EEG-based cerebral arterial thrombosis detection and evaluation system
By developing an ischemic stroke detection and evaluation system based on EEG, the existing detection methods have been solved, and efficient, convenient and non-invasive early detection and evaluation have been achieved, which significantly improves the accuracy and reliability of the detection and reduces the incidence and disability of stroke.
Patent Information
- Application Number
- CN202510090203.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-21
- Publication Date
- 2025-05-30
AI Technical Summary
The existing ischemic stroke detection methods have limitations such as low resolution, poor portability and injury, which lead to difficulties in early detection and delayed the optimal treatment opportunity.
A ischemic stroke detection and evaluation system based on EEG is developed, using signal acquisition modules of 2 forehead channels and 2 ear clip channels, combining the upper computer design module, signal preprocessing module and mixed prediction and classification model to realize wireless data transmission and personalized evaluation.
It has achieved efficient, convenient and non-invasive early detection and evaluation, which has significantly improved the accuracy and reliability of the detection. It can be performed in homes or community health stations and other places, helping to detect potential disease risks in the early stage and reducing the incidence and disability of stroke.
Smart Images
Figure CN120052920A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of ischemic stroke detection and evaluation, and specifically provides an EEG-based ischemic stroke detection and evaluation system. Background Art
[0002] Ischemic stroke refers to the general term for necrosis of brain tissue caused by stenosis or occlusion of the cerebral blood supply artery and insufficient cerebral blood supply. It is due to the blood supply disorder in a local brain tissue area, resulting in ischemic and hypoxic pathological necrosis of the brain tissue, and then presenting neurological ischemic manifestations.
[0003] The functional activities of the brain are achieved through the electrical signal conduction of nerve cells. These electrical signals are recorded as electroencephalograms on the scalp by precise electronic instruments. Generally, the waveforms of electroencephalograms have fixed waveform and rhythm characteristics. However, when the brain is in an ischemic state, due to insufficient blood supply, nerve cells will be affected, resulting in changes in electrophysiological activities, and then affecting the rhythm and waveform of brain waves. When ischemic stroke occurs, electroencephalograms can capture these changes in electrophysiological activities. Specifically, the brain tissue in the infarct area undergoes necrosis, and the bioelectrical activity will weaken or even disappear, while the neurons with impaired function around the infarct may generate various abnormal discharges. These abnormal discharges will be manifested as abnormal rhythms and waveforms on the electroencephalogram. Through electroencephalogram detection, these abnormal changes can be observed, thus helping to diagnose ischemic stroke.
[0004] However, due to the many limitations of the commonly used detection methods in current technologies, such as low resolution, poor portability, invasiveness, etc., it is difficult to detect early, delaying the best treatment opportunity. Therefore, developing an efficient, convenient, and non-invasive early detection and evaluation system has important clinical significance and social value. Summary of the Invention
[0005] The present invention aims to provide a stroke detection and evaluation system with high practicability, convenience, and suitability for various places such as families and community health stations, which can break through the limitation that the risk detection and evaluation of high-risk groups are limited to hospital professional equipment, help patients in a healthy state discover potential disease risks in a timely manner, intervene and treat early, and reduce the incidence and disability rate of stroke.
[0006] To solve the above problems, the present invention provides an EEG-based ischemic stroke detection and evaluation system for monitoring the EEG signals of the left and right prefrontal lobes of the brains of ischemic stroke patients, including:
[0007] Signal acquisition module: The system includes 2 frontal channels and 2 earclip channels. The signals of the earclip channels are mainly used as references to ensure accurate signal acquisition. The TGAM EEG module is used to integrate the analog front-end and digital signal processing structures. The system input noise of up to 10 mV ensures that the impact of the hardware on EEG signal acquisition is minimized. Its core includes the STM32F1 series microcontrollers. The signals of the left and right prefrontal lobes of the brain and the left and right earlobes are collected by using two TGAM modules in parallel. The left prefrontal lobe and left ear of the brain respectively correspond to the EEG signal input and REF reference input of TGAM module 1; the right prefrontal lobe and right ear of the brain respectively correspond to the EEG signal input and REF reference input of TGAM module 2. The data collected by the TGAM modules are respectively transmitted to the STM32 minimum system board. The power supply of the signal acquisition module uses two 3.3V lithium batteries in series, and the voltage is rectified and regulated through a power supply voltage stabilization circuit to supply power to the system.
