Epilepsy characteristic waveform detection method and equipment applied to epileptic seizure prediction

Abstract

The invention provides an epilepsy characteristic waveform detection method, device and equipment applied to epileptic seizure prediction, a medium and a product, and is applied to the technical field of medical device.The method comprises the steps that electroencephalogram signals of a target in different frequency bands are obtained; based on weight data learned by the frequency spectrum information, performing frequency spectrum processing on the electroencephalogram signals of each frequency band to obtain time domain signals of the target in different frequency bands; respectively carrying out time feature extraction on the time domain signal of each frequency band on different scales to obtain time feature data; inputting the time characteristic data into a constructed dynamic space-time diagram convolutional network, and capturing functional connection change of each region of the brain in the epileptic seizure process and / or propagation change of an epileptic characteristic waveform on a time axis to obtain space-time characteristic data; and performing classification prediction about epilepsy feature waveforms on the spatial-temporal feature data to obtain a waveform detection result. According to the invention, the problem of low epilepsy characteristic waveform detection precision in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of medical technology, and more specifically, to a method and device for detecting epilepsy characteristic waveforms used in epileptic seizure prediction. Background Technology

[0002] The detection of waveforms such as spikes and ripples in electroencephalogram (EEG) signals is crucial in the prediction of epileptic seizures. Currently, methods for detecting epileptic characteristic waveforms either fail to adapt to the dynamic changes in spectral characteristics during epileptic seizures or struggle to simultaneously capture the local propagation patterns and long-range dependencies of epileptic characteristic waveforms, ultimately resulting in low accuracy in detecting epileptic characteristic waveforms. Summary of the Invention

[0003] This application provides a method, apparatus, electronic device, storage medium, and computer program product for detecting epilepsy characteristic waveforms in epileptic seizure prediction, which can solve the problem of low accuracy in detecting epilepsy characteristic waveforms in related technologies. The technical solutions are as follows: According to one aspect of this application, a method for detecting epilepsy feature waveforms in epileptic seizure prediction is provided. The method includes: acquiring electroencephalogram (EEG) signals of a target at different frequency bands; the EEG signals are used to characterize the spectral information of the epilepsy feature waveforms; performing spectral processing on the EEG signals at each frequency band based on weight data learned from the spectral information to obtain time-domain signals of the target at different frequency bands; extracting time features from the time-domain signals at each frequency band at different scales to obtain time feature data; the time-domain signal of one frequency band corresponds to one scale; inputting the time feature data into a constructed dynamic spatiotemporal graph convolutional network to capture changes in functional connectivity of different brain regions and / or propagation changes of epilepsy feature waveforms on the time axis during epileptic seizures to obtain spatiotemporal feature data; the dynamic spatiotemporal graph convolutional network includes a spatial network constructed based on the spatial correlation of different brain regions and a temporal network generated based on the time dependence of epilepsy feature waveforms; and performing classification prediction on the spatiotemporal feature data regarding epilepsy feature waveforms to obtain waveform detection results.

[0004] According to one aspect of this application, an epilepsy characteristic waveform detection device for epileptic seizure prediction is provided. The device includes: an electroencephalogram (EEG) signal acquisition module for acquiring EEG signals of a target in different frequency bands; the EEG signals are used to characterize spectral information of epilepsy characteristic waveforms; an adaptive spectrum processing module for performing spectral processing on the EEG signals of each frequency band based on weights learned from the spectral information, to obtain time-domain signals of the target in different frequency bands; and a time feature extraction module for extracting time features from the time-domain signals of each frequency band at different scales, to obtain time features. Data; a time-domain signal of one frequency band corresponds to one scale; a spatiotemporal feature extraction module is used to input the time feature data into a constructed dynamic spatiotemporal graph convolutional network to capture the functional connectivity changes of various brain regions and / or the propagation changes of epileptic feature waveforms on the time axis during epileptic seizures, thereby obtaining spatiotemporal feature data; the dynamic spatiotemporal graph convolutional network includes a spatial network constructed based on the spatial correlation of various brain regions and a temporal network generated based on the time dependence of epileptic feature waveforms; a waveform classification and prediction module is used to classify and predict the epileptic feature waveforms based on the spatiotemporal feature data, thereby obtaining waveform detection results.

[0005] According to one aspect of this application, an electronic device includes at least one processor and at least one memory, wherein the memory stores a computer program that, when executed by the processor, implements the epilepsy feature waveform detection method described above for epileptic seizure prediction.

[0006] According to one aspect of this application, a storage medium storing a computer program that, when executed by one or more processors, implements the epilepsy feature waveform detection method described above for epileptic seizure prediction.

[0007] According to one aspect of this application, a computer program product includes a computer program that, when executed by one or more processors, implements the epilepsy feature waveform detection method described above for use in epileptic seizure prediction.

