seizure and the interictal period from the network level, and analyzes the connectivity difference between the lesion area and the non-lesion area from the node level; adopts a difference quantification method to quantify the difference size of the causal flow of each lead between the interictal period and the initial stage of seizure, and sequentially identifies the seizure main frequency band, the seizure side and the seizure lead based on the difference size, so as to realize the lesion positioning and side identification of epilepsy patients.

Description

A scalp EEG-based epilepsy focus localization system Technical Field

[0001] This invention belongs to the field of medical electronic technology, specifically relating to an epilepsy lesion localization system based on scalp electroencephalography. Background Technology

[0002] Epilepsy is a chronic neurological disorder caused by abnormal electrical activity in groups of neurons in the brain, affecting more than 1% of the global population. Epileptic seizures are sudden and recurrent, severely damaging not only the patient's physical health but also their mental well-being, significantly reducing their quality of life. For refractory focal epilepsy, preoperative evaluation is necessary to determine the location and extent of the epileptic focus before lesion resection surgery. Multichannel electroencephalography (EEG) is crucial for studying the mechanisms of epilepsy development and spread, providing a reference for preoperative evaluation, identifying key points in epilepsy transmission, and predicting surgical outcomes.

[0003] In recent years, numerous studies have focused on using signal analysis and processing techniques to analyze electroencephalogram (EEG) signals to improve the accuracy of epileptic focus localization. A significant portion of these studies has concentrated on quantitatively characterizing the interactions and causal relationships between neuronal populations. Epilepsy is a result of brain network dysfunction; therefore, measuring changes in brain state through alterations in connectivity between different regions is more consistent with the principles of epileptic seizures. Brain network functional connectivity allows for the spatial study of causal relationships between different neurophysiological events, contributing to a deeper understanding of the interactions between various brain regions.

[0004] Functional connectivity analysis based on Granger causality measurements is an effective method for constructing brain functional networks and localizing epileptic foci. In 2017, Ioannis introduced the concept of effective inflow, building upon Partial Directed Coherence (PDC) and Directed Transfer Function (DTF), and used General Partial Directed Coherence (GPDC) and Directed Connectivity (DC) to measure the connectivity of different brain regions. Peter analyzed intracranial EEG during seizures by calculating Adaptive Partial Directed Coherence (APDC) and Adaptive Directed Transfer Function (ADTF), constructed a brain functional network matrix, and analyzed the differences between the brain functional network matrices of seizure and non-seizure areas based on graph theory measurement methods. In 2018, he further analyzed the impact of time series normalization and the number of leads used to construct the brain functional network on localization performance. The study showed that time series normalization can significantly improve localization accuracy, while increasing the number of leads has little effect on localization performance. Based on GPDC, Adkinson A calculated graph theory measurement parameters such as degree centrality and eigenvector centrality, and focused on analyzing the connectivity and centrality of brain functional networks.

[0005] (1) Current research on epileptic focus localization mainly focuses on the ictal phase and preictal phase, because abnormalities in the focus area are easier to detect during an attack. However, the mechanism of epileptic seizures is very complex, and abnormal discharges may also be present in leads during the interictal phase. Existing studies have not fully considered the relevant information between the interictal and ictal phases, thus the localization results contain certain errors.

[0006] (2) Existing studies only analyze the differences in EEG signals during the attack or between attacks, and lack methods to quantify the magnitude of the differences, thus failing to develop an effective method for lesion identification.

[0007] (3) Existing research is mainly based on intracranial electroencephalogram (EEG) signals, which must be collected by professionals in a medical setting by inserting electrodes into the brain. The collection conditions are very demanding. In addition, long-term collection can greatly restrict patients' daily activities, and is not only expensive, but also carries a certain risk of infection. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide an epilepsy lesion localization system based on scalp EEG, which can realize the lateralization and localization of lesions in epilepsy patients. An epilepsy lesion localization system based on scalp EEG includes a data preprocessing module, a causal flow matrix construction module, a feature extraction module, a dominant frequency band selection module, a lateralization selection module, and a lead localization module;

[0009] The data preprocessing module is used for:

[0010] For each patient's seizure, the EEG signal before the seizure was extracted as the interictal sample; a segment of EEG signal before and after the seizure was extracted as the initial seizure sample, with the same length as the interictal sample.

