Epileptic seizure detection method and epileptic seizure detection device based on EEG signal covariance matrix
By calculating the covariance matrix of multi-channel EEG signals as input to deep neural networks, the problem of high computational complexity of deep learning epilepsy detection methods in the prior art is solved, and the calculation is greatly reduced without reducing the detection accuracy The effect of complexity.
Patent Information
- Application Number
- CN202411601215.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-11
- Publication Date
- 2025-02-18
AI Technical Summary
Existing deep learning-based seizure detection methods have challenges in computing complexity, making it difficult to implement detection on wearable devices, and reducing the size of the model to reduce computational complexity can affect detection accuracy.
By calculating the covariance matrix of multi-channel EEG signals, the covariance matrix is used as the input of the deep neural network to reduce the input signal dimension, thereby reducing the computational complexity of the deep neural network.
While ensuring the accuracy of epilepsy detection, it greatly reduces the inference calculation complexity of deep learning in epilepsy detection, improves the detection performance in small sample scenarios, and enhances the model's adaptability to different signal lengths.
Smart Images
Figure CN119453934A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of biomedicine, and in particular to an epileptic seizure detection method based on an EEG signal covariance matrix and a detection device thereof. Background Art
[0002] Epilepsy is a common neurological disease that affects more than 65 million people worldwide. The main feature of epilepsy is the repeated and unpredictable interruption of normal brain function, namely epileptic seizures. Although epilepsy as a chronic disease is not fatal in itself, patients may lose consciousness during an attack, leading to serious consequences that may even be life-threatening. Therefore, timely detection of epileptic seizures is crucial to take preventive measures and prevent patients from suffering serious consequences.
[0003] Electroencephalogram (EEG) signals are often used to diagnose epileptic seizures. Traditional EEG-based epileptic seizure detection methods usually include two stages: feature extraction and classification, among which feature extraction has a particularly important impact on detection performance. These features mainly include time domain features, frequency domain features, and time-frequency domain features, and their design is highly dependent on expert experience. The quality of feature design is directly related to the effectiveness of epilepsy detection. However, due to differences in EEG signal features between different patients, and even the EEG signals of the same patient in different states may change significantly, these features still have limitations in dealing with EEG signal changes.
[0004] In recent years, with the development of deep learning technology, deep neural networks have shown unique advantages in epilepsy detection. Compared with manually designed features, deep learning methods can automatically learn more unique and robust features from raw EEG data, thereby achieving performance superior to traditional machine learning methods. Therefore, it is particularly necessary to build a deep epilepsy detection model that can learn features directly from raw EEG data, which can avoid the limitations of manual feature engineering.
[0005] Although deep learning has achieved satisfactory classification performance in epileptic seizure detection, the high computational requirements of deep neural networks remain a major challenge in practical applications, especially when implementing epilepsy detection on wearable devices. To reduce the computational burden, existing methods mainly reduce the computational complexity by reducing the model size. Although these methods reduce the computational pressure, the reduction in the size of the neural network also leads to a decrease in the accuracy of epilepsy detection. Therefore, how to reduce the computational complexity of epilepsy detection methods based on deep learning without affecting the accuracy of epilepsy detection remains a challenge that needs to be solved. Summary of the invention
[0006] Based on the above analysis, the present invention aims to provide a method for detecting epileptic seizures based on deep learning of the covariance matrix of EEG signals, which reduces the dimension of the input signals through the covariance matrix, thereby reducing the computational complexity of the deep neural network.