[0008] The beneficial effects of the technical solution provided in this application are: In the above technical solution, not only is the adaptability limitation of traditional fixed-band filtering overcome through a dynamic frequency band enhancement mechanism—specifically, by learning weight data based on the spectral information of the epilepsy characteristic waveforms represented by the EEG signals of the target in different frequency bands—the spectral processing of the EEG signals in each frequency band is performed to obtain the time-domain signals of the target in different frequency bands. Furthermore, a multi-scale time feature extraction mechanism is used to simultaneously capture diverse time patterns such as transient spikes and continuous ripples in the epilepsy characteristic waveforms. Specifically, time feature extraction is performed on the time-domain signals of each frequency band at different scales to obtain time feature data. This is then combined with dynamic… The dynamic spatiotemporal graph convolutional network effectively characterizes the dynamic evolution and temporal propagation characteristics of brain networks during epileptic seizures. Specifically, it inputs temporal feature data into the dynamic spatiotemporal graph convolutional network to capture changes in functional connectivity of different brain regions and / or propagation changes of epileptic feature waveforms along the time axis during epileptic seizures, obtaining spatiotemporal feature data. Finally, it performs classification and prediction of epileptic feature waveforms based on this spatiotemporal feature data, obtaining waveform detection results indicating three types of epileptic feature waveforms: "normal," "spikes," and "ripples." This effectively solves the problem of low accuracy in detecting epileptic feature waveforms in related technologies. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram based on the implementation environment involved in this application; Figure 2 This is a hardware structure diagram of a computer device according to an exemplary embodiment; Figure 3 This is a flowchart illustrating a method for detecting epileptic feature waveforms in epileptic seizure prediction, according to an exemplary embodiment. Figure 4 This is a schematic diagram of a multi-band dynamic adaptive spatiotemporal graph convolutional network model according to an exemplary embodiment; Figure 5 yes Figure 3 A flowchart of step 310 in one embodiment corresponds to the following example; Figure 6 yes Figure 3 A flowchart of step 330 in one embodiment corresponds to the following example; Figure 7 yes Figure 3 A flowchart of step 350 in one embodiment corresponds to the following example; Figure 8 This is a flowchart illustrating the construction process of a dynamic spatiotemporal graph convolutional network according to an exemplary embodiment; Figure 9 This is a structural block diagram of an epilepsy feature waveform detection device applied in epilepsy seizure prediction, according to an exemplary embodiment. Figure 10 This is a structural block diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0011] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0012] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this disclosure means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0013] As mentioned earlier, waveforms such as spikes and ripples in electroencephalogram (EEG) signals are crucial for predicting epileptic seizures. Spikes are characterized by brief, sharp negative deflections, lasting approximately 20–70 milliseconds; ripples are slightly longer and have greater amplitude, lasting approximately 70–200 milliseconds. The frequency, location, and morphology of these waveforms are directly related to epileptic seizures, and accurate detection of them is essential for epilepsy diagnosis and early intervention.

[0014] However, current methods for detecting epilepsy characteristic waveforms still suffer from low accuracy. On the one hand, due to over-reliance on fixed-band filtering or global spectrum analysis, the analysis strategy cannot be adjusted according to the dynamic changes in the spectrum characteristics during epileptic seizures, resulting in insufficient enhancement of key spectrum components and susceptibility to noise interference such as electromyography artifacts, ultimately leading to an inability to adapt to the dynamic changes in the spectrum characteristics during epileptic seizures.

[0015] On the other hand, existing temporal modeling methods (such as single-scale LSTM and fixed-kernel CNN) are computationally complex and sensitive to noise, making it difficult to simultaneously capture the local propagation patterns (such as waveform diffusion over short time) and long-term dependencies (such as signal correlations between different brain regions) of epileptic feature waveforms, ultimately limiting the accuracy of spatial feature extraction.

[0016] As can be seen from the above, the relevant technologies still have the drawback of low accuracy in detecting epilepsy characteristic waveforms.

[0017] Therefore, the epilepsy characteristic waveform detection method provided in this application can be applied to the epilepsy seizure prediction process, effectively improving the accuracy of epilepsy characteristic waveform detection. Accordingly, the epilepsy characteristic waveform detection method is applicable to an epilepsy characteristic waveform detection device, which can be deployed on an electronic device. The electronic device can be a computer device configured with a von Neumann architecture, such as a desktop computer, laptop computer, server, etc.; the electronic device can also refer to a portable mobile electronic device, such as a smartphone, tablet computer, etc.

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0019] Figure 1 This is a schematic diagram of an implementation environment for a method of detecting epileptic characteristic waveforms used in epileptic seizure prediction. It should be noted that this implementation environment is merely an example adapted to the present invention and should not be considered as providing any limitation on the scope of the invention.

[0020] The implementation environment includes an epilepsy characteristic waveform detection system 100, which can be deployed end-to-end, specifically including a data acquisition end 110 and a server end 130.

[0021] Specifically, the acquisition terminal 110 can also be considered as an EEG signal acquisition device. This EEG signal acquisition device can be worn by the target to acquire EEG signals from at least one EEG channel of the target. The target can refer to an object that can generate EEG signals under the stimulation of the EEG signal acquisition device; for example, the object can be a person.

[0022] Server 130 can be a desktop computer, laptop computer, server, or other computer device, or it can be a computer cluster consisting of multiple servers, or even a cloud computing center consisting of multiple servers. Server 130 is used to provide backend services, such as, but not limited to, epilepsy characteristic waveform detection services.

[0023] The server 130 and the acquisition terminal 110 establish a network communication connection in advance via wired or wireless means, and data transmission between the server 130 and the acquisition terminal 110 is realized through this network communication connection. The transmitted data includes, but is not limited to, the target's raw electroencephalogram (EEG) signals, etc.

[0024] In one application scenario, through the interaction between the acquisition terminal 110 and the server terminal 130, the acquisition terminal 110 can upload the acquired raw EEG signal of the target to the server terminal 130 to request the server terminal 130 to provide epilepsy characteristic waveform detection service.

[0025] For server 130, after receiving the raw EEG signal of the target uploaded by acquisition terminal 110, it calls the epilepsy feature waveform detection service to perform epilepsy feature waveform detection on the raw EEG signal of the target, and predicts epileptic seizures for the target based on the detected waveform detection results, thereby solving the problem of low accuracy of epilepsy feature waveform detection in related technologies.

[0026] Please see Figure 2 , Figure 2 This is a hardware structure diagram of a computer device according to an exemplary embodiment. The computer device is suitable for… Figure 1 The server 130 in the implementation environment is shown.

[0027] It should be noted that this computer device is merely an example adapted to this application and should not be construed as providing any limitation on the scope of use of this application. For example, the various embodiments of this application can also be implemented using portable mobile electronic devices (such as smartphones). Furthermore, this computer device should not be interpreted as requiring or depending on any particular feature. Figure 2 One or more components of the exemplary computer device 200 shown.

[0028] The hardware structure of computer device 200 can vary considerably due to differences in configuration or performance, such as Figure 2 As shown, the computer device 200 includes: a power supply 210, an interface 230, at least one memory 250, and at least one central processing unit (CPU) 270.

[0029] Specifically, power supply 210 is used to provide operating voltage to the various hardware devices on computer equipment 200.

[0030] Interface 230 includes at least one wired or wireless network interface 231 for interacting with external devices. For example, to perform... Figure 1 The diagram shows the interaction between the acquisition terminal 110 and the server terminal 130 in the implementation environment.

[0031] Of course, in other examples adapted in this application, interface 230 may further include at least one serial-to-parallel conversion interface 233, at least one input / output interface 235, and at least one USB interface 237, etc. Figure 2 As shown, this does not constitute a specific limitation.

