An Automatic Detection Method For Epileptic EEG Signals Based On Empirical Mode Decomposition And Permutation Entropy

Epilepsy is a common neurological disorder characterized by abnormal discharges,high instability,and increased high-frequency content in electroencephalographic(EEG)signals.The current research focuses mainly on the processing,analysis,and classification of epileptic EEG signals,aiming to develop new decomposition and classification algorithms to better understand and identify the characteristics of epileptic EEG signals.In order to improve the accuracy and efficiency of epileptic EEG signal analysis,this study starts with the computer algorithm for processing epileptic EEG signals,focusing on the time-domain signal processing algorithm called Empirical Mode Decomposition(EMD).It explores an automatic detection method for epileptic EEG signals based on improved EMD combined with Refined Composite Multiscale Permutation Entropy(RCMPE)as a feature extraction technique.Two new signal processing algorithms are proposed,and three automatic classification algorithm models for epileptic EEG signals are designed.(1)Addressing the issue of traditional permutation entropy algorithms increasing the probability of uncertain entropy values,,a feature extraction method based on EMD and RCMPE is proposed.RCMPE is applied to each Intrinsic Mode Function(IMF)to extract nonlinear information with temporal and spectral characteristics.This fusion method can better distinguish the features among different frequency components and reduce the impact of mode mixing.Experimental results using the epilepsy public dataset from the University of Bonn show an accuracy of 90.3%,specificity of96.0%,sensitivity of 85.0%,and an AUC of 0.98.(2)Addressing the issue of modal aliasing and endpoint effects in traditional EMD algorithms,a feature extraction method based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and RCMPE is proposed.By introducing an ensemble of noise and adaptive methods,CEEMDAN can accurately extract the intrinsic mode functions of signals,improving the discriminability of the fused feature vectors and providing a more reliable and stable foundation for subsequent feature extraction and classification.The experimental results demonstrate an accuracy of 93.3%,specificity of 97.0%,sensitivity of 86.0%,and an AUC of 0.9432.(3)Addressing the issue of lack of discriminability in feature selection and suboptimal model parameters in EMD-related algorithms,a constrained CEEMDAN decomposition method based on Elastic Net regularization is proposed.By combining L1 and L2 regularization to construct a constrained model for empirical mode decomposition,the weights of irrelevant or redundant features are reduced to zero,thereby improving the discriminability and predictive performance of the features,reducing the influence of redundant features,and optimizing the decomposition results and predictive performance of CEEMDAN.The combination of Elastic Net regularization and CEEMDAN allows the signal decomposition algorithm to better adapt to the nonlinear and non-stationary characteristics of epileptic EEG signals.The experimental results demonstrate an accuracy of 96.1%,specificity of98.0%,sensitivity of 89.0%,and an AUC of 0.9848.