EEG signal detection method based on abnormal discharge waveforms, epilepsy detection system

Abstract

This invention discloses a method for detecting electroencephalogram (EEG) signals based on abnormal discharge waveforms and an epilepsy detection system. The method includes: acquiring multiple abnormal discharge waveforms; obtaining abnormal discharge templates through morphological clustering, i.e., classifying multiple abnormal discharge templates according to the correlation coefficient between the cluster centroids of each abnormal discharge waveform; selecting abnormal matching templates from the multiple abnormal discharge templates, i.e., selecting several abnormal discharge templates as abnormal matching templates based on the abnormal discharge detection distribution; and detecting online EEG signals based on the abnormal matching templates. By acquiring multiple abnormal discharge waveforms and then performing morphological clustering to obtain abnormal discharge templates, frequent modifications to the feature extraction process can be avoided, improving the generalization ability of clustering; further, selecting the optimal n abnormal discharge templates as abnormal matching templates based on the abnormal discharge detection distribution can further improve the detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) signal detection technology, specifically to an EEG signal detection method and an epilepsy detection system based on abnormal discharge waveforms. Background Technology

[0002] Epilepsy is a chronic brain disorder characterized by recurrent seizures. It is caused by abnormal electrical discharges in brain neurons, and seizures are characterized by their recurrence and brevity. Electroencephalography (EEG) signals contain rich information about brain activity. High-frequency oscillating signals, which are brain electrical activities at high gamma rhythms (>80Hz), are closely related to epilepsy and are a reliable biomarker for epilepsy. They can reflect the severity of epilepsy and are helpful in evaluating the effectiveness of epilepsy treatment, assessing susceptibility to epilepsy, and predicting seizures. Therefore, the detection of EEG signals can predict epileptic seizures to some extent.

[0003] In related technologies, nonlinear dynamic models, EEG complexity, and abnormal discharge (spike) frequency detection methods are used to predict epileptic seizures. These methods use all EEG data, including physiological and pathological spikes, to generate features during training. However, due to the large differences in EEG discharge waveforms of different subjects at different times, it is impossible to accurately retain pathological spikes related to seizures, resulting in insufficient accuracy of input features and low accuracy in recognizing epileptic seizures. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for detecting electroencephalogram (EEG) signals based on abnormal discharge waveforms and an epilepsy detection system.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] In a first aspect, the present invention provides a method for detecting electroencephalogram (EEG) signals, comprising:

[0007] Multiple abnormal discharge waveforms were obtained;

[0008] Morphological clustering is used to obtain abnormal discharge templates, that is, multiple abnormal discharge templates are classified according to the correlation coefficient between the cluster centroids of each abnormal discharge waveform.

[0009] The process involves selecting abnormal matching templates from multiple abnormal discharge templates, i.e., selecting several abnormal discharge templates as abnormal matching templates based on the distribution of abnormal discharge detection.

[0010] Detecting online EEG signals based on abnormal matching templates.

[0011] In one example, the classification of multiple abnormal discharge templates includes:

[0012] Identify existing clusters based on abnormal discharge waveforms;

[0013] A sliding correlation is performed between the abnormal discharge waveform after the existing cluster and the cluster centroid of the existing cluster. The sliding correlation coefficient during the sliding process is recorded, and the maximum sliding correlation coefficient is taken as the correlation coefficient between the abnormal discharge waveform and the cluster centroid. If the correlation coefficient is greater than a set threshold, the abnormal discharge waveform is assigned to the existing cluster; otherwise, the abnormal discharge waveform is taken as a new cluster.

[0014] Iterate until all abnormal discharge waveforms have been classified.

[0015] In one example, if multiple correlation coefficients between an abnormal discharge waveform and the cluster centroid exceed a set threshold, the abnormal discharge waveform is assigned to the existing cluster with the highest correlation coefficient.

