Design and analysis of epileptic EEG framework using automated artifact removal techniques

Abstract

The work presents the design and analysis of an Epileptic seizure-based EEG framework using automated physiological artifact removal techniques. An efficient muscle and ocular artifact based identification and correction framework is developed to improve the overall performance. The key concern while designing the artefact identification technique is the accuracy, adaptivity and time consumption. To identify and remove the muscle artifacts, the entropy-based M-DDTW (Manhattan derivative dynamic time warping) technique is proposed where the local distance for time warping as well as its derivative-based warping technique has been modified by optimizing Manhattan and Canberra distances. The sample entropy is used for reference generation. The threshold is calculated from simulated and semi-simulated datasets and tested on real Epileptic EEG data. A combination of HOS(higher order statistics) based adaptive thresholding and SMDC-VMD is proposed to mitigate the effect of ocular artifacts. The analytical performance for both the proposed identification techniques is done using the confusion matrix. For the detection and prediction of EpilepticEEG, a machine learning-based VMD-MRMR technique is presented. Statistical and entropy-based features are extracted in the VMD (variational mode decomposition) domain and MRMR (minimum redundancy and maximum relevance) is used to extract the optimal feature set for different scenarios. The Epileptic EEG pattern sometimes looks similar to physiological artefacts, especially ocular artefacts, resulting in false alarms. That is why efforts have been made to differentiate epileptic EEG epochs from the ocular-contaminated epochs. The hyperparameters of classifiers are trained and validated using Bayesian optimization. newline