Online adaptive prediction method and system for epileptic seizure

Abstract

The invention discloses an online adaptive prediction method and system for epileptic seizure, and relates to the technical field of medical signal processing and artificial intelligence, and the method comprises the steps: obtaining a multi-channel electroencephalogram signal of a target user; constructing an epileptic seizure online adaptive prediction model according to spatial constraint independent component analysis, a brain function network and a transfer learning mechanism; inputting the multi-channel electroencephalogram signals into an online adaptive epileptic seizure prediction model, and performing state identification and prediction before epileptic seizure to obtain a prediction result; and performing online early warning judgment according to a prediction result, and triggering an alarm if an early warning condition is met. According to the method, the problem of unstable early warning effect caused by epilepsy electroencephalogram data variability is solved, clinical accurate and rapid online epileptic seizure prediction is realized, and a basis is provided for treatment of refractory epilepsy and research on a seizure mechanism of the refractory epilepsy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical signal processing and artificial intelligence, in particular to an online adaptive seizure prediction method and system. BACKGROUND

[0002] Epilepsy is a clinical syndrome caused by various causes, characterized by paroxysmal, transient, repetitive and stereotyped central nervous system dysfunction caused by abnormal discharge of highly synchronized neurons in the brain. It is the second most stubborn disease after cerebrovascular disease. Moreover, refractory epilepsy is extremely harmful and lacks effective treatment. Long-term frequent seizures can cause brain function damage, and even threaten life in severe cases. Therefore, seizure prediction is crucial.

[0003] In recent years, research on seizure prediction has made some progress, laying a theoretical foundation for the clinical application of seizure prediction, but there is still a considerable distance from clinical application. At present, most of the algorithms for seizure prediction are based on public data sets on the Internet, which lack sufficient data samples and the data information is not universal, and the heterogeneity of epilepsy is not considered, so the algorithm performance is not convincing for clinical big data. The International League Against Epilepsy considers epilepsy as a brain network connection disorder, highlighting the role of brain network in epilepsy research. And most of the existing seizure prediction methods based on multiple variables do not consider the spatial synchronization and connectivity information inside the brain, and are not highly related to the study of the pathogenesis of epilepsy.

[0004] Therefore, there is an urgent need for an online adaptive seizure prediction method and system that focuses on the difficult problem of seizure prediction based on the accumulation of related epilepsy electroencephalogram signal research, to achieve accurate and rapid online seizure prediction. SUMMARY

[0005] The purpose of the present application is to provide an online adaptive seizure prediction method and system, which solves the problem of unstable early warning effect caused by the variability of epilepsy electroencephalogram data, thereby achieving accurate and rapid online seizure prediction.

[0006] To achieve the above-mentioned purpose, the present application provides an online adaptive seizure prediction method, comprising the following steps: S1, acquiring multi-channel electroencephalogram signals of a target user; S2, constructing an online adaptive seizure prediction model according to fusion spatial constraint independent component analysis, brain function network and transfer learning mechanism; S3, inputting the multi-channel electroencephalogram signals into the online adaptive seizure prediction model to identify and predict the pre-seizure state and obtain the prediction result; S4, according to the prediction result, online early warning judgment is carried out, and if the early warning condition is met, an alarm is triggered.

[0007] Preferably, the specific content of constructing the online adaptive seizure prediction model according to the fusion space constraint independent component analysis, brain function network and transfer learning mechanism in S2 comprises: Obtaining the multi-channel historical electroencephalogram (EEG) signals and the seizure conditions of a user; According to the multi-channel historical electroencephalogram (EEG) signals and the seizure conditions, constructing a training set and a test set; According to the fusion space constraint independent component analysis, constructing an EEG noise estimation module; According to the brain function network, constructing an EEG feature information mapping module; According to the transfer learning mechanism, constructing an EEG information prediction module; Connecting the EEG noise estimation module, the EEG feature information mapping module and the EEG information prediction module to obtain an initial model of the online adaptive seizure prediction model; Training and testing the initial model based on the training set and the test set, and taking the trained and tested initial model as the online adaptive seizure prediction model.

