An ocular artefacts removal technique for interictal epileptiform discharges detection in epilepsy

Abstract

Electroencephalography (EEG) is the cornerstone of diagnostic testing for patients with epilepsy. However, EEG recording is easily contaminated by eye movement artefacts and other non-cerebral origin signals which affect the performance of available epileptic discharges (EDs) detection algorithm. The aim of this paper is to describe a new technique to improve the performance of an available EDs detection algorithm. An independent component analysis (ICA) based algorithm, Fast ICA is developed to extract eye movement artefacts from 10 hours sleep EEG data acquired from the National University of Malaysia Medical Center (PPUKM) database using the electrooculogram (EOG) of the corresponding subject. A clean EEG data without this artefact is then reconstructed. Data before and after the removal of ocular artefacts using ICA will be used for the following process to make a comparison on their statistical measure in the validation stage. A pre-trained support vector machine (SVM) model is used as a classifier between normal and epileptic discharge segments. The segments that contain EDs are further processed using a developed ED detection algorithm to count the number of EDs occurrence in a particular EEG channel. The EEG channels that contained EDs and the number of EDs occurrence are validated by a neurologist from PPUKM. Qualitative results show that a significant improvement of EDs detection sensitivity from 0% to 100% after the removal of ocular artefacts. Nevertheless, EDs detection specificity drops slightly from 81.3% to 75% while the accuracy remains unchanged at 76%. The drop on specificity is believed due to the two ways of contamination between EEG and EOG when EOG is used in the ocular artefact removal process. However, it helps in balancing the value of sensitivity and specificity to an optimum level which is preferable when introduced to the clinical environment.