Abnormal brain discharge detection method, system, device, equipment and product

Abstract

The embodiments of the present disclosure provide a method, system, device, equipment and product for detecting abnormal brain discharges, which relate to the field of artificial intelligence technology. The method includes: acquiring EEG data of the subject; performing inter-channel feature extraction on the EEG data to obtain the inter-channel mutual information features of the EEG data; performing intra-channel feature extraction on the EEG data based on the inter-channel mutual information features to obtain the intra-channel features of each channel in the EEG data; and determining the abnormal discharge detection results of the subject based on multiple intra-channel features. The present disclosure performs feature extraction on the EEG data, which can not only capture the mutual information between different channels of the EEG data, but also extract the time-frequency domain information of a single channel, so that the abnormal discharge detection results have higher detection accuracy.

Description

Background Art

[0002] This section is intended to provide a background or context to the embodiments of the disclosure that are recited in the claims. No description herein is admitted to be prior art by inclusion in this section.

[0003] Abnormal discharges in the human brain are caused by abnormal activity of brain cells and can cause neurological diseases such as epilepsy. Electroencephalograms are an important diagnostic tool for the diagnosis and treatment of epilepsy, and epileptic discharges are important markers of diseases such as epilepsy. With the increasing awareness and attention paid to diseases such as epilepsy, the number of EEG monitoring has also increased. Due to the heavy workload of analyzing EEG data and the low consistency of experts' judgments on EEG epileptic discharges, the demand for automatic EEG analysis is increasing. By establishing standard judgments, the workload can be reduced, human bias can be reduced, and accuracy can be improved. The development of artificial intelligence has made it possible to realize the idea of ​​automatic EEG analysis. Summary of the invention

[0004] However, the accuracy of detecting abnormal discharges in the human brain needs to be improved.

[0005] In related technologies, artificial intelligence technology (such as deep learning models) is used to detect epileptic discharge signals. Existing deep learning models generally cannot take into account time information, spatial information and the mutual information of EEG channels at the same time, so that the information in the EEG signals is not fully utilized. There is still room for improvement in the accuracy of the algorithm model.

[0006] To this end, the present invention proposes an improved abnormal brain discharge detection method to perform feature extraction on EEG data, which can not only capture the mutual information between different channels of EEG data, but also extract the time and frequency domain information of a single channel, so that the abnormal discharge detection result has higher detection accuracy.

[0007] In this context, embodiments of the present disclosure are intended to provide a method for detecting abnormal brain discharges, a system for detecting abnormal brain discharges, an apparatus for detecting abnormal brain discharges, an electronic device, a computer-readable storage medium, and a computer program product.

[0008] In a first aspect of the embodiments of the present disclosure, a method for detecting abnormal brain discharges is provided, comprising: acquiring EEG data of a subject; performing inter-channel feature extraction on the EEG data to obtain inter-channel mutual information features of the EEG data; performing intra-channel feature extraction on the EEG data based on the inter-channel mutual information features to obtain intra-channel features of each channel in the EEG data; and determining an abnormal discharge detection result of the subject based on a plurality of the intra-channel features.

[0009] In one embodiment of the present disclosure, the method of obtaining EEG data of the subject includes: obtaining an initial EEG signal of the subject, dividing the initial EEG signal into a plurality of time windows; performing feature transformation processing on the initial EEG signal in each of the time windows to obtain the EEG data, wherein the EEG data is a time-frequency spectrum diagram of the initial EEG signal.

[0010] In one embodiment of the present disclosure, the method also includes: performing initial feature extraction processing on the EEG data to obtain an initial feature map; performing dimensionality transformation processing on the initial feature map, adjusting the channel dimension position of the initial feature map, and obtaining dimensionality transformation data of each channel; performing convolution processing on multiple dimensionality transformation data respectively to obtain reduced dimensionality feature data of each channel; performing data expansion processing on the reduced dimensionality feature data of each channel to obtain one-dimensional channel data corresponding to each channel.

[0011] In one embodiment of the present disclosure, the inter-channel feature extraction of the EEG data to obtain the inter-channel mutual information feature of the EEG data includes: acquiring one-dimensional channel data corresponding to each channel in the EEG data; encoding the multiple one-dimensional channel data to extract the inter-channel feature of the EEG data to obtain the inter-channel mutual information feature.

[0012] In one embodiment of the present disclosure, the inter-channel mutual information feature includes a first inter-channel mutual information feature and a second inter-channel mutual information feature, and the encoding processing of the multiple one-dimensional channel data to perform inter-channel feature extraction on the EEG data to obtain the inter-channel mutual information feature includes: encoding the multiple one-dimensional channel data to obtain the first inter-channel mutual information feature; convolution processing the first inter-channel mutual information feature to obtain a reduced-dimensional channel feature; encoding processing the reduced-dimensional channel feature to obtain the second inter-channel mutual information feature.

[0013] In one embodiment of the present disclosure, the intra-channel feature extraction of the EEG data based on the inter-channel mutual information feature to obtain the intra-channel feature of each channel in the EEG data includes: acquiring an initial feature map corresponding to the EEG data; performing intra-channel feature extraction on the initial feature map to obtain initial single-channel features of each of the channels; determining the channel weight corresponding to each of the channels based on the inter-channel mutual information feature; and determining the intra-channel feature of each of the channels according to the channel weight and the initial single-channel feature.

[0014] In one embodiment of the present disclosure, determining the channel weights corresponding to each of the channels based on the inter-channel mutual information features includes: decoding the inter-channel mutual information features to obtain channel decoding data corresponding to each channel in the EEG data; determining the channel value of each channel based on the channel decoding data, and using the channel value as the channel weight of each of the channels.

[0015] In one embodiment of the present disclosure, determining the intra-channel features of each channel according to the channel weight and the initial single-channel features includes: acquiring the initial single-channel features obtained through the current convolution processing; adjusting the channel weight of the initial single-channel features according to the channel weight; and determining the intra-channel features based on the adjusted channel weight and the initial single-channel features.

[0016] In a second aspect of the disclosed embodiments, a system for detecting abnormal brain discharges is provided, comprising: a data determination module for acquiring EEG data of a subject, performing initial feature extraction processing on the EEG data, and obtaining an initial feature map; a channel weight determination module for determining one-dimensional channel data corresponding to each channel according to the initial feature map, encoding multiple one-dimensional channel data, and obtaining inter-channel mutual information features of the EEG data; and determining channel weights corresponding to each of the channels based on the inter-channel mutual information features; a fusion convolution module for performing intra-channel feature extraction on the initial feature map, obtaining initial single-channel features of each of the channels, and determining intra-channel features of each of the channels according to the channel weights and the initial single-channel features; and an abnormality detection module for determining abnormal discharge detection results of the subject according to multiple intra-channel features.

[0017] In one embodiment of the present disclosure, the channel weight determination module includes: a dimensionality transformation unit, which is used to perform dimensionality transformation processing on the initial feature map, adjust the channel dimension position of the initial feature map, and obtain dimensionality transformation data of each channel; a one-dimensional convolution unit, which is used to perform convolution processing on multiple dimensionality transformation data respectively to obtain reduced dimensionality feature data of each channel; a data expansion unit, which is used to perform data expansion processing on the reduced dimensionality feature data of each channel to obtain one-dimensional channel data corresponding to each channel.

[0018] In one embodiment of the present disclosure, the inter-channel mutual information feature includes a first inter-channel mutual information feature and a second inter-channel mutual information feature, and the channel weight determination module also includes a first encoding unit and a second encoding unit, and the second encoding unit includes a convolution layer; the first encoding unit is used to encode the multiple one-dimensional channel data based on the attention mechanism to obtain the first inter-channel mutual information feature; the convolution layer is used to convolve the first inter-channel mutual information feature to obtain a reduced-dimensional channel feature; the second encoding unit is used to encode the reduced-dimensional channel feature based on the attention mechanism to obtain a second inter-channel mutual information feature.

[0019] In one embodiment of the present disclosure, the fused convolution module includes: a fused convolution unit, used to extract intra-channel features from the initial feature map to obtain initial single-channel features of each channel; and a feature determination unit, used to determine the intra-channel features of each channel based on the channel weights corresponding to each channel in the initial feature map and the initial single-channel features.