[0008] Preferably, the specific content of constructing the EEG noise estimation module according to the fusion space constraint independent component analysis comprises: Filtering the multi-channel historical electroencephalogram (EEG) signals to obtain filtered signals; According to the spatial constraint matrix and the fusion space constraint independent component analysis, separating the filtered signals to obtain filtered signal components; Performing wavelet threshold denoising processing on the filtered signal components to obtain noise components; Performing projection inverse transform on the noise components to obtain noise estimators; Removing the noise estimators from the filtered signals to obtain pure EEG signals, and obtaining the EEG noise estimation module.

[0009] Preferably, the specific content of constructing the EEG feature information mapping module according to the brain function network comprises: Performing rhythm extraction on the pure EEG signals by frequency slice wavelet transform to obtain rhythmic EEG signals; Establishing a multi-track autoregressive module based on the rhythmic EEG signals, and solving the multi-track autoregressive module coefficients by least squares method; Based on the partial directed coherence method, constructing a brain function network graph and a connectivity matrix according to the multi-track autoregressive module coefficients, and taking the brain function network graph and the connectivity matrix as feature images of the mapped EEG information to obtain the EEG feature information mapping module.

[0010] Preferably, according to the mechanism of transfer learning, the specific content of the electroencephalogram information prediction module comprises: According to the feature image, a bidirectional long short-term memory network is constructed by applying inter-field joint knowledge transfer to obtain the electroencephalogram information prediction module.

[0011] Preferably, the bidirectional long short-term memory network is based on the structural optimization of the long short-term memory network, and specifically, the connection from the input gate to the output gate is added in the long short-term memory network structure, and the input gate and the forgetting gate are merged into an update gate.

[0012] Preferably, the online early warning judgment in S4 comprises: A preset early warning threshold n and a prediction number s; the n is an integer greater than 1; the s is an integer greater than or equal to 0; If s=0, it indicates that the seizure online adaptive model does not detect the electroencephalogram information of the seizure precursor; If s≥n, the prediction result is identified as the pre-seizure state, and an alarm is triggered; If s<n, the prediction result is identified as the pre-seizure state, and the prediction is continued until s≥n, and an alarm is triggered.

[0013] The present application provides a seizure online adaptive prediction system, comprising: A signal acquisition module for acquiring multi-channel electroencephalogram signals of a target user; A model construction module for constructing a seizure online adaptive prediction model according to the fusion of spatial constraint independent component analysis, brain function network and transfer learning mechanism; An online prediction module for inputting the multi-channel electroencephalogram signals into the seizure online adaptive prediction model to identify and predict the pre-seizure state and obtain a prediction result; An online early warning module for making online early warning judgment according to the prediction result, and triggering an alarm if the early warning condition is met.

[0014] The present application provides a computer device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to realize the content of the above-mentioned seizure online adaptive prediction method.

[0015] The present application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to realize the content of the above-mentioned seizure online adaptive prediction.

[0016] In summary, the epilepsy seizure online adaptive prediction method and system provided by the application has the following beneficial effects compared with the prior art: the epilepsy seizure online adaptive prediction model constructed according to the spatial constraint independent component analysis, the brain function network and the transfer learning mechanism fully explores the change process of the brain function network topology structure of different epilepsy patients at different times, solves the problems of the diversity of the epilepsy electroencephalogram, the large amount of noise contained in the electroencephalogram signal and the complexity and variability of the electroencephalogram data, and in combination with the online early warning mechanism, solves the problem of unstable early warning effect caused by the variability of the epilepsy electroencephalogram data, so that the online epilepsy seizure prediction is realized with clinical accuracy and rapidness, and a basis is provided for the treatment of intractable epilepsy and the research on the seizure mechanism.