[0011] The causal flow matrix construction module is used for:

[0012] The EEG signal was divided into 5 sub-bands, and samples were taken at intervals for interictal and early-onset samples. For each sampling time t, the elements of the causal matrix M for each lead were calculated, as follows:

[0013]

[0014] In the formula, f1 and f2 represent the lower and upper limits of each sub-band, respectively; H ij (f,t) represents the transfer function between lead i and lead j at sampling time t at frequency f, and K represents the number of leads; M ij (t) represents the intensity of information flow from lead j to lead i, with a value range of [0,1]. At a certain moment, the sum of information flowing into a certain lead from other leads is 1.

[0015] In the causal matrix M, all elements on the main diagonal are set to 0, and elements with values ​​greater than a set threshold th are set to 1, while those less than the set threshold th are set to 0, resulting in a binary matrix M. bin ;

[0016] Based on the direction of information flow, calculate the in-degree DE of each lead in each frequency band. in and out-degree DE out The in-degree is the sum of all element values ​​in the row containing the lead within that sub-band, and the out-degree is the sum of all element values ​​in the column containing the lead. The causal current value of this lead is: CF = DE. out -DE in ;

[0017] The feature extraction module is used for:

[0018] For EEG signal samples from all sampling times in each lead of each frequency band during the interictal period, the causal flow average th was calculated. nonseizureThis serves as a reference value for the interictal interval in this lead within this frequency band.

[0019] For all EEG signal samples taken at all sampling times in each lead of each frequency band during the initial stage of the attack, each sample value was compared with the threshold value. nonseizure Compare and calculate all values ​​less than th. nonseizure The ratio m1 of the number of values ​​to the total number of samples:

[0020] For all EEG signal samples taken at all sampling times in each lead of each frequency band during the initial stage of the attack, calculate all values ​​less than th. nonseizure The mean of the sample values ​​is m2; multiply m1 and m2, and denote the final extracted feature as fea1;

[0021] The feature values ​​fea1 extracted from each lead in each frequency band for each patient are stacked to obtain the feature matrix fea1, which is used to characterize the degree of difference in the patient's causal flow parameters between the early stage of the attack and the interictal period.

[0022] The main frequency band selection module is used for:

[0023] Based on the extracted feature matrix fea1, F is calculated by summing the eigenvalues ​​of all leads in each frequency band. bandh This is used to characterize the degree of difference between each frequency band in the early stage of an attack and the interictal period; a total of 5 values ​​were obtained, and the frequency band corresponding to the maximum value was selected as the dominant frequency band during the patient's epileptic seizure;

[0024] The side selection module is used for:

[0025] After the dominant frequency band of the patient's seizure is selected, lateral localization is performed based on the feature matrix under the dominant frequency band. Let the feature matrix under the dominant frequency band be fea2, with a dimension of 1*K. The summation of the elements belonging to the left lead in matrix fea2 yields F. left Summing the values ​​of the elements belonging to the right lead in matrix fea2 yields F. right According to F left and F right The size of the lesion can be used to determine whether it is located on the left or right side of the brain.

[0026] The lead positioning module is used for:

[0027] Based on the selected dominant frequency band and side, the characteristic values ​​of all leads in the selected side under the dominant frequency band are compared. The lead corresponding to the maximum value is the lead with the greatest change between the initial stage of the seizure and the interictal period. The brain region corresponding to this lead is the location of the lesion in this seizure.

[0028] Preferably, the data preprocessing module extracts a 15-second segment of EEG signal at least 1 hour before the onset of the seizure as an interictal sample, and extracts an EEG signal from 7 seconds before the onset of the seizure to 8 seconds after the onset of the seizure as an early seizure sample, with the same length as the interictal sample.

[0029] Preferably, in the data preprocessing module, the first 5 seconds of each EEG signal are used for the adaptation phase of the Kalman filter and are discarded after the calculation is completed, so the actual signal length analyzed is 10 seconds; a 0.5Hz high-pass filter and a 50Hz band-stop filter are used to remove baseline drift and power frequency interference of each EEG signal.