[0032] The memory 250 serves as a carrier for resource storage and can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored on it include the operating system 251, application programs 253, and data 255, etc., and the storage method can be temporary storage or permanent storage.

[0033] The operating system 251 is used to manage and control the various hardware devices and application programs 253 on the computer device 200, so as to enable the central processing unit 270 to perform calculations and processing on the massive data 255 in the memory 250. It can be Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0034] Application 253 is a computer program formed by computer-readable instructions based on operating system 251 to perform at least one specific task, and may include at least one module ( Figure 2 (Not shown), each module can contain corresponding computer-readable instructions. For example, the epilepsy characteristic waveform detection device can be considered as an application 253 deployed on computer device 200.

[0035] Data 255 can be photos, pictures, etc. stored on a disk, or it can be EEG signals, waveform detection results, etc., stored in memory 250.

[0036] The central processing unit 270 may include one or more processors and is configured to communicate with the memory 250 via at least one communication bus to read computer programs stored in the memory 250, thereby performing calculations and processing on massive amounts of data 255 stored in the memory 250. For example, an epilepsy characteristic waveform detection method may be implemented by the central processing unit 270 reading an application program 253 stored in the memory 250.

[0037] Furthermore, this application can also be implemented through hardware circuits or a combination of hardware circuits and software. Therefore, the implementation of this application is not limited to any specific hardware circuit, software, or combination thereof.

[0038] Please see Figure 3 This application provides a method for detecting epileptic characteristic waveforms in epileptic seizure prediction. This method is applicable to electronic devices, for example, the electronic device may be... Figure 1The server 130 in the implementation environment is shown. The hardware structure of this electronic device can be as follows: Figure 2 As shown.

[0039] In the following method embodiments, for ease of description, the execution subject of each step of the method is an electronic device, but this does not constitute a specific limitation.

[0040] like Figure 3 As shown, the method may include the following steps: Step 310: Acquire the target's EEG signals in different frequency bands.

[0041] Among them, electroencephalogram (EEG) signals are used to characterize the spectral information of epileptic characteristic waveforms. It should be understood that EEG signals in different frequency bands are different, and the spectral information they characterize will also be different. Accordingly, the epileptic characteristic waveforms they characterize will also be different.

[0042] In some embodiments, the frequency band includes a low-frequency band, a high-frequency band, and a full-frequency band. The low-frequency band has a frequency range of 1–70 Hz, the high-frequency band has a frequency range of 80 Hz or higher, and the full-frequency band has a frequency range of 1 Hz or higher. In some embodiments, the characteristic waveform of epilepsy includes at least one of spikes and ripples. Thus, in one possible implementation, the low-frequency EEG signal mainly contains spikes, the high-frequency EEG signal mainly contains ripples, and the full-frequency EEG signal retains complete spectral information, i.e., mainly contains spikes and ripples.

[0043] In some embodiments, the target's EEG signals at different frequency bands are obtained by preprocessing the target's raw EEG signals. In some embodiments, data preprocessing includes, but is not limited to, segmentation, filtering, data standardization, etc., and is not limited here.

[0044] Please refer back to Figure 1 ,exist Figure 1In this embodiment, when the acquisition end 110 is an EEG signal acquisition device, the target's raw EEG signal can be acquired by having the target wear the EEG signal acquisition device. Based on this, the acquisition of the target's EEG signals in different frequency bands can originate from the target's raw EEG signals acquired in real-time by the EEG signal acquisition device, or from the target's raw EEG signals acquired by the EEG signal acquisition device during a historical time period pre-stored in the electronic device. Therefore, for the electronic device, after the EEG signal acquisition device acquires the target's raw EEG signals, it can perform data preprocessing on the raw EEG signals in real-time, or it can pre-store the raw EEG signals and then perform data preprocessing. For example, data preprocessing can be performed when the electronic device's task processing rate is low, or data preprocessing can be performed according to the operator's instructions. Thus, the epilepsy characteristic waveform detection in this embodiment can target the target's raw EEG signals acquired in real-time, or it can target the target's raw EEG signals acquired during a historical time period; no specific limitation is made here.

[0045] Step 330: Based on the weight data learned from the spectral information, perform spectral processing on the EEG signals of each frequency band to obtain the time-domain signals of the target in different frequency bands.

[0046] As mentioned earlier, electroencephalogram (EEG) signals are used to characterize the spectral information of epilepsy characteristic waveforms. Different frequency bands of EEG signals differ, resulting in variations in the epilepsy characteristic waveforms they represent, and consequently, different spectral information. Based on this, this embodiment proposes a dynamic frequency band enhancement mechanism for EEG signals at different target frequency bands, thereby enhancing the effective components related to epilepsy in the EEG signals.

[0047] In some embodiments, the dynamic frequency band enhancement mechanism can also be considered as multi-band adaptive spectrum processing, which refers to the spectral weighting of the spectral information of the epilepsy characteristic waveforms represented by EEG signals in different frequency bands based on weight data learned from spectral information.

[0048] In some embodiments, the weight data is learned based on the spectral information represented by EEG signals in different frequency bands. In one possible implementation, for each frequency band, the weight data can be assigned weights corresponding to several sub-bands, which can be further subdivided from the frequency band. For example, the low-frequency band can be further divided into low-frequency, mid-frequency, and high-frequency sub-bands based on its energy intensity; correspondingly, the weight data for the low-frequency band can be assigned weights corresponding to these sub-bands. In this approach, by learning the spectral information represented by EEG signals in different frequency bands, the weight data can adapt to the spectral changes during epileptic seizures in real time, resulting in stronger spectral adaptability. This is beneficial for better suppressing noise interference in EEG signals, thereby improving the subsequent feature representation of useful information in the EEG signals.

[0049] In other words, through the dynamic frequency band enhancement mechanism, the time domain signal of the target is essentially enhanced, which is the brain electrical signal of the effective components related to epilepsy.

[0050] Step 350: Extract time features from the time-domain signals of each frequency band at different scales to obtain time feature data.

[0051] In this context, a time-domain signal in one frequency band corresponds to one scale.