[0017] The technical method of the application is further described below through the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The flowchart of the epilepsy seizure online adaptive prediction method of the application is shown in Figure 1. Figure 2 The structural diagram of the electroencephalogram noise estimation module in the application is shown in Figure 2. Figure 3 The architecture diagram of the electroencephalogram feature information mapping module in the application is shown in Figure 3. Figure 4 The structural diagram of the electroencephalogram information prediction module in the application is shown in Figure 4. Figure 5 The structural diagram of the online early warning mechanism in the application is shown in Figure 5. Figure 6 The module diagram of the epilepsy seizure online adaptive prediction system of the application is shown in Figure 6. DETAILED DESCRIPTION

[0019] The technical method of the application is further described below through the drawings and examples. It should be noted that: unless otherwise specifically stated, the relative arrangement, numerical expression and numerical value of the components and steps set forth in these examples do not limit the scope of the application.

[0020] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way limiting to the scope of the application or its applications or uses.

[0021] Techniques, systems, and devices known to those of ordinary skill in the relevant art can not be discussed in detail herein, but should be considered as part of the description if appropriate.

[0022] In all examples shown and discussed herein, any specific values should be interpreted as merely illustrative, and not as a limitation. Thus, other examples of the exemplary embodiments can have different values.

[0023] Unless otherwise defined, technical terms or scientific terms used in the present application shall have the meanings that are commonly understood by a person of ordinary skill in the art to which the present application belongs.

[0024] As shown in Figure 1 A method for online adaptive prediction of epileptic seizures, comprising: Step S1, acquiring multi-channel electroencephalogram signals of a target user. The multi-channel electroencephalogram signals are current multi-channel electroencephalogram signal data of the target user. The multi-channel electroencephalogram signals are complex due to the complex mechanism of epileptic seizures, and the epileptic electroencephalogram presents diversity. This diversity is reflected in that the electroencephalogram features during epileptic seizures of different epileptic patients present significant individual differences. Moreover, a large amount of noise is contained in the electroencephalogram signals. Traditional epileptic electroencephalogram pattern recognition methods often analyze features according to the form and characteristics of a part of signals, and apply the same features to all data. The effect is very unstable for different individuals. Such a method is only suitable for cases with strong signal pattern consistency, while clinical patients are different, and electroencephalogram data are often complex and variable, which are typical non-consistent data. Therefore, the traditional method is difficult to achieve good results. Therefore, the present application constructs an online adaptive prediction model of epileptic seizures.

[0025] Step S2, constructing an online adaptive prediction model of epileptic seizures according to the fusion space constraint independent component analysis, brain function network and transfer learning mechanism. The epileptic electroencephalogram signal decoding method with adaptive learning ability is the key to realize clinical epileptic warning. Therefore, the present application designs the algorithm from three aspects of signal denoising, feature analysis and state prediction according to the characteristics of complex epileptic electroencephalogram pattern and easy noise interference, and constructs an online adaptive prediction model of epileptic seizures.

[0026] Further, step S2 can be replaced by steps S201 to S207, specifically as follows: Step S201, acquiring multi-channel historical electroencephalogram signals and epileptic seizure conditions of a user. The multi-channel historical electroencephalogram signals are historical data of the multi-channel electroencephalogram signals of the user.

[0027] Step S202, constructing a training set and a test set according to the multi-channel historical electroencephalogram signals and the epileptic seizure conditions.

[0028] Step S203, constructing an electroencephalogram noise estimation module according to the fusion space constraint independent component analysis. The present application constructs an electroencephalogram noise estimation module based on the fusion space constraint independent component analysis, which is to construct a constraint matrix by using prior knowledge, and to add a space constraint matrix to the traditional fast independent component analysis algorithm (FastICA) to reduce the calculation amount and enhance the robustness of the method. Then, the wavelet threshold de-noising is applied reversely, and the advantages of the spatially constrained independent component analysis (SCICA) and the traditional filter are fused to establish a noise estimation module, and the module structure is as shown in Figure 2 . Figure 2 , is a mixing matrix, wherein, is a mixing matrix of unknown prior components. SC-I CS is a spatially constrained independent component. is the final de-noised electroencephalogram signal. X ( t ) is an input multi-channel historical electroencephalogram signal, is a spatially constrained independent component obtained after transformation.