[0016] In one example, the sliding correlation includes:

[0017] The next abnormal discharge waveform is slid towards the abnormal discharge waveform corresponding to the cluster centroid, with each slid step being one sampling interval; after each slid, the sliding correlation coefficient between the overlapping part of the next abnormal discharge waveform and the abnormal discharge waveform corresponding to the cluster centroid and the cluster centroid is calculated.

[0018] In one example, multiple abnormal discharge waveforms are obtained based on the energy characteristics of historical EEG signals; wherein, the energy characteristics of the historical EEG signals include: global discharge coefficient, line length, and envelope.

[0019] In one example, abnormal matching templates are selected from multiple abnormal discharge templates based on the abnormal discharge detection distribution, including:

[0020] Remove abnormal discharge templates whose number of abnormal discharges detected before the onset time is less than a set threshold;

[0021] In the remaining abnormal discharge templates, the median distance between the abnormal discharge detection time and the onset time of each abnormal discharge template is calculated. The median distances are arranged in ascending order, and the first n abnormal discharge templates are selected as abnormal matching templates.

[0022] In one example, the detection of online EEG signals based on anomaly matching templates includes:

[0023] The historical EEG signals are detected using the abnormal matching templates, and the abnormal discharge features identified by each abnormal matching template are statistically analyzed. The abnormal discharge features are used to train a classifier, and the trained classifier is used for real-time detection of EEG signals.

[0024] In one example, the abnormal discharge features include: the number of abnormal discharges identified by each abnormal matching template and their proportion in the total number.

[0025] In one example, the moment when the energy characteristics of historical EEG signals exceed a set threshold is marked as an abnormal discharge; data from the time before and after the abnormal discharge moment are extracted as the abnormal discharge waveform.

[0026] In a second aspect, the present invention provides an epilepsy detection system, comprising:

[0027] The processor runs the electroencephalogram (EEG) signal detection method to obtain the detection results at the time of the epileptic seizure.

[0028] A classifier is used to output the detection results.

[0029] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the electroencephalogram (EEG) signal detection method.

[0030] The beneficial effects of this invention are that the EEG signal detection method and epilepsy detection system based on abnormal discharge waveforms of this invention, by acquiring multiple abnormal discharge waveforms and then performing morphological clustering on the abnormal discharge waveforms to obtain abnormal discharge templates, can avoid frequent modifications in the feature extraction process and improve the generalization ability of clustering; furthermore, by selecting the optimal n abnormal discharge templates as abnormal matching templates based on the abnormal discharge detection distribution, the detection accuracy can be further improved. This method can not only be used for early warning detection before epileptic seizures, but also for evaluating treatment effects. Attached Figure Description

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] Figure 1 This is a flowchart of the electroencephalogram (EEG) signal detection method of the present invention.

[0033] Figure 2 This is a schematic diagram of the abnormal discharge moment identification method of the present invention.

[0034] Figure 3 This is a schematic diagram related to the sliding of the present invention.

[0035] Figure 4 This is a schematic diagram showing that the distribution of abnormal discharge detection times in this invention has no obvious trend.

[0036] Figure 5 This is a schematic diagram showing that the distribution of abnormal discharge detection times is concentrated near the time of onset.

[0037] Figure 6This is the percentage distribution of discharge identified by the n abnormal matching templates in this invention.

[0038] Figure 7 This is a schematic diagram of the confusion matrix of the classification results of this invention.

[0039] Figure 8 This is a graph showing the comparison of the number of abnormalities detected before and after treatment according to the present invention. Detailed Implementation

[0040] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0041] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, features defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0042] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0043] like Figure 1 As shown, the EEG signal detection method of this embodiment includes: Step S1, obtaining multiple abnormal discharge waveforms. Step S2, obtaining abnormal discharge templates through morphological clustering, that is, classifying multiple abnormal discharge templates according to the correlation coefficient between the cluster centroids of each abnormal discharge waveform. Step S3, selecting abnormal matching templates from multiple abnormal discharge templates, that is, selecting several abnormal discharge templates as abnormal matching templates based on the abnormal discharge detection distribution. Step S4, detecting online EEG signals according to the abnormal matching templates.