[0020] In a third aspect of the embodiments of the present disclosure, an abnormal brain discharge detection device is provided, comprising: a data acquisition module for acquiring EEG data of a subject; a first feature extraction module for performing inter-channel feature extraction on the EEG data to obtain inter-channel mutual information features of the EEG data; a second feature extraction module for performing intra-channel feature extraction on the EEG data based on the inter-channel mutual information features to obtain intra-channel features of each channel in the EEG data; and a detection result determination module for determining the abnormal discharge detection result of the subject based on a plurality of the intra-channel features.

[0021] In one embodiment of the present disclosure, the data acquisition module includes a data acquisition unit, which is used to: acquire the initial EEG signal of the subject, and divide the initial EEG signal into multiple time windows; perform feature transformation processing on the initial EEG signal in each time window to obtain the EEG data, and the EEG data is a time-frequency spectrum diagram of the initial EEG signal.

[0022] In one embodiment of the present disclosure, the abnormal brain discharge detection device also includes an initial processing module, which is used to: perform initial feature extraction processing on the EEG data to obtain an initial feature map; perform dimensionality transformation processing on the initial feature map, adjust the channel dimension position of the initial feature map, and obtain dimensionality transformation data of each channel; perform convolution processing on multiple dimensionality transformation data respectively to obtain reduced dimensionality feature data of each channel; perform data expansion processing on the reduced dimensionality feature data of each channel to obtain one-dimensional channel data corresponding to each channel.

[0023] In one embodiment of the present disclosure, the first feature extraction module includes a first feature extraction unit, which is used to: obtain one-dimensional channel data corresponding to each channel in the EEG data; encode the multiple one-dimensional channel data to perform inter-channel feature extraction on the EEG data to obtain the inter-channel mutual information feature.

[0024] In one embodiment of the present disclosure, the inter-channel mutual information feature includes a first inter-channel mutual information feature and a second inter-channel mutual information feature, and the first feature extraction unit includes a first feature extraction sub-unit, which is used to: encode the multiple one-dimensional channel data to obtain the first inter-channel mutual information feature; convolve the first inter-channel mutual information feature to obtain a reduced-dimensional channel feature; encode the reduced-dimensional channel feature to obtain the second inter-channel mutual information feature.

[0025] In one embodiment of the present disclosure, the second feature extraction module includes a second feature extraction unit, which is used to: obtain an initial feature map corresponding to the EEG data; perform intra-channel feature extraction on the initial feature map to obtain initial single-channel features of each of the channels; determine the channel weights corresponding to each of the channels based on the mutual information features between the channels; and determine the intra-channel features of each of the channels based on the channel weights and the initial single-channel features.

[0026] In one embodiment of the present disclosure, the second feature extraction unit includes a weight determination subunit, which is used to: decode the inter-channel mutual information features to obtain channel decoding data corresponding to each channel in the EEG data; determine the channel value of each channel according to the channel decoding data, and use the channel value as the channel weight of each channel.

[0027] In one embodiment of the present disclosure, the second feature extraction unit includes a second feature extraction sub-unit, which is used to: obtain the initial single-channel feature obtained through the current convolution processing; adjust the channel weight of the initial single-channel feature according to the channel weight; and determine the intra-channel feature based on the adjusted channel weight and the initial single-channel feature.

[0028] In a fourth aspect of the embodiments of the present disclosure, an electronic device is provided, comprising: a processor; and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the abnormal brain discharge detection method as described above is implemented.

[0029] According to a fifth aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the abnormal brain discharge detection method according to any one of the above items.

[0030] According to a sixth aspect of an embodiment of the present disclosure, there is provided a computer program product, comprising a computer program, which, when executed by a processor, implements any one of the above-mentioned methods for detecting abnormal brain discharges.

[0031] According to the technical solution of the embodiment of the present disclosure, on the one hand, feature extraction of EEG data can not only capture the mutual information between different channels of EEG data, but also extract the time-frequency domain information of a single channel, thereby realizing comprehensive and sufficient extraction of features of EEG data from multiple angles, which are used in the determination process of abnormal discharge detection results. On the other hand, the spatial mutual information of EEG data and the time-frequency domain information of a single channel are effectively used at the same time, so that the abnormal discharge detection results have higher detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, in which:

[0033] Figure 1A A schematic diagram showing a system architecture in this exemplary embodiment.

[0034] Figure 1B A schematic diagram showing another system architecture in this exemplary embodiment.

[0035] Figure 2 The flowchart of the abnormal brain discharge detection method according to some embodiments of the present disclosure is schematically shown;

[0036] Figure 3 Schematically shows a structural diagram of an abnormal brain discharge detection system according to some embodiments of the present disclosure;

[0037] Figure 4 Schematically shows a structural diagram of a Transformer module according to some embodiments of the present disclosure;

[0038] Figure 5 Schematically illustrates a structural diagram of MBConv and Fused-MBConv convolution blocks according to some embodiments of the present disclosure;

[0039] Figure 6 A schematic block diagram of an abnormal brain discharge detection device according to some embodiments of the present disclosure is schematically shown;

[0040] Figure 7 A schematic diagram schematically illustrates a storage medium according to an example embodiment of the present disclosure;

[0041] Figure 8 A block diagram of an electronic device according to an exemplary embodiment of the invention is schematically shown.

[0042] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts. DETAILED DESCRIPTION

[0043] The principles and spirit of the present disclosure will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present disclosure, and are not intended to limit the scope of the present disclosure in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0044] Those skilled in the art will appreciate that the embodiments of the present disclosure may be implemented as a system, device, apparatus, method or computer program product. Therefore, the present disclosure may be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0045] According to an embodiment of the present disclosure, a method for detecting abnormal brain discharge, an abnormal brain discharge detection device, a medium and an electronic device are proposed.

[0046] In this article, it is important to understand that the terms involved, such as the convolutional network EfficientNetV2, is an optimized deep learning model that is particularly suitable for processing large amounts of image data. It uses depthwise separable convolution to reduce model complexity while improving model performance through reasonable network depth, width, and resolution. The Squeeze-and-Excitation module (SE module) is a network structure improvement that aims to improve the performance of Convolutional Neural Networks (CNN) by adaptively recalibrating the feature responses of channels within the model.

[0047] In addition, any number of elements in the drawings is for illustration and not limitation, and any naming is only for distinction and does not have any limiting meaning.

[0048] The principle and spirit of the present disclosure are explained in detail below with reference to several representative embodiments of the present disclosure. SUMMARY OF THE INVENTION

[0050] In one technical solution, a convolutional model such as EfficientNetV2 is used for epileptiform discharge signal detection. The EfficientNetV2 model can be simply summarized as a stack of MBCON convolutional blocks and Fused-MBCon convolutional blocks. The MBCON convolutional block contains a depth-separable convolutional layer, a 1*1 point convolution, and an optional SE module and batch normalization layer, aiming to reduce the number of parameters and computational complexity of the model. The Fused-MBCon convolutional block uses a 3*3 convolutional layer to replace the depth-separable convolutional layer and the 1*1 point convolutional layer. Its main advantage is that it significantly reduces the number of parameters and computational cost while maintaining or improving model performance. By fusing multiple operation steps, Fused-MBConv can reduce information loss between convolutional layers and improve the efficiency of feature extraction.

[0051] However, the SE module generates statistics of feature channels through a global average pooling operation (squeeze), and then learns how to effectively use these statistics to highlight (or "excite") useful features and suppress less useful features through a fully connected neural network (excitation). Subsequently, these calibrated weights are used to adjust each channel of the original feature map. Since the SE module operates independently on each channel, it does not capture the dependencies between features across different spatial locations very well.

[0052] Based on the above content, the basic idea of ​​the present disclosure is to obtain the EEG data of the subject; perform inter-channel feature extraction on the EEG data to obtain the inter-channel mutual information features of the EEG data; perform intra-channel feature extraction on the EEG data based on the inter-channel mutual information features to obtain the intra-channel features of each channel in the EEG data; and determine the abnormal discharge detection results of the subject based on multiple intra-channel features. By performing feature extraction on the EEG data, not only can the mutual information between different channels of the EEG data be captured, but also the time-frequency domain information of a single channel can be extracted, so that the abnormal discharge detection results have higher detection accuracy.

[0053] After introducing the basic principles of the present disclosure, various non-limiting embodiments of the present disclosure are described in detail below.

[0054] Application Scenario Overview

[0055] It should be noted that the following application scenarios are only shown to facilitate understanding of the spirit and principle of the present disclosure, and the embodiments of the present disclosure are not limited in this respect. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario.

