An electroencephalogram monitoring system and method for neurology

Abstract

This invention provides an electroencephalogram (EEG) monitoring system and method for neurology, comprising: an EEG acquisition module configured to acquire EEG signals and preprocess the EEG signals to acquire a first EEG signal; an EEG processing module configured to remove artifacts from the first EEG signal to acquire a second EEG signal; a feature extraction module configured to acquire a feature vector of the second EEG signal; and an EEG recognition module configured to output a recognition result based on the feature vector. This invention accurately removes electrooculogram (EOG) artifacts from the first EEG signal and obtains a clean second EEG signal through independent component analysis and DWT decomposition. This invention improves the system's stability and the accuracy of the recognition results by employing convolutional neural networks with multiple different activation functions to recognize the feature vector composed of sample entropy.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) monitoring technology, and more particularly to an EEG monitoring system and method for use in neurology. Background Technology

[0002] Electroencephalography (EEG) signals record the electrical activity of brain neurons and contain a wealth of physiological information. Long-term monitoring and analysis of EEG signals are of great significance for neurological research. To improve the accuracy and reliability of signal classification during neurological EEG monitoring, the removal of electrooculography (EOG) artifacts from EEG signals is crucial. However, existing neurological EEG monitoring systems have the following shortcomings: they do not accurately remove EOG artifacts from EEG signals, resulting in the inability to acquire clean EEG signals; convolutional neural networks use only a single activation function, leading to inaccurate recognition results and insufficient system stability. Summary of the Invention

[0003] Purpose of the invention: To address the problems existing in the prior art, the present invention provides an electroencephalogram (EEG) monitoring system and method for use in neurology.

[0004] Technical solution: On one hand, the present invention provides an electroencephalogram (EEG) monitoring system for neurology, the system comprising:

[0005] An EEG acquisition module is configured to acquire EEG signals and preprocess the EEG signals to acquire a first EEG signal.

[0006] An EEG processing module, configured to remove artifacts from the first EEG signal and acquire a second EEG signal;

[0007] A feature extraction module, configured to acquire feature vectors of the second EEG signal;

[0008] An EEG recognition module is configured to output a recognition result based on the feature vector.

[0009] Furthermore, the specific process of acquiring the second EEG signal includes:

[0010] Independent component analysis was performed on the first EEG signal;

[0011] The independent components containing artifacts are determined based on the detection threshold;

[0012] Perform DWT decomposition on independent components containing artifacts;

[0013] The artifact components are determined based on the boundary values ​​of the low-frequency approximation components and high-frequency detail components of the DWT decomposition.

[0014] The DWT decomposition results after removing artifact components are reconstructed.

[0015] The reconstructed independent components are then reconstructed together with the remaining independent components that do not contain artifacts to obtain the second EEG signal.

[0016] Furthermore, the specific process of obtaining the feature vector of the second EEG signal includes:

[0017] The second EEG signal was decomposed into MEMD to obtain the IMF components after decomposition of each channel;

[0018] Calculate the correlation coefficient between each IMF component and the corresponding channel of the second EEG signal;

[0019] Calculate the sample entropy of IMF components whose correlation coefficient is greater than a preset threshold, and form a feature vector from the sample entropy.

[0020] Furthermore, the specific process of outputting the recognition result based on the feature vector includes:

[0021] The feature vector is input into a convolutional neural network for recognition. The convolutional neural network uses different activation functions to perform recognition and output the recognition results.

[0022] Furthermore, the activation function is Sigmoid, ReLU, or Leaky ReLU.

[0023] On the other hand, the present invention provides a method for electroencephalogram (EEG) monitoring in neurology, the method being applied to the aforementioned EEG monitoring system for neurology, the method comprising the following steps:

[0024] Step S1: Acquire EEG signals and preprocess the EEG signals to obtain a first EEG signal;

[0025] Step S2: Perform independent component analysis on the first EEG signal;

[0026] Step S3: Determine the independent components containing artifacts based on the detection threshold;

[0027] Step S4: Perform DWT decomposition on the independent components containing artifacts;

[0028] Step S5: Determine the artifact components based on the boundary values ​​of the low-frequency approximation components and high-frequency detail components of the DWT decomposition.

[0029] Step S6: Reconstruct the DWT decomposition results after removing artifact components;

[0030] Step S7: Reconstruct the reconstructed independent components with the remaining independent components without artifacts to obtain the second EEG signal.

