Nerve brain discharge anomaly detection method and device, electronic equipment and readable medium
By mapping multi-channel EEG signals to heterogeneous graphs and training the model using graph convolutional neural networks, the problem of insufficient accuracy in EEG detection in existing technologies is solved, achieving higher accuracy and robustness in detecting abnormal brain discharges.
Patent Information
- Application Number
- CN202411162568.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies have insufficient accuracy in EEG detection, especially in multi-channel EEG signal analysis, which fails to fully explore the deep correlation between channels, resulting in insufficient accuracy and robustness in detecting abnormal brain discharges.
By employing graph neural network technology, multi-channel EEG signals are mapped onto a heterogeneous graph. The model is trained using a graph convolutional neural network, and abnormal brain discharges are detected by mining the deep correlations between different channels.
It improves the accuracy and robustness of detecting abnormal brain discharges, can more comprehensively characterize the state of brain activity, and reduces the cost of manual re-examination.
Smart Images

Figure CN119033383A_ABST
Abstract
Description
Background Art
[0002] This section is intended to provide a background or context to embodiments of the invention that are recited in the claims. No description herein is admitted to be prior art by inclusion in this section.
[0003] Electroencephalogram (EEG) is a physiological indicator that records neuronal activity in the biological brain through electrical signals. Neuronal activity is the basic physiological activity of the brain, so by collecting brain waves, the physiological state of the brain of the subject can be analyzed. Due to the complexity of brain physiological activities and the frequent electrical signals between neurons, how to detect electrical signals corresponding to specific physiological activities in a large number of brain waves - such as abnormal electrical signals corresponding to epilepsy or other neurological diseases in brain waves - is the main research direction in the field of brain wave detection.
[0004] At present, the basic method is to distinguish EEG data based on manual experience. This solution relies on expert experience, has high labor costs, low consistency, and insufficient reliability of results. Alternatively, a standard automated analysis system can be established to construct relevant strategies and rules based on expert experience, EEG data, analysis results, etc., or to train neural network models to determine whether there are abnormal brain discharges in the EEG.
[0005] However, the above-mentioned standard automated analysis schemes are usually directly based on multi-channel EEG analysis, and the level of feature mining is relatively shallow. They do not fully represent the global correlation of multi-channel EEG signals, resulting in insufficient accuracy in abnormal brain discharge detection. In addition, further review by experts is sometimes required, which reduces labor costs to a limited extent.
[0006] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention
[0007] Based on the above problems, the embodiments of the present disclosure provide a method for detecting abnormal brain discharges in a neural network, a device for detecting abnormal brain discharges in a neural network, an electronic device, and a computer-readable storage medium, which can fully explore the deep-level information of the interaction between channels in multi-channel EEG signals and improve the accuracy and robustness of abnormal brain discharge detection.
[0008] According to the first aspect of the embodiments of the present disclosure, a graph neural brain discharge abnormality detection method is disclosed, which may include: collecting multi-channel EEG signals of a subject under test; mapping the multi-channel EEG signals into a heterogeneous graph based on the correlation between the EEG signals of each channel in the multi-channel EEG signals, wherein a node in the heterogeneous graph corresponds to the EEG signal of one channel, and an edge indicates that there is a correlation between the EEG signals of the channels corresponding to the nodes at both ends; inputting the heterogeneous graph into a brain discharge abnormality detection model to obtain a brain discharge abnormality detection result output by the brain discharge abnormality detection model; wherein the brain discharge abnormality detection model is trained based on a graph convolutional neural network.
[0009] In one embodiment, based on the above scheme, collecting multi-channel EEG signals of the subject under test includes: collecting multi-channel EEG signals from multiple locations on the scalp of the subject under test; and preprocessing the multi-channel EEG signals.
[0010] In one embodiment, based on the above scheme, the preprocessing includes at least one of resampling, filtering, noise data removal, and numerical normalization.
[0011] In one embodiment, based on the aforementioned scheme, based on the correlation between the EEG signals of each channel in the multi-channel EEG signals, the multi-channel EEG signals are mapped to a heterogeneous graph for representation, including: determining the correlation between the EEG signals of each channel in the multi-channel EEG signals, and constructing a correlation matrix; on the basis of the correlation matrix, determining the correlation relationship between the EEG signals of each channel according to the degree of correlation reflected by the correlation, and constructing an adjacency matrix; constructing a heterogeneous graph based on the adjacency matrix; wherein the situation where there is a correlation relationship includes at least one of the following: the degree of correlation reflected by the correlation is greater than the degree of correlation reflected by the correlation threshold; the degree of correlation reflected by the correlation is at a preset position in the correlation degree ranking of the correlation matrix.
[0012] In one embodiment, based on the aforementioned scheme, the correlation between the EEG signals of each channel in the multi-channel EEG signal is determined, and a correlation matrix is constructed, including: using the product difference method to calculate the Pearson correlation coefficient between the EEG signals of each channel in the multi-channel EEG signal; and constructing the correlation matrix based on the Pearson correlation coefficient.
[0013] In one embodiment, based on the aforementioned scheme, the training steps of the abnormal brain discharge detection model are as follows: obtain training samples, the training samples include heterogeneous graph samples and abnormal brain discharge labels corresponding to the heterogeneous graph samples; the heterogeneous graph samples are used to represent the mapping of the correlation between the EEG signals of each channel in the multi-channel EEG signal samples, and the abnormal brain discharge labels are used to characterize the annotation results of the multi-channel EEG signal samples; the training samples are input into the graph convolutional neural network to be trained to obtain the abnormal brain discharge detection results output by the graph convolutional neural network; based on a preset loss function, the loss value of the difference between the abnormal discharge detection results and the abnormal brain discharge labels is determined; the network parameters of the graph convolutional neural network are updated according to the loss value, and it is iterated until the graph convolutional neural network meets the convergence conditions to obtain the abnormal brain discharge detection model.
[0014] In one embodiment, based on the aforementioned solution, the preset loss function is a contrastive learning loss function.
[0015] According to the second aspect of the embodiment of the present disclosure, a graph neural brain discharge abnormality detection device is disclosed, and the device includes: a signal acquisition unit, used to collect multi-channel EEG signals of a subject under test; a graph representation unit, used to map the multi-channel EEG signals into a heterogeneous graph for representation based on the correlation between the EEG signals of each channel in the multi-channel EEG signals, wherein a node in the heterogeneous graph corresponds to the EEG signal of one channel, and an edge represents that the EEG signals of the channels corresponding to the nodes at both ends have a correlation relationship; a graph model prediction unit, used to input the heterogeneous graph into a brain discharge abnormality detection model to obtain a brain discharge abnormality detection result output by the brain discharge abnormality detection model; wherein the brain discharge abnormality detection model is trained based on a graph convolutional neural network.