[0030] Preferably, the causal flow matrix construction module divides the sub-frequency bands as follows: {δ: 0.5-4Hz; θ: 4-8Hz; α: 8-13Hz; β: 13-30Hz; γ: 30-128Hz}, for a total of 5 sub-frequency bands.

[0031] Preferably, the causal flow matrix construction module uses every 20 sampling points as a sampling time.

[0032] Preferably, the selection criterion for the threshold th is as follows: among all connections in the causal matrix except for self-connected edges, 40% of the connections with stronger connection strength are selected and retained; for a causal matrix of dimension K*K, after setting the 26 elements on the diagonal to zero, all elements are sorted in descending order, and the value of the 270th element is selected as the threshold th.

[0033] Preferably, the main frequency band selection module, when selecting the frequency band... After sorting in descending order, if the difference between the maximum and the second largest value does not exceed α% of the maximum value, the frequency band corresponding to the maximum and the second largest value is selected as the main frequency band of the patient's attacks; the value range of α is [0.5, 3].

[0034] Preferably, the side selection module: if F left >F right If the changes are more pronounced in the left leads than in the right leads, the lesion is likely located on the left side; conversely, if the changes are more pronounced in the right leads, the lesion is more likely located on the left side. right >F left If the changes are more pronounced in the right lead than in the left lead, the lesion is likely located on the right side.

[0035] Preferably, the side selection module: when |F left -F right |<0.02×max(F left ,F rightAt this time, it was believed that there was no significant difference between the left and right hemispheres in the dominant frequency band of the patient, which may be a generalized seizure with abnormalities in both hemispheres. At this time, the determination of the seizure side will not be made. The subsequent analysis will be conducted on all leads of the left and right hemispheres.

[0036] Preferably, in the lead localization module, when the lesion is located deep inside the brain, the area transmitted to the scalp may not be limited to one lead. Therefore, after sorting the feature values ​​of all leads in descending order, if the difference between the maximum value and the second largest value does not exceed 2% of the maximum value, the maximum value and the second largest value can be considered to be very close. At this time, the leads corresponding to the maximum value and the second largest value are selected as the abnormal leads of the patient.

[0037] The present invention has the following beneficial effects:

[0038] The purpose of this invention is to provide an epilepsy lesion localization system based on scalp EEG. Based on Granger causality theory, the system establishes the brain functional network of the patient during the interictal and ictal periods, respectively. It integrates effective information from different time periods, analyzes the connectivity differences between the initial ictal period and the interictal period at the network level, and analyzes the connectivity differences between the lesion area and non-lesion areas at the node level. It uses a difference quantification method to quantify the difference in causal flow of each lead between the interictal period and the initial ictal period, and identifies the dominant seizure frequency band, seizure side, and seizure lead sequentially based on the difference magnitude, thereby achieving lateralization and localization of the epilepsy lesion in the patient. Attached Figure Description

[0039] Figure 1 is a system block diagram of the epilepsy lesion localization system based on scalp EEG;

[0040] Figure 2 is a flowchart of the lesion localization method;

[0041] Figure 3 is a schematic diagram of the feature extraction principle. Detailed Implementation

[0042] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] [Overexcitation and synchronization of brain neurons may lead to abnormal discharges during or between seizures. Abnormal EEG signals are more pronounced during seizures but are easily affected by limb movements during the seizure. EEG signals during the interictal period have relatively less noise but are difficult to capture abnormal discharge signals.] This invention analyzes brain functional connectivity during seizures and between seizures to identify significant differences and proposes an effective lesion localization method based on these quantified differences. This lesion localization method uses causal flow indices as the evaluation basis, with the difference between each lead at the onset of a seizure and during the interictal period as the evaluation index. It employs a selection scheme of "frequency band selection—lateral selection—lead selection" to gradually narrow down the lesion area, achieving precise lateralization and localization. This method quantifies the difference between each lead in each frequency band and the onset of a seizure during the interictal period through feature extraction. Then, it determines the frequency band and laterality of the seizure through dominant frequency band selection and lateral selection. Finally, given the dominant frequency band and laterality have been selected, the lead with the largest difference is chosen as the lesion area, providing a solution to the limitations of existing research in clinical application. As shown in Figure 1, the epilepsy lesion localization system based on scalp EEG of the present invention includes a data preprocessing module, a causal flow matrix construction module, a feature extraction module, a main frequency band selection module, a lateral selection module, and a lead localization module.