[0044] In other words, this embodiment first extracts abnormal discharge waveforms of different morphologies from the EEG signal, and then performs morphological clustering on the abnormal discharge waveforms (i.e., grouping abnormal discharge waveforms with similar / identical morphologies into the same category) to obtain different abnormal discharge templates. Then, several superior templates are selected from multiple abnormal discharge templates as abnormality matching templates for EEG signal detection. This embodiment obtains abnormality matching templates based on morphological clustering, which avoids frequent modifications during feature extraction and helps reduce the computational load of the detection process. By performing morphological clustering based on abnormal discharge waveforms and constructing abnormality matching templates to detect EEG signals, the accuracy of abnormality detection can be improved.

[0045] Specifically, abnormal discharge waveforms can be obtained based on the energy characteristics of historical EEG signals. For example, the energy characteristics of historical EEG signals include: global discharge coefficient, line length, and envelope. It should be noted that the energy characteristics of EEG signals change over time, with each moment having corresponding energy characteristics. Moments where the energy characteristics exceed a set threshold are marked as abnormal discharges, and data from periods before and after the abnormal discharge moment are extracted as the abnormal discharge waveform. Taking the global discharge coefficient as an example... Figure 2 As shown, a threshold for the global discharge coefficient is set (this threshold = the mean of the global discharge coefficient + 3 * variance). Moments where the global discharge coefficient exceeds the threshold are marked as abnormal discharges. A signal segment is extracted 1 second before and 1 second after the sampling moment exceeding the threshold, forming an abnormal discharge waveform. This allows for the acquisition of multiple abnormal discharge waveforms.

[0046] This embodiment obtains abnormal discharge templates through morphological clustering, specifically including: acquiring existing clusters based on abnormal discharge waveforms; performing a sliding correlation between abnormal discharge waveforms following existing clusters and the centroids of existing clusters; recording the sliding correlation coefficients during the sliding process; and using the maximum sliding correlation coefficient as the correlation coefficient between the abnormal discharge waveform and the cluster centroid. If the correlation coefficient is greater than a set threshold, the abnormal discharge waveform is assigned to an existing cluster; otherwise, the abnormal discharge waveform is assigned to a new cluster. This process is iterated until all abnormal discharge waveforms are classified. When multiple correlation coefficients between an abnormal discharge waveform and a cluster centroid exceed the set threshold, the abnormal discharge waveform is assigned to the existing cluster with the highest correlation coefficient. The sliding correlation includes: sliding the next abnormal discharge waveform towards the abnormal discharge waveform corresponding to the cluster centroid, with each sliding step being a sampling interval; calculating the sliding correlation coefficient between the overlapping portion of the next abnormal discharge waveform and the abnormal discharge waveform corresponding to the cluster centroid and the cluster centroid after each sliding step.