[0016] According to the third aspect of an embodiment of the present disclosure, an electronic device is disclosed, including: a processor; and a memory, on which computer-readable instructions are stored. When the computer-readable instructions are executed by the processor, the method for detecting abnormal brain discharges in a neural network as disclosed in the first aspect is implemented.
[0017] According to the fourth aspect of an embodiment of the present disclosure, a computer program medium is disclosed, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a computer, the computer executes the method for detecting abnormal brain discharges in the neural network disclosed in the first aspect of the present disclosure.
[0018] The disclosed embodiments provide a graph neural brain discharge anomaly detection method, a graph neural brain discharge anomaly detection device, an electronic device, and a computer-readable storage medium. The method can represent the correlation between channels of the collected multi-channel EEG signals with a heterogeneous graph, wherein a node in the heterogeneous graph corresponds to an EEG signal of a channel, and an edge represents that there is a correlation between the EEG signals of the channels corresponding to the nodes at both ends; on this basis, a brain discharge anomaly detection model obtained by training a graph convolutional neural network is used to predict the heterogeneous graph to obtain a brain discharge anomaly detection result. Compared with the prior art, the embodiments of the present disclosure are implemented to represent the deep correlation between different channel signals in the multi-channel brain signal by a graph, and the graph neural network is trained on the basis of the graph representation to perform brain discharge anomaly detection, which can dig out the deep and hidden correlation between different channels of the brain signal, and characterize the brain activity state from a global and deep level, so as to improve the accuracy and robustness of brain discharge anomaly detection.
[0019] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.
[0020] It is to be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] 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:
[0022] Figure 1A A schematic diagram showing a system architecture for detecting abnormal brain discharges in a neural network is shown.
[0023] Figure 1B A schematic diagram showing another system architecture for detecting abnormal brain discharges in a neural network.
[0024] Figure 2 Shown is one of the flowcharts of a method for detecting abnormal brain discharges in a neural network according to an example embodiment of the present disclosure.
[0025] Figure 3 Shown is a second flowchart of a method for detecting abnormal brain discharges in accordance with an example implementation of the present disclosure.
[0026] Figure 4 Shown is a schematic diagram of a multi-channel EEG signal according to an example embodiment of the present disclosure.
[0027] Figure 5 Shown is a schematic diagram of a heterogeneous graph constructed based on multi-channel EEG signals according to an example embodiment of the present disclosure.
[0028] Figure 6 Shown is a flowchart of the training steps of a brain discharge abnormality detection model according to an example embodiment of the present disclosure.
[0029] Figure 7 FIG. 4 is a schematic diagram showing a contrastive learning loss principle according to an exemplary embodiment of the present disclosure.
[0030] Figure 8 Shown is a structural block diagram of a device for detecting abnormal brain discharges according to an optional example implementation of the present disclosure.
[0031] Fig. 9 Shown is an electronic device according to an alternative example embodiment of the present disclosure.
[0032] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts. DETAILED DESCRIPTION
[0033] 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.
[0034] Those skilled in the art will appreciate that the embodiments of the present disclosure may be implemented as a 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.
[0035] According to the embodiments of the present disclosure, a method for detecting abnormal brain discharge in a neural network, a device for detecting abnormal brain discharge in a neural network, an electronic device, and a computer-readable storage medium are proposed.
[0036] Any number of elements in the drawings is for illustration and not limitation, and any naming is for distinction only and does not have any limiting meaning.
[0037] 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
[0039] At present, the existing technology has the problem of insufficient accuracy in detecting abnormal brain discharges. Specifically, in the scheme of related technology, the electroencephalogram and neural network technology are integrated, mainly to directly extract multi-channel EEG signals, and continuously update the neural network parameters through forward propagation and back propagation, so that the model can classify multi-channel EEG signals to identify possible abnormal brain discharges. However, the EEG signals of each channel are collected separately, and the waveform and potential characteristics are extracted on each channel, ignoring the globality of the EEG signals of the tested object, resulting in insufficient information mining, so that the accuracy of abnormal brain discharge detection needs to be improved.
[0040] Based on the above problems, the present disclosure provides a graph neural network brain discharge abnormality detection method, a graph neural network brain discharge abnormality detection device, an electronic device and a computer-readable storage medium. The scheme of the present disclosure is implemented to represent the deep correlation between different channel signals in multi-channel brain signals in a graph, and on the basis of the graph representation, a graph neural network is trained to detect brain discharge abnormalities, which can mine the deep and hidden correlation between different channels of brain signals, and characterize the brain activity state from a global and deep level, so as to improve the accuracy and robustness of brain discharge abnormality detection.
[0041] Application Scenario Overview
[0042] 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.
[0043] The embodiments of the present disclosure can be applied to abnormal brain discharge detection and auxiliary diagnosis and treatment scenarios of suspected patients. For example, if a user is suspected of having epilepsy and needs to detect the user's brain electrical activity state, especially whether there is abnormal brain discharge, the abnormal brain discharge method of the present disclosure can be used for detection. The following is a specific description of the application scenario in combination with the system architecture.
[0044] Figure 1A A schematic diagram of a system architecture for abnormal brain discharge of a neurological disorder is shown. The system architecture includes a subject 101, a biomedical monitoring device 102, and a processing device 103. When abnormal brain discharge of the subject 101 needs to be detected, the biomedical monitoring device 102 can be set for the subject 101 to collect corresponding data, and the data can be processed by the processing device 103.
[0045] 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 object 101 to collect electrical signals. The main device 1024 aggregates and processes the electrical signals to obtain original 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 multiple parts of the scalp of the object 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.
[0046] The biomedical monitoring device 102 can be connected to the processing device 103 in communication, such as through a wired or wireless communication link, so that the biomedical monitoring device 102 sends the collected data to the processing device 103. The processing device 103 can execute the abnormal brain discharge method of the present exemplary embodiment, process the acquired multi-channel EEG signal, and obtain the abnormal brain discharge detection result of the subject 101. In one embodiment, the processing device 103 may include a display, which can display the abnormal brain discharge detection result and the multi-channel EEG signal.
[0047] Any two or more of the biomedical monitoring device 102 and the processing device 103 may be integrated into the same device. Exemplarily, the biomedical monitoring device 102 and the processing device 103 may be integrated into the same device, or more specifically, the main device 1024 of the biomedical monitoring device 102 may serve as the processing device 103. The biomedical monitoring device 102 may include a head-mounted device, in which detection electrodes 1021, 1022, 1023 may be provided, and when the subject 101 wears the head-mounted device, the head contacts the detection electrodes 1021, 1022, 1023.