[0029] Specifically, the content of step S203 includes: The original electroencephalogram (EEG), i.e. the multi-channel historical electroencephalogram signal, is subjected to a filtering operation to obtain a filtered signal. Specifically, after the multi-channel historical electroencephalogram signal is subjected to a band-pass filter such as a Butterworth band-pass filter, a filtered signal with a frequency band of 0.5Hz-40Hz is obtained y ( t ); According to the space constraint matrix and the SCICA, the filtered signal is subjected to separation processing to obtain a filtered signal component.

[0030] The filtered signal component is subjected to wavelet threshold de-noising processing to obtain a noise component .

[0031] The noise component is subjected to projection inverse transformation to obtain a noise estimation , wherein, is a mixing matrix of known prior components.

[0032] The noise estimation in the filtered signal is removed to obtain a pure electroencephalogram signal, i.e. a de-noised EEG, and an electroencephalogram noise estimation module is obtained.

[0033] The present application is based on the traditional electroencephalogram constraint condition, accurately estimates the noise state, obtains relatively pure electroencephalogram data, suppresses and weakens the interference of non-Gaussian noise and artifacts on electroencephalogram signal analysis, and lays a solid foundation for revealing and extracting effective feature information in subsequent electroencephalogram signal analysis.

[0034] Step S204, constructing an electroencephalogram feature information mapping module according to the brain function network. In order to increase the acquisition degree of effective information in the electroencephalogram of different patients, the present application constructs an electroencephalogram feature information mapping module. The brain function network can directly reflect the change of the coupling dependence relationship between each channel before and after the seizure. The electroencephalogram feature information mapping module is constructed by first extracting the rhythm of the pure electroencephalogram signal by frequency-slicing wavelet transform (FSWT), and then constructing the brain function network based on the partial directed coherence method (PDC). The brain function network is first established by a multivariate autoregressive (MVAR) module, the least square method is used to solve the coefficients and error coefficients of the MVAR module, and then the Fourier transform is used to convert it to the frequency domain, the causal relationship of information transmission between the multi-channel electroencephalogram signal is calculated by PDC, and the brain function network diagram and the connected matrix are drawn as the feature image to map the electroencephalogram information. The structure of the electroencephalogram feature information mapping module is as shown in Figure 3 .

[0035] Specifically, the content of step S204 includes: The frequency-slicing wavelet transform is used to extract the rhythm of the pure electroencephalogram signal, and the rhythm electroencephalogram signal is obtained. The expression of the frequency-slicing wavelet transform is: ; Wherein, t is a time parameter, ω is a center frequency, k is an index of the frequency slice, π is a circular constant, W f ( t , ω , k ) is the time-frequency feature after the frequency-slicing wavelet transform, u is an integral variable in the frequency domain, e iut is a complex exponential function, is the Fourier transform of the signal, is the conjugate Fourier transform of the mother wavelet.

[0036] The expression of the selected slice function is: ; wherein, is a rectangular slice function, ω 0 is a cut-off frequency of the rectangular slice function.

[0037] ; wherein, sin is a sine function.

[0038] .

[0039] .

[0040] A multi-track autoregressive module is established for the rhythmic electroencephalogram signal, and the coefficients of the multi-track autoregressive module are solved by the least square method. The expression of the multi-track autoregressive module is: ; wherein, n is a time point, x 1( n ) is a signal value of the first electroencephalogram channel at the time n , x N ( n ) is a signal value of the N th electroencephalogram channel at the time n , r is a lag order, r =1, 2, …, p , p is a positive integer, A r is a coefficient of the multi-track autoregressive module, w 1( n ) is a residual term of the first channel, w N ( n ) is a residual term of the N th channel.