[0047] The process of morphological clustering is illustrated below with an example. Assume we obtain 10 abnormal discharge waveforms, denoted as abnormal discharge waveforms w1 to w2. 10Randomly select the abnormal discharge waveform w1 as the existing cluster C1. At this point, the centroid of the existing cluster C1 is the average value of the abnormal discharge waveform w1. When calculating the sliding correlation, first slide the abnormal discharge waveform w2 towards the abnormal discharge waveform w1 of the existing cluster C1. The step size of each slide is one sampling interval. With each slide, there will be an overlap between the abnormal discharge waveform w2 and the abnormal discharge waveform w1 (e.g., ...). Figure 3 As shown, the sliding correlation coefficient between the overlapping portion of the abnormal discharge waveform w2 and the centroid of the existing cluster C1 is calculated. It can be understood that a sliding correlation coefficient can be calculated for each sliding step; therefore, multiple sliding correlation coefficients can be obtained between the abnormal discharge waveform w2 and the centroid of the existing cluster C1 when calculating the sliding correlation. Among these multiple sliding correlations, the largest sliding correlation coefficient is selected as the correlation coefficient between the abnormal discharge waveform w2 and the centroid of the existing cluster C1. If the correlation coefficient is greater than a set threshold (e.g., 0.9), the abnormal discharge waveform w2 is assigned to the existing cluster C1; otherwise, the abnormal discharge waveform w2 is assigned to the new cluster C2. Assuming the abnormal discharge waveform w2 is assigned to the new cluster C2, the centroid of the new cluster C2 is the average value of the abnormal discharge waveform w2 data. Then, the abnormal discharge waveform w3 is correlated with the centroids of existing clusters C1 and C2 respectively. Assuming the correlation coefficient between abnormal discharge waveform w3 and the centroid of existing cluster C1 is 0.95, and the correlation coefficient between it and the centroid of existing cluster C2 is 0.7, then abnormal discharge waveform w3 is classified into existing cluster C1. The remaining abnormal discharge waveforms w4 to w2 are then classified into cluster C1. 10 The above process is repeated to calculate the sliding correlation between the abnormal discharge waveform and the centroid of the existing clusters, thus completing the classification. When multiple correlation coefficients between the abnormal discharge waveform and the cluster centroid exceed a set threshold, the abnormal discharge waveform is assigned to the existing cluster with the highest correlation coefficient. For example, the abnormal discharge waveform w 10 The correlation coefficients between the abnormal discharge waveform and the existing cluster centroids are 0.91, 0.94, and 0.99, respectively. 10 It is assigned to the existing cluster corresponding to 0.99. It should be noted that when an abnormal discharge waveform is added to an existing cluster, the cluster centroid is updated to the average value of all abnormal discharge waveforms within the cluster. The abnormal discharge waveforms in epileptic EEG signals are highly variable. By using a sliding correlation coefficient method, the discharges with the most similar waveforms can be directly clustered into one category, while simultaneously aligning different discharges.

[0048] After classifying multiple abnormal discharge waveforms, a total of m abnormal discharge templates can be obtained. Different abnormal discharge templates will yield different results in anomaly identification, so it is necessary to filter these m templates. Correct templates will detect a higher density of abnormal discharges closer to the time of the epileptic seizure, while detecting fewer abnormal discharges farther away. Therefore, abnormal matching templates can be selected from multiple abnormal discharge templates based on the distribution of abnormal discharge detection. Specifically, this involves: removing abnormal discharge templates whose number of detected abnormal discharges before the seizure time is less than a set threshold; and for the remaining abnormal discharge templates, calculating the median distance between the abnormal discharge detection time and the seizure time for each template, arranging the median distances in ascending order, and selecting the top n abnormal discharge templates as abnormal matching templates.

[0049] like Figure 4 As shown, using the abnormal discharge template, the distribution of abnormal discharge detection times within 1800 seconds prior to the epileptic seizure showed no obvious trend. Figure 5 As shown, using abnormal discharge template two, the detection times of abnormal discharges within 1800 seconds before the onset of an epileptic seizure are mostly concentrated close to the seizure time. When screening templates, templates with fewer than a set threshold (e.g., 20) of abnormal discharges detected before the seizure time can be eliminated first. Then, among the remaining abnormal discharge templates, the median distance between the detected abnormal discharge time and the seizure time is calculated for each template. These median distances are then arranged in ascending order, and the top n (e.g., n can be 5) abnormal discharge templates are selected as the abnormal matching templates. It can be understood that the smaller the median distance, the closer the abnormal discharge detection time is to the seizure time. Thus, the optimal n abnormal discharge templates can be selected for online EEG signal detection.