[0048] Figure 1BA schematic diagram of another system architecture for abnormal brain discharge of the graph neuroscience is shown. The system architecture includes a subject 101, a biomedical monitoring device 102, a data transceiver device 104, and a server 105. The biomedical monitoring device 102 can be connected to the data transceiver device 104 for communication, and the data transceiver device 104 is connected to the server 105 for communication. Thus, after the data transceiver device 104 obtains the multi-channel EEG signal, it sends it to the server 105. The server 105 may include any form of data processing server such as a cloud server and a distributed server. After obtaining the multi-channel EEG signal sent by the data transceiver device 104, the server 105 obtains the abnormal brain discharge detection result of the subject 101 by executing the graph neuroscience abnormal brain discharge method in this exemplary embodiment. In one embodiment, the server 105 can return the abnormal brain discharge detection result to the data transceiver device 104 for display on the data transceiver device 104 or the biomedical monitoring device 102.
[0049] Figure 1B The system architecture shown is suitable for portable scenarios. For example, the subject 101 can use a portable biomedical monitoring device 102 to collect data at home, office, or any other place, and the data transceiver device 104 sends the data to the server 105 to detect abnormal brain discharges.
[0050] Exemplary Methods
[0051] In combination with the above application scenarios, refer to Figure 2 To describe the method for detecting abnormal brain discharges in a graph neural network according to an exemplary embodiment of the present disclosure.
[0052] See also Figure 2 , Figure 2 FIG. 1 is a flow chart of a method for detecting abnormal brain discharges according to an exemplary embodiment of the present disclosure. Figure 2 As shown in the figure, the method for detecting abnormal neural brain discharges may include the following steps 201 to 203.
[0053] Step 201: Collect multi-channel EEG signals of the subject.
[0054] Among them, the subject refers to an object that can generate brain electrical activity and needs to be detected for abnormal brain discharge. Multi-channel EEG signals are spontaneous and rhythmic electrical activities of brain cell groups recorded by electrodes, which can be obtained by amplifying and recording the spontaneous potentials of multiple parts of the cerebral cortex from the scalp of the subject. Multi-channel EEG signals can be recorded in the form of an electroencephalogram, which is a "potential-time" relationship plane diagram recorded with potential as the vertical axis and time as the horizontal axis. In multi-channel EEG signals, by recording EEG signals at multiple different locations in the brain, it is possible to record the global activity information of the brain more comprehensively and completely, and improve its spatial resolution.
[0055] Step 202: Based on the correlation between the EEG signals of each channel in the multi-channel EEG signal, the multi-channel EEG signal is mapped into a heterogeneous graph for representation, wherein a node in the heterogeneous graph corresponds to the EEG signal of one channel, and an edge indicates that the EEG signals of the channels corresponding to the nodes at both ends are associated with each other.
[0056] In the disclosed embodiment, correlation refers to the degree of correlation between EEG signals of different channels, which can be expressed by calculating the correlation coefficient between EEG signals of two channels in a multi-channel EEG signal. Generally, when the calculated correlation coefficient indicates that the correlation between the EEG signals of the two channels is high, it can be considered that there is a correlation relationship; or, when the calculated correlation coefficient indicates that the correlation between the EEG signals of the two channels is low, it can be considered that there is no correlation relationship, wherein the degree of correlation can be divided according to the actual analysis requirements, and the disclosed embodiment does not impose specific restrictions on this.
[0057] On this basis, the disclosed embodiment can construct nodes for EEG signals of different channels in a heterogeneous graph, and construct edges when there is a correlation between the EEG signals of the channels corresponding to the two end nodes, so as to map the multi-channel EEG signals to the heterogeneous graph. In the heterogeneous graph, each node can contain the EEG signal of the corresponding channel, and the edge can represent the correlation between the EEG signals of different channels, thereby not only providing the EEG signal itself collected by each channel, but also describing the influence of the interaction between channels.
[0058] Step 203: input the heterogeneous graph into the abnormal brain discharge detection model to obtain the abnormal brain discharge detection result output by the abnormal brain discharge detection model;
[0059] Among them, the abnormal brain discharge detection model is trained based on the graph convolutional neural network.
[0060] In the disclosed embodiment, the abnormal brain discharge detection model is obtained by training based on a graph convolutional neural network, which can extract graph feature representations based on heterogeneous graphs and perform prediction and classification based on the graph feature representations. On this basis, after the heterogeneous graph constructed based on multi-channel EEG signals is input into the abnormal brain discharge detection model, the abnormal brain discharge detection model can extract the graph feature representation of the heterogeneous graph through graph readout aggregation and aggregation, and make predictions based on the graph feature representations and output the corresponding abnormal brain discharge detection results. Prediction is made based on the graph feature representation extracted from the heterogeneous graph, so that the prediction process is based on more sufficient, comprehensive and in-depth information, which not only takes into account the characteristics of single-channel EEG signals, but also considers the influence of the interaction between different channels from a global perspective, which can effectively improve the accuracy of abnormal brain discharge detection and avoid the cost and inefficiency of manual re-examination.
[0061] See also Figure 3 , Figure 3 FIG. 2 shows a flow chart of a method for detecting abnormal brain discharges according to an exemplary embodiment of the present disclosure. Figure 3 As shown in the figure, the method for detecting abnormal neural brain discharges may include the following steps 301 to 306.
[0062] Step 301: Collect multi-channel EEG signals from multiple locations on the scalp of the subject.
[0063] In the embodiment of the present disclosure, detection electrodes can be set at multiple locations on the scalp of the subject to be measured to collect multi-channel EEG signals. Based on the influence of data acquisition and processing conditions and detection accuracy requirements, the number and position of the locations where the detection electrodes are set on the scalp of the subject to be measured can be adaptively adjusted, and one or more detection electrodes can be set at each location, and the embodiment of the present disclosure does not impose specific restrictions on this.
[0064] See also Figure 4 , Figure 4 The diagram shows a multi-channel EEG signal according to an example embodiment of the present application. In the form of a visualized EEG, each waveform corresponds to a single-channel EEG signal collected by a detection electrode within the collection time, indicating the waveform change of the potential of the EEG signal at that part over time.
[0065] Step 302: pre-process the multi-channel EEG signal.
[0066] In the disclosed embodiment, based on the acquisition of multi-channel EEG signals, in order to improve data quality, reduce the interference of background noise, or adapt the data format to the interpretation requirements of the abnormal brain discharge detection model, the EEG signals can be preprocessed.
[0067] In an optional method embodiment of the present disclosure, the preprocessing includes at least one of resampling, filtering, noise data removal, and numerical normalization.
[0068] The preprocessing may include resampling. Resampling may change the sampling rate of the EEG signal and convert the non-uniformly sampled signal into a uniformly sampled signal. For example, the EEG monitoring data may be resampled at 500 Hz.
[0069] Among them, preprocessing may include filtering. Filtering can remove interference from AC signals and non-EEG signal frequency bands in EEG signals. Exemplarily, 50Hz notch filtering can be performed on multi-channel EEG signals to reduce interference from AC signals, and bandpass filtering (such as using a 0.1-70Hz frequency band) can be performed to reduce interference from non-EEG signal frequency bands.