[0041] The expression of the coefficient of the multi-track autoregressive module is: ; wherein, a 11 ( r ) is an influence coefficient of the first electroencephalogram channel on the current signal of itself at a lag r , a 1N ( r ) is an influence coefficient of the N th electroencephalogram channel on the current signal of the first channel at a lag r , aN1 ( r The first brainwave pathway is in lag. r At that moment, for the first N The influence coefficient of the current signal in each channel a NN ( r ) is the first N Individual brainwave channels are lagging r The influence coefficient of a given moment on its current signal.

[0042] Based on the partial-directed coherence method, a brain functional network map and connectivity matrix are constructed according to the coefficients of the multi-track autoregressive module. These brain functional network map and connectivity matrix are then used as feature images to map EEG information, resulting in EEG feature information mapping modules. Specifically, based on the partial-directed coherence method, a connectivity matrix is ​​constructed according to the coefficients of the multi-track autoregressive module. A threshold is selected to binarize the connectivity matrix, and then the brain functional network map is drawn at selected electrode locations, ultimately yielding the brain functional network map and connectivity matrix.

[0043] The expression for the partial orientation coherence function is: ; in, P xy ( f () is a partial orientation coherence function used to quantify EEG channels. x and y The intensity of causal information transmission at frequency f Frequency domain coefficient matrix A y ( f The conjugate transpose of ) A y ( f (electroencephalography channel) y The corresponding frequency domain coefficient matrix. Where: ; in, A xy ( f (electroencephalography channel) x and y In frequency f The frequency domain coefficient matrix below, f For frequency variables, I It is the identity matrix. e -πirf It is a complex exponential function.

[0044] The application is based on the rhythm characteristics of the electroencephalogram signal, and a brain function network is evaluated according to a multi-track autoregressive module to explore the change process of the brain function network topology of the epilepsy patients in different periods under different coherence analysis strategies, and a brain electrical feature information mapping module with universal significance is constructed.

[0045] In step S205, an electroencephalogram information prediction module is constructed according to a transfer learning mechanism. In view of the individual differences between patients and the sparseness of the epilepsy state, the deep electroencephalogram information prediction module based on transfer learning is constructed to solve the problem of different distributions of training data and test data. In the theoretical framework of deep learning, the feature image (brain function network graph and connected matrix) is taken as the input, a bidirectional long short-term memory network (BiLSTM) is constructed, the inter-field joint knowledge transfer is used, the information of the existing data is fully utilized to complete the classification of the target domain data, and the adaptability of the algorithm to different patient data is improved. In order to enable the long short-term memory network (LSTM) to better learn the long-term historical information, an optimization method of the LSTM is proposed, the connection from the input gate to the output gate is increased, and the input gate and the forgetting gate are combined into an update gate, so that the automatic learning ability of the model to the electroencephalogram feature is improved, and the robustness is enhanced. The structure of the electroencephalogram information prediction module is as shown in Figure 4 Figure 4 F 1, F 2,..., F i-1 , F i is input layer data, L 1, L 2,..., L i-1 , L i is an LSTM hidden layer state, I 1, I 2,..., I i-1 , I i is an attention weighted feature.

[0046] Specifically, the content of step S205 includes: According to the feature image, a bidirectional long short-term memory network is constructed by using inter-field joint knowledge transfer to obtain the electroencephalogram information prediction module. The bidirectional long short-term memory network is based on the structural optimization of the long short-term memory network, and specifically, the connection from the input gate to the output gate is increased in the long short-term memory network structure, and the input gate and the forgetting gate are combined into an update gate. ​​

[0047] The diversity and variability of EEG states in epilepsy patients lead to a mismatch between the source domain used for training the prediction model and the target domain used for testing, resulting in poor performance of the EEG information prediction module in the target domain. To improve the adaptability and learning ability of the EEG information prediction module to data from different patients, this invention constructs an EEG information prediction module based on transfer learning. Through domain-joint knowledge transfer and deep optimization networks, it automatically learns and summarizes the inherent patterns of feature maps, proposes an optimization method for the prediction network, expands the algorithm's adaptability to different scenarios, and improves its predictive performance, achieving adaptive prediction of epilepsy EEG signals from different patients.