[0044] 1. Data Preprocessing Module

[0045] For each patient's seizure, a 15-second segment of EEG signal at least 1 hour before the seizure onset was extracted as the interictal sample. Additionally, a segment of EEG signal from 7 seconds before the seizure onset to 8 seconds after the seizure onset was extracted as the initial seizure sample, with the same length as the interictal sample. The first 5 seconds of each signal segment were used for the Kalman filter adaptation phase and discarded after calculation, resulting in an actual analyzed signal length of 10 seconds. Brain functional networks were then constructed for both the interictal and initial seizure samples. Scalp EEG is easily affected by skull conduction and scalp friction. To avoid errors caused by noise in subsequent analysis, a 0.5Hz high-pass filter and a 50Hz band-stop filter were used to remove baseline drift and power line interference from each EEG channel.

[0046] 2. Causal Flow Matrix Construction Module

[0047] (1) Causal matrix

[0048] After preprocessing, the coefficients of the fourth-order time-varying autoregressive model are obtained using the Kalman filtering method, from which the transition matrix H of the time-varying system can be obtained. The iterative process of Kalman filtering is described in detail in the literature "Huo Yajun, Research on Epilepsy Brain Network Based on Intracranial EEG [D], University of Electronic Science and Technology of China, 2014". The basic frequency of epileptic seizures varies greatly among different patients, and there are also differences between different seizures of the same patient. In order to fully describe the functional connectivity in different frequency bands during the early stage of a seizure, this invention is based on Korzeniewska (KORZENIEWSKA A, M, M, et al. Determination of information flow direction among brain structures by a modified directed transfer function (dDTF) method [J]. Journal of Neuroscience Methods, 2003, 125(1): 195-207.) normalizes the DTF in different frequency bands to obtain the causal connectivity matrix of each sub-band. The sub-bands are divided into five sub-bands: {δ: 0.5-4Hz; θ: 4-8Hz; α: 8-13Hz; β: 13-30Hz; γ: 30-128Hz}. The causal matrix is ​​denoted as M, and the elements in the matrix are represented as follows:

[0049]

[0050] In the formula, f1 and f2 represent the lower and upper limits of each sub-band, respectively; H ij (f,t) represents the transfer function between lead i and lead j at sampling time t at frequency f, and K represents the number of leads, which is 26 in this invention. M ij The value range is [0,1], and at a certain moment, the sum of information flowing into a certain lead from other leads is 1. Considering the computational complexity of the causal matrix and taking into account the rate of change of the EEG signal, the causal matrix is ​​constructed every 20 sampling points. For the extracted data segment of 10 seconds, a total of 129 causal matrix samples were calculated, each sample containing 5 causal matrices M, corresponding to 5 sub-frequency bands. Subsequent matrix binarization and graph theory analysis were performed separately for each of the 5 sub-frequency bands.

[0051] (2) Matrix binarization

[0052] Based on the information flow direction of the causal matrix M, M ijThis represents the intensity of information flow from lead j to lead i. If we consider lead i and lead j as two vertices, then M ij This can characterize the degree of connection between these two vertices. To avoid the influence of changes in the lead's own state, M... ij All self-connected edges are set to 0, meaning all elements on the main diagonal are 0. Assume there exists a threshold th that satisfies the following condition (2):

[0053]

[0054] Then the causal matrix M can be transformed into a binary matrix M. bin Each element in the matrix represents whether a directed connection exists between the j-th lead and the i-th lead. When the value is 1, it indicates the existence of a directed connection from the j-th lead to the i-th lead; otherwise, when... A value of 0 indicates that the directed connection from lead j to lead i does not exist. The threshold th is selected based on the following criteria: among all connections in the causal matrix except for self-connections, 40% of the stronger connections are retained. For a 26*26 causal matrix, after setting the 26 diagonal elements to zero, all elements are sorted in descending order, and the value of the 270th element is selected as the threshold th. 40% of the valid connections are retained in both the interictal and early ictal causal matrices to avoid errors caused by differences in data acquisition equipment between patients and differences in baseline levels at different times for the same patient.