[0070] Among them, preprocessing can include removing noise data. Noise data may be caused by factors such as power supply and poor electrode contact. Removing noise data can improve data quality and the accuracy of detection results. Exemplarily, channel data with more noise can be removed. Exemplarily, the potential difference of multi-channel EEG signals can be calculated based on the reference electrode set during the acquisition process to reduce the impact of noise. In an optional method embodiment of the present disclosure, the selection of the reference electrode includes but is not limited to the following methods: using a unipolar lead, setting a detection electrode and an irrelevant electrode on the object under test, and the irrelevant electrode can be set at a position such as the earlobe, which is equivalent to using the irrelevant electrode as the reference electrode. The collected multi-channel EEG signal includes the potential difference of each channel relative to the irrelevant electrode, which is preprocessed to obtain the initial characteristic data of the EEG signal; using a bipolar lead method, no irrelevant electrode is set on the object under test, and each detection electrode uses other detection electrodes as reference electrodes. The collected multi-channel EEG signal includes the potential difference of each channel relative to other detection electrodes; average reference, setting detection electrodes and grounding electrodes on the object under test, then the collected multi-channel EEG signal includes the potential difference of each channel relative to the ground end, and one or more of the detection electrodes of the multi-channel EEG signal can be used as the reference electrode, and the potential difference between each channel and the reference electrode is calculated for the preprocessed multi-channel EEG signal. For example, the c detection electrodes of the electroencephalogram machine are set on the scalp of the subject to be measured, and any one of the c detection electrodes can be selected as the reference electrode, or multiple detection electrodes can be used as reference electrodes, and the average value of the preprocessed multi-channel EEG signal is calculated as the reference value, and the difference between the preprocessed multi-channel EEG signal of each channel and the reference value is calculated to obtain the potential difference of each channel. In one embodiment, different detection electrodes can also be selected as reference electrodes, and the potential difference of each channel is calculated under different reference electrodes, and then the average value is calculated.
[0071] For example, preprocessing may include numerical normalization. Numerical normalization can map data of different modalities and types into the same suitable numerical range for unified processing. Exemplarily, the multi-channel EEG signal may be dimensionless first and then normalized into a suitable numerical range, such as by multiplying by a corresponding coefficient for numerical mapping to complete the numerical normalization.
[0072] Step 303: determine the correlation between the EEG signals of each channel in the multi-channel EEG signal, and construct a correlation matrix.
[0073] In the embodiment of the present disclosure, the correlation matrix is a matrix that represents the correlation between the EEG signals of each channel. When the correlation is represented by the correlation coefficient, the correlation matrix of the multi-channel EEG signal is composed of the correlation coefficients between the EEG signals of each channel.
[0074] For example, the collected multi-channel EEG signal has a total of n channels, and the EEG signal of each channel can be represented as x1, x2, ···, x n , any x i With x j The correlation coefficient is denoted as ρ ij (i, j = 1, 2, ... n), then the correlation matrix R is constructed as shown in the following formula (1):
[0075]
[0076] In an optional method embodiment of the present disclosure, step 303 may include the following steps A1 to A2.
[0077] Step A1: Calculate the Pearson correlation coefficient between the EEG signals of each channel in the multi-channel EEG signal by using the product difference method.
[0078] In the embodiment of the present disclosure, the Pearson correlation coefficient can be used to determine the correlation between the EEG signals of each channel in the multi-channel EEG signal. Among the various optional correlation coefficients, the Pearson correlation coefficient is used to measure the linear correlation, and its value is between -1 and 1, and is determined by calculating the quotient of the product of the covariance and the standard deviation using the product difference method. Specifically, the product difference method calculates the Pearson correlation coefficient ρ ij It can be expressed as the following formula (2):
[0079]
[0080] Among them, cov is the covariance and σ is the standard deviation.
[0081] Step A2: construct a correlation matrix based on the Pearson correlation coefficient.
[0082] In the disclosed embodiment, the Pearson correlation coefficient is used to represent the correlation between the brain signals of each channel, and the calculated Pearson correlation coefficient is substituted into the above formula (1) to construct the corresponding correlation matrix.
[0083] In the embodiments of the present disclosure, the Pearson correlation coefficient is used for example only. Those skilled in the art may also select other correlation coefficients to represent the correlation between EEG signals of each channel in a multi-channel EEG signal based on actual needs, and the present disclosure does not impose any specific limitation on this.
[0084] Step 304: Based on the correlation matrix, the correlation relationship between the EEG signals of each channel is determined according to the degree of correlation reflected by the correlation, and an adjacency matrix is constructed.
[0085] The existence of an association relationship includes at least one of the following:
[0086] The degree of association reflected by the correlation is greater than the degree of association reflected by the association threshold;
[0087] The degree of association reflected by the correlation is at a preset position in the correlation degree ranking of the correlation matrix.
[0088] In the disclosed embodiment, the higher the degree of association reflected by the correlation, the more correlated the EEG signals of the two channels are. Conversely, the lower the degree of association reflected by the correlation, the less correlated the EEG signals of the two channels are. In practical applications, it can be considered that there is a correlation between the EEG signals of the two channels that are relatively more correlated.
[0089] In an optional method embodiment of the present disclosure, an association threshold may be set to evaluate whether the correlation indicates that there is a correlation relationship between the EEG signals of the corresponding two channels, wherein the association threshold may represent a lower limit of the degree of correlation when there is a correlation relationship, and when the degree of correlation represented by the correlation is lower than the degree of correlation represented by the association threshold, it may be considered that there is no correlation relationship between the EEG signals of the corresponding two channels; and, when the degree of correlation represented by the correlation is greater than or equal to the degree of correlation represented by the association threshold, it may be considered that there is a correlation relationship between the EEG signals of the corresponding two channels.
[0090] In an optional embodiment of the present disclosure, the correlation between the EEG signals of two channels can also be sorted based on the correlation matrix to determine whether there is a correlation relationship. The preset positions in the sorting from high to low correlation can be the first n positions, and conversely, the preset positions in the sorting from low to high correlation can be the last m positions. Among them, n and m can be fixed values, or they can be dynamically set according to the number of channels, or they can be set according to actual needs such as measurement conditions and detection accuracy, and the embodiments of the present disclosure do not impose specific restrictions on this.
[0091] In an optional embodiment of the present disclosure, the aforementioned scheme for determining the correlation relationship based on the correlation degree evaluation reflected by the correlation may also be combined. When the correlation degree reflected by the correlation between the EEG signals of the two channels is greater than the correlation degree reflected by the correlation threshold, and the correlation relationship is determined to exist when it is at a preset position in the correlation degree ranking of the correlation matrix. The embodiment of the present disclosure does not impose specific restrictions on this.
