Research On EEG Data Feature Extraction And Stroke Risk Classification Prediction

Cerebral Stroke,also known as "stroke",is a common cerebrovascular disease.The incidence of stroke is increasing year by year among Chinese residents.It is characterized by rapid onset and urgent course of disease.with high clinical morbidity,mortality and disability.Stroke is mainly divided into two types: hemorrhagic and ischemic,which have different pathogenesis.The electroencephalogram(EEG)is a reflection of the electrophysiological activity of brain nerve cells and contains a lot of physiological information of the human body.The EEG data before and after stroke onset show significant differences,and the data of different types of strokes also show different characteristics.Therefore,EEG-based stroke incidence prediction is an effective tool for efficient stroke diagnosis.Currently,the diagnosis of stroke risk prediction based on EEG data is mainly adopts the method of manual interpretation,which is easily influenced by the subjective judgment and experience of clinicians.In recent years,with the development of artificial intelligence technology in smart medicine,the use of modern signal processing means to predict stroke onset on EEG data has become a research hotspot.which helps to accelerate the realization of intelligent medical assistant diagnosis of stroke.It has been found that EEG data is a non-stationary time series with nonlinear dynamics characteristics,so using nonlinear dynamics methods to extract appropriate features from EEG data and select effective classification models can help doctors to diagnose the stroke type quickly and take the effective treatment measures promptly.Therefore,a Multifractal Detrended fluctuation Analysis(MF-DFA)and entropy based method for feature extraction of EEG data from stroke is introduced in this paper.compares and analyzes the performance of single classifier and integrated classifier for stroke onset prediction of the extracted EEG features to obtain better stroke classification prediction results.The main work of this paper is as follows.First: The study of EEG signal feature extraction in stroke based on multiple fractal detrended fluctuation analysis MF-DFA.Firstly,the generalized Hurst index spectrum,the scalar index spectrum and the multifractal spectrum were solved for two types of stroke EEG data,respectively.The multiple fractal characteristics of stroke EEG data were demonstrated,and four fractal parameters were extracted from the three spectra as fractal feature vectors.The four fractal feature vectors were extracted from the EEG data of 366 stroke patients,and a single decision tree was used as the prediction model for stroke classification prediction study.The experimental results showed that the optimal performance of stroke classification prediction was achieved when the maximum value of the generalized Hurst index was used as the feature vector.Furthermore,four fractal feature vectors were extracted from each of the 8leads of 366 patients,and a single decision tree was used for classification and identification,The experimental results showed that among the 8 leads of stroke patients,the best performance of stroke classification prediction was obtained by feature extraction for lead C4.Second: A study on the classification and prediction of stroke incidence risk based on multi-feature fusion of EEG data.Firstly,four single entropy feature values were introduced for feature extraction of two types of stroke EEG data,and then the wavelet packet decomposition theory was introduced.The wavelet packet decomposition was combined with single entropy features to extract four hierarchical entropy feature values of two types of stroke EEG data.After that,the obtained single entropy feature values and the hierarchical entropy feature values were fused with stroke multiple fractal feature values respectively,and the C4.5 single decision tree classification prediction model was established for stroke incidence risk classification prediction.The experimental results showed that the maximum value of generalized Hurst index of stroke EEG signal,the four fractal eigenvalues of C4 leads and hierarchical fuzzy entropy values were fused to achieve the best performance in stroke onset risk classification prediction.Third: A study on the application of integrated extreme learning machine in stroke onset risk classification prediction.In order to obtain better performance of stroke onset risk classification prediction,the extreme learning machine prediction model and the Adaboost integrated classification model were introduced,and the Adaboost-based integrated decision tree classification prediction model and the integrated extreme learning machine classification prediction model were designed.Single and fused features of EEG data were used as input to compare and analyze the performance differences of decision tree,extreme learning machine,and integrated decision tree classification prediction model and integrated extreme learning machine classification prediction model.The experimental results showed that the best performance of the integrated extreme learning machine was obtained based on the single and fused features of stroke EEG data.