[0056] The embodiments of the present disclosure can be applied to the detection of suspected patients and related scenarios of auxiliary diagnosis and treatment. For example, a user is suspected of having epilepsy. During the treatment in the hospital, the doctor needs to understand the user's EEG state, especially whether there is abnormal discharge in the brain. The abnormal discharge detection method of the human brain disclosed in the present disclosure can be used for detection. The application scenario is specifically described below in combination with the system architecture.

[0057] Figure 1A A schematic diagram of a system architecture for abnormal brain discharge detection is shown. The system architecture includes a test object 101, a biomedical monitoring device 102, a video acquisition device 103, and a processing device 104. When abnormal brain discharge detection is required for the test object 101, the biomedical monitoring device 102 and the video acquisition device 103 can be set for the test object 101 to collect corresponding data, and the data is processed by the processing device 104.

[0058] The biomedical monitoring device 102 may include detection electrodes 1021, 1022, 1023 and a main device 1024. The detection electrodes 1021, 1022, 1023 may contact different parts of the subject 101 to collect electrical signals. The main device 1024 aggregates and processes the electrical signals to obtain biomedical monitoring data or biomedical characteristic data. Exemplarily, the biomedical monitoring device 102 may be an electroencephalogram (EEG) monitoring device, and the detection electrodes 1021, 1022, 1023 may be fixed to various locations on the head of the subject 101 to collect multi-channel EEG signals. Of course, Figure 1A The biomedical monitoring device 102 shown is merely exemplary and may also include a helmet, a hat, bedding, packaged portable detection electrodes, and other components, which are not limited in the present disclosure.

[0059] The video acquisition device 103 is used to collect video monitoring data of the object 101, such as a surveillance camera, a mobile phone with a shooting function, etc., which can be set at a position where the camera faces the object 101, and continuously monitors and shoots to obtain video monitoring data. In one embodiment, the video acquisition device 103 may include a dual-channel camera, which respectively shoots the face and body of the object to be tested, so as to respectively collect face video monitoring data and body video monitoring data.

[0060] The biomedical monitoring device 102 and the video acquisition device 103 can be connected to the processing device 104 for communication, such as through a wired or wireless communication link, so that the biomedical monitoring device 102 and the video acquisition device 103 send the collected data to the processing device 104. The processing device 104 can execute the human brain abnormal discharge detection method in this exemplary embodiment, process the acquired multimodal data, and obtain the abnormal discharge detection result of the subject. In one embodiment, the processing device 104 may include a display, which can display the abnormal discharge detection result, and can also display one or more of the biomedical monitoring data and the video monitoring data.

[0061] Any two or more of the biomedical monitoring device 102, the video acquisition device 103, and the processing device 104 can be integrated into the same device. Exemplarily, the biomedical monitoring device 102 and the processing device 104 can be integrated into the same device, or more specifically, the main device 1024 of the biomedical monitoring device 102 can serve as the processing device 104. Alternatively, the video acquisition device 103 can be integrated into the biomedical monitoring device 102 or the processing device 104, such as the biomedical monitoring device 102 can include a head-mounted device, in which detection electrodes 1021, 1022, and 1023 can be set. When the subject 101 wears the head-mounted device, the head contacts the detection electrodes 1021, 1022, and 1023. A fixed bracket can also be set on the head-mounted device, and the other end of the fixed bracket is fixedly placed on the video acquisition device 103 (such as a mobile phone can be fixedly placed by a clamp), and the video acquisition device 103 faces the face and body of the subject 101, and can capture and obtain video monitoring data.

[0062] Figure 1BA schematic diagram of another system architecture for abnormal brain discharge detection is shown. The system architecture includes a subject 101, a biomedical monitoring device 102, a video acquisition device 103, a data transceiver device 105, and a server 106. The biomedical monitoring device 102 and the video acquisition device 103 can be connected to the data transceiver device 105 for communication, and the data transceiver device 105 is connected to the server 106 for communication. Thus, after the data transceiver device 105 obtains the biomedical monitoring data or biomedical feature data, face video data, and body key point data, it sends it to the server 106. The server 106 may include any form of data processing server such as a cloud server and a distributed server. After obtaining the multimodal data sent by the data transceiver device 105, the server 106 obtains the abnormal discharge detection result of the subject by executing the abnormal brain discharge detection method in this exemplary embodiment. In one embodiment, the server 106 can return the abnormal discharge detection result to the data transceiver device 105 to display it on the data transceiver device 105 or the biomedical monitoring device 102.

[0063] Figure 1B The system architecture shown is suitable for portable scenarios. For example, the subject 101 can use the portable biomedical monitoring device 102 and the video acquisition device 103 at home, office, or any other place. The data transceiver device 105 sends the data to the server 106 to realize abnormal brain discharge detection. In this way, the user can learn the abnormal brain discharge detection results without leaving home. When encountering problems such as suspected epilepsy, the obtained abnormal brain discharge detection results can be sent to the doctor to help the doctor judge the condition.

[0064] In one embodiment, Figure 1A or Figure 1B The system architecture shown may also include a motion capture device, which may include one or more sensors, which are bound to the key parts of the body of the subject 101 to sense the movement of each part and collect the key point data of the body of the subject 101. For example, the sensor can be bound to the arm joints, fingers and palms of the subject 101 (such as the sensor is set at the fingers and palms of the motion capture gloves, when the subject 101 wears the motion capture gloves, the sensor is located at the fingers and palms of the subject 101), and the motion capture device can collect the real-time position data of the arm joints, fingers and palms of the subject 101 to obtain the key point data of the body. Of course, the present disclosure does not limit the number of sensors and the bound body parts. In addition to the above-mentioned arm joints, fingers and palms, the sensors of the motion capture device can be bound to other key parts of the subject 101 according to specific needs to collect the corresponding key point data of the body.

[0065] It should be understood that the application scenario shown in Figure 1 is only an example in which the embodiments of the present disclosure can be implemented. The scope of application of the embodiments of the present disclosure is not limited in any aspect by the application scenario.

[0066] Exemplary Methods

[0067] In the following, we will refer to the application scenario in Figure 1. Figure 2 To describe the abnormal brain discharge detection method according to the exemplary embodiment of the present disclosure. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principle of the present disclosure, and the embodiments of the present disclosure are not limited in this respect. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario.

[0068] The present disclosure first provides a method for detecting abnormal brain discharges. The method can be executed by a terminal device or a server, and the present disclosure does not make any special limitation on this. In this exemplary embodiment, the method is described by taking a server executing the method as an example.

[0069] Reference Figure 2 As shown, the abnormal brain discharge detection method may include the following steps S210 to S240:

[0070] Step S210, obtaining the EEG data of the subject.

[0071] In some example embodiments of the present disclosure, the EEG data of the subject may be obtained by using an electroencephalogram (EEG) acquisition device to record the electrical activity of neurons in the cerebral cortex by placing electrodes on the subject's scalp, and converting these records into graphical or digital data.

[0072] Obtain the EEG data of the subject, such as using the data directly collected by the EEG acquisition device as the EEG data of the subject, or preprocessing the initial EEG data collected by the EEG acquisition device. Use the EEG data of the subject as the data basis for subsequent determination of abnormal discharge detection results.

[0073] Step S220, extracting inter-channel features of the EEG data to obtain inter-channel mutual information features of the EEG data.

[0074] In some example embodiments of the present disclosure, inter-channel feature extraction may be a process of extracting features from mutual information between multiple channels of EEG data. Inter-channel mutual information features may be features used to characterize mutual information between different channels of EEG data.

[0075] Different EEG channels in EEG data represent the spatial positions of the head electrical signals recorded by them. When the brain produces different epileptic lesions, the activation areas of the brain are different, and the strength of the activated electrical signals is also different, which makes rich mutual information between EEG channels. In order to solve the problem that related solutions can only extract the features of each channel in EEG data, but cannot well capture the features across different spatial positions (such as between different channels), the present disclosure extracts the features between different channels in EEG data, obtains the mutual information features between channels of EEG data, and uses it in the process of determining abnormal discharge detection results.

[0076] Step S230, extracting intra-channel features of the EEG data based on the inter-channel mutual information features to obtain intra-channel features of each channel in the EEG data.

[0077] In some example embodiments of the present disclosure, intra-channel feature extraction may be a process of extracting features from a single channel of EEG data. Intra-channel features may be features used to characterize time and frequency information of a single channel in EEG data.