[0092] For example, when the correlation coefficient is used to represent the correlation, if the larger the correlation coefficient is, the higher the correlation degree between the EEG signals of the two channels is, then it can be considered that when the correlation coefficient is greater than or equal to the correlation threshold, the correlation degree represented by the correlation is greater than or equal to the correlation degree represented by the correlation threshold, and it is considered that there is a correlation relationship between the EEG signals of the two channels; by sorting the correlation coefficients from large to small, it can be determined that there is a correlation relationship between the EEG signals of the two channels in the first n positions. On this basis, an adjacency matrix is constructed according to whether there is a correlation relationship.
[0093] In the disclosed embodiment, the adjacency matrix may represent the correlation between the EEG signals of each channel in the form of a two-dimensional array, where "0" indicates that there is no correlation, and "1" indicates that there is a correlation.
[0094] For example, after collecting eleven channels of EEG signals x1, x2, ···, x 11 Based on the correlation matrix, the correlation matrix is constructed based on the correlation between the EEG signals of each channel. Then, (x1, x2), (x2, x3), (x3, x4), (x3, x5), (x2, x5), (x4, x8), (x4, x9), (x4, x5), (x5, x7), (x5, x6), (x6, x8) are determined on the basis of the correlation matrix. 10 ),(x7,x9),(x9,x 11 ),(x 10 , x 11 ) There is an association relationship. The existence of an association relationship can be determined based on the set association threshold, based on the fact that the correlation degree reflected by the correlation between the 14 groups of EEG signals of the two channels is greater than the correlation degree reflected by the association threshold; or it can be determined that there is an association relationship between the EEG signals of the two channels based on the fact that the correlation degree reflected by the correlation is sorted from large to small, and the first 14 groups are used as the preset ranking.
[0095] The adjacency matrix A is constructed based on the above association relationship as follows:
[0096] <![CDATA[x1]]> <![CDATA[x2]]> <![CDATA[x3]]> <![CDATA[x4]]> <![CDATA[x5]]> <![CDATA[x6]]> <![CDATA[x7]]> <![CDATA[x8]]> <![CDATA[x9]]> <![CDATA[x 10 ]]> <![CDATA[x 11 ]]> <![CDATA[x1]]> 0 1 0 0 0 0 0 0 0 0 0 <![CDATA[x2]]> 1 0 1 0 1 0 0 0 0 0 0 <![CDATA[x3]]> 0 1 0 1 1 0 0 0 0 0 0 <![CDATA[x4]]> 0 0 1 0 1 0 0 1 1 0 0 <![CDATA[x5]]> 0 1 1 1 0 1 1 0 0 0 0 <![CDATA[x6]]> 0 0 0 0 1 0 0 0 0 1 0 <![CDATA[x7]]> 0 0 0 0 1 0 0 0 1 0 0 <![CDATA[x8]]> 0 0 0 1 0 0 0 0 0 0 0 <![CDATA[x9]]> 0 0 0 1 0 0 1 0 0 0 1 <![CDATA[x 10 ]]> 0 0 0 0 0 1 0 0 0 0 1 <![CDATA[x 11 ]]> 0 0 0 0 0 0 0 0 1 1 0
[0097] Step 305: construct a heterogeneous graph based on the adjacency matrix.
[0098] In the embodiment of the present disclosure, a heterogeneous graph representing the correlation between EEG signals of each channel can be constructed based on the adjacency matrix, wherein each node represents the EEG signal of a channel, and the existence of a connected edge between two nodes indicates that there is a correlation between the EEG signals of the corresponding two channels. Therefore, on the basis of providing the EEG signals collected by each channel, the influence of the interaction between the EEG signals of each channel is also mined and represented.
[0099] For example, see Figure 5 , Figure 5 FIG. 1 is a schematic diagram of a heterogeneous graph constructed based on multi-channel EEG signals according to an exemplary embodiment of the present disclosure. Figure 5 As shown, based on the aforementioned adjacency matrix A, a heterogeneous graph is constructed including node 1, node 2, ..., node 11, and 14 edges respectively connecting (node 1, node 2), (node 2, node 3), (node 3, node 4), (node 3, node 5), (node 2, node 5), (node 4, node 8), (node 4, node 9), (node 4, node 5), (node 5, node 7), (node 5, node 6), (node 6, node 10), (node 7, node 9), (node 9, node 11), (node 10, node 11).
[0100] Step 306: input the heterogeneous graph into the abnormal brain discharge detection model to obtain the abnormal brain discharge detection result output by the abnormal brain discharge detection model.
[0101] Among them, the abnormal brain discharge detection model is trained based on the graph convolutional neural network.
[0102] In the embodiment of the present disclosure, step 306 may correspond to the relevant description of the aforementioned step 203, and will not be described again here to avoid repetition.
[0103] See also Figure 6 , Figure 6 FIG. 1 is a flow chart showing the training steps of a brain discharge abnormality detection model according to an exemplary embodiment of the present disclosure. Figure 6 As shown, training the abnormal brain discharge model may include the following steps 601 to 604.
[0104] Step 601, obtaining training samples, the training samples including heterogeneous graph samples and abnormal brain discharge labels corresponding to the heterogeneous graph samples; the heterogeneous graph samples are used to represent the mapping of the correlation between the EEG signals of each channel in the multi-channel EEG signal samples, and the abnormal brain discharge labels are used to characterize the annotation results of the multi-channel EEG signal samples.
[0105] In the disclosed embodiment, on the basis of collecting multi-channel EEG signal samples through an electroencephalograph, the training samples can be constructed based on the multi-channel EEG signal samples. Among them, the multi-channel EEG signal samples can be pre-processed to determine the correlation between the EEG signals of each channel, and the correlation is mapped to the heterogeneous graph samples composed of nodes and edges, and then the heterogeneous graph samples are labeled according to the brain discharge state of the object under test corresponding to the multi-channel EEG signal samples to obtain abnormal brain discharge labels. The abnormal brain discharge label can characterize the labeling results of the heterogeneous graph samples. When the brain discharge state is normal, the constructed heterogeneous graph samples have no abnormal brain discharge label and are negative samples; when the brain discharge state is abnormal, the constructed heterogeneous graph samples are labeled with abnormal brain discharge labels and are positive samples. The heterogeneous graph samples labeled as "not sure whether it is abnormal discharge" can also be discarded.
[0106] In an optional method embodiment of the present disclosure, the multi-channel EEG signal samples collected from the brain of the subject can also be cut, and the multi-channel EEG signal samples of the complete time period can be cut into different segments of fixed length, and a corresponding heterogeneous graph sample can be constructed in each segment, and then abnormal discharges can be annotated to obtain abnormal brain discharge labels to enrich the number of samples. The time period in which each heterogeneous graph sample is located on the original multi-channel EEG signal sample can also be recorded.
[0107] On this basis, heterogeneous graph samples and corresponding abnormal brain discharge labels form a set of supervised data, and a large amount of supervised data forms a training data set to train the graph convolutional neural network to obtain a brain discharge abnormality detection model.