[0055] (3) Graph Analysis

[0056] The brain connectivity of each patient during the interictal period and the initial stage of an attack was measured at both the node level and the network level. The node attributes calculated included degree and causal flow.

[0057] (i) degree (DE)

[0058] The degree calculation is based on the binary matrix M under each frequency band. bin Calculations are performed to determine the in-degree DE of each lead in each frequency band, based on the direction of information flow. in and out-degree DE out As shown in equations (3) and (4) below. In-degree represents the number of directed connections from other leads to this lead; the larger the in-degree, the greater the amount of information flowing into this lead. Out-degree represents the number of directed connections from this lead to other leads; the larger the out-degree, the greater the amount of information flowing out of this lead.

[0059]

[0060]

[0061] In the above formula, N represents the binary matrix M. bin The number of rows and columns;

[0062] (ii) Causal flow (CF)

[0063] The causal flow is calculated as the difference between the out-degree and in-degree of the lead, as shown in Equation (5), which represents the relative outflow and inflow of information in each lead of the brain network. If the causal flow value of a lead is positive, its outflow trend is higher than its inflow trend, and the lead is the causal source; conversely, if the causal flow value of a lead is negative, its inflow trend is higher than its outflow trend, and the lead is the causal sink.

[0064] CF=DE out -DE in (5)

[0065] 3. Feature Extraction Module

[0066] For the extracted data segments with a length of 10 s during the interictal and episodic periods, the causal flow matrix has dimensions of 5*26*129; where 5 represents the number of sub-bands, 26 represents the number of leads, and 129 represents the number of samples. The following features are extracted from the causal flow matrix:

[0067] (1) For 129 samples in each lead of each frequency band during the interictal period, the causal current average th was calculated. nonseizure As a reference value for the interictal interval of this lead in this frequency band, it is shown as the red line in Figure 3;

[0068] (2) For the 129 samples in each lead of each frequency band during the early stage of the attack, the value of each sample was compared with the th nonseizure The comparison is shown in Figure 3. All values ​​less than th are calculated. nonseizure The ratio m1 of the number of values ​​to the total number of samples:

[0069]

[0070] (3) For 129 samples in each lead of each frequency band during the initial stage of the attack, calculate all samples less than th. nonseizure The mean m2 of the sample values:

[0071] m2 = avg(CF) ictal <th nonseizure (7)

[0072] (4) Multiply m1 and m2, and denote the final extracted feature as fea1:

[0073] fea1=m1×m2(8)

[0074] Feature extraction was performed separately for each lead in each frequency band. The feature values ​​fea1 extracted for each lead in each frequency band for each patient were stacked to obtain a feature matrix fea1 with a dimension of 5*26, which was used to characterize the degree of difference in the patient's causal flow parameters between the early stage of an attack and the interictal period.

[0075] 4. Main frequency band selection module

[0076] Epileptic seizures are closely related to EEG frequency. Patients with different lesion locations exhibit varying fundamental frequencies of EEG signals during seizures. Selecting the dominant frequency band can help preliminarily delineate lesions and avoid the influence of outliers. Based on the extracted feature matrix fea1, the eigenvalues ​​of all leads within each frequency band are calculated. This is used to characterize the difference in each frequency band between the initial stage of an epileptic seizure and the interictal period. A total of 5 values ​​were obtained, and the frequency band corresponding to the maximum value was selected as the dominant frequency band during the patient's epileptic seizure. The characteristic values ​​of all leads in the h-th frequency band are expressed as follows:

[0077]

[0078] Considering that changes in the early stages of an attack in some patients may be scattered across multiple frequency bands, therefore, in the analysis of... After sorting in descending order, if the difference between the maximum and the second largest value does not exceed α% of the maximum value, the maximum and the second largest value can be considered very close. In this case, the frequency bands corresponding to the maximum and the second largest values ​​are selected as the main frequency band for the patient's seizures. The value range of α is [0.5, 3], and this invention selects α as 2.