[0052] In some embodiments, temporal feature data is used to describe epileptic feature waveforms of varying durations captured at different scales on different EEG channels. In one possible implementation, the temporal feature data can be represented as a temporal feature matrix with dimensions C×T×D, where C is the number of EEG channels, T is the time step, and D is the feature dimension (also representing the number of scales).

[0053] In some embodiments, temporal feature extraction at different scales can be achieved using different convolutional kernels, i.e., one convolutional kernel corresponds to one scale. In one possible implementation, different convolutional kernels can refer to convolutional kernels of different widths. For example, three convolutional kernels of different widths can correspond to sliding windows of 0.5 seconds, 0.25 seconds, and 0.125 seconds at a sampling rate of 250Hz, respectively, to capture epileptic feature waveforms with different durations, such as spikes with a duration of 20-70 milliseconds or ripples with a duration of 70-200 milliseconds.

[0054] Step 370: Input the temporal feature data into the constructed dynamic spatiotemporal graph convolutional network to capture the changes in functional connectivity of different brain regions and / or the propagation changes of epileptic feature waveforms on the time axis during epileptic seizures, thereby obtaining spatiotemporal feature data.

[0055] It should be understood that during an epileptic seizure, the characteristics of EEG signals can be reflected not only in abnormal functional connections between different brain regions, but also in the temporal propagation patterns of the signals. Therefore, capturing the dynamic changes in brain network connections and the temporal propagation characteristics of signals is crucial for understanding the mechanisms of epileptic seizures.

[0056] Based on this, this embodiment utilizes a dynamic spatiotemporal graph convolutional network to extract features from temporal feature data from both temporal and spatial dimensions, thereby achieving accurate feature representation of epileptic characteristic waveforms. Specifically, feature extraction in the spatial dimension refers to capturing changes in functional connectivity between different brain regions during an epileptic seizure, thus accurately representing abnormal functional connections between these regions. Feature extraction in the temporal dimension refers to capturing the propagation changes of epileptic characteristic waveforms along the time axis, thus accurately representing the propagation pattern of signals over time during an epileptic seizure.

[0057] In some embodiments, spatiotemporal feature data is used to describe the brain region distribution characteristics and temporal evolution patterns of epileptic feature waveforms, providing accurate data for subsequent classification and prediction tasks. In one possible implementation, the spatiotemporal feature data can be represented as a spatiotemporal feature matrix with dimensions C×T×D, where C is the number of EEG channels, T is the time step, and D is the feature dimension (also representing the number of scales). In some embodiments, the dynamic spatiotemporal graph convolutional network includes a spatial network constructed based on the spatial correlations of different brain regions and a temporal network generated based on the temporal dependence of epileptic feature waveforms. The spatial network can capture changes in functional connectivity across different brain regions during a seizure, while the temporal network can capture the propagation changes of epileptic feature waveforms along the time axis. Thus, by cascading the spatial and temporal networks, the spatial correlations and temporal dependencies of EEG signals during a seizure can be dynamically modeled, ultimately achieving a deep fusion of features from both the temporal and spatial dimensions.

[0058] Step 390: Classify and predict the characteristic waveforms of epilepsy based on the spatiotemporal feature data to obtain the waveform detection results.

[0059] The waveform detection results indicate the category of epileptic characteristic waveforms to which the target's EEG signal belongs. In other words, the waveform detection results reflect which category of epileptic characteristic waveforms can be detected in the target's EEG signal.

[0060] In some embodiments, classification prediction refers to calculating the probability that spatiotemporal feature data corresponds to different epileptic feature waveform categories. In one possible implementation, epileptic feature waveform categories include, but are not limited to, normal, spike, and ripple.

[0061] In some embodiments, classification prediction is achieved using a softmax function.

[0062] For example, spatiotemporal feature data is input into a softmax function to calculate P1, P2, and P3. Here, P1 represents the probability that the spatiotemporal feature data corresponds to the normal category, P2 represents the probability that the spatiotemporal feature data corresponds to the spike category, and P3 represents the probability that the spatiotemporal feature data corresponds to the ripple category. If P1 is maximized, the waveform detection result for the target's EEG signal showing a normal category of epileptic characteristic waveforms is obtained. Similarly, if P2 is maximized, the waveform detection result for the target's EEG signal showing a spike category of epileptic characteristic waveforms is obtained; and if P3 is maximized, the waveform detection result for the target's EEG signal showing a ripple category of epileptic characteristic waveforms is obtained.

[0063] Of course, in other embodiments, to ensure sufficient detection accuracy of epilepsy feature waveforms, a detection threshold can be set for classification prediction. Only when the probability of the epilepsy feature waveform category corresponding to the spatiotemporal feature data exceeds the detection threshold can the final waveform detection result be obtained. This embodiment does not constitute a specific limitation in this regard. For example, if P1, P2, and P3 do not exceed the detection threshold, the waveform detection result cannot be obtained.

[0064] Through the above process, not only is the adaptability limitation of traditional fixed frequency band filtering overcome by using a dynamic frequency band enhancement mechanism, but also a multi-scale time feature extraction mechanism is used to simultaneously capture diverse time patterns such as transient spikes and continuous ripples in epileptic characteristic waveforms. Combined with a dynamic spatiotemporal graph convolutional network, the dynamic evolution and time propagation characteristics of brain networks during epileptic seizures are effectively characterized, ultimately improving the detection accuracy of epileptic characteristic waveforms.

[0065] Please see Figure 4 , Figure 4 This is a schematic diagram of a multi-band dynamic adaptive spatiotemporal graph convolutional network model in an exemplary embodiment. Figure 4 The multi-band dynamic adaptive spatiotemporal graph convolutional network model includes a data preprocessing module, a multi-band adaptive spectrum processing module, a multi-scale temporal feature extraction module, a dynamic spatiotemporal graph convolution module, and an epileptic event classification module.

[0066] The system comprises the following modules: a data preprocessing module for segmenting, filtering, and standardizing the raw EEG signals of the target to obtain EEG signals at different frequency bands; a multi-band adaptive spectrum processing module for performing multi-band adaptive spectrum processing on the target's EEG signals at different frequency bands to enhance the effective components related to epilepsy; a multi-scale temporal feature extraction module for extracting the temporal features of the target's EEG signals at different frequency bands using multi-scale convolutional kernels; a dynamic spatiotemporal graph convolution module for constructing a dynamic spatiotemporal graph convolutional network to capture changes in functional connectivity across brain regions and / or the propagation changes of epileptic characteristic waveforms along the time axis during epileptic seizures, obtaining spatiotemporal feature data; and an epileptic event classification module for classifying and predicting epileptic characteristic waveforms (spikes, ripples, or normal) from the spatiotemporal feature data to obtain waveform detection results.