[0108] Step 602: input the training samples into the graph convolutional neural network to be trained, and obtain the abnormal brain discharge detection result output by the graph convolutional neural network.
[0109] In the disclosed embodiment, the graph convolutional neural network to be trained can extract graph feature representations of heterogeneous graph samples. The graph convolutional neural network (GCN) can extract graph feature representations of heterogeneous graph samples, and predict the abnormal brain discharge detection results corresponding to the heterogeneous graph based on the graph feature representations. Among them, the abnormal brain discharge detection results output by the graph convolutional neural network to be trained based on the graph feature representations of the heterogeneous graph samples can be in a probabilistic form, which is used to characterize the positive sample and negative sample classification to which the input heterogeneous graph samples belong. There may be a certain deviation between the abnormal brain discharge detection results output by the graph convolutional neural network to be trained and the annotated abnormal brain discharge labels, indicating that the prediction results of the abnormal brain discharge detection model are not accurate enough and the parameters need to be further adjusted.
[0110] Step 603: Determine the loss value of the difference between the abnormal discharge detection result and the abnormal brain discharge label based on a preset loss function.
[0111] In the embodiment of the present disclosure, a loss value can be used to characterize the difference between the abnormal discharge detection result and the abnormal brain discharge label, and guide the graph convolutional neural network to iterate in the direction of reducing the loss value. The calculation of the loss value depends on the type of preset loss function used. Those skilled in the art can select the corresponding loss function according to the actual model training requirements, and the embodiment of the present disclosure does not impose specific restrictions on this.
[0112] Step 604: Update the network parameters of the graph convolutional neural network according to the loss value, iterate until the graph convolutional neural network meets the convergence conditions, and obtain a brain discharge abnormality detection model.
[0113] In the disclosed embodiment, the convergence condition refers to the condition that the graph convolutional neural network completes convergence in accordance with the training expectations to determine the expected abnormal brain discharge detection model. During the training process, the convergence condition may be that the loss value is less than or equal to a preset threshold, and the convergence condition is not met when the loss value is greater than the preset threshold; or, the convergence condition may be that the number of iterations of the graph convolutional neural network reaches a preset number, and the convergence condition is not met when the number of iterations of the graph convolutional neural network is still less than the preset number. When the graph convolutional neural network does not meet the convergence condition, its model parameters can be updated according to the loss value, so that the graph convolutional neural network iterates in the direction of reducing the loss value.
[0114] In an optional method embodiment of the present disclosure, the preset loss function may be a contrastive learning loss function.
[0115] In the disclosed embodiment, the contrastive learning loss function is a loss function used for self-supervised learning. Based on the idea of information theory, the similarity between positive samples and negative samples can be compared to guide model learning. In the calculation of the contrastive learning loss function, when the feature is similar to the only positive sample and is not similar to other negative samples, the loss value is low. On the contrary, if it is similar to other negative samples, the loss value should be high to guide the model parameters to update in the direction of reducing the loss value. The contrastive learning loss function can be used to bring the distance between samples of the same type closer in the mapping space and to distance samples of different types to guide model learning. Please refer to Figure 7 , Figure 7 FIG. 2 is a schematic diagram of the contrastive learning loss principle according to an exemplary embodiment of the present application. Figure 7 As shown in the figure, for the input positive sample A, it is initially closer to the negative sample in the mapping space, but farther away from the positive sample. At this time, the calculated loss value is larger, guiding the model to reduce the loss value. After the parameters are updated, the positive sample A is closer to the positive sample in the mapping space, and the expected detection result is output.
[0116] The disclosed embodiment provides a graph neural brain discharge abnormality detection method that can represent the correlation between channels of the collected multi-channel EEG signals with a heterogeneous graph, in which a node corresponds to the EEG signal of a channel, and an edge represents that the EEG signals of the channels corresponding to the nodes at both ends have a correlation relationship; on this basis, a brain discharge abnormality detection model obtained by training a graph convolutional neural network is used to predict the heterogeneous graph to obtain the brain discharge abnormality detection result. Compared with the prior art, the implementation of the disclosed embodiment represents the deep correlation between different channel signals in the multi-channel brain signal by a graph, and the graph neural network is used to train the detection model for brain discharge abnormality detection on the basis of the graph representation, which can dig out the deep and hidden correlation between different channels of the brain signal, and characterize the brain activity state from a global and deep level, so as to improve the accuracy and robustness of brain discharge abnormality detection.
[0117] Exemplary Media
[0118] After introducing the method according to the exemplary embodiment of the present disclosure, next, a medium according to the exemplary embodiment of the present disclosure will be described.
[0119] In some possible 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 method for detecting abnormal brain discharges in the neural network according to various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification.
[0120] Specifically, the processor of the device is used to implement the following steps when executing the program code: collecting multi-channel EEG signals of the object under test; based on the correlation between the EEG signals of each channel in the multi-channel EEG signals, mapping the multi-channel EEG signals into a heterogeneous graph for representation, wherein a node in the heterogeneous graph corresponds to the EEG signal of one channel, and an edge indicates that there is a correlation between the EEG signals of the channels corresponding to the nodes at both ends; inputting the heterogeneous graph into a brain discharge abnormality detection model to obtain a brain discharge abnormality detection result output by the brain discharge abnormality detection model; wherein the brain discharge abnormality detection model is obtained based on the training of a graph convolutional neural network.
[0121] In some embodiments of the present disclosure, the processor of the device is further used to implement the following steps when executing the program code: collecting multi-channel EEG signals from multiple locations on the scalp of the subject; and preprocessing the multi-channel EEG signals.
[0122] In some embodiments of the present disclosure, preprocessing includes at least one of resampling, filtering, noise data removal, and numerical normalization.
[0123] In some embodiments of the present disclosure, the processor of the device is also used to implement the following steps when executing the program code: determining the correlation between the EEG signals of each channel in the multi-channel EEG signal, and constructing a correlation matrix; on the basis of the correlation matrix, determining the correlation relationship between the EEG signals of each channel according to the degree of correlation reflected by the correlation, and constructing an adjacency matrix; constructing a heterogeneous graph based on the adjacency matrix; wherein the situation where there is a correlation relationship includes at least one of the following: the degree of correlation reflected by the correlation is greater than the degree of correlation reflected by the correlation threshold; the degree of correlation reflected by the correlation is in a preset position in the correlation degree ranking of the correlation matrix.
[0124] In some embodiments of the present disclosure, when the processor of the device executes the program code, it is also used to implement the following steps: using the product difference method to calculate the Pearson correlation coefficient between the EEG signals of each channel in the multi-channel EEG signal; and constructing a correlation matrix based on the Pearson correlation coefficient.