[0008] Host computer design module: According to the preset program, send instructions to the lower computer to start signal acquisition, and read the continuous data stream of brain electrical activities in real time. The collected raw data is encoded and packed to optimize the data transmission efficiency, and wireless data transmission to the host computer is achieved through the Bluetooth module. After receiving the data, the host computer uses PyQt5 to develop a human-computer interaction interface to display the signal waveform in real time, and saves the processed data to a document to complete the prediction and classification of EEG signals.
[0009] Signal preprocessing module: The collected signals are preprocessed by using the cubic spline interpolation method optimized by artificial bee colony and the wavelet transform algorithm based on kurtosis to further improve the curve fitting quality and smoothness, avoid overfitting and eliminate the jitter noise existing in the EEG signals, thereby significantly improving the quality and efficiency of EEG signal processing and facilitating subsequent feature extraction.
[0010] Hybrid prediction and classification model building module: Using advanced deep learning technologies, a composite model integrating a temporal convolutional network and a bidirectional long short-term memory network is constructed to optimize the prediction performance of electroencephalogram (EEG) signals. A composite model of a convolutional network (CNN) and a least squares support vector machine (LSSVM) is constructed to optimize the classification performance of EEG signals.
[0011] Personalized Evaluation Module: Using the methodology of EEG feature extraction, entropy value indicators such as information entropy, energy entropy, sample entropy, and attention entropy are selected to construct a time-domain feature matrix. This matrix is designed to comprehensively capture the complexity and dynamic characteristics of EEG signals. Further, frequency-domain analysis indicators such as variance, mean, power spectral density, and kurtosis are used to construct a frequency-domain feature matrix to deeply analyze the performance and characteristics of EEG signals in the frequency domain. In addition, by extracting spatial features such as corner points, smoothness, trends, and periodicity, the spatial attributes of EEG signals are comprehensively understood. Then, a graph theory method is introduced to construct an EEG signal network. The network structure characteristics of EEG signals are evaluated by defining key nodes and boundaries, calculating clustering coefficients, average path lengths, network densities, centralities, etc.; the graph theory method is applied to analyze the topological structure of the network and identify community structures and core-edge structures in the network; thus, starting from this, the dynamic characteristics and spatial distribution of EEG signals are explained, and the relationship between EEG signals and the disease state and degree of illness of ischemic stroke is explained in combination with clinical knowledge. Finally, modeling based on three parameters of time-frequency-space is realized to guide the detection and evaluation of EEG for different individuals.
[0012] Beneficial Effects: In actual clinical experiments, a total of 15 subjects were detected by this product, and 8 of them were consistent with the results of clinical detection methods. This fully shows that this product has high accuracy and reliability in terms of detection consistency.
[0013] Based on a dual-channel EEG acquisition device, the present invention proposes to develop a detection and evaluation system for ischemic stroke. It can not only significantly improve the quality and efficiency of medical services, but also effectively reduce the overall medical cost. Through the application of machine learning algorithms, potential risk factors of ischemic stroke can be more accurately identified and analyzed, realizing early warning and personalized intervention for high-risk populations, thereby greatly reducing the incidence and disability rate of stroke. Description of the Drawings
[0014] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments.
[0015] Attached Figure 1 is a schematic diagram of the signal acquisition module of the present invention;
[0016] Attached Figure 2 is a schematic diagram of the overall design of the present invention. Specific Embodiments
[0017] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0018] As attached Figure 1 and attached Figure 2 As shown, an ischemic stroke detection and evaluation system based on EEG is used to monitor the EEG signals of the left and right prefrontal lobes of the brain of patients with ischemic stroke, including:
[0019] The signal acquisition module is used to collect signals from the left and right frontal lobes and left and right earlobes of the subject's brain, and transmit the collected signals to the host computer in a time-sharing manner for signal processing;
[0020] The host computer design module is used to send instructions to the lower computer to start signal acquisition. The collected raw data is optimized and transmitted to the host computer through the Bluetooth module. After receiving the data, the host computer processes and displays the signal waveform and saves the processed data to a document.