[0078] By extracting features from EEG data, we can obtain the inter-channel mutual information features of EEG data. In order to extract both inter-channel features and intra-channel features of EEG data, the present disclosure proposes a parallel extraction scheme. While extracting inter-channel features from EEG data, we also extract intra-channel features from EEG data, and apply the inter-channel mutual information features to the intra-channel feature extraction process of EEG data to obtain the intra-channel features of each channel in the EEG data. EEG data in a single channel has rich frequency information, which may represent various operating states of the body, such as eye movement signals and sleep signals.

[0079] Step S240, determining the abnormal discharge detection result of the object under test according to the characteristics in the multiple channels.

[0080] In some example embodiments of the present disclosure, the abnormal discharge detection result may be a detection result of whether abnormal discharge exists in the EEG data of the subject.

[0081] By extracting the intra-channel features in the above manner and integrating the features between different channels in the EEG data, and analyzing multiple intra-channel features in the EEG data, the abnormal discharge detection result of the subject can be determined, that is, whether there is abnormal discharge in the EEG data of the subject.

[0082] In the abnormal brain discharge detection method provided in this example embodiment, on the one hand, feature extraction of EEG data can not only capture the mutual information between different channels of EEG data, but also extract the time-frequency domain information of a single channel, thereby realizing comprehensive and sufficient extraction of features of EEG data from multiple angles, and using it in the determination process of abnormal discharge detection results. On the other hand, the spatial mutual information of EEG data and the time-frequency domain information of a single channel are effectively used at the same time, so that the abnormal discharge detection results have higher detection accuracy.

[0083] Next, the above steps of this exemplary embodiment are described in more detail.

[0084] In one embodiment of the present disclosure, for step S210, obtaining EEG data of the subject includes: obtaining the initial EEG signal of the subject, dividing the initial EEG signal into multiple time windows; performing feature transformation processing on the initial EEG signal in each time window to obtain EEG data, which is a time-frequency spectrum diagram of the initial EEG signal.

[0085] Among them, the EEG data of the subject may be data obtained by performing time-frequency domain transformation processing on the initial EEG signal of the subject. The initial EEG signal may be signal data collected by an EEG acquisition device. The time-frequency domain transformation processing may be a processing process of converting the initial EEG signal into a time-frequency domain representation. The time window may be a local time window applied to the time-frequency transformation processing of the initial EEG signal. The feature transformation processing may be a processing process of performing time-domain and frequency-domain transformation on the initial EEG signal. The time-frequency spectrum diagram, also known as the spectrogram, may be a time-frequency spectrum diagram drawn by first performing Fourier transformation on the initial EEG signal, and then using the horizontal axis as time, the vertical axis as frequency, and using color to represent the amplitude. The time-frequency spectrum diagram can represent the changes in the frequency and amplitude of the initial EEG signal over time in one diagram.

[0086] The reason why EEG technology occupies an important position in the study of abnormal brain discharges (such as epilepsy research) is partly due to its high temporal resolution. This feature enables EEG to accurately capture rapid changes in brain activity, including those transient potential fluctuations that occur with the onset of epilepsy. However, a major challenge in using EEG signals for epileptiform discharge detection is the presence of different physiological signals as noise. The physiological information contained in EEG signals is very rich and difficult to distinguish. It is very difficult to distinguish epileptiform discharge information from many signals. Therefore, in the research process, it is generally necessary to extract the original time signal into spatiotemporal (frequency) features.

[0087] For the initial EEG signal directly collected by the EEG acquisition device, it is necessary to perform data preprocessing on it to obtain the corresponding time-frequency spectrum. For example, the present disclosure can use Short-Time Fourier Transform (STFT) to perform data preprocessing on the initial EEG signal. STFT is a powerful tool that allows the time domain signal to be converted into a time-frequency domain representation, which provides a balance between the time domain and the frequency domain.

[0088] The specific processing process is as follows: STFT applies a local time window to the initial EEG signal, divides the initial EEG signal into multiple time windows, and then performs Fourier transform on the signals in these local windows to obtain the spectrum information in different time windows. The present disclosure uses STFT with divided time windows for feature extraction, converts the relationship between the initial EEG signal and the time change into a spectrum with respect to the time-frequency change, which can make the abnormal discharge detection result determined based on the extracted features in the subsequent feature extraction process have a higher accuracy.

[0089] For example, in this embodiment, when performing STFT on the initial EEG signal, each segment of EEG data is 4s long, and the 4s EEG signal is subjected to STFT transformation, the time window is set to 0.5s, and the update step is updated every 0.0625 seconds. The initial EEG signal is subjected to time-frequency transformation using the above time window to obtain a time-frequency spectrum of the initial EEG signal as the EEG data of the subject.

[0090] In one embodiment of the present disclosure, initial feature extraction processing is performed on EEG data to obtain an initial feature map; dimensionality transformation processing is performed on the initial feature map to adjust the channel dimension position of the initial feature map to obtain dimensionality transformation data of each channel; convolution processing is performed on multiple dimensionality transformation data to obtain dimensionality reduction feature data of each channel; data expansion processing is performed on the dimensionality reduction feature data of each channel to obtain one-dimensional channel data corresponding to each channel.

[0091] Among them, the initial feature map can be a feature map obtained by performing initial feature extraction processing on the EEG data. The dimension transformation processing can be a processing process for adjusting the position of certain feature dimensions in the initial feature map. The channel dimension position can be the position of the channel dimension feature in the initial feature map. The dimension transformation data can be data obtained by adjusting the channel dimension position in the initial feature map. The reduced dimension feature data can be data obtained by performing convolution processing on the dimension transformation data. The one-dimensional channel data can be data obtained by flattening the data of each channel of the reduced dimension feature data (such as one-dimensional expansion processing).

[0092] The EEG data of the subject is used as input data, and the input data is processed with initial feature extraction. For example, the input data is first processed by a CNN convolution, the purpose of which is to perform preliminary feature extraction on the original input data and obtain the corresponding initial feature map. The data after the above processing (initial feature map) is copied into two copies for subsequent processing. One copy of the data continues to extract intra-channel features along the original path, while the other copy is sent to the inter-channel feature extraction module to extract inter-channel features of the EEG data and obtain inter-channel mutual information features.

[0093] Before performing inter-channel feature extraction on the initial feature map data, a dimensional change is performed on the initial feature map to move the channel dimension to the second position. For example, the initial feature map can usually contain dimensions such as the number of samples (batchside), the number of feature maps, channels, frequency dimensions, and time dimensions. The "channel" dimension is moved before the "number of feature maps" to obtain the dimensional transformation data of each channel. This operation places the feature information of each channel in the latter two dimensions, and the dimensional rearrangement provides the basis for the next processing steps.

[0094] After obtaining the dimensionally transformed data of multiple channels, the multiple dimensionally transformed data are convolved separately, such as a layer of convolution on the dimensionally transformed data, using a 1x1 convolution kernel and setting the step size to 1, in order to further refine the feature information of each channel and obtain reduced-dimensional feature data. For the obtained reduced-dimensional feature data, the data of each channel is flattened, and the features of each EEG channel are converted into a one-dimensional data (1, X) to prepare the data for subsequent inter-channel feature extraction. The purpose of this processing method is to provide a deeper level of mining for the unique features of each EEG channel, and through dimensional transformation and feature stretching, it provides suitable input for inter-channel feature extraction of the data; it enables the inter-channel feature extraction module to effectively learn and evaluate the relative importance of different EEG channels, and then optimize the overall data processing and information flow path by automatically adjusting the channel weights.

[0095] In one embodiment of the present disclosure, for step S220, inter-channel feature extraction is performed on the EEG data to obtain inter-channel mutual information features of the EEG data, including: obtaining one-dimensional channel data corresponding to each channel in the EEG data; encoding multiple one-dimensional channel data to extract inter-channel features of the EEG data to obtain inter-channel mutual information features.

[0096] The processed one-dimensional channel data can be sent to the inter-channel feature extraction module, which can include a CNN convolution layer and an encoder (Transformer) encoding layer. For example, the module can be composed of a combination of multiple CNN convolution layers and Transformer encoding layers. The above structure is used to encode the one-dimensional channel data of each channel to extract the inter-channel features of the EEG data and obtain the inter-channel mutual information features.

