An electroencephalogram epilepsy detection system based on determined learning feature extraction

Abstract

The application relates to the technical field of electroencephalogram epilepsy detection, and discloses an electroencephalogram epilepsy detection system based on determined learning feature extraction, which comprises a training and verification module configured to: acquire electroencephalogram signals of multiple channels; for the electroencephalogram signals of each channel, call a data processing module to perform preprocessing, obtain multiple segments, and divide the segments into a training set and a verification set; for each segment, obtain a feature vector; for the electroencephalogram signals of each channel, train a classifier by using the training set; select an optimal channel by using the verification set; and a detection module configured to: acquire the electroencephalogram signals of the optimal channel, call the data processing module to perform preprocessing, and obtain multiple segments; call a signal expansion module to expand each segment; call a determined learning module and a feature extraction module to perform determined learning feature extraction by using a neural network, obtain a feature vector, and put the feature vector into the trained classifier to perform epilepsy detection. The epilepsy detection is accurate and fast.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electroencephalogram epilepsy detection, and particularly relates to an electroencephalogram epilepsy detection system based on determined learning feature extraction. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] Epilepsy is a common chronic neurological disease caused by abnormal neuron firing, affecting more than 50 million patients worldwide, of which more than 20% are ineffective to drug treatment. These seizures can cause loss of physical control, cognitive impairment, memory loss, and ultimately loss of life. Since these seizures are sudden and unpredictable, patients have always been at risk. Electroencephalogram can reflect brain activity by recording the electrical physiological signals of neurons. Therefore, electroencephalogram has been widely used in the diagnosis and treatment of seizures. However, manual diagnosis of epilepsy requires experienced neurologists to analyze a large amount of electroencephalogram signals, which can take several hours to make a diagnosis. In addition, it is impractical for doctors to monitor patients all the time. Therefore, automatic epilepsy detection technology and real-time fast prediction methods are crucial for epilepsy patients, and this field is rapidly developing thanks to the progress of artificial intelligence.

[0004] Feature extraction plays a core role in epilepsy detection, which converts the measured electroencephalogram signal values into more relevant and information-rich features to improve epilepsy recognition ability. Most feature extraction methods extract time domain, frequency domain and time-frequency domain features of electroencephalogram signals, and time domain features mainly include mean, variance, entropy, etc. Time domain features reflect the time-varying characteristics of electroencephalogram signals, but these features are sensitive to noise. Frequency domain features can avoid noise problems because noise is mainly concentrated in fixed frequency bands, and frequency domain features can filter out the impact of noise on epilepsy detection, and frequency domain features mainly include spectral power, spectral entropy, median frequency, etc. Frequency domain features can reduce the impact of noise, but the non-stationarity of electroencephalogram signals makes them insensitive to local change patterns of electroencephalogram signals within a certain time period. Therefore, time-frequency domain features that combine time-varying and frequency characteristics can achieve better detection performance. Common methods of time-frequency methods include wavelet decomposition and empirical mode decomposition, etc.

[0005] Time domain, frequency domain and time-frequency domain features usually reflect the characteristics of a certain specific attribute, and many researchers need to integrate several different features for epilepsy detection. It is a problem to determine which features should be combined to accurately and quickly detect epilepsy in electroencephalogram signals; extracting too many features will increase the calculation time and may cause redundancy, while extracting too few features cannot fully represent the epilepsy pattern and reduce the recognition ability. SUMMARY

[0006] In order to solve the problems in the prior art, the application provides an electroencephalogram (EEG) epilepsy detection system based on deterministic learning feature extraction, which can extract nonlinear dynamic features of single-channel short-time interval EEG signals and realize accurate and rapid epilepsy detection.

[0007] In a first aspect, the application provides an electroencephalogram (EEG) epilepsy detection system based on deterministic learning feature extraction.