[0079] 5. Side Selection Module

[0080] After the dominant seizure frequency band for the patient is selected, lateral localization is performed based on the feature matrix under the dominant frequency band. Let the feature matrix under the dominant frequency band be fea2, with a dimension of 1*26. The differences between the left and right leads during the initial seizure and the interictal period are calculated separately.

[0081]

[0082]

[0083] If F left >F right If the changes are more pronounced in the left leads than in the right leads, the lesion is likely located on the left side; conversely, if the changes are more pronounced in the right leads, the lesion is more likely located on the left side. right >F leftIn this case, the changes in the right lead are more pronounced than those in the left lead, indicating that the lesion is likely located on the right side. When |F left -F right |<0.02×max(F left ,F right At this time, it was believed that there was no significant difference between the left and right hemispheres in the dominant frequency band of the patient, which may be a generalized seizure with abnormalities in both hemispheres. At this time, the determination of the seizure side will not be made. The subsequent analysis will be conducted on all leads of the left and right hemispheres.

[0084] 6. Lead positioning module

[0085] Based on the selected dominant frequency band and laterality, the characteristic values ​​of all leads in the selected laterality within the dominant frequency band are compared. The lead corresponding to the maximum value is the lead with the greatest change between the initial stage of the seizure and the interictal period, and the brain region corresponding to this lead is the location of the lesion in this seizure. When the lesion is located deep inside the brain, the area transmitted to the scalp may not be limited to one lead. Therefore, after sorting the characteristic values ​​of all leads in descending order, if the difference between the maximum and the second largest value does not exceed 2% of the maximum value, the maximum and the second largest value can be considered very close. In this case, the leads corresponding to both the maximum and the second largest values ​​are selected as the abnormal leads for this patient.

[0086] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A system for locating epileptic lesions based on scalp electroencephalography (EEG), characterized in that, The system includes a data preprocessing module, a causal flow matrix construction module, a feature extraction module, a main frequency band selection module, a lateral selection module, and a lead localization module. The data preprocessing module is used to: extract the EEG signal before each seizure for each patient, as an interictal sample; and extract a segment of EEG signal before and after the seizure, as an early seizure sample, with the same length as the interictal sample. The causal flow matrix construction module is used to: divide the EEG signal into 5 sub-frequency bands, perform interval sampling on the interictal and early seizure samples, and calculate the elements of the causal matrix M for each lead at each sampling time t, represented as: In the formula, and These represent the lower and upper limits of each sub-band, respectively. Let f represent the transfer function between lead i and lead j at sampling time t at frequency f, and K represent the number of leads; The signal represents the intensity of information flow from lead j to lead i, with a value range of [0,1]. At a certain moment, the sum of information flowing into a lead from other leads is 1. In the causal matrix M, all elements on the main diagonal are 0, and elements with values ​​greater than a set threshold th are set to 1, while those less than the set threshold th are set to 0, resulting in a binary matrix. Based on the direction of information flow, calculate the in-degree of each lead in each frequency band. Japanese output The in-degree is the sum of all element values ​​in the row containing the lead within that sub-band, and the out-degree is the sum of all element values ​​in the column containing the lead. The causal current value of this lead is: The feature extraction module is used to: calculate the causal average value of EEG signal samples from all sampling times in each lead of each frequency band during the interictal period. This serves as a reference value for the interictal interval in this lead within this frequency band. For all EEG signal samples taken at all sampling times in each lead of each frequency band during the initial stage of the attack, each sample value was compared with... Compare and calculate all values ​​less than The ratio of the number of values ​​to the total number of samples. For all EEG signal samples taken at all sampling times in each lead of each frequency band during the initial stage of the attack, calculate all samples less than... mean of sample values ； Will and Multiply, and denote the final extracted features as ; Feature values ​​extracted for each lead in each frequency band for each patient Stacking them together yields the feature matrix. This is used to characterize the degree of difference in causal flow parameters of patients during the early stages of an attack compared to the interictal period; the main frequency band selection module is used to: based on the extracted feature matrix The result is obtained by summing the eigenvalues ​​of all leads in each frequency band. This is used to characterize the degree of difference between each frequency band in the early stage of an attack and the interictal period; a total of 5 values ​​were obtained, and the frequency band corresponding to the maximum value was selected as the dominant frequency band during the patient's epileptic seizure; The lateral selection module is used to: perform lateral localization based on the feature matrix under the main frequency band after the patient's seizure frequency band is selected; let the feature matrix under the main frequency band be denoted as... For a matrix with dimension 1*K, Summing the values ​​of the elements belonging to the left lead, we get For matrices Summing the values ​​of the elements belonging to the right lead, we get... ;according to and The size relationship determines whether the lesion is located on the left or right side of the brain; the lead localization module is used to: compare the feature values ​​of all leads on the selected side within the selected dominant frequency band based on the selected dominant frequency band, and the lead corresponding to the maximum value is the lead with the greatest change between the initial stage of the attack and the interictal period, and the brain region corresponding to this lead is the location of the lesion in this attack; the side selection module: if If the changes are more pronounced in the left leads than in the right leads, the lesion is likely located on the left side; conversely, if the changes are more pronounced in the right leads than in the left leads, the lesion is more likely located on the left side. In this case, the changes in the right lead are more pronounced than those in the left lead, and the lesion may be located on the right side. The side selection module: when At that time, it was believed that there was no significant difference between the left and right hemispheres in the dominant frequency band of the patient, which may be a generalized seizure with abnormalities in both hemispheres. At this time, the determination of the seizure side was not made, and the subsequent analysis will be conducted on all leads of the left and right hemispheres.