[0050] Online EEG signal detection based on anomaly matching templates includes: detecting historical EEG signals using anomaly matching templates and statistically analyzing the abnormal discharge features identified by each template; training a classifier using these abnormal discharge features, and then using the trained classifier for real-time EEG signal detection. Abnormal discharge features include the number of abnormal discharges identified by each template and their proportion in the total number of discharges. When detecting historical EEG signals using anomaly matching templates, the historical EEG signals are divided into multiple 10-minute segments. Each segment is identified using n anomaly matching templates, and the number of abnormal discharges detected by the n templates in each segment is recorded. For example, taking five anomalous matching templates as an example, in each signal segment, let the number of anomalous discharges detected by the five anomalous matching templates be a, b, c, d, and e, respectively. Then the proportions of anomalous discharges detected by each anomalous matching template are: a / (a+b+c+d+e), b / (a+b+c+d+e), c / (a+b+c+d+e), d / (a+b+c+d+e), and e / (a+b+c+d+e). Figure 6 As shown, at locations farther from the epileptic seizure time, the distribution of abnormal discharges detected by the five abnormal matching templates varies considerably (different colors represent the detection percentage of different templates). However, at locations closer to the epileptic seizure time, the distribution of abnormal discharges detected by the five abnormal matching templates is relatively stable, with templates one and five having a lower proportion, while templates two, three, and four have a higher proportion.

[0051] After obtaining the abnormal discharge characteristics, the classifier can be trained using these characteristics. A signal segment is extracted every 10 minutes, and the number and proportion of abnormal discharges detected by each abnormal matching template in the signal segment are obtained as training features. The first hour before each attack is defined as the pre-attack period, and the corresponding abnormal discharge characteristics are extracted as pre-attack training data. Data within one day before and after the attack are marked as indeterminate data. The data remaining after removing the pre-attack and indeterminate data is used as the inter-attack data, and the corresponding abnormal discharge characteristics are extracted as inter-attack training data. Figure 7 As shown, the Naive Bayes method is used to train the pre-identification classifier, and the detection accuracy of the trained pre-identification classifier can reach 98.8%.

[0052] Abnormal discharges occurring before or after epileptic seizures indicate that the patient is in a seizure-prone state. Electrical stimulation therapy can help patients escape this state, but the short-term treatment effect is difficult to determine. By identifying abnormal discharges and statistically analyzing the number of abnormal discharges before and after treatment, it is possible to confirm whether electrical stimulation therapy reduces the occurrence of abnormal states, thereby determining the treatment effect. Figure 8As shown, data from the stable period several days after a seizure, and 5 hours before and after treatment, were used to detect abnormal states using a pre-seizure identification classifier, and the number of detected abnormalities was compared. The figure shows that the average number of abnormal detections before treatment was 0.75, while the average number after treatment was 0.2, a significant reduction, demonstrating the effectiveness of the treatment plan. In other words, the detection method in this embodiment can be used not only for early warning of epilepsy but also for evaluating treatment effectiveness.

[0053] The detection method provided in this embodiment obtains abnormal discharge waveforms and performs morphological clustering on the abnormal discharge waveforms to obtain different abnormal discharge templates. Then, it selects the optimal n abnormal matching templates for detecting abnormal discharges. This avoids frequent modifications in the feature extraction process, which helps reduce the amount of computation. On the other hand, it improves the generalization ability of clustering and increases the accuracy of the detection results.

[0054] This invention also provides an epilepsy detection system, comprising: a processor and a classifier. The processor is used for an electroencephalogram (EEG) signal detection method to obtain detection results at the time of an epileptic seizure; the classifier is used to output the detection results. The processor may include one or more single-core or multi-core processors, and may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor may be a microprocessor or any conventional processor. Furthermore, the processor may be implemented using an integrated circuit chip. The classifier can be obtained through model training.

[0055] The present invention also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of an electroencephalogram (EEG) signal detection method. The storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0056] In summary, the EEG signal detection method and epilepsy detection system based on abnormal discharge waveforms of the present invention, by acquiring multiple abnormal discharge waveforms and then performing morphological clustering on these waveforms to obtain abnormal discharge templates, can avoid frequent modifications in the feature extraction process and improve the generalization ability of clustering. Furthermore, by selecting the optimal n abnormal discharge templates based on the abnormal discharge detection distribution as abnormal matching templates, the detection accuracy can be further improved. This method can be used not only for early warning detection before epileptic seizures but also for evaluating treatment effectiveness.