[0067] The following is combined Figure 4 The multi-band dynamic adaptive spatiotemporal graph convolutional network model shown is described in detail below. Please see Figure 5 In one exemplary embodiment, step 310 may include the following steps: Step 311: Obtain the target's raw EEG signals.

[0068] Step 313: Use a sliding window with a second set window length to segment the raw EEG signal.

[0069] The second window length can be flexibly set according to the actual needs of the application scenario, as long as each segment of the original EEG signal can completely contain the potential spikes or ripples. There is no limitation here. For example, the second window length can be set to 1 second.

[0070] Step 315: Input each segment of the original EEG signal into different bandpass filters to obtain the target's EEG signals in different frequency bands.

[0071] One bandpass filter corresponds to one frequency band.

[0072] Specifically, in the data preprocessing stage, all raw EEG signals from the target dataset need to be segmented. The raw EEG signals are divided into segments according to a 1-second time window to ensure that each segment fully contains potential spikes (20-70ms) or ripples (70-200ms). Then, different bandpass filters are used to further decompose the segmented raw EEG signals into three frequency bands: low-frequency band (1-70Hz), high-frequency band (above 80Hz), and full-frequency band (above 1Hz). Among them, the low-frequency band (1-70Hz) mainly contains spikes, the high-frequency band (above 80Hz) mainly contains ripples, and the full-frequency band retains complete spectral information for comprehensive analysis. Finally, z-score normalization can be performed on the EEG signals of each frequency band.

[0073] The above embodiments can eliminate individual differences between different EEG channels in different targets and different brain regions, which is beneficial to improving the generalization ability of the model.

[0074] Please see Figure 6 In one exemplary embodiment, step 330 may include the following steps: Step 331: Perform time-frequency transformation processing on the EEG signals of each frequency band to obtain the frequency domain signals of each frequency band.

[0075] The time-frequency transformation processing includes, but is not limited to, Fast Fourier Transform, Wavelet Transform, and Short-Time Fourier Transform, etc., without any specific limitation here.

[0076] Step 333: For each frequency band, calculate the spectral energy distribution of the frequency domain signal to quantify the energy intensity of the frequency band in which the frequency domain signal is located.

[0077] In other words, each frequency domain signal needs to go through steps 333 to 337 to finally obtain the time domain signal of the target in different frequency bands. For example, after going through steps 333 to 337, the low-frequency band frequency domain signal is converted into a low-frequency band time domain signal.

[0078] Step 335: Based on the energy intensity and weight data of the frequency band where the frequency domain signal is located, perform spectral weighting processing on the frequency domain signal in the frequency band where the frequency domain signal is located.

[0079] In one possible implementation, step 335 may include the following steps: Step 3351: Based on the energy intensity of the frequency band where the frequency domain signal is located, divide the frequency band where the frequency domain signal is located into several sub-frequency bands.

[0080] Step 3353: Calculate the mask corresponding to each sub-frequency band for filtering the effective components of each sub-frequency band, and assign corresponding weights to each sub-frequency band based on the weight data.

[0081] Step 3355: Perform element-wise multiplication on the mask corresponding to each sub-frequency band, the weight corresponding to each sub-frequency band, and the frequency domain signal to complete the spectral weighting processing of the frequency domain signal.

[0082] Step 337: The weighted frequency domain signal is converted into the time domain signal of the target in the frequency band of the frequency domain signal by inverse time-frequency transformation.

[0083] Among them, inverse time-frequency transform processing refers to the inverse process of time-frequency transform processing, which can be inverse fast Fourier transform, etc., and is not limited here.

[0084] Specifically, the EEG signals of the three frequency bands after data preprocessing are first transformed to the frequency domain using FFT to obtain the corresponding frequency domain signals, denoted as X. fft Then, calculate X. fft The spectral energy distribution of X is normalized using the following formula to quantify X. fft Energy intensity E in each frequency band norm : ; Where E is the original energy, E min E max These represent the minimum and maximum energy values, respectively.

[0085] To further extract effective components from EEG signals of different frequency bands, firstly, the quartiles (Q1, Q2, Q3) of the normalized energy are calculated, and X is then... fft The spectrum is further divided into three sub-bands: low, medium, and high. Then, a corresponding mask (M) is generated for each sub-band. low M mid M high ), used to screen the effective components of each sub-band, and based on X fft The weight data learned from the spectral information represented by the EEG signals in the corresponding frequency band is used to assign learnable weights W to the three sub-frequency bands respectively. low W mid W high X can be achieved through the following formula. fft Spectral weighting processing is used to enhance the effectiveness of epilepsy-related cost-saving measures. ; Here, ⊙ represents element-wise multiplication.

[0086] Finally, the weighted frequency domain signal Xweighted Perform an IFFT and convert back to the time domain to obtain the processed time-domain signal X. out This is used for subsequent time feature extraction.

[0087] Under the above embodiments, the EEG signal is converted to the frequency domain by fast Fourier transform, the frequency band is divided according to the energy distribution quartiles, and the epilepsy-related frequency band is dynamically enhanced and noise interference is suppressed by learningable frequency band weight coefficients. The optimized time domain signal is reconstructed by inverse fast Fourier transform, overcoming the adaptability limitations of traditional fixed frequency band filtering methods.

[0088] Please see Figure 7 In one exemplary embodiment, step 350 may include the following steps: Step 351: Input the time-domain signals of each frequency band into convolution kernels of different scales to extract time features and obtain time features of different scales.

[0089] In this context, a time feature at one scale corresponds to a frequency band. That is, a time feature at one scale corresponds to a convolution kernel at one scale, and a convolution kernel at one scale corresponds to a time-domain signal in a frequency band. For example, the input width of a low-frequency time-domain signal corresponds to a convolution kernel with a 0.5-second time window at a 250Hz sampling rate, yielding the time feature at the corresponding scale.