[0007] An epileptic seizure detection method based on the covariance matrix of EEG signals disclosed by the present invention includes the following steps:
[0008] Step 1: Collect and obtain a multi-channel EEG signal dataset
[0009] Collect multi-channel EEG signals during epileptic seizures and multi-channel EEG signals without epileptic seizures respectively. The number of channels is C, and each signal is a discretized digital signal sequence with a length of N, constituting a multi-channel EEG signal dataset A;
[0010] Step 2: Expand the signal dataset based on channel swapping and noise injection
[0011] For each signal in dataset A, randomly select two channels, swap the signals of the two channels to obtain a new multi-channel signal, add noise to the signals of each channel to obtain a noisy multi-channel signal, and add it to dataset A. Repeat this process multiple times to obtain an expanded EEG signal dataset B;
[0012] Step 3: Calculate the covariance matrix of the multi-channel EEG signals
[0013] Randomly select a signal from the expanded dataset B, randomly intercept a signal segment with a length of M from this signal, where 1 < M ≤ N, calculate the covariance matrix of this signal segment, repeat this process multiple times, and the length M of each intercepted signal segment is random and satisfies 1 < M ≤ N, thereby obtaining a new covariance matrix dataset C;
[0014] Step 4: Construct an epileptic seizure detection training set and a validation set
[0015] Randomly divide the covariance matrix dataset C into 2 sets according to a certain ratio as the training set and the validation set for deep learning training. Each sample data in the training set and the validation set includes the covariance matrix and the corresponding sample label. The sample label is determined according to whether the EEG signal corresponding to the acquisition comes from the epileptic seizure period. If it comes from the EEG signal collected during epileptic seizures, the label is "epilepsy", otherwise it is "normal";
[0016] Step 5: Construct and train an epileptic seizure detection convolutional neural network
[0017] Construct a convolutional neural network model for epilepsy detection. The size of the convolutional kernel in the first layer of the convolutional neural network is P×P, where 1 < P ≤ C. The number of hidden nodes in the final classification layer is 2, and the confidence level is output through softmax. Use the training set and the validation set to train the convolutional neural network to obtain a trained epilepsy detection convolutional neural network model;
[0018] Step 6: Obtain the multi-channel EEG signal to be detected
[0019] Collect and obtain the EEG signal to be detected;
[0020] Step 7: Calculate the covariance matrix of the EEG signal to be detected
[0021] Calculate the covariance matrix of the signal to be detected;
[0022] Step 8: Perform epilepsy judgment based on the covariance matrix of the signal to be detected
[0023] Input the calculated covariance matrix of the signal to be detected into the trained epilepsy detection convolutional neural network model, calculate the output confidence level, and select the category with a high confidence level as the epilepsy detection result.
[0024] An epilepsy seizure detection device for an epilepsy seizure detection method based on the covariance matrix of EEG signals according to any one of claims 1 to 3, characterized in that it includes the following modules: an EEG signal acquisition module for collecting and obtaining a multi-channel EEG signal to be detected; a covariance matrix calculation module for calculating the covariance matrix of the EEG signal to be detected; an epilepsy intelligent detection module for inputting the calculated covariance matrix of the signal to be detected into the trained epilepsy detection convolutional neural network model, calculating the output confidence level, and taking the category with a high confidence level as the epilepsy detection result.
[0025] Compared with the prior art, the present invention can at least achieve the following beneficial effects:
[0026] 1. By calculating the covariance matrix of the multi-channel EEG signal and using the covariance matrix as the input of the deep neural network, the present invention greatly reduces the inference calculation complexity of epilepsy detection deep learning while ensuring the accuracy of epilepsy detection.
[0027] 2. By expanding the signal data set through channel exchange and noise injection, the present invention reduces the acquisition burden of labeled epilepsy samples for deep learning methods and improves the epilepsy detection performance in small sample scenarios.
[0028] 3. By randomly intercepting the multi-channel EEG signal in length to calculate the covariance matrix to construct the training set, the present invention improves the adaptability of the epilepsy detection model to different signal lengths during actual deployment and use.
[0029] Other features and advantages of the present invention will be described in the following description, and some advantages may become apparent from the description, or may be understood by practicing the present invention. The purpose and other advantages of the present invention may be realized and obtained through the contents particularly pointed out in the description and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.
[0031] Figure 1 The present invention is a flow chart of an epileptic seizure detection method based on the EEG signal covariance matrix. DETAILED DESCRIPTION
[0032] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.