[0008] The EEG epilepsy detection system based on deterministic learning feature extraction comprises:

[0009] The training and verification module is configured to: acquire EEG signals of multiple channels; for the EEG signals of each channel, call the data processing module for preprocessing to obtain multiple segments, and divide the segments into a training set and a verification set; for each segment, sequentially call the signal expansion module, the deterministic learning module and the feature extraction module to obtain a feature vector; for the EEG signals of each channel, train a classifier based on the feature vector using the training set; and select an optimal channel based on the feature vector using the verification set.

[0010] The detection module is configured to: acquire EEG signals of the optimal channel, and call the data processing module for preprocessing to obtain multiple segments; call the signal expansion module to expand each segment; call the deterministic learning module and the feature extraction module to perform deterministic learning feature extraction using a neural network to obtain a feature vector; and put the feature vector into the trained classifier for epilepsy detection.

[0011] Further, the preprocessing comprises: after band-pass filtering the EEG signals, segmenting the EEG signals into a plurality of segments.

[0012] Further, the signal expansion module is configured to: for each segment, observe a high-order state thereof using a high-gain observer, and expand the single-channel EEG signal segment into two-dimensional data by combining the original state and the high-order state.

[0013] Further, after expanding each segment, a state trajectory of each segment is a curve on a two-dimensional plane.

[0014] Further, the deterministic learning module is configured to: perform deterministic learning modeling on the expanded segment, and obtain a constant radial basis neural network storing system dynamics knowledge along the state trajectory, wherein the system dynamics knowledge is stored in a neighborhood of the state trajectory.

[0015] Further, all activated neurons are contained in the neighborhood, and non-activated neurons are outside the neighborhood.

[0016] Further, the feature extraction module is configured to: extract nonlinear dynamic features at a center of a neuron of the neural network as the feature vector.

[0017] Further, the optimal channel is selected by using the validation set.

[0018] In a second aspect, the present application further provides an electronic device comprising:

[0019] a memory for non-transitorily storing computer readable instructions; and

[0020] a processor for running the computer readable instructions,

[0021] wherein the computer readable instructions, when run by the processor, perform the following steps:

[0022] obtaining electroencephalogram signals of the optimal channel and calling a data processing module for preprocessing to obtain a plurality of segments;

[0023] calling a signal expansion module to expand each segment;

[0024] calling a determination learning module and a feature extraction module to use a neural network for determination learning and feature extraction to obtain a feature vector;

[0025] putting the feature vector into a trained classifier for epilepsy detection.

[0026] wherein the training of the classifier and the obtaining of the optimal channel comprise: obtaining electroencephalogram signals of a plurality of channels; for the electroencephalogram signals of each channel, calling a data processing module for preprocessing to obtain a plurality of segments and dividing them into a training set and a validation set; for each segment, sequentially calling a signal expansion module, a determination learning module and a feature extraction module to obtain a feature vector; for the electroencephalogram signals of each channel, training a classifier based on the feature vector using the training set; and selecting the optimal channel based on the feature vector using the validation set.

[0027] In a third aspect, the present application further provides a storage medium for non-transitorily storing computer readable instructions, wherein when the non-transitory computer readable instructions are executed by a computer, the following steps are performed:

[0028] obtaining electroencephalogram signals of the optimal channel and calling a data processing module for preprocessing to obtain a plurality of segments;

[0029] calling a signal expansion module to expand each segment;

[0030] calling a determination learning module and a feature extraction module to use a neural network for determination learning and feature extraction to obtain a feature vector;

[0031] putting the feature vector into a trained classifier for epilepsy detection.

[0032] The training of the classifier and the acquisition of the optimal channel comprise: acquiring electroencephalogram signals of multiple channels; for the electroencephalogram signals of each channel, calling a data processing module for preprocessing to obtain multiple segments, and dividing the segments into a training set and a verification set; for each segment, sequentially calling a signal expansion module, a determination learning module and a feature extraction module to obtain a feature vector; for the electroencephalogram signals of each channel, training a classifier based on the feature vector and using the training set; and selecting an optimal channel based on the feature vector and using the verification set.

[0033] Compared with the prior art, the brain electrical epilepsy detection system based on determination learning feature extraction has the following beneficial effects:

[0034] The brain electrical epilepsy detection system based on determination learning feature extraction can extract nonlinear dynamic characteristics of single-channel short-time interval electroencephalogram signals, and realize accurate and rapid epilepsy detection.