2. The epilepsy lesion localization system based on scalp EEG as described in claim 1, characterized in that, The data preprocessing module extracts a 15-second segment of EEG signal at least 1 hour before the onset of the seizure as an interictal sample. It also extracts EEG signals from 7 seconds before the onset of the seizure to 8 seconds after the onset of the seizure as an early seizure sample, with the same length as the interictal sample.

3. The epilepsy lesion localization system based on scalp EEG as described in claim 2, characterized in that, The data preprocessing module uses the first 5 seconds of each EEG signal for the adaptation phase of the Kalman filter, which is discarded after calculation, resulting in an actual signal length of 10 seconds. A 0.5 Hz high-pass filter and a 50 Hz band-stop filter are used to remove baseline drift and power frequency interference from each EEG signal.

4. The epilepsy lesion localization system based on scalp EEG as described in claim 1, characterized in that, The causal flow matrix construction module divides the sub-bands as follows: {δ: 0.5-4 Hz; θ: 4-8 Hz; α: 8-13 Hz; β: 13-30 Hz; γ: 30-128 Hz}, for a total of 5 sub-bands.

5. The epilepsy lesion localization system based on scalp EEG as described in claim 1, characterized in that, The causal flow matrix construction module uses every 20 sampling points as a sampling moment.

6. The epilepsy lesion localization system based on scalp EEG as described in claim 1, characterized in that, The selection criteria for the threshold th are as follows: among all connections in the causal matrix except for self-connected edges, 40% of the connections with stronger connection strength are selected and retained; for a causal matrix of dimension K*K, after setting the 26 elements on the diagonal to zero, all elements are sorted in descending order, and the value of the 270th element is selected as the threshold th.

7. The epilepsy lesion localization system based on scalp EEG as described in claim 1, characterized in that, The main frequency band selection module, in the case of After sorting in descending order, if the difference between the maximum and the second largest value does not exceed α% of the maximum value, the frequency bands corresponding to the maximum and the second largest values ​​are selected as the main frequency bands of the patient's attacks; the value range of α is [0.5, 3].

8. The epilepsy lesion localization system based on scalp EEG as described in claim 1, characterized in that, When the lesion is located deep inside the brain, the area transmitted to the scalp may not be limited to one lead. Therefore, after sorting the feature values ​​of all leads in descending order, if the difference between the maximum and the second largest value does not exceed 2% of the maximum value, the leads corresponding to the maximum and the second largest values ​​are selected as the abnormal leads of the patient.

Images