[0031] Furthermore, the second EEG signal is decomposed into MEMD components to obtain the IMF components after decomposition of each channel.

[0032] Furthermore, the correlation coefficient between each IMF component and the corresponding channel of the second EEG signal is calculated.

[0033] Further, the sample entropy of IMF components with correlation coefficients greater than a preset threshold is calculated, and the sample entropy is used to form a feature vector; the feature vector is then input into a convolutional neural network for recognition.

[0034] Furthermore, the convolutional neural network uses different activation functions to perform recognition and output recognition results.

[0035] Technical Effects: Compared with existing technologies, this invention has the following advantages: This invention provides an electroencephalogram (EEG) monitoring system and method for neurology, comprising: an EEG acquisition module configured to acquire EEG signals and preprocess the EEG signals to acquire a first EEG signal; an EEG processing module configured to remove artifacts from the first EEG signal to acquire a second EEG signal; a feature extraction module configured to acquire a feature vector of the second EEG signal; and an EEG recognition module configured to output a recognition result based on the feature vector. This invention accurately removes oculomotor artifacts from the first EEG signal and obtains a clean second EEG signal through independent component analysis and DWT decomposition. This invention improves the system's stability and the accuracy of the recognition results by employing convolutional neural networks with multiple different activation functions to recognize the feature vector composed of sample entropy. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A structural diagram of an electroencephalogram (EEG) monitoring system for neurology provided in an embodiment of the present invention;

[0038] Figure 2 This is a flowchart of an electroencephalogram (EEG) monitoring method for neurology, provided as an embodiment of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely one unit of embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0040] As one embodiment of the present invention, such as Figure 1 As shown, this embodiment provides an electroencephalogram (EEG) monitoring system for neurology, the system comprising:

[0041] An EEG acquisition module is configured to acquire EEG signals and preprocess the EEG signals to acquire a first EEG signal.

[0042] An EEG processing module, configured to remove artifacts from the first EEG signal and acquire a second EEG signal;

[0043] A feature extraction module, configured to acquire feature vectors of the second EEG signal;

[0044] An EEG recognition module is configured to output a recognition result based on the feature vector.

[0045] Furthermore, the specific process of acquiring the second EEG signal includes:

[0046] Independent component analysis was performed on the first EEG signal;

[0047] The independent components containing artifacts are determined based on the detection threshold;

[0048] Perform DWT decomposition on independent components containing artifacts;

[0049] The artifact components are determined based on the boundary values ​​of the low-frequency approximation components and high-frequency detail components of the DWT decomposition.

[0050] The DWT decomposition results after removing artifact components are reconstructed.

[0051] The reconstructed independent components are then reconstructed together with the remaining independent components that do not contain artifacts to obtain the second EEG signal.

[0052] Furthermore, the specific process of obtaining the feature vector of the second EEG signal includes:

[0053] The second EEG signal was decomposed into MEMD to obtain the IMF components after decomposition of each channel;

[0054] Calculate the correlation coefficient between each IMF component and the corresponding channel of the second EEG signal;

[0055] Calculate the sample entropy of IMF components whose correlation coefficient is greater than a preset threshold, and form a feature vector from the sample entropy.

[0056] Furthermore, the specific process of outputting the recognition result based on the feature vector includes:

[0057] The feature vector is input into a convolutional neural network for recognition. The convolutional neural network uses different activation functions to perform recognition and output the recognition results.

[0058] Furthermore, the activation function is Sigmoid, ReLU, or Leaky ReLU.

[0059] As another embodiment of the present invention, such as Figure 2 As shown, this embodiment provides a method for electroencephalogram (EEG) monitoring in neurology, which is applied to the aforementioned EEG monitoring system for neurology. The method includes the following steps:

[0060] Step S1: Acquire EEG signals and preprocess the EEG signals to obtain a first EEG signal;

[0061] Specifically, the acquired EEG signals undergo preprocessing. The purpose of preprocessing is to reduce the number of estimated parameters, decrease the computational load, and accelerate the decomposition process. In this embodiment, the acquired EEG signals are first centered and then whitened.

[0062] Centralization processing involves removing the mean from the raw EEG signal. The specific steps are as follows:

[0063]

[0064] Where X represents the raw EEG signal collected, and E(X) represents the mean of the raw EEG signal. This is the centered observation signal. The specific steps for whitening the centered observation signal are as follows:

[0065]

[0066] Where V is the whitening matrix and Z represents the first EEG signal.