[0035] The brain electrical epilepsy detection system based on determination learning feature extraction extracts nonlinear dynamic characteristics of electroencephalogram signals based on determination learning, the extracted features aim to improve the ability to distinguish epilepsy and normal brain state, and can detect epilepsy based on single-channel electroencephalogram, and only extract one feature, thereby simplifying the detection process.

[0036] The advantages of the additional aspects of the present application will be partially given in the following description, or be understood through the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0037] The drawings accompanying the specification integrated into part of the present application serve to provide further understanding of the present application, the illustrative embodiments of the present application and the description thereof serve to explain the present application, and do not constitute an improper limitation on the present application.

[0038] Figure 1 A data flow diagram of the brain electrical epilepsy detection system based on determination learning feature extraction of the first embodiment;

[0039] Figure 2 An electroencephalogram signal diagram of a normal state of the first embodiment;

[0040] Figure 3 An electroencephalogram signal diagram of an epilepsy state of the first embodiment;

[0041] Figure 4 A neighborhood visualization model diagram of the first embodiment;

[0042] Figure 5 A feature distribution diagram of an epilepsy state of the first embodiment;

[0043] Figure 6 A feature distribution diagram of a normal state of the first embodiment. DETAILED DESCRIPTION

[0044] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0045] In the case of no conflict, the embodiments in the application and the features in the embodiments can be combined with each other.

[0046] All data acquisition of this embodiment is based on legal regulations and user consent, and legal application of data.

[0047] Term explanation:

[0048] Deterministic learning theory: a dynamic system identification and dynamic pattern recognition mechanism, which can learn the internal dynamics of the system and store the learned knowledge as the constant weight of the radial basis function (RBF) network, and the dynamic system identification and dynamic pattern recognition process is based on the RBF network with these constant weights. Deterministic learning has made many achievements in dynamic pattern recognition, and has been used for early detection of myocardial ischemia based on myocardial kinetic graph, which can identify different system dynamics of normal and myocardial ischemia according to system state trajectory.

[0049] Embodiment one

[0050] Dynamic system is increasingly used in artificial intelligence to model time series data, providing valuable feature information for identification and classification. Human brain is a complex nonlinear dynamic system, and the generated electroencephalogram signal is a chaotic time series. Therefore, nonlinear dynamic features can be detected from the source of electroencephalogram signal to detect seizures, and detection of epilepsy from the perspective of system dynamics is a promising research.

[0051] The embodiment provides an electroencephalogram epilepsy detection system based on deterministic learning feature extraction, which extracts nonlinear dynamic features of electroencephalogram signal based on deterministic learning theory, and then completes epilepsy detection by using support vector machine.

[0052] The electroencephalogram epilepsy detection system based on deterministic learning feature extraction provided by the embodiment extracts the internal dynamic features of electroencephalogram signal, compared with other feature-based epilepsy detection systems, and detects epilepsy from the root of electroencephalogram generation, has higher resolution for electroencephalogram signal with short time length, realizes fast and accurate epilepsy detection; and only one feature is extracted, without other subsequent processing process, realizing simple and efficient epilepsy detection.

[0053] As Figure 1As shown, an electroencephalogram epilepsy detection system based on deterministic learning feature extraction includes a data processing module, a signal expansion module, a deterministic learning module, a feature extraction module, a training and verification module, and a detection module.

[0054] The data processing module is configured to preprocess the electroencephalogram signals.

[0055] As shown, a normal state electroencephalogram signal graph is as shown in Figure 2 As shown, an electroencephalogram signal graph in an epilepsy state is as shown in Figure 3

[0056] The preprocessing includes band-pass filtering the electroencephalogram signals at 0.1-50 Hz, and then segmenting them into single-channel time intervals of 1 second.

[0057] The signal expansion module is configured to use a high-gain observer to observe the high-order state of all electroencephalogram signal samples, and expand the single-channel electroencephalogram signals into two dimensions.