[0090] Step 353: Concatenate the time features of each scale according to the time dimension to obtain a multi-scale time feature matrix, which serves as the time feature data.

[0091] Among them, the temporal feature data is used to describe the epileptic feature waveforms of different durations captured at different scales on different EEG channels.

[0092] Specifically, for each frequency band's time-domain signal, after processing with different convolution kernels, corresponding feature representations will be obtained.

[0093] First, regarding the time-domain signals X output by the multi-band adaptive spectrum processing module, including low-frequency, high-frequency, and full-band signals... out These time-domain signals X from different frequency bands out By inputting three convolutional kernels respectively, the temporal features F1, F2, and F3 at different scales are extracted using the following formula: ; Among them, Conv i Let be the i-th convolutional kernel, BN be the batch normalization, and ReLU be the activation function. It should be noted that the widths of the three convolutional kernels can correspond to time windows of 0.5 seconds, 0.25 seconds, and 0.125 seconds at a sampling rate of 250Hz, respectively, to capture epileptic waveform features with different durations.

[0094] Then, the time features F1, F2, and F3 at different scales are concatenated along the time dimension to obtain the multi-scale time feature matrix X. temp The dimensions are C×T×D, where C is the number of EEG channels, T is the time step, and D is the feature dimension.

[0095] Under the above embodiments, for reconstructed signals of different frequency bands, multiple one-dimensional convolutional kernels with different time scales are used to extract time features in parallel and splice and fuse them along the time dimension, so as to simultaneously capture diverse time patterns such as transient spikes and continuous ripples in epileptic waveforms.

[0096] Please see Figure 8 In an exemplary embodiment, prior to step 350, the method may further include the following steps: Step 410: Construct a spatial network to capture changes in functional connectivity across different brain regions during an epileptic seizure.

[0097] Among them, spatial networks are used to extract features in the spatial dimension, thereby accurately representing the abnormal functional connections between different brain regions during an epileptic seizure.

[0098] In one possible implementation, step 410 may include the following steps: Step 411: Create a multi-channel feature matrix based on the correlation between multiple EEG channels in different brain regions.

[0099] Step 413: Based on the multi-channel feature matrix, a correlation matrix is ​​generated by calculating the Pearson correlation coefficient between all EEG channels at each time step.

[0100] Step 415: Introduce a learnable weight matrix to the correlation matrix to enhance its discriminative power and generate an initial adjacency matrix.

[0101] Step 415: Construct a symmetric normalized Laplace matrix based on the initial adjacency matrix to generate a spatial Laplace matrix, which serves as a spatial network for capturing functional connectivity changes in different brain regions during an epileptic seizure.

[0102] Specifically, firstly, regarding the multi-scale time feature matrix X... temp At each time step, the Pearson correlation coefficient between all EEG channels is calculated, generating a correlation matrix Cor∈R C×C This matrix quantifies the strength of functional connectivity between any two brain regions at a specific time, with positive values ​​indicating cooperative activity and negative values ​​indicating antagonistic relationships.

[0103] Meanwhile, to enhance the discriminative power of the correlation matrix, a learnable weight matrix W is introduced based on the correlation matrix. channel ∈R C×C The initial adjacency matrix A is generated using the following formula. channel : ; ReLU is the activation function.

[0104] Then, in the initial adjacency matrix A channel Based on this, a symmetric normalized Laplace matrix is ​​further constructed, and the spatial Laplace matrix L is generated using the following formula: ; Where D is the degree matrix and I is the identity matrix.

[0105] In this approach, based on spatial networks, it is possible to accurately capture the transient functional connectivity changes of different brain regions during epileptic seizures and eliminate the influence of differences in channel degrees on feature propagation, thereby achieving standardized spatial relationship representation.

[0106] Step 430: Construct a temporal network to capture the propagation changes of epileptic characteristic waveforms on the time axis during an epileptic seizure.

[0107] Among them, the temporal network is used to extract features in the time dimension, so as to accurately express the propagation pattern of signals changing over time during an epileptic seizure.

[0108] In one possible implementation, step 430 may include the following steps: Step 431: Map the multi-channel feature matrix to a single-channel time series, and generate a time difference matrix and learnable residual terms based on the single-channel time series at adjacent time steps.

[0109] Step 433: Combine the temporal difference matrix, the learnable residual terms, and the sparse mask generated by the sliding window based on the first set window length to generate the temporal adjacency matrix.

[0110] Step 435: Use the temporal adjacency matrix as a temporal network to capture the propagation changes of epileptic characteristic waveforms on the time axis during a seizure.

[0111] Specifically, to simplify the complexity of time modeling while preserving the core time variation trend, the multi-channel feature matrix X is transformed using the following linear transformation formula. temp Mapped to a single-channel time series X single : ; Among them, W proj ∈RD×1 For the projection weights, b proj For bias.

[0112] Then, to reduce computational complexity and focus on local time dependencies, a sparse mask M is generated based on a sliding window ω (with values ​​ranging from 3 to 5 time steps, which can adapt to the typical duration of epileptic waveforms). sparse , of which M sparse [i, j]=1 if and only if |ij| ≤ω (only the association between adjacent time steps is retained).

[0113] Finally, in the sparse mask M sparse Based on this, combined with the time difference matrix D diff With learnable residual term W time ∈R C×C The temporal adjacency matrix A is generated using the following formula. time : .

[0114] Where ReLU is the activation function, and ⊙ represents element-wise multiplication.

[0115] In this approach, based on a time network, the propagation pattern of epileptic waveforms on the time axis can be accurately captured, such as the rapid appearance and disappearance of spikes and the continuous spread of ripples. At the same time, key time dependencies are extracted through sparse dynamic modeling.

[0116] Step 450: The spatial network and the temporal network are fused, and a dynamic spatiotemporal graph convolutional network is generated through graph convolution operations.

[0117] In other words, based on dynamic spatiotemporal graph convolutional networks, the spatial correlation and temporal dependence of EEG signals during epileptic seizures can be dynamically modeled, ultimately achieving deep fusion of features in the temporal and spatial dimensions.