[0048] Step S206: Connect the EEG noise estimation module, the EEG feature information mapping module, and the EEG information prediction module to obtain the initial model of the online adaptive prediction model for epileptic seizures.

[0049] Step S207: Train and test the initial model based on the training set and test set, and use the trained and tested initial model as an online adaptive prediction model for epileptic seizures.

[0050] Step S3: Input the multi-channel EEG signals into the online adaptive prediction model for epileptic seizures to identify and predict the pre-seizure state and obtain the prediction results.

[0051] Step S4: Based on the prediction results, perform online early warning judgment. If the early warning conditions are met, trigger an alarm. The specific content of the online early warning judgment includes: The preset warning threshold n and the number of predictions s are defined. Where n is an integer greater than 1 and s is an integer greater than or equal to 0.

[0052] If s=0, it means that the online adaptive model for epileptic seizures did not detect EEG information of pre-epileptic aura.

[0053] If s≥n, the prediction result is identified as a pre-seizure state, triggering an alarm.

[0054] If s < n, the prediction result is identified as the pre-seizure state, and prediction continues until s ≥ n, at which point an alarm is triggered.

[0055] like Figure 5 As shown, the multi-channel EEG signals are processed through a sliding window and EEG decoding to output prediction results. The prediction results are then used to determine whether they match the premonitory signs of an attack. If the prediction is determined to be a premonitory sign of an attack, an alarm is generated, allowing epilepsy to be effectively controlled before an attack occurs.

[0056] This invention is based on an online adaptive prediction model for epileptic seizures, combined with an online early warning mechanism for epileptic seizures. It can simulate clinical scenarios and demonstrates the effectiveness of this invention in early warning of epileptic seizures.

[0057] This invention provides an online adaptive prediction system for epileptic seizures, such as... Figure 6 As shown, it includes: The signal acquisition module is used to acquire multi-channel EEG signals from the target user.

[0058] The model building module is used to construct an online adaptive prediction model for epileptic seizures based on fusion spatial constraint independent component analysis, brain functional networks, and transfer learning mechanisms.

[0059] The online prediction module is used to input multi-channel EEG signals into the online adaptive prediction model for epileptic seizures, identify and predict the pre-seizure state, and obtain prediction results.

[0060] The online early warning module is used to make online early warning judgments based on the prediction results, and to trigger an alarm if the early warning conditions are met.

[0061] The present invention provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the content of an online adaptive prediction method for epileptic seizures.

[0062] The present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the content of an online adaptive prediction method for epileptic seizures.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical methods of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical methods of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical methods to deviate from the spirit and scope of the technical methods of the present invention.

Claims

1. An online adaptive prediction method for epileptic seizures, characterized in that, The online adaptive prediction method for epileptic seizures includes the following steps: S1. Acquire multi-channel EEG signals from the target user; S2. Based on fusion spatial constraint independent component analysis, brain functional networks, and transfer learning mechanisms, an online adaptive prediction model for epileptic seizures is constructed. S3. Input multi-channel EEG signals into the online adaptive prediction model for epileptic seizures to identify and predict the pre-seizure state and obtain prediction results. S4. Based on the prediction results, make an online early warning judgment. If the early warning conditions are met, trigger an alarm.

