Detection of epileptic seizures based on electroencephalogram using enhanced signal processing and soft computing techniques

Abstract

A myriad of epilepsies indicate that its arousal could be due to manifold reasons but the newlineentirety culminates into a communal phenomenon the seizure Requirement of an accurate and newlinereliable seizure detector is highly indispensable not only for the better understanding of seizure but newlinealso a boon for medicos who are subjected to fatigue in visualizing a continuous and enormous longterm recording Despite substantial innovations in an automated detection of epilepsy over the past decade the quench for an appropriate blend of methodology emphasizes the need for an appropriate seizure detection systems A broad range of strategies is being investigated thoroughly and an effort is put forth with utmost intervention to yield an approach for seizure detection Automated seizure detection system fundamentally aims in providing a consistent algorithm by excavating indepth knowledge of dynamical properties of the signal and clinical domains EEG signals are often obscured by the presence of artifacts which in turn disrupts correct newlinediagnosis and analysis Preprocessing stage comprises of implementing a conventional notch filter newlinefollowed by spatially constrained Independent Component Analysis along with Wavelet denoising newlinetechnique to improve the quality of EEG signal Five types of artifacts were analyzed and removed newlineNeed for extracting relevant information in EEG was satisfied by a transformation called Fast Walsh newlineHadamard Transform FWHT and Hanning window were used to isolate the EEG bands alpha beta delta theta and gamma The Welsh power spectrum estimator was used to calculate the spectrum and the power in delta alpha and beta band These values were used to form the feature vector that was used as input to the next phase The proposed classification algorithms Hybrid Extreme Learning Machine HELM and Fast Adaptive Neuro Fuzzy Inference System FANFIS used for detecting the presence of seizure in EEG signal Modified Levenberg Marquardt algorithm MLM was incorporated with the classifiers to reduce the time complexity of the