[0118] Specifically, an improved graph convolution formula is used to obtain the spatiotemporal feature matrix X. conv ∈R C×T ×D : ; Where L is the spatial Laplacian matrix, which controls the propagation of features in the brain region network; A time W is a time adjacency matrix that constrains the flow of features along the time axis. g ∈R D×D Let b be the graph convolution weight matrix, and b be the bias term, which realizes feature dimension transformation and nonlinear mapping.

[0119] In this approach, based on dynamic spatiotemporal graph convolutional networks, changes in the brain network state are reflected in real time. Combined with sparse temporal modeling to focus on key propagation patterns, this solves the problem that traditional static graph models cannot adapt to the spatiotemporal dynamic characteristics of EEG signals during epileptic seizures.

[0120] In the above process, a functional connectivity matrix is ​​constructed based on the inter-channel Pearson correlation coefficient in the spatial dimension, and a dynamic brain network adjacency matrix is ​​generated by combining learnable weights; in the temporal dimension, a self-attention mechanism is used to extract temporal contextual relationships and construct a sparse temporal relationship matrix; finally, the spatial and temporal adjacency matrices are fused, and a unified modeling and deep fusion of spatiotemporal features is achieved through graph convolution operations, which further improves the detection accuracy and generalization ability of epilepsy feature waveforms.

[0121] In one application scenario, the dynamic spatiotemporal graph convolutional network constructed in this embodiment was validated on both private and public datasets. The private dataset contained scalp EEG records from 18 patients with epilepsy, totaling approximately 2.78 hours, with 19 channels and a sampling rate of 250Hz, and annotated with two types of epileptic events (spikes and ripples). The public TUEV dataset contained a subset of the TUH EEG corpus, annotated with six types of epileptic events (spikes and sharp waves, generalized periodic epileptiform discharges, focal periodic epileptiform discharges, eye movements, artifacts, and background), with 23 channels and a sampling rate of 256Hz. The evaluation metrics used were precision, recall, and F1 score. The verification results show that, in this application scenario, the detection precision of epilepsy feature waveforms on the private dataset is 0.9845, the recall rate is 0.9882, and the F1 score is 0.9863. On the public dataset, the detection precision is 0.9590, the recall rate is 0.9690, and the F1 score is 0.9619. This fully demonstrates the feasibility and effectiveness of the epilepsy feature waveform detection method based on the dynamic spatiotemporal graph convolutional network. Therefore, the epilepsy feature waveform detection method provided in the embodiments of this application can provide guidance for neurosurgeons in the field of epilepsy seizure diagnosis and treatment. This not only helps to reduce the reliance on manual feature extraction and expert experience but also helps to improve the efficiency of electroencephalogram analysis.

[0122] Compared to related technologies, the proposed dynamic frequency band weight adjustment can adapt to the spectral changes during epileptic seizures in real time, with better noise suppression and stronger spectral adaptability than traditional fixed-band filtering. Compared to single-scale LSTM, the multi-scale convolutional kernel design can simultaneously capture waveform features as short as 0.125 seconds and as long as 0.5 seconds, providing more comprehensive temporal features. Compared to fixed-connection GNNs, the dynamic brain network and temporal relationship matrix can reflect the real-time changes in brain region connectivity and temporal propagation during epileptic seizures. The proposed model's detection process for epileptic characteristic waves such as spikes and ripples is more automated, providing doctors with accurate epileptic event localization and classification results, reducing diagnostic time and dependence on doctors' experience.

[0123] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0124] The following are embodiments of the apparatus described in this application, which can be used to execute the epilepsy characteristic waveform detection method involved in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the method embodiments of the epilepsy characteristic waveform detection method involved in this application.

[0125] Please see Figure 9 This application provides an epilepsy feature waveform detection device 900 for epilepsy seizure prediction, including but not limited to: an electroencephalogram (EEG) signal acquisition module 910, an adaptive spectrum processing module 930, a time feature extraction module 950, a spatiotemporal feature extraction module 970, and a waveform classification prediction module 990.

[0126] The EEG signal acquisition module 910 is used to acquire EEG signals of the target at different frequency bands. EEG signals are used to characterize the spectral information of epileptic characteristic waveforms.

[0127] The adaptive spectrum processing module 930 is used to perform spectrum processing on the EEG signals of each frequency band based on the weights of each frequency band learned from the spectrum information, so as to obtain the time domain signals of the target in different frequency bands.

[0128] The time feature extraction module 950 is used to extract time features from the time-domain signals of each frequency band at different scales to obtain time feature data. The time-domain signal of one frequency band corresponds to one scale.

[0129] The spatiotemporal feature extraction module 970 is used to input temporal feature data into the constructed dynamic spatiotemporal graph convolutional network to capture changes in functional connectivity of different brain regions and / or propagation changes of epileptic feature waveforms along the time axis during epileptic seizures, thereby obtaining spatiotemporal feature data. The dynamic spatiotemporal graph convolutional network includes a spatial network constructed based on the spatial correlation of different brain regions and a temporal network generated based on the temporal dependence of epileptic feature waveforms.

[0130] The waveform classification and prediction module 990 is used to classify and predict epilepsy characteristic waveforms from spatiotemporal feature data to obtain waveform detection results.

[0131] It should be noted that the epilepsy characteristic waveform detection device provided in the above embodiments is only illustrated by the division of the above functional modules when performing epilepsy characteristic waveform detection. In actual applications, the above functions can be assigned to different functional modules as needed. That is, the internal structure of the epilepsy characteristic waveform detection device will be divided into different functional modules to complete all or part of the functions described above.

[0132] Furthermore, the epilepsy characteristic waveform detection device and the epilepsy characteristic waveform detection method provided in the above embodiments belong to the same concept, and the specific way in which each module performs its operation has been described in detail in the method embodiments, and will not be repeated here.

[0133] Please see Figure 10 This application provides an electronic device 4000, which may include: a desktop computer, a laptop computer, a server, etc.

[0134] exist Figure 10 In this context, the electronic device 4000 includes at least one processor 4001 and at least one memory 4003.

[0135] Data interaction between the processor 4001 and the memory 4003 can be achieved through at least one communication bus 4002. This communication bus 4002 may include a path for transmitting data between the processor 4001 and the memory 4003. The communication bus 4002 can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus 4002 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not indicate that there is only one bus or one type of bus.

[0136] Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of this application.

[0137] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0138] The memory 4003 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing computer programs having instruction or data structure forms and accessible by the electronic device 4000, but not limited to these.