[0058] The high-gain observer can estimate the unknown system dynamics that are difficult to measure by using known system states, thereby obtaining high-order state information of the system. The original system state and the high-order system state are combined to expand the single-channel electroencephalogram signals into two-dimensional data.

[0059] The deterministic learning module is configured to perform deterministic learning modeling on the expanded electroencephalogram signal samples, whose state trajectory is a curve on a two-dimensional plane. Along this state trajectory, a constant radial basis neural network that stores system dynamics knowledge can be obtained. These system dynamics knowledge are stored in the neighborhood near the state trajectory.

[0060] The deterministic learning modeling process is as follows: for the high-gain expanded electroencephalogram signal samples, the corresponding electroencephalogram dynamics system dynamics is

[0061]

[0062] Where x represents the system state, p represents the system parameter vector, and F(x; p) is an unknown system dynamics function. The state trajectory of this dynamic system is ζ Deterministic learning can use the radial basis neural network along the state trajectory to model F(x; p):

[0063]

[0064] Where w is the radial basis neural network weight, s is the Gaussian function, M is the number of neurons, and ∈ is a small estimation error. ​The average neural network weight value in a period of time after the weight value converges, and the calculation method is t a to t b The average value of the weight value in a period of time:

[0065]

[0066] Wherein, is the estimated value of the neural network weight W. In this way, the determination learning modeling process of estimating F(x; p) is completed.

[0067] After the modeling is completed, whether the electroencephalogram sample is in the epilepsy state is distinguished according to the different constant neural network distributions; the system state function of the electroencephalogram sample in the epilepsy state is expressed as F E (x; p), and the system state function of the electroencephalogram sample in the normal state is expressed as F N (x; p), and the two kinds of system states are determined and modeled:

[0068]

[0069]

[0070] Wherein, the subscript E represents the epilepsy state, and the subscript N represents the normal state.

[0071] The constant neural network of the electroencephalogram sample in the epilepsy state has a similar distribution with the constant neural network , and the constant neural network of the electroencephalogram sample in the normal state has a similar distribution with the constant neural network , therefore, the state of the electroencephalogram sample is distinguished according to the constant neural network .

[0072] Wherein, the constant neural network is a local accurate estimation of the original system dynamics function F(x; p), and the dynamics knowledge of the system is stored in the neighborhood along the system state trajectory φ ζ ; assuming that the maximum accurate estimation distance is d>0, then the neighborhood can be described as:

[0073]

[0074] All activated neurons are contained in the neighborhood, and the unactivated neurons are outside the neighborhood, and the weight value thereof is 0. The visualization model diagram of the neighborhood is shown in Figure 4 .

[0075] ​The feature extraction module is configured to extract the system dynamics at the neuron centers of the constant neural network storing the system dynamics knowledge as a feature vector, instead of the original EEG signal segment.

[0076] wherein the system dynamics at the neuron centers are extracted as features, and the calculation of the system dynamics at the neuron centers is as follows:

[0077]

[0078] wherein Z center represents all neuron centers, the neurons in the neighborhood are activated, and the weights of the neurons are not 0, and the neurons outside the neighborhood are not activated, and the weights of the neurons are 0, and the position of the corresponding EEG signal sample in the neural network can be determined according to whether the extracted feature value is 0 or not, and the discrimination degree of the samples in different states is increased, so the system dynamics at all activated and non-activated neuron centers are extracted as feature vectors. In this way, the nonlinear dynamics feature extraction based on deterministic learning is completed. The extracted nonlinear dynamics features are as shown in Figure 5 , Figure 6 , Figure 5 is a feature distribution graph of the epilepsy state, Figure 6 is a feature distribution graph of the normal state.

[0079] The training and verification module is configured to: after the data processing module pre-processes the EEG signal, label each segment in the EEG signal as an epilepsy seizure or non-seizure according to whether the epilepsy occurs or not, take each segment as a sample, divide the segments into a training set and a verification set, and use the training set and the verification set for training the classifier and selecting the optimal channel respectively; call the signal expansion module, the deterministic learning module and the feature extraction module to complete the feature extraction of the EEG signal segments in all the training set and the verification set, and convert the EEG signal segments into feature vectors; train the support vector machine classifier using the samples in the training set, and select the optimal channel using the samples in the verification set.