[0021] Signal preprocessing module, used to improve the curve fitting quality and smoothness in EEG processing and eliminate the jitter noise widely present in EEG signals;
[0022] Prediction and classification module, used to achieve high-precision prediction and high-precision classification of EEG signals;
[0023] The personalized evaluation module is used to evaluate the randomness and regularity of EEG signal changes, calculate the instantaneous phase difference of dual-channel EEG signals, reflect the spatial domain characteristics of EEG signals, and construct a connection matrix based on the EEG network to guide EEG personalized detection and evaluation.
[0024] The signal acquisition module includes:
[0025] The TGAM EEG module from NeuroSky is used to integrate the analog front end and digital signal processing structure, and the maximum input noise of the system is set to 10mV to ensure that the hardware has minimal impact on EEG signal acquisition;
[0026] The STM32F1 series microcontroller amplifies, eliminates noise, and processes DAC inside the chip, and then uses the NeuroSkyeSense algorithm to convert brain waves into the concentration and meditation level of the subject's current mental state.
[0027] The power supply of the signal acquisition module adopts:
[0028] Two 3.3V lithium batteries are connected in series, and the power supply voltage is rectified and regulated through a power supply voltage stabilizing circuit to supply power to this system.
[0029] The host computer design module designs the host computer through Labview. The host computer sends instructions to the lower computer according to the preset program, starts signal acquisition, and reads the continuous data stream of electroencephalogram activities in real time. The collected original data is encoded and packed to optimize the data transmission efficiency, and wireless data transmission to the host computer is realized through the Bluetooth module.
[0030] The data of the Bluetooth module is transmitted to the host computer. After receiving the data, the host computer immediately processes and displays the signal waveform. The system performs filtering processing on the waveform to effectively remove noise and interference. Through Fourier transform, the time-domain data is converted to the frequency domain to deeply analyze the frequency characteristics of the signal. Finally, the system saves the processed frequency-domain waveform data to a document.
[0031] The signal preprocessing module includes:
[0032] The cubic spline interpolation method optimized by artificial bee colony. First, the original EEG is processed by low-pass filtering to remove high-frequency noise components, and then the cubic spline interpolation algorithm is used to further smooth the signal to ensure the continuity and smoothness of the curve. On this basis, the artificial bee colony optimization algorithm is introduced to optimize the spline interpolation function.
[0033] The wavelet transform algorithm based on kurtosis is used to denoise the EEG signal through the wavelet transform algorithm based on kurtosis to eliminate the persistent jitter noise widely existing in the EEG signal.
[0034] The process of the cubic spline interpolation method optimized by artificial bee colony includes:
[0035] At the beginning of the optimization, the algorithm randomly generates a set of candidate solutions as the initial spline interpolation parameters. Subsequently, the algorithm simulates the foraging behavior of bees, iteratively updates the spline interpolation parameters, explores a better solution space. In each iteration, according to the foraging effect of bees, the candidate solutions are updated, the parameters with poor effects are eliminated, and the better parameters are retained. This process is repeated continuously until the predetermined number of iterations is reached or a satisfactory solution is found.
[0036] The wavelet transform algorithm based on kurtosis includes:
[0037] Wavelet filtering, based on the discrete wavelet transform of single-channel data and the analysis of the obtained wavelet coefficients and their changes over time. After this method performs wavelet decomposition, the kurtosis value of the wavelet coefficients of each layer will be calculated. If the kurtosis value is larger than the preset threshold, the highest wavelet coefficient will be set to zero, and further calculations will be performed until the kurtosis value is less than the threshold, and then the signal after removing motion artifacts is reconstructed using the inverse wavelet transform.
