ICA and SVM clustering applied to remove ocular artifacts from electroencephalography

Abstract

Brain behavior study has taken attention by the scientific world in last years. Comprehension of cerebral processes is made by means of data obtained through several techniques such as: electrocorticography, magnetoencephalography, functional and structural imaging, and electroencephalography (EEG). This latter is the most used due to its low cost and minimal risk. However, recording of cerebral information through EEG is affected by different artifactual sources, which influence their posterior data processing. Inside non-cerebral sources, the potentials caused by ocular movements during tracking and fixing tasks have the greatest impact. For this reason, a procedure to identify and subtract the ocular artifacts from EEG signals is needed. In this work, an automatic method to remove artifacts is presented. The algorithm is compounded of four main stages: a) the separation of the EEG and artifacts sources by means of Independent Component Analysis (ICA); b) the characterization of the EEG and artifact components using complexity features; c) the components classification through a Hybrid Support Vector Machine (SVM)-External clustering; and d) the reconstruction of the EEG free of artifacts. The method has an overall accuracy of 85.9%, the elimination of ocular artifacts is 85.68% and the preservation of EEG information is 85.9%. The entire algorithm was written in Python.