[0080] wherein the deterministic learning feature extraction is performed on all EEG signal samples, and the extracted feature vector x i represents the original sample, the label of the sample is y i , if the sample is in the epilepsy state, y i = 1, and if the sample is in the normal state, y i = -1.

[0081] wherein the support vector machine classifier is trained using the samples in the training set:

[0082]

[0083] subject toi (w·x i +)≥1-ξ i

[0084] ξ i ≥0,=1,…,

[0085] Where, ξ i Let b represent the error, n represent the bias, w represent the number of samples, and C represent the weight vector. The problem can be transformed into its dual problem as follows:

[0086]

[0087]

[0088] Where, α i is the Lagrange multiplier.

[0089] The above problem can be solved by using a quadratic optimization method to obtain the optimal value of α.

[0090] The method for selecting the optimal channel using the validation set is as follows: Single-channel EEG signal samples from the training set are extracted using predetermined learning features, transformed into feature vectors, and a support vector machine (SVM) classifier is trained. In the EEG signal samples of the validation set, features are extracted using the same channel as those in the training set, transformed into feature vectors, and then the trained SVM is used for classification. The classification method is as follows:

[0091]

[0092] For a new input test sample x, if the output value is 1, it is classified as an epileptic state; if it is -1, it is classified as a normal state. The classification accuracy can be obtained by comparing it with the true label value. This verification process is repeated on all channels, and the channel with the highest classification accuracy is selected as the optimal channel. Epilepsy detection is performed using the EEG signal from this single channel.

[0093] The detection module is configured to: acquire test set EEG signals and call the data processing module to perform the same preprocessing process on the acquired EEG signals; call the signal expansion module to expand to two dimensions through a high-gain observer; then call the deterministic learning module and the feature extraction module to perform deterministic learning and extract features; finally, feed the signals into a trained support vector machine classifier for epilepsy detection.

[0094] The test process is: collecting the electroencephalogram of the test set, performing 0.1-50Hz band-pass filtering, then selecting the electroencephalogram of the optimal channel, and segmenting into electroencephalogram test samples with a time length of 1 second; the test sample is extended to two dimensions by high-gain observation, and then the same size radial basis neural network as the training set is used for determining learning feature extraction, the signal value of the test set sample is converted into a feature vector, and finally put into the support vector machine trained by the training set sample using the optimal channel for classification, if the output result is 1, the test set sample is in the epilepsy state, if the output result is-1, the sample is in the normal state.

[0095] The embodiment realizes a detection accuracy of 97.3%, a detection sensitivity of 98.57%, and a detection specificity of 95.05%. Under the computer configuration of Intel(R) Core(TM) i7-11700@2.50GHz CPU processor, 16GB RAM, when facing the electroencephalogram with a time length of 1 second, the detection time is only 0.35 seconds, which has the ability of rapid and accurate detection of epilepsy.

[0096] In the embodiment, deterministic learning is introduced to identify the difference in brain activity between epilepsy and normal state. The brain in the epilepsy state and the normal state has different system dynamics. Deterministic learning can approximate unknown system dynamics and store them as constant weight radial basis function networks. Therefore, the system dynamics knowledge obtained based on deterministic learning is used to detect seizures.

[0097] The embodiment proposes feature extraction based on deterministic learning, which is used to extract the nonlinear dynamic features of the electroencephalogram. The extracted features aim to improve the ability to distinguish epilepsy and normal brain state; the seizures can be detected based on single-channel electroencephalogram, only one feature is extracted, and the detection process is simplified.

[0098] Embodiment two

[0099] The embodiment also provides an electronic device, comprising: one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected with the memory, the one or more computer programs are stored in the memory, and when the electronic device is running, the processor executes the one or more computer programs stored in the memory, so that the electronic device executes the following steps:

[0100] Obtaining the electroencephalogram of the optimal channel, and calling the data processing module for preprocessing to obtain a plurality of segments;

[0101] Calling the signal expansion module to expand each segment;

[0102] The determination learning module and the feature extraction module are called to use a neural network to determine learning feature extraction, and a feature vector is obtained.