[0033] like Figure 1 As shown, the present invention discloses an epileptic seizure detection method based on the EEG signal covariance matrix, comprising the following steps:
[0034] Step 1: Acquire a multi-channel EEG signal dataset
[0035] Multi-channel EEG signals during epileptic seizures and multi-channel EEG signals without epileptic seizures are collected respectively, the number of channels is C, and each signal is a discretized digital signal sequence with a length of N, forming a multi-channel EEG signal data set A.
[0036] Step 2: Augment the signal dataset based on channel swapping and noise injection
[0037] For each signal in data set A, two channels are randomly selected, the signals of the two channels are exchanged to obtain a new multi-channel signal, noise is added to the signal of each channel to obtain the noisy multi-channel signal, and added to data set A. The process is repeated multiple times to obtain the expanded EEG signal data set B; wherein, the noise injection process is to first generate a Gaussian distribution sequence w(n) with a mean of 0 and a variance of 1, n = 0, 1, ..., N-1, and then multiply it by the amplitude a to obtain aw(n), where a is a random number greater than 0 and less than 1, and finally add aw(n) to a channel of the signal.
[0038] Step 3: Calculate the covariance matrix of the multi-channel EEG signal
[0039] Randomly select a signal from the expanded dataset B, randomly intercept a signal segment of length M from this signal, where 1 < M ≤ N, calculate the covariance matrix of this signal segment, and repeat this process multiple times. Each time, the length M of the intercepted signal segment is random and satisfies 1 < M ≤ N, thus obtaining a new covariance matrix dataset C. Among them, the element c in the i-th row and j-th column of the EEG signal covariance matrix ij The calculation process is as follows:
[0040]
[0041] where x i (n) and x j (n) are the EEG signals of the i-th channel and the j-th channel respectively, and n = 0, 1,..., M - 1.
[0042] Step 4: Construct the epilepsy detection training set and validation set
[0043] Randomly divide the covariance matrix dataset C into two sets according to a certain ratio as the training set and validation set for deep learning training. Each sample data in the training set and validation set includes the covariance matrix and the corresponding sample label. The sample label is determined according to whether the corresponding EEG signal during acquisition comes from the epileptic seizure period. If it comes from the EEG signal collected during an epileptic seizure, the label is "epilepsy", otherwise it is "normal".
[0044] Step 5: Construct and train the epilepsy detection convolutional neural network
[0045] Construct a convolutional neural network model for epilepsy detection. The size of the convolutional kernel in the first layer of the convolutional neural network is P×P, where 1 < P ≤ C, and the number of hidden nodes in the last classification layer is 2. After softmax, the confidence level is output. Use the training set and validation set to train the convolutional neural network to obtain the trained epilepsy detection convolutional neural network model.
[0046] Step 6: Obtain the multi-channel EEG signal to be detected
[0047] Collect and obtain the EEG signal to be detected.
[0048] Step 7: Calculate the covariance matrix of the EEG signal to be detected
[0049] Calculate the covariance matrix of the signal to be detected. The element c in the i-th row and j-th column of the EEG signal covariance matrix ij The calculation process is as follows:
[0050]
[0051] where x i (n) and x j(n) are the EEG signals of the i-th channel and the j-th channel respectively, n=0, 1,…, L-1, L is the signal length.
[0052] Step 8: Make epilepsy judgment based on the covariance matrix of the signal to be detected
[0053] The calculated covariance matrix of the signal to be detected is input into the trained epilepsy detection convolutional neural network model, the output confidence is calculated, and the category with high confidence is selected as the epilepsy detection result.
[0054] The present invention also provides an epileptic seizure detection device based on the epileptic seizure detection method of the EEG signal covariance matrix, and the device comprises the following modules:
[0055] EEG signal acquisition module, collecting and acquiring multi-channel EEG signals to be detected;
[0056] A covariance matrix calculation module is used to calculate the covariance matrix of the EEG signal to be detected;
[0057] The epilepsy intelligent detection module inputs the calculated covariance matrix of the signal to be detected into the trained epilepsy detection convolutional neural network model, calculates the output confidence, and takes the category with high confidence as the epilepsy detection result.