[0097] Through the Transformer encoding layer, the inter-channel feature extraction module can learn the mutual information between channels. This learning process allows the feature extraction module to adaptively recalibrate the feature response of the channel, thereby significantly improving the overall performance of anomaly detection. In addition, the Transformer encoding layer can accurately capture and process the complex relationships and dynamics between channels through its built-in attention mechanism, and can effectively process and analyze EEG data.

[0098] In one embodiment of the present disclosure, multiple one-dimensional channel data are encoded to extract inter-channel features of EEG data to obtain inter-channel mutual information features, including: encoding multiple one-dimensional channel data to obtain first inter-channel mutual information features; convolution processing the first inter-channel mutual information features to obtain reduced-dimensional channel features; encoding the reduced-dimensional channel features to obtain second inter-channel mutual information features.

[0099] The first inter-channel mutual information feature may be a feature of the inter-channel mutual information obtained after encoding one-dimensional channel data once. The reduced-dimensional channel feature may be a feature obtained by convolution processing the first inter-channel mutual information feature. The second inter-channel mutual information feature may be a feature of the inter-channel mutual information obtained after multiple convolution processing and encoding processing are performed on the first inter-channel mutual information feature.

[0100] The inter-channel feature extraction module is composed of multiple CNN convolutional layers and Transformer encoding layers, for example, multiple structures including one CNN convolutional layer and four Transformer encoding layers. It is worth noting that there is no CNN convolutional layer before the first Transformer encoding module in the network structure of the overall inter-channel feature extraction module. The one-dimensional channel data is encoded only by the Transformer encoding layer structure, and the encoding processing result obtained is used as the first inter-channel mutual information feature.

[0101] For the obtained first channel-to-channel mutual information feature, the above data is convolved and encoded respectively. For example, when the data flows into the feature extraction structure composed of the CNN convolution layer and the Transformer encoding layer, the first channel-to-channel mutual information feature is first convolved to obtain the reduced-dimensional channel feature. It is first processed through a CNN layer, and the data is in the form of (number of samples, number of channels, number of channel features). The convolution kernel size used here is 3x3, and the step size is set to (2,1). The main purpose of the CNN convolution layer in this step is to reduce the number of channels, so as to achieve alignment of the EEG channel dimension with the FusedMBcon convolution block during data propagation.

[0102] After data alignment, it is sent to the Transformer encoding layer, which encodes the dimension-reduced channel features to obtain the second channel-to-channel mutual information features. The mutual information between channels can be learned through the Transformer encoding layer model. After the data is processed by the Transformer encoding module, it will also be copied into two copies. The first copy of the data continues along the original data flow path and is sent to the next Transformer encoding module. This process will continue until the data passes through all Transformer encoding modules. The continuous data transmission mechanism ensures that the feature extraction module can deeply learn and extract the complex features of the data while maintaining the consistency and efficiency of the data processing flow.

[0103] In one embodiment of the present disclosure, for step S230, intra-channel features of EEG data are extracted based on inter-channel mutual information features to obtain intra-channel features of each channel in the EEG data, including: obtaining an initial feature map corresponding to the EEG data; performing intra-channel feature extraction on the initial feature map to obtain initial single-channel features of each channel; determining the channel weight corresponding to each channel based on the inter-channel mutual information features; and determining the intra-channel features of each channel according to the channel weights and the initial single-channel features.

[0104] The initial single channel feature may be a feature of a single channel obtained by performing only intra-channel feature extraction on the initial feature map, and the channel weight may be a calculation weight corresponding to each channel during the feature extraction process.

[0105] After the data is processed by the Transformer encoding module, another copy of the data will be used for the intra-channel feature extraction process. The present disclosure adopts an intra-channel feature extraction module to extract intra-channel features from the initial feature map. For example, the intra-channel feature extraction module also adopts the EfficientNetV2 structure in the related technical solution to obtain the initial single-channel features of each channel.

[0106] At the same time, the inter-channel mutual information feature between multiple channels is taken as a factor to be considered in the intra-channel feature extraction. In the process of intra-channel feature extraction, the inter-channel mutual information feature is used as one of the influencing factors to determine the channel weights corresponding to each channel, and the channel weights of different channels are adjusted. Then, the intra-channel features of each channel are determined based on the channel weights and the initial single channel features. This mechanism optimizes the overall data processing efficiency and performance of EEG data.

[0107] In one embodiment of the present disclosure, the channel weights corresponding to each channel are determined based on the mutual information features between channels, including: decoding the mutual information features between channels to obtain channel decoding data corresponding to each channel in the EEG data; determining the channel value of each channel according to the channel decoding data, and using the channel value as the channel weight of each channel.

[0108] The channel decoding data may be feature data obtained by decoding the extracted inter-channel mutual information features. The channel value may be a specific value of each channel.

[0109] After the Transformer encoding module processes the other data, it is sent to the decoding module, which is mainly composed of multiple layers of CNN. The CNN layer and pooling in the decoding module use a one-dimensional convolution kernel (for example, a 1x3 convolution kernel) and a step size of (1,2) to ensure that the number of channels remains unchanged.

[0110] After decoding the inter-channel mutual information features, channel decoding data is obtained and compressed into one-dimensional data. At this time, the value of each channel is regarded as the channel weight of the channel, and these channel weights are passed back to the Fused-MBcon convolution block. In the TransEfficientNet model, the output of each convolution block will receive the channel weight output from the inter-channel feature extraction module as the data basis for intra-channel feature extraction for different channels. This mechanism not only enhances the ability to adapt to the importance of different channel features, but also optimizes the overall data processing efficiency and model performance.

[0111] In one embodiment of the present disclosure, the intra-channel features of each channel are determined based on the channel weights and the initial single-channel features, including: obtaining the initial single-channel features obtained through the current convolution processing; adjusting the channel weights of the initial single-channel features based on the channel weights; and determining the intra-channel features based on the adjusted channel weights and the initial single-channel features.

[0112] The initial single-channel feature is obtained by convolutionally processing the initial feature map using the convolution block structure in the intra-channel feature extraction module. The decoding module can obtain the channel weights transmitted by the inter-channel feature extraction module, and adjust the weights of the initial single-channel features of each channel according to the received channel weights. For example, the specific weight adjustment method is: intra-channel feature = (output * (1 + channel weight)), so as to achieve dynamic adjustment of channel weights. Through this design, the feature extraction module can not only effectively process and analyze EEG data, but also automatically adjust the channel weights according to the characteristics and importance of the data, further improving the accuracy and efficiency of processing.

[0113] In summary, the abnormal brain discharge detection method disclosed in the present invention obtains the EEG data of the subject; performs inter-channel feature extraction on the EEG data to obtain the inter-channel mutual information features of the EEG data; performs intra-channel feature extraction on the EEG data based on the inter-channel mutual information features to obtain the intra-channel features of each channel in the EEG data; and determines the abnormal discharge detection result of the subject according to multiple intra-channel features. On the one hand, feature extraction on the EEG data can not only capture the mutual information between different channels of the EEG data, but also extract the time-frequency domain information of a single channel, thereby achieving comprehensive and sufficient extraction of the features of the EEG data from multiple angles, which are used in the process of determining the abnormal discharge detection results. On the other hand, the spatial mutual information of the EEG data and the time-frequency domain information of a single channel are effectively utilized at the same time, so that the abnormal discharge detection results have higher detection accuracy.

[0114] Exemplary Systems

[0115] After introducing the method of the exemplary embodiment of the present disclosure, next, refer to Figure 3 An abnormal brain discharge detection system according to an exemplary embodiment of the present disclosure is described.

[0116] exist Figure 3 In the embodiment, the abnormal brain discharge detection system 300 may include: a data determination module 310, a channel weight determination module 320, a fusion convolution module 330 and an abnormality detection module 340.

[0117] Among them: the data determination module 310 is used to obtain the EEG data of the subject, perform initial feature extraction processing on the EEG data, and obtain an initial feature map; the channel weight determination module 320 is used to determine the one-dimensional channel data corresponding to each channel according to the initial feature map, encode multiple one-dimensional channel data, and obtain the channel mutual information characteristics of the EEG data; and determine the channel weight corresponding to each channel based on the channel mutual information characteristics; the fusion convolution module 330 is used to extract intra-channel features on the initial feature map to obtain the initial single-channel features of each channel, and determine the intra-channel features of each channel according to the channel weight and the initial single-channel features; the abnormality detection module 340 is used to determine the abnormal discharge detection result of the subject according to multiple intra-channel features.