[0103] The feature vector is put into the trained classifier to perform epilepsy detection.

[0104] The training of the classifier and the acquisition of the optimal channel include: acquiring electroencephalogram signals of multiple channels; for the electroencephalogram signals of each channel, calling a data processing module to perform preprocessing to obtain multiple segments, and dividing the segments into a training set and a verification set; for each segment, sequentially calling a signal expansion module, a determination learning module and a feature extraction module to obtain a feature vector; for the electroencephalogram signals of each channel, training a classifier based on the feature vector using the training set; and selecting an optimal channel based on the feature vector using the verification set.

[0105] It should be understood that in the embodiments, the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), ready-to-program gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any other conventional processor.

[0106] The memory can include read-only memory and random access memory, and provide instructions and data to the processor, and a portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0107] In the implementation process, each step of the above method can be completed by integrated logic circuits of hardware in the processor or instructions in the form of software.

[0108] The method in the first embodiment can be directly embodied as hardware processor execution completion, or executed by a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the art such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and combines the hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0109] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software mode depends on the specific application of the technical solution and the design constraints. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0110] Embodiment three

[0111] The embodiment also provides a computer readable storage medium for storing computer instructions, when the computer instructions are executed by a processor, the following steps are performed:

[0112] The electroencephalogram signal of the optimal channel is acquired, and a data processing module is called to perform preprocessing to obtain a plurality of segments;

[0113] A signal expansion module is called to expand each segment;

[0114] A determination learning module and a feature extraction module are called, and neural network is used to perform determination learning feature extraction to obtain a feature vector;

[0115] The feature vector is put into a trained classifier to perform epilepsy detection.

[0116] The training of the classifier and the acquisition of the optimal channel include: acquiring electroencephalogram signals of a plurality of channels; for the electroencephalogram signal of each channel, a data processing module is called to perform preprocessing to obtain a plurality of segments, and the segments are divided into a training set and a validation set; for each segment, a signal expansion module, a determination learning module and a feature extraction module are called in turn to obtain a feature vector; for the electroencephalogram signal of each channel, a classifier is trained based on the feature vector and the training set; and the optimal channel is selected based on the feature vector and the validation set.

[0117] The above only describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An electroencephalogram seizure detection system based on deterministic learning feature extraction, characterized in that, The method comprises the following steps: a training and verification module configured to obtain electroencephalogram signals of multiple channels; for the electroencephalogram signals of each channel, calling a data processing module for preprocessing to obtain multiple segments and dividing them into a training set and a verification set; for each segment, sequentially calling a signal expansion module, a learning determination module and a feature extraction module to obtain a feature vector; for the electroencephalogram signals of each channel, training a classifier based on the feature vector using the training set; based on the feature vector, selecting an optimal channel using the verification set; the preprocessing comprises: performing band-pass filtering of 0.1-50Hz on the electroencephalogram signals, and then segmenting them into single-channel segments with a time interval of 1 second; a detection module configured to obtain electroencephalogram signals of the optimal channel and call a data processing module for preprocessing to obtain multiple segments; call a signal expansion module to expand each segment; call a learning determination module and a feature extraction module to use a neural network for learning determination and feature extraction to obtain a feature vector; and put the feature vector into the trained classifier for epilepsy detection; the signal expansion module is configured to: for each segment, use a high-gain observer to observe its high-order state, and expand the single-channel electroencephalogram signal segment into two-dimensional data by combining the original state with the high-order state; after expanding each segment, the state trajectory of each segment is a curve on a two-dimensional plane; the learning determination module is configured to: perform learning determination modeling on the expanded segment, obtain a constant radial basis neural network storing system dynamics knowledge along the state trajectory, and the system dynamics knowledge is stored in a neighborhood of the state trajectory; all activated neurons are contained in the neighborhood, and non-activated neurons are outside the neighborhood.

