Identifying evoked potential response patterns using independent component analysis and unsupervised learning

Abstract

Independent Component Analysis(ICA) is a pre-processing step widely used in brain studies. One of the most common problems in artifact elimination or brain activity related studies is the ordering and identification of the independent components(ICs). In this work, a novel procedure is proposed which combines ICA decomposition at trial level with an unsupervised learning algorithm (K-means) at participant level in order to enhance the related signal patterns which might represent interesting brain waves. The feasibility of this methodology is evaluated with EEG data acquired with participants performing on the Halstead Category Test. The analysis shows that it is possible to find the Feedback Error Negativity (FRN) Potential at single-trial level and relate its characteristics with the performance of the participant based on their knowledge of the abstract principle underlying the task.