Automatic Seizure Detection and Classification Techniques for the Diagnosis of Epilepsy using Scalp Electroencephalogram EEG Signals

Abstract

The electroencephalogram (EEG) is the preferred scientific approach for recording newlinebrain electrical activity. The EEG can assist doctors in diagnosing a variety of newlineneurological diseases, most significantly epilepsy. A seizure is an unexpected event newlinein the brain which occurs due to a rapid change in the electrical activity of the neurons. newlineThe repeated unprovoked seizures are evidence of the most common neurological newlinedisease known as epilepsy. EEG is used as a bio-marker to diagnose epilepsy and it newlineallows highly trained neurological experts to recognize the occurrence of seizures. It newlineis a time-consuming process for a specialist. In comparison with a visual inspection of newlineEEG signals by doctors, machine learning techniques are more consistent. Thus, automatic newlineEEG seizure detection has a great deal of interest in neuro-informatics among newlineresearchers. This work has introduced different automatic seizure classification models newlinewith machine learning and deep learning techniques. Initially, a Machine learning newlinemodel with Empirical Mode Decomposition (EMD) based dynamic features is proposed newlineto classify seizure and healthy subjects. Subsequently, various deep learning architectures newlinesuch as a 1D Convolutional Neural Network (CNN)-Fuzzy c-means (FCM), a 1D newlineCNN-Fuzzy Support Vector Machine (SVM), a Hierarchical attention mechanism with newlineCNN and Soft-transition shrinkage based CNN-SVM are proposed to classify seizure newlinesubjects directly from the raw EEG signals. Automated quantitative analysis to differentiate newlinebetween healthy and ictal, between healthy and inter-ictal, and between ictal newlineand inter-ictal has been implemented. This study was done on two datasets, one is a newlinepublicly available Bonn university database and another is real-time data collected in newlineRamesh hospitals, India. Furthermore, the experimental results were clinically validated. newlineLower computation time and better accuracy achieved depict the effectiveness of newlinethe proposed methods to be used in the classification of healthy and seizure subjects. newline newline