2. The electroencephalogram seizure detection system based on deterministic learning feature extraction as claimed in claim 1, wherein, the feature extraction module is configured to extract nonlinear dynamic features at the center of the neural network neurons as the feature vector.

3. The electroencephalogram seizure detection system based on deterministic learning feature extraction as claimed in claim 1, wherein, When selecting the optimal channel using the verification set, the channel with the highest classification accuracy is selected as the optimal channel.

4. An electronic device, characterized in that it comprises: a memory for non-transitory storage of computer readable instructions; and a processor for running the computer readable instructions, wherein the computer readable instructions, when executed by the processor, perform the following steps: obtain electroencephalogram signals of the optimal channel and call a data processing module for preprocessing to obtain multiple segments; call a signal expansion module to expand each segment; call a learning determination module and a feature extraction module to use a neural network for learning determination and feature extraction to obtain a feature vector; put the feature vector into the trained classifier for epilepsy detection. The training of the classifier and the acquisition of the optimal channel comprise: acquiring electroencephalogram signals of multiple channels; for the electroencephalogram signals of each channel, calling a data processing module to perform preprocessing to obtain multiple segments, and dividing the segments into a training set and a verification set; for each segment, sequentially calling a signal expansion module, a deterministic learning module and a feature extraction module to obtain a feature vector; for the electroencephalogram signals of each channel, training a classifier based on the feature vector and using the training set; and selecting an optimal channel based on the feature vector and using the verification set; the preprocessing comprises: performing band-pass filtering on the electroencephalogram signals at 0.1-50 Hz, and then segmenting the signals into segments of single channels with a time interval of 1 second; The signal expansion module is configured to: for each segment, observe a high-order state thereof using a high-gain observer, and expand the segment of the single-channel electroencephalogram signal into two-dimensional data by combining the original state with the high-order state; after expansion of each segment, a state trajectory of each segment is a curve on a two-dimensional plane; The deterministic learning module is configured to: perform deterministic learning modeling on the expanded segment, and obtain a constant radial basis neural network storing system dynamics knowledge along the state trajectory, the system dynamics knowledge being stored in a neighborhood of the state trajectory; all activated neurons are contained in the neighborhood, and non-activated neurons are outside the neighborhood.

5. A storage medium characterized by being non-transitory The computer-readable instructions are stored in the computer-readable storage medium, and when the computer-readable instructions are executed by a computer, the following steps are performed: acquiring electroencephalogram signals of an optimal channel, and calling a data processing module to perform preprocessing to obtain multiple segments; calling a signal expansion module to expand each segment; calling a deterministic learning module and a feature extraction module to perform deterministic learning feature extraction using a neural network to obtain a feature vector; putting the feature vector into a trained classifier to perform epilepsy detection; The training of the classifier and the acquisition of the optimal channel comprise: acquiring electroencephalogram signals of multiple channels; for the electroencephalogram signals of each channel, calling a data processing module to perform preprocessing to obtain multiple segments, and dividing the segments into a training set and a verification set; for each segment, sequentially calling a signal expansion module, a deterministic learning module and a feature extraction module to obtain a feature vector; for the electroencephalogram signals of each channel, training a classifier based on the feature vector and using the training set; and selecting an optimal channel based on the feature vector and using the verification set; the preprocessing comprises: performing band-pass filtering on the electroencephalogram signals at 0.1-50 Hz, and then segmenting the signals into segments of single channels with a time interval of 1 second; The signal expansion module is configured to: for each segment, observe a high-order state thereof using a high-gain observer, and expand the segment of the single-channel electroencephalogram signal into two-dimensional data; after expansion of each segment, a state trajectory of each segment is a curve on a two-dimensional plane; The determining learning module is configured to perform determining learning modeling on the extended segment, and obtain a constant radial basis neural network storing system dynamics knowledge along the state trajectory, the system dynamics knowledge being stored in a neighborhood of the state trajectory; all activated neurons are contained in the neighborhood, and non-activated neurons are outside the neighborhood.

Images