[0038] The prediction and classification module includes:
[0039] A composite model of a temporal convolutional network and a bidirectional long short-term memory network. First, the temporal convolutional network is used to preliminarily process the EEG signal to identify and extract the key features of the signal in the time domain. Subsequently, these features are passed to the bidirectional long short-term memory network, which further deepens the feature extraction process through its bidirectional parallel processing mechanism and captures the deep dynamic characteristics of the EEG signal in the time series.
[0040] A composite model of a convolutional neural network and a least squares support vector machine. First, the convolutional network is used to preliminarily process the EEG signal to identify and extract the key features of the signal in the space domain. Subsequently, these features are passed to the least squares support vector machine to further deepen the feature extraction process and capture the deep dynamic characteristics of the EEG signal in the time series.
[0041] The personalized evaluation module includes:
[0042] Time domain analysis, a feature extraction method based on permutation entropy. By calculating the permutation entropy, it describes the time complexity of the EEG signal and evaluates the randomness and regularity of the EEG signal changes.
[0043] Entropy includes:
[0044] Information entropy, which solves the problem of quantifying information and can describe the irregularity and complexity of signals in the field of information theory.
[0045] Permutation entropy, which uses entropy to describe the changes in the sorting pattern of complex time series, mainly for non-linear time series changes rather than analyzing amplitude or phase.
[0046] Frequency domain analysis, a feature extraction method based on phase difference. By calculating the phase difference feature, the instantaneous phase difference of the two-channel EEG signal is calculated.
[0047] Phase difference includes:
[0048] Phase is a measure used to describe the waveform change of a signal. Phase difference is the phase difference between two or more vectors. Usually, the instantaneous phase of a vector can be obtained by using the Hilbert transform.
[0049] Spatial analysis, a feature extraction method based on permutation mutual information. By introducing the permutation mutual information method, the dynamic coupling and information transfer between different signal vectors are quantified, thereby reflecting the spatial domain characteristics of the EEG signal.
[0050] Mutual information and permutation entropy include:
[0051] Mutual information is used in information theory to evaluate the degree of dependence between two variables, while the permutation entropy theory can reflect the degree of pattern change in a system. The mutual information based on permutation entropy is a symbolized mutual information algorithm that can reflect the dynamic coupling and information transmission between vectors x(t) and y(t) under different degrees of pattern change;
[0052] In the scheme design, by integrating the characteristics of the time-frequency-space three dimensions, a connection matrix based on the EEG network is constructed by introducing graph theory methods to define network nodes, boundaries, and calculate network attributes, and three-dimensional modeling of time-frequency-space is carried out to guide the personalized detection and evaluation of EEG;
[0053] The graph theory methods include:
[0054] Apply graph theory methods to analyze the topological structure of the network, identify the community structure and core-edge structure in the network, and thus use this as a starting point to explain the dynamic characteristics and spatial distribution of EEG signals, and combine clinical knowledge to explain the relationship between EEG signals and the disease state and degree of illness of ischemic stroke. Finally, a model based on the three parameters of time-frequency-space is realized to guide the detection and evaluation of EEG for different individuals.
[0055] In this embodiment, two TGAM modules are connected in parallel to collect signals from the left and right prefrontal lobes and the left and right earlobes of the brain. The left prefrontal lobe and left ear of the brain respectively correspond to the EEG signal input and REF reference input of TGAM module 1; the right prefrontal lobe and right ear of the brain respectively correspond to the EEG signal input and REF reference input of TGAM module 2. The data collected by the TGAM modules are respectively transmitted to the STM32 minimum system board, and through the data sending end of the Bluetooth module, the collected signals are transmitted to the host computer for signal processing in a time-sharing manner. The host computer sends instructions to the lower computer according to a preset program to start signal collection and real-time read the continuous data stream of brain electrical activities. The collected original data is encoded and packed to optimize the data transmission efficiency, and wireless data transmission to the host computer is realized through the Bluetooth module.
[0056] As shown in the appendix Figure 2 After receiving the data, the host computer immediately processes and displays the signal waveform. Through Fourier transform, the time-domain data is converted to the frequency domain to deeply analyze the frequency characteristics of the signal. Finally, the system saves the processed frequency-domain waveform data to a document for subsequent research and data sharing.