[0058] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.
[0059] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
Claims
1. An epileptic seizure detection method based on the EEG signal covariance matrix, characterized in that: It includes the following steps: Step 1: Collect and obtain a multi-channel EEG signal dataset Collect multi-channel EEG signals during epileptic seizures and multi-channel EEG signals without epileptic seizures respectively. The number of channels is C, and each signal is a digital signal sequence after discretization, with a length of N, forming a multi-channel EEG signal dataset A; Step 2: Expand the signal dataset based on channel swapping and noise injection For each signal in dataset A, randomly select two channels, swap the signals of the two channels to obtain a new multi-channel signal, add noise to the signals of each channel to obtain a multi-channel signal with added noise, and add it to dataset A. Repeat this process multiple times to obtain an expanded EEG signal dataset B; Step 3: Calculate the covariance matrix of the multi-channel EEG signal Randomly select a signal from the expanded dataset B, randomly intercept a signal segment with a length of M from this signal, where 1 < M ≤ N, calculate the covariance matrix of this signal segment, repeat this process multiple times, and the length M of the signal segment intercepted each time is random and satisfies 1 < M ≤ N, thus obtaining a new covariance matrix dataset C; Step 4: Construct an epilepsy detection training set and a validation set Randomly divide the covariance matrix dataset C into 2 sets according to a certain ratio as the training set and the validation set for deep learning training. Each sample data in the training set and the validation set includes the covariance matrix and the corresponding sample label. The sample label is determined according to whether the EEG signal corresponding to the acquisition comes from the epileptic onset period. If it comes from the EEG signal collected during epileptic seizures, the label is "epilepsy", otherwise it is "normal"; Step 5: Construct and train an epilepsy detection convolutional neural network Construct a convolutional neural network model for epilepsy detection. The size of the convolutional kernel in the first layer of the convolutional neural network is P×P, where 1 < P ≤ C, and the number of hidden nodes in the last classification layer is 2. After softmax, the confidence is output. Use the training set and the validation set to train the convolutional neural network to obtain a trained epilepsy detection convolutional neural network model; Step 6: Obtain the multi-channel EEG signal to be detected Collect and obtain the EEG signal to be detected; Step 7: Calculate the covariance matrix of the EEG signal to be detected Calculate the covariance matrix of the signal to be detected; Step 8: Make an epilepsy judgment based on the covariance matrix of the signal to be detected Input the calculated covariance matrix of the signal to be detected into the trained epilepsy detection convolutional neural network model, calculate the output confidence, and select the category with a high confidence as the epilepsy detection result.
2. The epileptic seizure detection method based on the EEG signal covariance matrix according to claim 1, characterized in that: In the noise injection process in Step 2, first generate a Gaussian distribution sequence w(n) with a mean of 0 and a variance of 1, where n = 0, 1,..., N - 1, then multiply it by an amplitude a to get aw(n), where a is a random number greater than 0 and less than 1, and finally add aw(n) to a certain channel of the signal.
3. The epileptic seizure detection method based on the EEG signal covariance matrix according to claim 1, characterized in that: The element c in the i-th row and j-th column of the EEG signal covariance matrix in steps 3 and 7 is ij The calculation process is: where x i (n) and x j (n) are the EEG signals of the i-th channel and the j-th channel respectively, n=0, 1, ..., L-1, L is the signal length.
4. An epileptic seizure detection device based on the epileptic seizure detection method based on the EEG signal covariance matrix according to any one of claims 1 to 3, characterized in that: It includes the following modules: An EEG signal acquisition module that collects and obtains the multi-channel EEG signal to be detected; a covariance matrix calculation module that calculates the covariance matrix of the EEG signal to be detected; The epilepsy intelligent detection module inputs the calculated covariance matrix of the signal to be detected into the trained epilepsy detection convolutional neural network model, calculates the output confidence, and takes the category with high confidence as the epilepsy detection result.