Background rejection techniques for automated spike detection in diagnosis of epilepsy

Abstract

Epilepsy is regarded as one of the most common neuro-physiological disorders characterised by recurrent, involuntary, paroxysmal seizure activity in the brain. Although some causes of epilepsy are known, the majority of the causes are still unknown and under research. So far epilepsy is diagnosed manually based on the detection of unusual spikes present in the spatial-temporal characteristics of the brain signals measured using Electroencephalogram (EEG). However, this becomes erroneous due to presence of artefacts and also the random nature of the spike size and shape. In order to overcome these issues, we need a reliable system that can automatically detect spikes and thus diagnose epilepsy. This dissertation work involves the classification of Spike and non-spikes from the EEG data of patients and rejecting the background. Two major areas of Background Rejection focussed in this project are feature based rejection using feature pool and Classifier based rejection using Machine Learning techniques. Using Background Rejection, at each stage, best features are identified and a feature ranking table is formulated after a cascade of rejection. The feature selection method is validated by building a single classifier using these features and the results are compared.