[0057] To further improve the signal quality, the cubic spline interpolation method is optimized by the artificial bee colony optimization algorithm to enhance the curve fitting quality and smoothness in EEG processing. First, the original EEG is processed by low-pass filtering to remove high-frequency noise components. Then, the cubic spline interpolation algorithm is used to further smooth the signal to ensure the continuity and smoothness of the curve. On this basis, the artificial bee colony optimization algorithm is introduced to optimize the spline interpolation function. At the beginning of the optimization process, the algorithm randomly generates a set of candidate solutions, that is, the initial spline interpolation parameters. Subsequently, the algorithm simulates the foraging behavior of bees and explores a better solution space by iteratively updating the spline interpolation parameters. In each iteration, according to the foraging effect of bees, the candidate solutions are updated, the poor-performing parameters are eliminated, and the better parameters are retained. This process is repeated continuously until the predetermined number of iterations is reached or a satisfactory solution is found. In this way, the artificial bee colony optimization algorithm can ensure that the interpolation curve closely fits the actual data and avoid the phenomenon of overfitting, thus significantly improving the quality and efficiency of EEG signal processing.
[0058] Furthermore, to effectively remove persistent jitter noise and interference and enhance the quality and accuracy of EEG signals for subsequent feature extraction, a wavelet transform algorithm based on kurtosis is used for EEG signal denoising. After wavelet decomposition by this method, the kurtosis value of the wavelet coefficients of each layer is calculated. If the kurtosis value is larger than the preset threshold, the highest wavelet coefficient is set to zero, and the calculation is further performed until the kurtosis value is less than the threshold, and then the signal after removing motion artifacts is reconstructed by inverse wavelet transform. The final data kurtosis value is set to 3.3.
[0059] Furthermore, the prediction and classification of EEG signals are optimized. In the time domain, first, a temporal convolutional network is used to preliminarily process the EEG signals to identify and extract the key features of the signals in the time domain. Subsequently, these features are passed to a bidirectional long short-term memory network, which further deepens the feature extraction process through its bidirectional parallel processing mechanism and captures the deep dynamic characteristics of EEG signals in the time series. The integration of this bidirectional information flow significantly enhances the model's ability to understand the change trend of EEG signals, thus achieving high-precision prediction of EEG signals. In the spatial domain, first, a convolutional network is used to preliminarily process the EEG signals to identify and extract the key features of the signals in space. Subsequently, these features are passed to LSSVM to further deepen the feature extraction process and capture the deep dynamic characteristics of EEG signals in the time series. The integration of this fused information flow significantly enhances the model's ability to understand the change trend of EEG signals, thus achieving high-precision classification of EEG signals, and the classification accuracy reaches 99.8%. By mixing the features extracted by these two models, the complexity and diversity of EEG signals can be more comprehensively captured, thus more accurately classifying different degrees of illness of ischemic stroke patients.
[0060] A personalized evaluation scheme based on multi-dimensional feature extraction is given for the classified signals. First, by calculating entropy value indicators such as information entropy, energy entropy, sample entropy, and attention entropy to describe the time complexity of EEG signals, evaluate the randomness and regularity of EEG signal changes, and construct a time-domain feature matrix. Secondly, calculate the instantaneous phase difference of dual-channel EEG signals through phase difference features, and use frequency-domain analysis indicators such as variance, mean, power spectral density, and kurtosis to construct a frequency-domain feature matrix to deeply analyze the performance and characteristics of EEG signals in the frequency domain. Finally, by introducing the method of sorted mutual information, spatial features such as corner points, smoothness, trends, and periodicity are extracted to quantify the dynamic coupling and information transfer between different signal vectors, thereby reflecting the spatial-domain features of EEG signals. Combining the above features, an EEG signal network is constructed by introducing graph theory methods. Evaluate the network structure characteristics of EEG signals by defining key nodes and boundaries, calculating clustering coefficients, average path lengths, network densities, centralities, etc.; apply graph theory methods to analyze the topological structure of the network, identify community structures and core-edge structures in the network; thus, starting from this, explain the dynamic characteristics and spatial distribution of EEG signals, and combine clinical knowledge to explain the relationship between EEG signals and the disease state and degree of illness of ischemic stroke. Finally, realize modeling based on three parameters of time-frequency-space to guide the detection and evaluation of EEG of different individuals.