[0067] Step S2: Perform independent component analysis on the first EEG signal;

[0068] Specifically, the FastICA algorithm based on negative entropy is used to solve the first EEG signal.

[0069] The definition of negative entropy J(y) is:

[0070] J(y)=H(y gauss )-H(y) (3)

[0071] Among them, y gauss Let be a Gaussian distributed random variable with the same covariance matrix as y, and H(·) represent the differential entropy of the variable. Since the direct computation of negative entropy is very complex, in practical calculations, a simplified estimate based on expectation is generally used for negative entropy. In FastICA, the target negative entropy function is approximated as:

[0072] J(y) = [E{G(y)} - E{G(v)}] 2 (4)

[0073] Where v is a Gaussian random variable with zero mean and unit variance, and G can be any non-quadratic function.

[0074] Let the separation matrix to be solved be W. Substituting it into the objective negative entropy function, we can obtain:

[0075] J G (W)=[E{G(W T Z)}-E{G(v)}] 2 (5)

[0076] Step S3: Determine the independent components containing artifacts based on the detection threshold;

[0077] Specifically, the detection threshold is determined based on the standard deviation of the separation matrix W, and independent components containing artifacts are screened out.

[0078] Step S4: Perform DWT decomposition on the independent components containing artifacts;

[0079] Step S5: Determine the artifact components based on the boundary values ​​of the low-frequency approximation components and high-frequency detail components of the DWT decomposition.

[0080] Specifically, a two-layer DWT is used to decompose the independent component S containing artifacts to obtain the high-frequency detail part D1, the high-frequency detail part D2, and the low-frequency approximation part A2, i.e.:

[0081] S=D1+D2+A2 (6)

[0082] More specifically, if the boundary value of the low-frequency approximation of A2 exceeds the preset value, then the low-frequency approximation of A2 is determined to be an artifact component.

[0083] Step S6: Reconstruct the DWT decomposition results after removing artifact components;

[0084] Specifically, the reconstructed component S' consists of the D1 high-frequency detail and the D2 high-frequency detail, i.e.:

[0085] S'=D1+D2 (7)

[0086] Step S7: Reconstruct the reconstructed independent components with the remaining independent components without artifacts to obtain the second EEG signal.

[0087] Specifically, the reconstructed component S' is reconstructed together with the remaining independent components free of artifacts to obtain the second EEG signal. Since the frequency of electrooculography (EOG) artifacts is relatively fixed and low, DWT decomposition can separate the fixed low-frequency EOG artifacts from the original signal. Combining FastICA with DWT decomposition allows for rapid identification of EOG artifacts, thus obtaining a clean second EEG signal. Experimental comparative analysis shows that, compared to EMD and EWT decomposition methods, DWT decomposition more accurately identifies EOG artifacts.

[0088] Furthermore, the second EEG signal is decomposed into MEMD components to obtain the IMF components after decomposition of each channel.

[0089] Specifically, since EEG signals are multi-channel, we first perform multi-channel synchronous joint analysis on the second EEG signals from different channels to ensure that the IMF components are matched in quantity and scale while obtaining common patterns between different channels.

[0090] Furthermore, the correlation coefficient between each IMF component and the corresponding channel of the second EEG signal is calculated.

[0091] Specifically, the correlation coefficient C A,B The calculation method is as follows:

[0092]

[0093] Where B is the IMF component, A is the corresponding second EEG signal of the IMF channel, and N is the length of the signal sequence.

[0094] Further, the sample entropy of IMF components with correlation coefficients greater than a preset threshold is calculated, and the sample entropy is used to form a feature vector; the feature vector is then input into a convolutional neural network for recognition.

[0095] Specifically, the sample entropy of IMF components with a correlation coefficient greater than 0.1 is calculated, and the sample entropy is used to form a feature vector.

[0096] Furthermore, the convolutional neural network uses different activation functions to perform recognition and output recognition results.