2. The online adaptive prediction method for epileptic seizures according to claim 1, characterized in that, The specific content of S2, which constructs an online adaptive prediction model for epileptic seizures based on fusion spatial constraint independent component analysis, brain functional networks, and transfer learning mechanisms, includes: Acquire users' multi-channel historical EEG signals and epileptic seizure data; Training and testing sets were constructed based on multi-channel historical EEG signals and epileptic seizure data. Based on fusion spatial constraint independent component analysis, an EEG noise estimation module is constructed; Based on brain functional networks, a brain electroencephalogram (EEG) feature information mapping module was constructed. Based on the transfer learning mechanism, an EEG information prediction module was constructed; By connecting the EEG noise estimation module, the EEG feature information mapping module, and the EEG information prediction module, an initial model for the online adaptive prediction model of epileptic seizures is obtained. The initial model is trained and tested based on the training and test sets, and the trained and tested initial model is used as an online adaptive prediction model for epileptic seizures.

3. The online adaptive prediction method for epileptic seizures according to claim 2, characterized in that, Based on fusion spatially constrained independent component analysis, the specific content of the EEG noise estimation module includes: The multi-channel historical EEG signals were filtered to obtain the filtered signal; Based on the spatial constraint matrix and the fusion spatial constraint independent component analysis, the filtered signal is separated to obtain the filtered signal components; Wavelet threshold denoising is performed on the filtered signal components to obtain the noise components; Perform an inverse projection transformation on the noise components to obtain the noise estimate; The noise estimate in the filtered signal is removed to obtain a clean EEG signal, and an EEG noise estimation module is obtained.

4. The online adaptive prediction method for epileptic seizures according to claim 3, characterized in that, Based on brain functional networks, the specific content of constructing the EEG feature information mapping module includes: Frequency slice wavelet transform was used to extract the rhythm from the pure EEG signal to obtain the rhythmic EEG signal; A multi-track autoregressive module was established using rhythmic EEG signals as the object, and the coefficients of the multi-track autoregressive module were solved by the least squares method. Based on the partial orientation coherence method, a brain functional network diagram and connectivity matrix are constructed according to the coefficients of the multi-track autoregressive module. The brain functional network diagram and connectivity matrix are used as feature images to map EEG information, thus obtaining the EEG feature information mapping module.

5. The online adaptive prediction method for epileptic seizures according to claim 4, characterized in that, Based on the transfer learning mechanism, the specific content of constructing the EEG information prediction module includes: Based on the feature images, a bidirectional long short-term memory network is constructed by applying inter-domain joint knowledge transfer to obtain the EEG information prediction module.

6. The online adaptive prediction method for epileptic seizures according to claim 5, characterized in that, The bidirectional long short-term memory network is based on the structural optimization of the long short-term memory network. Specifically, it adds a connection from the input gate to the output gate in the long short-term memory network structure and merges the input gate and the forget gate into an update gate.

7. The online adaptive prediction method for epileptic seizures according to claim 2, characterized in that, Online early warning judgments in S4 include: A preset warning threshold n and prediction count s are defined; where n is an integer greater than 1 and s is an integer greater than or equal to 0. If s=0, it means that the online adaptive model for epileptic seizures did not detect EEG information of epileptic premonitory signals. If s≥n, the prediction result is identified as a pre-seizure state, triggering an alarm; If s < n, the prediction result is identified as the pre-seizure state, and prediction continues until s ≥ n, at which point an alarm is triggered.

8. An online adaptive prediction system for epileptic seizures, characterized in that, include: The signal acquisition module is used to acquire multi-channel EEG signals from the target user. The model building module is used to construct an online adaptive prediction model for epileptic seizures based on fusion spatial constraint independent component analysis, brain functional networks, and transfer learning mechanisms. The online prediction module is used to input multi-channel EEG signals into the online adaptive prediction model for epileptic seizures to identify and predict the pre-seizure state and obtain prediction results. The online early warning module is used to make online early warning judgments based on the prediction results, and to trigger an alarm if the early warning conditions are met.

9. A computer device, comprising: The memory, the processor, and the computer program stored in the memory and executable on the processor are characterized in that the processor executes the computer program to implement the content of the online adaptive prediction method for epileptic seizures according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the content of the online adaptive prediction method for epileptic seizures as described in any one of claims 1-7.

Images