[0061] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.
Claims
1. An ischemic stroke detection and evaluation system based on EEG, used to monitor the EEG signals of the left and right prefrontal lobes of the brain of ischemic stroke patients, characterized in that: include: The signal acquisition module is used to collect signals from the left and right frontal lobes and left and right earlobes of the subject's brain, and transmit the collected signals to the host computer in a time-sharing manner for signal processing; The host computer design module is used to send instructions to the lower computer to start signal acquisition. The collected raw data is optimized and transmitted to the host computer through the Bluetooth module. After receiving the data, the host computer processes and displays the signal waveform and saves the processed data to a document. Signal preprocessing module, used to improve the curve fitting quality and smoothness in EEG processing and eliminate the jitter noise widely present in EEG signals; Prediction and classification module, used to achieve high-precision prediction and high-precision classification of EEG signals; The personalized evaluation module is used to evaluate the randomness and regularity of EEG signal changes, calculate the instantaneous phase difference of dual-channel EEG signals, reflect the spatial domain characteristics of EEG signals, and construct a connection matrix based on the EEG network to guide EEG personalized detection and evaluation.
2. The EEG-based ischemic stroke detection and assessment system as claimed in claim 1, characterized in that: The signal acquisition module comprises: The TGAM EEG module from NeuroSky is used to integrate the analog front end and digital signal processing structure, and the maximum input noise of the system is set to 10mV to ensure that the hardware has minimal impact on EEG signal acquisition; The STM32F1 series microcontroller amplifies, eliminates noise, and processes DAC inside the chip, and then uses the NeuroSkyeSense algorithm to convert brain waves into the concentration and meditation level of the subject's current mental state.
3. The EEG-based ischemic stroke detection and assessment system as claimed in claim 2, characterized in that: The power supply of the signal acquisition module adopts: Two 3.3V lithium batteries are connected in series, and the system is powered by a power supply voltage regulator circuit that rectifies and stabilizes the voltage.
4. The EEG-based ischemic stroke detection and assessment system as claimed in claim 3, characterized in that: The host computer design module uses Labview to design the host computer. The host computer sends instructions to the subordinate computer according to the preset program, starts signal acquisition, and reads the continuous data stream of EEG activity in real time. The collected raw data is encoded and packaged to optimize data transmission efficiency, and wireless data transmission to the host computer is realized through the Bluetooth module.
5. The EEG-based ischemic stroke detection and assessment system as claimed in claim 4, characterized in that: The data of the Bluetooth module is transmitted to the host computer. After receiving the data, the host computer immediately processes and displays the signal waveform. The system performs filtering on the waveform to effectively remove noise and interference. Through Fourier transform, the time domain data is converted to the frequency domain, thereby deeply analyzing the frequency characteristics of the signal. Finally, the system saves the processed frequency domain waveform data to a document.
6. The EEG-based ischemic stroke detection and assessment system as claimed in claim 5, characterized in that: The signal preprocessing module comprises: Cubic spline interpolation method under artificial bee colony optimization, first, the original EEG is processed by low-pass filtering to remove high-frequency noise components, and then the cubic spline interpolation algorithm is used to further smooth the signal to ensure the continuity and smoothness of the curve. On this basis, the artificial bee colony optimization algorithm is introduced to optimize the spline interpolation function; The kurtosis-based wavelet transform algorithm is used to perform EEG signal denoising and eliminate the widespread continuous jitter noise in EEG signals.