[0097] Specifically, the convolutional neural network includes a first convolutional neural network, a second convolutional neural network, and a third convolutional neural network. The activation function of the first convolutional neural network is Sigmoid, the activation function of the second convolutional neural network is ReLU, and the activation function of the third convolutional neural network is Leaky ReLU. Sigmoid is prone to gradient vanishing. While ReLU avoids this problem, its characteristics with negative inputs may cause "neuron death." Leaky ReLU adds a small constant to the left of the ReLU function, thus mitigating the "neuron death" problem. Therefore, this embodiment uses convolutional neural networks with three activation functions to identify feature vectors. Based on the three identification results, a voting mechanism is used to select the identification result with the most votes and output it. Compared with convolutional neural networks using a single activation function, this embodiment uses convolutional neural networks with three different activation functions for identification, improving the system's stability and the accuracy of the identification results.

[0098] This invention designs an electroencephalogram (EEG) monitoring system and method for neurology, comprising: an EEG acquisition module configured to acquire EEG signals and preprocess the EEG signals to obtain a first EEG signal; an EEG processing module configured to remove artifacts from the first EEG signal to obtain a second EEG signal; a feature extraction module configured to obtain a feature vector of the second EEG signal; and an EEG recognition module configured to output a recognition result based on the feature vector. This invention accurately removes electrooculogram (EOG) artifacts from the first EEG signal and obtains a clean second EEG signal through independent component analysis and DWT decomposition. This invention improves the system's stability and the accuracy of the recognition results by employing convolutional neural networks with multiple different activation functions to recognize the feature vector composed of sample entropy.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An electroencephalogram (EEG) monitoring system for use in neurology, characterized in that, include: An EEG acquisition module is configured to acquire EEG signals and preprocess the EEG signals to acquire a first EEG signal. An EEG processing module, configured to remove artifacts from the first EEG signal and acquire a second EEG signal; A feature extraction module, configured to acquire feature vectors of the second EEG signal; An EEG recognition module, configured to output a recognition result based on the feature vector; The specific process of obtaining the second EEG signal includes: Independent component analysis was performed on the first EEG signal; The independent components containing artifacts are determined based on the detection threshold; Perform DWT decomposition on independent components containing artifacts; The artifact components are determined based on the boundary values ​​of the low-frequency approximation components and high-frequency detail components of the DWT decomposition. The DWT decomposition results after removing artifact components are reconstructed. The reconstructed independent components are then reconstructed together with the remaining independent components that do not contain artifacts to obtain the second EEG signal. The second EEG signal was decomposed into MEMD to obtain the IMF components after decomposition of each channel; Calculate the correlation coefficient between each IMF component and the corresponding channel of the second EEG signal; Calculate the sample entropy of IMF components whose correlation coefficient is greater than a preset threshold, and form a feature vector from the sample entropy; The correlation coefficient The calculation method is as follows: in, For IMF components, This is the second EEG signal corresponding to the IMF channel. is the length of the signal sequence.

2. The system according to claim 1, characterized in that: The specific process of outputting the recognition result based on the feature vector includes: The feature vector is input into a convolutional neural network for recognition. The convolutional neural network uses different activation functions to perform recognition and output the recognition results.

3. The system according to claim 2, characterized in that: The activation functions are Sigmoid, ReLU, and Leaky ReLU.

4. A method for electroencephalogram (EEG) monitoring in neurology, wherein the method is applied to the EEG monitoring system for neurology as described in any one of claims 1-3, characterized in that, Includes the following steps: Step S1: Acquire EEG signals and preprocess the EEG signals to obtain a first EEG signal; Step S2: Perform independent component analysis on the first EEG signal; Step S3: Determine the independent components containing artifacts based on the detection threshold; Step S4: Perform DWT decomposition on the independent components containing artifacts; Step S5: Determine the artifact components based on the boundary values ​​of the low-frequency approximation components and high-frequency detail components of the DWT decomposition. Step S6: Reconstruct the DWT decomposition results after removing artifact components; Step S7: Reconstruct the reconstructed independent components with the remaining independent components without artifacts to obtain the second EEG signal; The second EEG signal was decomposed into MEMD to obtain the IMF components after decomposition of each channel; Calculate the correlation coefficient between each IMF component and the corresponding channel of the second EEG signal; calculate the sample entropy of IMF components with correlation coefficients greater than a preset threshold, and form a feature vector from the sample entropy; The correlation coefficient The calculation method is as follows: in, For IMF components, This is the second EEG signal corresponding to the IMF channel. is the length of the signal sequence.

5. The method according to claim 4, characterized in that: The feature vector is input into a convolutional neural network for recognition.

6. The method according to claim 5, characterized in that: The convolutional neural network uses different activation functions to perform recognition and output the recognition results.

Images