[0125] In some embodiments of the present disclosure, the processor of the device is also used to implement the following steps when executing the program code: obtaining training samples, the training samples including heterogeneous graph samples and abnormal brain discharge labels corresponding to the heterogeneous graph samples; the heterogeneous graph samples are used to represent the mapping of the correlation between the EEG signals of each channel in the multi-channel EEG signal samples, and the abnormal brain discharge labels are used to characterize the annotation results of the multi-channel EEG signal samples; the training samples are input into the graph convolutional neural network to be trained to obtain the abnormal brain discharge detection results output by the graph convolutional neural network; based on a preset loss function, the loss value of the difference between the abnormal discharge detection results and the abnormal brain discharge labels is determined; the network parameters of the graph convolutional neural network are updated according to the loss values, and the graph convolutional neural network is iterated until the graph convolutional neural network meets the convergence conditions to obtain a brain discharge abnormality detection model.
[0126] In some embodiments of the present disclosure, the preset loss function is a contrastive learning loss function.
[0127] It should be noted that the above-mentioned 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 thereof. More specific examples of readable storage media (a non-exhaustive list) 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 thereof.
[0128] A readable signal medium may include a data signal 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 of the above. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.
[0129] The program code contained on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.
[0130] 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 computing 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), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).
[0131] Exemplary Devices
[0132] After introducing the medium of the exemplary embodiment of the present disclosure, next, reference is made to Figure 8 A device for detecting abnormal brain discharges using a neurological system according to an exemplary embodiment of the present disclosure will be described.
[0133] See also Figure 8 , Figure 8 FIG. 2 is a block diagram of a device for detecting abnormal brain discharges according to an exemplary embodiment of the present disclosure. Figure 8As shown, a graph neural brain discharge abnormality detection device 800 of an example implementation of the present disclosure includes: a signal acquisition unit 801, used to acquire multi-channel EEG signals of a subject under test; a graph representation unit 802, used to map the multi-channel EEG signals into a heterogeneous graph for representation based on the correlation between the EEG signals of each channel in the multi-channel EEG signals, wherein a node in the heterogeneous graph corresponds to the EEG signal of one channel, and an edge represents that the EEG signals of the channels corresponding to the nodes at both ends have a correlation relationship; a graph model prediction unit 803, used to input the heterogeneous graph into a brain discharge abnormality detection model to obtain a brain discharge abnormality detection result output by the brain discharge abnormality detection model; wherein the brain discharge abnormality detection model is trained based on a graph convolutional neural network.
[0134] It can be seen that implementation Figure 7 The device shown can represent the correlation between channels of the collected multi-channel EEG signals with a heterogeneous graph, in which a node corresponds to the EEG signal of a channel, and an edge indicates that the EEG signals of the channels corresponding to the nodes at both ends are correlated; on this basis, a brain discharge abnormality detection model obtained by training a graph convolutional neural network is used to predict the heterogeneous graph to obtain the brain discharge abnormality detection result. Compared with the prior art, the implementation of the embodiments of the present disclosure can represent the deep correlation between different channel signals in the multi-channel brain signal with a graph, and the detection model of the graph neural network is trained on the basis of the graph representation to detect brain discharge abnormality, which can dig out the deep and hidden correlation between different channels of the brain signal, and characterize the brain activity state from a global and deep level, so as to improve the accuracy and robustness of brain discharge abnormality detection.
[0135] In one embodiment, based on the above solution, the signal acquisition unit 801 is specifically used to collect multi-channel EEG signals from multiple locations on the scalp of the subject; and pre-process the multi-channel EEG signals.
[0136] It can be seen that by implementing this optional embodiment, multi-channel EEG signals from multiple parts of the scalp of the subject can be collected, providing multi-dimensional brain activity information and improving the spatial resolution of the EEG signals.
[0137] In one embodiment, based on the above scheme, the preprocessing includes at least one of resampling, filtering, noise data removal, and numerical normalization.
[0138] It can be seen that by implementing this optional embodiment, the influence of some unnecessary data on the multi-channel EEG signals can be reduced by preprocessing the multi-channel EEG signals, thereby increasing the accuracy of abnormal brain discharge detection.
[0139] In one embodiment, based on the aforementioned scheme, the graph representation unit 802 is specifically used to determine the correlation between the EEG signals of each channel in the multi-channel EEG signal, and construct a correlation matrix; on the basis of the correlation matrix, determine the correlation relationship between the EEG signals of each channel according to the degree of correlation reflected by the correlation, and construct an adjacency matrix; construct a heterogeneous graph based on the adjacency matrix; wherein, the situation where there is a correlation relationship includes at least one of the following: the degree of correlation reflected by the correlation is greater than the degree of correlation reflected by the correlation threshold; the degree of correlation reflected by the correlation is in a preset position in the correlation degree ranking of the correlation matrix.
[0140] It can be seen that by implementing this optional embodiment, a heterogeneous graph can be constructed based on the correlation between channels in a multi-channel EEG signal. While providing the EEG signal of each channel, the influence of the interaction between different channels is also explored and expanded, further improving the accuracy of abnormal brain discharge detection.
[0141] In one embodiment, based on the above scheme, the graph representation unit 802 is specifically used to calculate the Pearson correlation coefficient between the EEG signals of each channel in the multi-channel EEG signal by using the product difference method; and construct a correlation matrix based on the Pearson correlation coefficient.
[0142] It can be seen that by implementing this optional embodiment, the Pearson correlation coefficient can be used to accurately represent the correlation between the EEG signals of each channel of the multi-channel EEG signal, thereby improving the accuracy of mining the association relationship.
[0143] In one embodiment, based on the aforementioned scheme, the graph neural brain discharge abnormality detection device 800 may also include: a sample acquisition unit, used to obtain training samples, the training samples include heterogeneous graph samples and brain discharge abnormality labels corresponding to the heterogeneous graph samples; a model training unit, used for heterogeneous graph samples to represent the mapping of the correlation between the EEG signals of each channel in the multi-channel EEG signal sample, and the brain discharge abnormality label is used to characterize the annotation results of the multi-channel EEG signal sample; and, inputting the training samples into the graph convolutional neural network to be trained to obtain the brain discharge abnormality detection result output by the graph convolutional neural network; then determining the loss value of the difference between the abnormal discharge detection result and the brain discharge abnormality label based on a preset loss function; and updating the network parameters of the graph convolutional neural network according to the loss value, iterating until the graph convolutional neural network meets the convergence conditions, and obtaining the brain discharge abnormality detection model.
[0144] It can be seen that the implementation of this optional embodiment can provide a brain discharge abnormality detection model trained based on heterogeneous graph samples, thereby improving the prediction accuracy of the brain discharge abnormality detection model.
[0145] In one embodiment, based on the aforementioned solution, the preset loss function is a contrastive learning loss function.
[0146] It can be seen that implementing this optional embodiment and calculating the loss value by the contrastive learning loss function to guide the model iteration direction can effectively improve the performance of the abnormal brain discharge detection model.