7. The EEG-based ischemic stroke detection and assessment system as claimed in claim 6, characterized in that: The process of the cubic spline interpolation method under artificial bee colony optimization includes: At the beginning of the optimization, the algorithm randomly generates a set of candidate solutions as the initial spline interpolation parameters. Then the algorithm simulates the foraging behavior of bees, updates the spline interpolation parameters through iteration, and explores a better solution space. In each iteration, the candidate solutions are updated according to the foraging effect of the bees, and the parameters with poor effects are eliminated, and the better parameters are retained. This process is repeated until the predetermined number of iterations is reached or a satisfactory solution is found.
8. The EEG-based ischemic stroke detection and assessment system as claimed in claim 7, characterized in that: The kurtosis-based wavelet transform algorithm includes: Wavelet filtering is based on the analysis of the obtained wavelet coefficients and their changes over time based on the discrete wavelet transform of a single channel data. After wavelet decomposition, this method will calculate the kurtosis value of the wavelet coefficients of each layer. If the kurtosis value is greater than the preset threshold, the highest wavelet coefficient will be set to zero, and further calculation will be performed until the kurtosis value is less than the threshold. The inverse wavelet transform is then used to reconstruct the signal after removing motion artifacts.
9. The EEG-based ischemic stroke detection and assessment system as claimed in claim 8, characterized in that: The prediction and classification module includes: The composite model of the temporal convolutional network and the bidirectional long short-term memory network first uses the temporal convolutional network to preliminarily process the EEG signal to identify and extract the key features of the signal in the time domain. Subsequently, these features are passed to the bidirectional long short-term memory network, which further deepens the feature extraction process through its bidirectional parallel processing mechanism and captures the deep dynamic characteristics of the EEG signal in the time series. The composite model of convolutional neural network and least squares support vector machine first uses convolutional network to perform preliminary processing on EEG signals to identify and extract the key spatial features of the signals. Subsequently, these features are passed to the least squares support vector machine to further deepen the feature extraction process and capture the deep dynamic characteristics of EEG signals in time series.
10. The EEG-based ischemic stroke detection and assessment system as claimed in claim 9, characterized in that: The personalized evaluation module includes: Time domain analysis, based on the feature extraction method of sorting entropy, describes the time complexity of EEG signals by calculating sorting entropy, and evaluates the randomness and regularity of EEG signal changes; The entropy includes: Information entropy solves the problem of quantitative measurement of information. In the field of information theory, it can describe the irregularity and complexity of signals. Sorting entropy uses entropy to describe the changes in the sorting pattern of complex time series, mainly targeting nonlinear time series changes rather than analyzing amplitude or phase; Frequency domain analysis, feature extraction method based on phase difference, calculates the instantaneous phase difference of dual-channel EEG signals through phase difference features; The phase difference includes: Phase is a measure used to describe the change of signal waveform. Phase difference is the phase difference between two or more vectors. Usually, the instantaneous phase of a vector can be obtained by Hilbert transform. Spatial analysis, a feature extraction method based on sorting mutual information, quantifies the dynamic coupling and information transfer between different signal vectors by introducing the sorting mutual information method, thereby reflecting the spatial domain characteristics of EEG signals; The mutual information and sorting entropy include: Mutual information is used in information theory to evaluate the degree of dependence between two variables, while the sorting entropy theory can reflect the degree of pattern change in the system. The mutual information based on sorting entropy is a symbolic mutual information algorithm that can reflect the dynamic coupling and information transfer of vectors x(t) and y(t) under different degrees of pattern change. The scheme design integrates the characteristics of the three dimensions of time, frequency and space, defines network nodes and boundaries by introducing graph theory, and calculates network properties to construct a connection matrix based on the EEG network, and conducts modeling in the three dimensions of time, frequency and space to guide EEG personalized detection and evaluation; The graph theory approach includes: Graph theory methods are used to analyze the topological structure of the network and identify the community structure and core-edge structure in the network. This is used as a starting point to explain the dynamic characteristics and spatial distribution of EEG signals. Clinical knowledge is combined to explain the relationship between EEG signals and the disease state and severity of ischemic stroke. Finally, modeling based on the three parameters of time, frequency and space is implemented to guide the detection and evaluation of EEG of different individuals.