[0057] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined by the scope of the claims.

Claims

1. A method of detecting an electroencephalographic signal, characterized by, include: Multiple abnormal discharge waveforms were obtained; Morphological clustering is used to obtain anomalous discharge templates, that is, multiple anomalous discharge templates are classified according to the correlation coefficient between the cluster centroids of various anomalous discharge waveforms, specifically including: Identify existing clusters based on abnormal discharge waveforms; A sliding correlation is performed between the abnormal discharge waveform after the existing cluster and the cluster centroid of the existing cluster. The sliding correlation coefficient during the sliding process is recorded, and the maximum sliding correlation coefficient is taken as the correlation coefficient between the abnormal discharge waveform and the cluster centroid. If the correlation coefficient is greater than a set threshold, the abnormal discharge waveform is assigned to the existing cluster; otherwise, the abnormal discharge waveform is taken as a new cluster. Iterate until all abnormal discharge waveforms have been classified. The process involves selecting abnormal matching templates from multiple abnormal discharge templates. Specifically, based on the distribution of abnormal discharge detections, several abnormal discharge templates are selected as abnormal matching templates. This includes: Remove abnormal discharge templates whose number of abnormal discharges detected before the onset time is less than a set threshold; In the remaining abnormal discharge templates, the median distance between the abnormal discharge detection time and the onset time of each abnormal discharge template is calculated. The median distances are arranged in ascending order, and the first n abnormal discharge templates are selected as abnormal matching templates. Detecting online EEG signals based on abnormal matching templates.

2. The electroencephalogram (EEG) signal detection method as described in claim 1, characterized in that, If multiple correlation coefficients between an abnormal discharge waveform and the cluster centroid exceed a set threshold, then the abnormal discharge waveform is assigned to the existing cluster with the highest correlation coefficient.

3. The electroencephalogram (EEG) signal detection method as described in claim 1, characterized in that, The sliding correlation includes: The next abnormal discharge waveform is slid towards the abnormal discharge waveform corresponding to the cluster centroid, with each slid step being one sampling interval; after each slid, the sliding correlation coefficient between the overlapping part of the next abnormal discharge waveform and the abnormal discharge waveform corresponding to the cluster centroid and the cluster centroid is calculated.

4. The electroencephalogram (EEG) signal detection method as described in claim 1, characterized in that, Multiple abnormal discharge waveforms were obtained based on the energy characteristics of historical EEG signals; among them... The energy characteristics of the historical EEG signals include: global discharge coefficient, line length, and envelope.

5. The electroencephalogram (EEG) signal detection method as described in claim 1, characterized in that, The detection of online EEG signals based on abnormal matching templates includes: Historical EEG signals are detected using the aforementioned abnormal matching templates, and the abnormal discharge features identified by each of the aforementioned abnormal matching templates are statistically analyzed. The abnormal discharge features are used to train a classifier, and the trained classifier is used for real-time detection of EEG signals.

6. The electroencephalogram (EEG) signal detection method as described in claim 5, characterized in that, The abnormal discharge characteristics include: the number of abnormal discharges identified by each abnormal matching template and their proportion in the total number.

7. The electroencephalogram (EEG) signal detection method as described in claim 1, characterized in that, The moment when the energy characteristics of historical EEG signals exceed a set threshold is marked as an abnormal discharge; Data from periods before and after the abnormal discharge were extracted to form the abnormal discharge waveform.

8. An epilepsy detection system characterized by, include: The processor runs the electroencephalogram (EEG) signal detection method as described in any one of claims 1-7 to obtain the detection results at the time of an epileptic seizure; A classifier is used to output the detection results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the electroencephalogram (EEG) signal detection method as described in any one of claims 1-7.

Images