[0147] The disclosed embodiment provides a graph neural brain discharge abnormality detection device, which can represent the correlation between channels of the collected multi-channel EEG signals with a heterogeneous graph, in which a node corresponds to the EEG signal of a channel, and an edge represents that the EEG signals of the channels corresponding to the nodes at both ends have a correlation relationship; on this basis, a brain discharge abnormality detection model obtained by training a graph convolutional neural network is used to predict the heterogeneous graph to obtain the brain discharge abnormality detection result. Compared with the prior art, the embodiment of the present disclosure is implemented to represent the deep correlation between different channel signals in the multi-channel brain signal by a graph, and the graph neural network is trained on the basis of the graph representation to detect brain discharge abnormality, which can dig out the deep and hidden correlation between different channels of the brain signal, and characterize the brain activity state from a global and deep level, so as to improve the accuracy and robustness of brain discharge abnormality detection.
[0148] It should be noted that although several modules or units of the neurological brain discharge abnormality detection device are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the present disclosure, the features and functions of two or more modules or units described above can be concretized 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 concretized.
[0149] Exemplary Electronic Devices
[0150] After introducing the method, medium, and apparatus according to the exemplary embodiment of the present disclosure, next, an electronic device according to another exemplary embodiment of the present disclosure is introduced.
[0151] 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 implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".
[0152] Refer to the following Fig. 9 An electronic device 900 according to such an embodiment of the present invention is described. Fig. 9 The electronic device 900 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.
[0153] like Fig. 9As shown, the electronic device 900 is in the form of a general computing device. The components of the electronic device 900 may include, but are not limited to: the at least one processing unit 910, the at least one storage unit 920, a bus 930 connecting different system components (including the storage unit 920 and the processing unit 910), and a display unit 940.
[0154] The storage unit stores program codes, which can be executed by the processing unit 910, so that the processing unit 910 executes the steps according to various exemplary embodiments of the present invention described in the above “Exemplary Method” section of this specification.
[0155] The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 921 and / or a cache memory unit 922 , and may further include a read-only memory unit (ROM) 923 .
[0156] The storage unit 920 may also include a program / utility 924 having a set (at least one) of program modules 925, such program modules 925 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include the reality of a network environment.
[0157] The bus 930 may include a data bus, an address bus, and a control bus.
[0158] The electronic device 900 may also communicate with one or more external devices 970 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), and such communication may be performed via an input / output (I / O) interface 950. Furthermore, the electronic device 900 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 960. As shown, the network adapter 960 communicates with other modules of the electronic device 900 via a bus 930. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.
[0159] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.
[0160] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible embodiments, various aspects of the present invention can also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to perform the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of the present specification.
[0161] 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 specific embodiments of the invention, 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 in a neural network, characterized in that: The method comprises: Collect multi-channel EEG signals of the tested object; Based on the correlation between the EEG signals of each channel in the multi-channel EEG signal, the multi-channel EEG signal is mapped into a heterogeneous graph for representation, wherein a node in the heterogeneous graph corresponds to an EEG signal of one channel, and an edge indicates that the EEG signals of the channels corresponding to the nodes at both ends are associated with each other; Inputting the heterogeneous graph into a brain discharge abnormality detection model to obtain a brain discharge abnormality detection result output by the brain discharge abnormality detection model; Among them, the abnormal brain discharge detection model is obtained based on graph convolutional neural network training.
2. The method according to claim 1, characterized in that: The method of collecting multi-channel EEG signals of the subject includes: Collecting multi-channel EEG signals from multiple locations on the scalp of the subject; The multi-channel EEG signal is preprocessed.
3. The method according to claim 2, characterized in that The preprocessing includes at least one of resampling, filtering, noise data removal, and numerical standardization.
4. The method according to claim 1, characterized in that The step of mapping the multi-channel EEG signal to a heterogeneous graph based on the correlation between the EEG signals of each channel in the multi-channel EEG signal comprises: Determine the correlation between the EEG signals of each channel in the multi-channel EEG signal, and construct a correlation matrix; On the basis of the correlation matrix, determining the correlation relationship between the EEG signals of each channel according to the degree of correlation reflected by the correlation, and constructing an adjacency matrix; Constructing the heterogeneous graph based on the adjacency matrix; The existence of the association relationship includes at least one of the following: The degree of association reflected by the correlation is greater than the degree of association reflected by the association threshold; The degree of association reflected by the correlation is at a preset position in the correlation degree ranking of the correlation matrix.
5. The method according to claim 4, characterized in that The step of determining the correlation between the EEG signals of each channel in the multi-channel EEG signals and constructing a correlation matrix includes: The Pearson correlation coefficient between the EEG signals of each channel in the multi-channel EEG signal is calculated by using a product difference method; The correlation matrix is constructed based on the Pearson correlation coefficient.
6. The method according to claim 1, characterized in that The training steps of the abnormal brain discharge detection model are as follows: Acquire training samples, wherein the training samples include heterogeneous graph samples and abnormal brain discharge labels corresponding to the heterogeneous graph samples; the heterogeneous graph samples are used to represent the mapping of correlations between EEG signals of each channel in a multi-channel EEG signal sample, and the abnormal brain discharge labels are used to characterize the annotation results of the multi-channel EEG signal sample; Inputting the training samples into a graph convolutional neural network to be trained, and obtaining abnormal brain discharge detection results output by the graph convolutional neural network; Determining a loss value of the difference between the abnormal discharge detection result and the abnormal brain discharge label based on a preset loss function; The network parameters of the graph convolutional neural network are updated according to the loss value, and the graph convolutional neural network is iterated until the graph convolutional neural network meets the convergence conditions to obtain the abnormal brain discharge detection model.
7. The method according to claim 6, characterized in that The preset loss function is a contrastive learning loss function.
8. A device for detecting abnormal brain discharges of a neuron, characterized in that: The device comprises: A signal acquisition unit, used for acquiring multi-channel EEG signals of the subject; A graph representation unit, for mapping the multi-channel EEG signal into a heterogeneous graph based on the correlation between the EEG signals of each channel in the multi-channel EEG signal, wherein a node in the heterogeneous graph corresponds to an EEG signal of one channel, and an edge indicates that the EEG signals of the channels corresponding to the nodes at both ends are associated with each other; A graph model prediction unit, used for inputting the heterogeneous graph into a brain discharge abnormality detection model to obtain a brain discharge abnormality detection result output by the brain discharge abnormality detection model; Among them, the abnormal brain discharge detection model is obtained based on graph convolutional neural network training.
9. An electronic device, characterized in that: include: processor; A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 7 by executing the executable instructions.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
Epilepsy detection system based on linear graph convolutional neural network
CN112294337A
Epilepsy detection system and device based on graph attention network and medium
CN112294338A
Epilepsy electroencephalogram signal monitoring method and system based on space-time attention mechanism
CN115590529A
Multi-resolution sleep stage classification method based on dynamic adaptive kernel graph neural network
CN116439672A
Newborn epilepsy detection method, device and equipment
CN117617908A