[0118] In order to overcome the problem that deep learning models in related solutions generally cannot take into account time information, spatial information and mutual information of EEG channels at the same time, so that the information in EEG signals is not fully utilized and the accuracy of the algorithm model still has room for improvement, this paper constructs a new abnormal brain discharge detection system with a new model structure (TransEfficientNet), which combines EfficientNetV2 with the Transformer encoding layer to improve model performance through parallel training. This model specifically introduces a channel weight determination module, which is based on a combination of CNN convolutional layers and Transformer encoding layers to capture and process the complex features of EEG data.

[0119] Specifically, the data determination module 310 performs initial feature extraction processing on the acquired EEG data of the subject to obtain an initial feature map. The EEG data can be data obtained after preprocessing such as video conversion of the initial EEG signal collected by the EEG acquisition device, and it is used in the subsequent feature extraction process. The data after video conversion processing is used to accelerate the convergence speed of the model while having a higher accuracy rate. For example, the EEG data 311 is input into the convolution layer 312, and the convolution layer 312 performs a CNN convolution processing on the EEG data 311 to obtain an initial feature map.

[0120] After obtaining the initial feature map, the data extracted through the initial feature is copied into two copies for subsequent processing. One copy of the data continues along the original path and is sent to the FusedMBcon convolution block of the fused convolution module 330, while the other copy of the data is sent to the channel weight determination module 320 proposed in the present disclosure. The channel weight determination module 320 performs dimension transformation, convolution, and data expansion processing on the initial feature map to obtain one-dimensional channel data corresponding to each channel; the encoding layer then encodes the multiple one-dimensional channel data to obtain the channel mutual information features of the EEG data. After extracting the channel mutual information features through the channel weight determination module 320, the channel mutual information features can also be used as the data basis to determine the channel weights corresponding to each channel.

[0121] The fused convolution module 330 can use the EfficientNetV2 network structure to extract intra-channel features from the initial feature map to obtain the initial single-channel features of each channel. In addition, the abnormal brain discharge detection system 300 can improve the processing performance of the entire system by extracting inter-channel features and intra-channel features in parallel. The fused convolution module 330 can adjust the weights of the initial single-channel features corresponding to multiple channels based on the channel weights determined by the channel weight determination module 320, and determine the intra-channel features of each channel.

[0122] The abnormality detection module 340 performs analysis and processing based on the multiple channel features output by the fused convolution module 330, such as convolution processing, pooling processing and full connection processing, to finally determine the abnormal discharge detection result of the object under test.

[0123] The abnormal brain discharge detection system disclosed in the present invention, on the one hand, captures the time and frequency information in a single channel by fusing the convolution structure of the convolution module, and captures the mutual information between different EEG channels through the Transformer encoder, so that the abnormal brain discharge detection system can more comprehensively and fully understand the characteristics of EEG data from multiple different angles, so that the model can obtain higher accuracy.

[0124] Next, the above-mentioned system structure of this exemplary embodiment will be described in more detail.

[0125] In one embodiment of the present disclosure, the channel weight determination module 320 includes: a dimensionality transformation unit, which is used to perform dimensionality transformation processing on the initial feature map, adjust the channel dimension position of the initial feature map, and obtain dimensionality transformation data of each channel; a one-dimensional convolution unit, which is used to perform convolution processing on multiple dimensionality transformation data respectively to obtain reduced dimensionality feature data of each channel; a data expansion unit, which is used to perform data expansion processing on the reduced dimensionality feature data of each channel to obtain one-dimensional channel data corresponding to each channel.

[0126] In the channel weight determination module 320, a dimensional change is first performed on the initial feature map to move the channel dimension in the initial feature map data to the second position. This operation places the feature information of each channel in the latter two dimensions. The dimensional rearrangement lays the foundation for the subsequent processing steps of the channel weight determination module 320.

[0127] After obtaining the dimension transformation data of multiple channels, a one-dimensional convolution unit is first used to perform convolution processing on the multiple dimension transformation data. For example, the dimension transformation data is subjected to a layer of convolution, a 1x1 convolution kernel is used, and the step size is set to 1, in order to further refine the feature information of each channel and obtain the reduced dimension feature data. Then, a data expansion unit is used to flatten the data of each channel in the reduced dimension feature data, and the feature of each EEG channel is converted into a one-dimensional data (1, X), so as to prepare the data for the subsequent inter-channel feature extraction.

[0128] The purpose of this processing method is to provide a deeper level of mining for the unique features of each EEG channel in the abnormal brain discharge detection system, and to provide suitable input for inter-channel feature extraction of the data through dimensionality transformation and feature stretching; enabling the inter-channel feature extraction module to effectively learn and evaluate the relative importance of different EEG channels, and then optimize the overall data processing and information flow path by automatically adjusting the channel weights.

[0129] In one embodiment of the present disclosure, the channel weight determination module 320 also includes a first encoding unit and a second encoding unit, the second encoding unit includes a convolution layer; the first encoding unit is used to encode multiple one-dimensional channel data based on an attention mechanism to obtain a first channel mutual information feature; the convolution layer is used to convolve the first channel mutual information feature to obtain a reduced-dimensional channel feature; the second encoding unit is used to encode the reduced-dimensional channel feature based on an attention mechanism to obtain a second channel mutual information feature.

[0130] refer to Figure 4 , Figure 4 The structure diagram of the Transformer module according to some embodiments of the present disclosure is schematically shown. Figure 4 It can be seen that the Transformer module includes a multi-head attention module (Multi-Head Attention) 410, an add & normalization layer (Add & Norm) 420 and a feed forward module (Feed Forward) 430. The input data is extracted through the features of the above network structure to obtain the corresponding output data.

[0131] The first encoding unit can be a network structure that only includes a Transformer module, such as a four-layer Transformer encoding layer, and the second encoding unit can be a network structure that includes a convolutional layer (CNN) and a Transformer module, such as a CNN convolutional layer and a four-layer Transformer encoding layer. The first encoding unit can encode multiple one-dimensional channel data based on the attention mechanism inside the Transformer module to obtain the first channel mutual information feature. The mutual information between channels can be learned through the Transformer encoding layer model. The Transformer encoding layer can accurately capture and process the complex relationships and dynamics between channels through its built-in attention mechanism.

[0132] After obtaining the first inter-channel mutual information feature, the obtained data will continue to be encoded. After the data is sent to the second encoding unit, the first inter-channel mutual information feature is first convolved by the convolution layer to obtain the reduced-dimensional channel feature. At this time, the data is in the form of (number of samples, number of channels, number of channel features). The convolution kernel size used here is 3x3, and the step size is set to (2,1). The main purpose of the CNN convolution layer in this step is to reduce the number of channels, so as to achieve alignment of the EEG channel dimension with the FusedMBcon convolution block during data propagation.

[0133] Then, the attention mechanism of the Transformer encoding layer in the second encoding unit is used to encode the reduced dimension channel features to obtain the second channel mutual information features. This learning process allows the model to adaptively recalibrate the feature response of the channel, thereby significantly improving the overall performance of the model. After the data is processed by the Transformer encoding module, it will also be copied into two copies. The first copy of the data continues along the original data flow path and is sent to the next Transformer encoding module. This process continues until the data passes through all Transformer encoding modules. The continuous data transmission mechanism ensures that the model can deeply learn and extract the complex features of the data while maintaining the consistency and efficiency of the data processing flow.

[0134] The other data is sent to the decoding module, such as Figure 3 As shown in the figure, the decoding module is mainly composed of multiple layers of CNN. The CNN layer and pooling in the decoding module use a one-dimensional convolution kernel (for example, a 1x3 convolution kernel) and a step size of (1,2) to ensure that the number of channels remains unchanged. After being processed by the decoding module, the data is compressed into one-dimensional data. At this time, the model regards the value of each channel as the weight of the channel and passes these channel weights back to the Fused-MBcon convolution block.

[0135] In one embodiment of the present disclosure, the fused convolution module 330 includes a fused convolution unit, which is used to extract intra-channel features from the initial feature map to obtain initial single-channel features of each channel; and a feature determination unit, which is used to determine the intra-channel features of each channel based on the channel weights corresponding to each channel in the initial feature map and the initial single-channel features.

[0136] In the TransEfficientNet model, the output of each convolutional block receives the channel weight from the channel weight model and adjusts it according to the weight (output * (1 + channel weight)), so as to achieve dynamic adjustment of the channel weight. This mechanism not only enhances the model's ability to adapt to the importance of different channel features, but also optimizes the overall data processing efficiency and model performance.