[0139] The memory 4003 stores a computer program, and the processor 4001 can read the computer program stored in the memory 4003 through the communication bus 4002.

[0140] The computer program is executed by one or more processors 4001 to implement the epilepsy feature waveform detection method for epileptic seizure prediction in the above embodiments.

[0141] Furthermore, this application provides a storage medium storing a computer program that is executed by one or more processors to implement the epilepsy feature waveform detection method described above for epileptic seizure prediction.

[0142] This application provides a computer program product, including a computer program that is executed by one or more processors to implement the epilepsy feature waveform detection method described above for epileptic seizure prediction.

[0143] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for detecting epileptic characteristic waveforms applied in epileptic seizure prediction, characterized in that, The method includes: Acquire electroencephalogram (EEG) signals of the target at different frequency bands; the EEG signals are used to characterize the spectral information of epileptic characteristic waveforms; Based on the weight data learned from the spectral information, the EEG signals of each frequency band are subjected to spectral processing to obtain the time-domain signals of the target in different frequency bands; Time features are extracted from the time-domain signals of each frequency band at different scales to obtain time feature data; the time-domain signal of one frequency band corresponds to one scale; The temporal feature data is input into the constructed dynamic spatiotemporal graph convolutional network to capture the changes in functional connectivity of different brain regions and / or the propagation changes of epileptic feature waveforms on the time axis during an epileptic seizure, thereby obtaining spatiotemporal feature data; the dynamic spatiotemporal graph convolutional network includes a spatial network constructed based on the spatial correlation of different brain regions and a temporal network generated based on the temporal dependence of epileptic feature waveforms. The spatiotemporal feature data is used to classify and predict epilepsy feature waveforms to obtain waveform detection results.

2. The method as described in claim 1, characterized in that, The weight data learned based on the spectral information is used to perform spectral processing on the EEG signals of each frequency band to obtain the time-domain signals of the target in different frequency bands, including: The EEG signals of each frequency band are subjected to time-frequency transformation processing to obtain the frequency domain signals of each frequency band; For each frequency band of the frequency domain signal, the spectral energy distribution of the frequency domain signal is calculated to quantify the energy intensity of the frequency band in which the frequency domain signal is located; Based on the energy intensity of the frequency band where the frequency domain signal is located and the weight data, the frequency domain signal is subjected to spectral weighting processing in the frequency band where the frequency domain signal is located; The weighted frequency domain signal is converted into the time domain signal of the target in the frequency band of the frequency domain signal by inverse time-frequency transformation.

3. The method as described in claim 2, characterized in that, The step of performing spectral weighting processing on the frequency domain signal in the frequency band based on the energy intensity of the frequency band where the frequency domain signal is located and the weight data includes: Based on the energy intensity of the frequency band where the frequency domain signal is located, the frequency band where the frequency domain signal is located is divided into several sub-frequency bands; Calculate the mask corresponding to each sub-frequency band for filtering the effective components of each sub-frequency band, and assign corresponding weights to each sub-frequency band based on the weight data; Element-wise multiplication is performed on the mask corresponding to each sub-frequency band, the weight corresponding to each sub-frequency band, and the frequency domain signal to complete the spectral weighting processing of the frequency domain signal.

4. The method as described in claim 1, characterized in that, The time-domain signals of each frequency band are subjected to time feature extraction at different scales to obtain time feature data, including: The time-domain signals of each frequency band are input into convolutional kernels of different scales to extract time features, resulting in time features of different scales; each time feature of a scale corresponds to a frequency band. By concatenating the time features at each scale according to the time dimension, a multi-scale time feature matrix is ​​obtained, which serves as the time feature data.

5. The method as described in claim 1, characterized in that, Before inputting the temporal feature data into the constructed dynamic spatiotemporal graph convolutional network to capture the functional connectivity changes in different brain regions and / or the propagation changes of epileptic feature waveforms along the time axis during an epileptic seizure, the method includes: A multi-channel feature matrix is ​​created based on the correlation between multiple EEG channels in different brain regions. Based on the multi-channel feature matrix, a correlation matrix is ​​generated by calculating the Pearson correlation coefficient between all EEG channels at each time step; A learnable weight matrix is ​​introduced into the correlation matrix to enhance its discriminative power, thereby generating an initial adjacency matrix. Based on the initial adjacency matrix, a symmetric normalized Laplace matrix is ​​constructed to generate a spatial Laplace matrix, which serves as the spatial network used to capture changes in functional connectivity across different brain regions during an epileptic seizure.

6. The method as described in claim 5, characterized in that, Before inputting the temporal feature data into the constructed dynamic spatiotemporal graph convolutional network to capture the functional connectivity changes in different brain regions and / or the propagation changes of epileptic feature waveforms along the time axis during an epileptic seizure, the method includes: The multi-channel feature matrix is ​​mapped to a single-channel time series, and a time difference matrix and learnable residual terms are generated based on the single-channel time series at adjacent time steps. A temporal adjacency matrix is ​​generated by combining the temporal difference matrix, the learnable residual term, and the sparse mask generated by a sliding window based on a first set window length. The time adjacency matrix is ​​used as the time network for capturing the propagation changes of epileptic characteristic waveforms on the time axis during an epileptic seizure.

7. The method according to any one of claims 1 to 6, characterized in that, The acquisition of the target's EEG signals in different frequency bands includes: Acquire the raw electroencephalogram (EEG) signals of the target; The original EEG signal is segmented using a sliding window with a second set window length; Each segment of the original EEG signal is input into a different bandpass filter to obtain the target's EEG signals in different frequency bands.

8. The method according to any one of claims 1 to 6, wherein the frequency band includes a low-frequency band, a high-frequency band, and a full-frequency band; the frequency range of the low-frequency band is 1 to 70 Hz, and the frequency range of the high-frequency band is above 80 Hz.

9. The method according to any one of claims 1 to 6, wherein the epileptic characteristic waveform includes at least one of spikes and ripples.

10. An electronic device comprising at least one processor and at least one memory, wherein, The memory stores a computer program, characterized in that, when the computer program is executed by the processor, it implements the method for detecting epileptic characteristic waveforms in epileptic seizure prediction as described in any one of claims 1 to 8.

Images