[0137] The fused convolution module 330 can be implemented using the EfficientNetV2 structure, which includes an MBConv convolution block and a Fused-MBConv convolution block. Figure 5 , Figure 5 The structure diagram of the MBConv convolution block and the Fused-MBConv convolution block according to some embodiments of the present disclosure is schematically shown. Figure 5 It can be seen that the network structure of the MBConv convolution block 331 includes a 1*1 convolution layer 510, a depth convolution layer 520, and a SE module 530; the network structure of the Fused-MBConv convolution block 332 includes a 1*1 convolution layer 510 and a SE module 530.

[0138] The initial feature map is subjected to feature extraction by the Fused-MBConv convolution block 332, and the corresponding convolution output value is output by the Fused-MBConv convolution block 332 as the initial single-channel feature. Then, the initial single-channel features of different channels are weighted and adjusted in combination with the channel weight output by the channel weight determination module 320 to obtain the intra-channel features of each channel. Subsequently, the abnormal brain discharge detection system 300 will perform convolution, pooling and full connection processing on the intra-channel features of multiple channels to obtain abnormal discharge detection results.

[0139] Exemplary Devices

[0140] After introducing the method of the exemplary embodiment of the present disclosure, next, refer to Figure 6 An abnormal brain discharge detection device according to an exemplary embodiment of the present disclosure is described.

[0141] exist Figure 6 In the embodiment, the abnormal brain discharge detection device 600 may include: a data acquisition module 610, a first feature extraction module 620, a second feature extraction module 630 and a detection result determination module 640.

[0142] Among them: the data acquisition module 610 is used to acquire the EEG data of the subject; the first feature extraction module 620 is used to perform inter-channel feature extraction on the EEG data to obtain the inter-channel mutual information features of the EEG data; the second feature extraction module 630 is used to perform intra-channel feature extraction on the EEG data based on the inter-channel mutual information features to obtain the intra-channel features of each channel in the EEG data; the detection result determination module 640 is used to determine the abnormal discharge detection result of the subject based on multiple intra-channel features.

[0143] In one embodiment of the present disclosure, the data acquisition module 610 includes a data acquisition unit, which is used to: acquire the initial EEG signal of the subject, divide the initial EEG signal into multiple time windows; perform feature transformation processing on the initial EEG signal in each time window to obtain EEG data, which is a time-frequency spectrum diagram of the initial EEG signal.

[0144] In one embodiment of the present disclosure, the abnormal brain discharge detection device 600 also includes an initial processing module, which is used to: perform initial feature extraction processing on the EEG data to obtain an initial feature map; perform dimensionality transformation processing on the initial feature map, adjust the channel dimension position of the initial feature map, and obtain dimensionality transformation data of each channel; perform convolution processing on multiple dimensionality transformation data respectively to obtain reduced dimensionality feature data of each channel; perform data expansion processing on the reduced dimensionality feature data of each channel to obtain one-dimensional channel data corresponding to each channel.

[0145] In one embodiment of the present disclosure, the first feature extraction module 620 includes a first feature extraction unit, which is used to: obtain one-dimensional channel data corresponding to each channel in the EEG data; encode multiple one-dimensional channel data to extract inter-channel features of the EEG data to obtain inter-channel mutual information features.

[0146] In one embodiment of the present disclosure, the inter-channel mutual information feature includes a first inter-channel mutual information feature and a second inter-channel mutual information feature, and the first feature extraction unit includes a first feature extraction sub-unit, which is used to: encode multiple one-dimensional channel data to obtain the first inter-channel mutual information feature; convolve the first inter-channel mutual information feature to obtain a reduced-dimensional channel feature; encode the reduced-dimensional channel feature to obtain a second inter-channel mutual information feature.

[0147] In one embodiment of the present disclosure, the second feature extraction module 630 includes a second feature extraction unit, which is used to: obtain an initial feature map corresponding to the EEG data; perform intra-channel feature extraction on the initial feature map to obtain initial single-channel features of each channel; determine the channel weight corresponding to each channel based on the mutual information features between channels; and determine the intra-channel features of each channel based on the channel weights and the initial single-channel features.

[0148] In one embodiment of the present disclosure, the second feature extraction unit includes a weight determination subunit, which is used to: decode the mutual information features between channels to obtain channel decoding data corresponding to each channel in the EEG data; determine the channel value of each channel based on the channel decoding data, and use the channel value as the channel weight of each channel.

[0149] In one embodiment of the present disclosure, the second feature extraction unit includes a second feature extraction sub-unit, which is used to: obtain the initial single-channel feature obtained through the current convolution processing; adjust the channel weight of the initial single-channel feature according to the channel weight; and determine the intra-channel feature based on the adjusted channel weight and the initial single-channel feature.

[0150] Since the various functional modules of the abnormal brain discharge detection device of the example embodiment of the present disclosure correspond to the steps of the example embodiment of the abnormal brain discharge detection method described above, for details not disclosed in the embodiment of the device of the present disclosure, please refer to the embodiment of the abnormal brain discharge detection method described above in the present disclosure, and no further details will be given here.

[0151] It should be noted that, although several modules or units of the abnormal brain discharge detection device are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0152] In a third aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the abnormal brain discharge detection method as described in the first aspect above is implemented.

[0153] Exemplary Media

[0154] After introducing the device of the exemplary embodiment of the present disclosure, next, refer to Figure 7 A storage medium according to an exemplary embodiment of the present disclosure is described.

[0155] In some embodiments, various aspects of the present disclosure may also be implemented as a medium on which program code is stored, which, when executed by a processor of a device, is used to implement the steps of the abnormal brain discharge detection method according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification.

[0156] For example, when the processor of the device executes the program code, the following may be implemented: Figure 2The step S210 described in the above is to obtain the EEG data of the subject; the step S220 is to extract inter-channel features of the EEG data to obtain the inter-channel mutual information features of the EEG data; the step S230 is to extract intra-channel features of the EEG data based on the inter-channel mutual information features to obtain the intra-channel features of each channel in the EEG data; and the step S240 is to determine the abnormal discharge detection result of the subject based on multiple intra-channel features.

[0157] refer to Figure 7 As shown, a program product 700 for implementing the above-mentioned abnormal brain discharge detection method or implementing the above-mentioned abnormal brain discharge detection method according to an embodiment of the present disclosure is described, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto.

[0158] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0159] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, wherein readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Readable signal media may also be any readable medium other than a readable storage medium.

[0160] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0161] Exemplary Computing Devices

[0162] After introducing the abnormal brain discharge detection method, abnormal brain discharge detection device, and storage medium according to the exemplary embodiments of the present disclosure, next, reference is made to Figure 8 An electronic device according to an exemplary embodiment of the present disclosure is described.

[0163] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware embodiments, complete software embodiments (including firmware, microcode, etc.), or embodiments combining hardware and software aspects, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0164] In some possible embodiments, the electronic device according to the present disclosure may include at least one processing unit and at least one storage unit. The storage unit stores program code, and when the program code is executed by the processing unit, the processing unit executes the steps of the abnormal brain discharge detection method according to various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification. For example, the processing unit may execute the following steps: Figure 2 In step S210, the EEG data of the subject is obtained; in step S220, inter-channel feature extraction is performed on the EEG data to obtain inter-channel mutual information features of the EEG data; in step S230, intra-channel feature extraction is performed on the EEG data based on the inter-channel mutual information features to obtain intra-channel features of each channel in the EEG data; and in step S240, the abnormal discharge detection result of the subject is determined based on multiple intra-channel features.

[0165] Reference below Figure 8 , the electronic device is exemplarily described in the form of a general-purpose computing device. It should be understood that Figure 8 The electronic device 800 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0166] like Figure 8 As shown, the electronic device 800 may include: a processor 810 , a memory 820 , a bus 830 , an I / O (input / output) interface 840 , a network adapter 850 , and a display 860 .

[0167] The memory 820 may include a volatile memory, such as a RAM 821, a cache unit 822, and may also include a non-volatile memory, such as a ROM 823. The memory 820 may also include one or more program modules 824, such program modules 824 include but are not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or a combination thereof may include the implementation of a network environment. For example, the program module 824 may include each module in the above-mentioned device.

[0168] The processor 810 may include one or more processing units. For example, the processor 810 may include an AP (Application Processor), a modem processor, a GPU (Graphics Processing Unit), an ISP (Image Signal Processor), a controller, an encoder, a decoder, a DSP (Digital Signal Processor), a baseband processor and / or an NPU (Neural-Network Processing Unit), etc.

[0169] The processor 810 can be used to execute executable instructions stored in the memory 820, such as executing the above-mentioned abnormal brain discharge detection method.

[0170] The bus 830 is used to realize the connection between different components of the electronic device 800, and may include a data bus, an address bus, and a control bus.

[0171] The electronic device 800 can communicate with one or more external devices 900 (eg, a keyboard, a mouse, an external controller, etc.) through the I / O interface 840 .

[0172] The electronic device 800 can communicate with one or more networks through the network adapter 850. For example, the network adapter 850 can provide mobile communication solutions such as 3G / 4G / 5G, or wireless communication solutions such as wireless LAN, Bluetooth, near field communication, etc. The network adapter 850 can communicate with other modules of the electronic device 800 through the bus 830.

[0173] The electronic device 800 can display a graphical user interface through the display 860, such as an interface that displays abnormal brain discharge detection results and EEG data.

[0174] although Figure 8Not shown, other hardware and / or software modules may also be provided in the electronic device 800, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0175] It should be noted that, although several units / modules or sub-units / modules of the abnormal brain discharge detection device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0176] In addition, although the operations of the disclosed method are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0177] Although the spirit and principle of the present disclosure have been described with reference to several specific embodiments, it should be understood that the present disclosure is not limited to the disclosed specific embodiments, and the division of various aspects does not mean that the features in these aspects cannot be combined to benefit, and such division is only for the convenience of expression. The present disclosure is intended to cover various modifications and equivalent arrangements included in the spirit and scope of the attached claims.

Claims

1. A method for detecting abnormal brain discharges, characterized in that: include: Obtaining EEG data of the subject; Performing inter-channel feature extraction on the EEG data to obtain inter-channel mutual information features of the EEG data; Extracting intra-channel features of the EEG data based on the inter-channel mutual information features to obtain intra-channel features of each channel in the EEG data; Determining an abnormal discharge detection result of the object under test according to the plurality of in-channel features; The step of obtaining the EEG data of the subject includes: Acquire an initial EEG signal of the subject, and divide the initial EEG signal into a plurality of time windows; The initial EEG signals in each of the time windows are subjected to feature transformation processing to obtain the EEG data, which is a time-frequency spectrum diagram of the initial EEG signals.

2. The method according to claim 1, characterized in that The method further comprises: Performing initial feature extraction processing on the EEG data to obtain an initial feature map; Performing dimension transformation processing on the initial feature map, adjusting the channel dimension position of the initial feature map, and obtaining dimension transformation data of each channel; Performing convolution processing on the plurality of dimensionally transformed data respectively to obtain dimensionally reduced feature data of each channel; Data expansion processing is performed on the dimension-reduced feature data of each channel to obtain one-dimensional channel data corresponding to each channel.

3. The method according to claim 1, characterized in that The extracting inter-channel features of the EEG data to obtain inter-channel mutual information features of the EEG data includes: Acquire one-dimensional channel data corresponding to each channel in the EEG data; The plurality of one-dimensional channel data are encoded to extract inter-channel features of the EEG data to obtain the inter-channel mutual information features.

4. The method according to claim 3, characterized in that The inter-channel mutual information feature includes a first inter-channel mutual information feature and a second inter-channel mutual information feature. The encoding process is performed on the plurality of one-dimensional channel data to extract inter-channel features of the EEG data to obtain the inter-channel mutual information feature, including: Performing encoding processing on the plurality of one-dimensional channel data to obtain the mutual information feature between the first channels; Performing convolution processing on the first inter-channel mutual information feature to obtain a dimension-reduced channel feature; The dimension-reduced channel features are encoded to obtain the second inter-channel mutual information features.

5. The method according to claim 1, characterized in that The extracting intra-channel features of the EEG data based on the inter-channel mutual information features to obtain intra-channel features of each channel in the EEG data includes: Obtaining an initial feature map corresponding to the EEG data; Performing intra-channel feature extraction on the initial feature map to obtain initial single-channel features of each channel; Determine the channel weight corresponding to each of the channels based on the mutual information feature between the channels; The intra-channel features of each of the channels are determined according to the channel weights and the initial single-channel features.

6. The method according to claim 5, characterized in that The determining the channel weight corresponding to each of the channels based on the inter-channel mutual information feature includes: Decoding the inter-channel mutual information features to obtain channel decoding data corresponding to each channel in the EEG data; A channel value of each channel is determined according to the channel decoding data, and the channel value is used as a channel weight of each channel.

7. The method according to claim 5, characterized in that Determining the intra-channel features of each channel according to the channel weight and the initial single-channel features includes: Get the initial single-channel features obtained after the current convolution processing; According to the channel weight, adjusting the channel weight of the initial single channel feature; The intra-channel features are determined based on the adjusted channel weights and the initial single-channel features.

8. A system for detecting abnormal brain discharges, characterized in that: include: A data determination module is used to obtain the EEG data of the subject, perform initial feature extraction processing on the EEG data, and obtain an initial feature map; A channel weight determination module, used to determine the one-dimensional channel data corresponding to each channel according to the initial feature map, encode the multiple one-dimensional channel data, and obtain the inter-channel mutual information characteristics of the EEG data; and determining a channel weight corresponding to each of the channels based on the mutual information characteristics between the channels; A fusion convolution module is used to extract intra-channel features from the initial feature map to obtain initial single-channel features of each channel, and determine the intra-channel features of each channel according to the channel weight and the initial single-channel features; An abnormality detection module, used for determining an abnormal discharge detection result of the tested object according to the characteristics in the plurality of channels; The data determination module is further used to obtain the initial EEG signal of the subject and divide the initial EEG signal into multiple time windows; The initial EEG signals in each of the time windows are subjected to feature transformation processing to obtain the EEG data, which is a time-frequency spectrum diagram of the initial EEG signals.

9. The system according to claim 8, characterized in that The channel weight determination module comprises: A dimension transformation unit, used to perform dimension transformation processing on the initial feature map, adjust the channel dimension position of the initial feature map, and obtain dimension transformation data of each channel; A one-dimensional convolution unit, used to perform convolution processing on the plurality of dimensionally transformed data respectively to obtain dimensionally reduced feature data of each channel; The data expansion unit is used to perform data expansion processing on the dimension reduction feature data of each channel to obtain one-dimensional channel data corresponding to each channel.

10. The system according to claim 8 or 9, characterized in that The inter-channel mutual information feature includes a first inter-channel mutual information feature and a second inter-channel mutual information feature, the channel weight determination module also includes a first encoding unit and a second encoding unit, and the second encoding unit includes a convolutional layer; The first encoding unit is used to encode the plurality of one-dimensional channel data based on an attention mechanism to obtain a first inter-channel mutual information feature; The convolution layer is used to perform convolution processing on the first inter-channel mutual information feature to obtain a dimension-reduced channel feature; The second encoding unit is used to encode the reduced-dimensional channel features based on the attention mechanism to obtain second inter-channel mutual information features.

11. The system according to claim 8, characterized in that The fused convolution module comprises: A fusion convolution unit is used to extract intra-channel features from the initial feature map to obtain initial single-channel features of each channel; The feature determination unit is used to determine the intra-channel features of each channel according to the channel weights corresponding to each channel in the initial feature map and the initial single-channel features.

12. A device for detecting abnormal brain discharge, characterized in that: include: A data acquisition module, used to acquire EEG data of the subject; A first feature extraction module is used to extract inter-channel features of the EEG data to obtain inter-channel mutual information features of the EEG data; A second feature extraction module is used to extract intra-channel features of the EEG data based on the inter-channel mutual information features to obtain intra-channel features of each channel in the EEG data; A detection result determination module, used to determine the abnormal discharge detection result of the detected object according to the multiple in-channel features; The data acquisition module is further used to acquire the initial EEG signal of the subject and divide the initial EEG signal into multiple time windows; The initial EEG signals in each of the time windows are subjected to feature transformation processing to obtain the EEG data, which is a time-frequency spectrum diagram of the initial EEG signals.

13. An electronic device, characterized in that: include: processor; as well as A memory having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by the processor, implement the abnormal brain discharge detection method as described in any one of claims 1 to 7.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the abnormal brain discharge detection method according to any one of claims 1 to 7 is implemented.

Images