Objective selection of EEG late potentials through residual dependence estimation of independent components

This paper presents a novel method to objectively select electroencephalographic (EEG) cortical sources estimated by independent component analysis (ICA) in event-related potential (ERP) studies. A proximity measure based on mutual information is employed to estimate residual dependences of the components that are then hierarchically clustered based on these residual dependences. Next, the properties of each group of components are evaluated at each level of the hierarchical tree by two indices that aim to assess both cluster tightness and physiological reliability through a template matching process. These two indices are combined in three different approaches to bring to light the hierarchical structure of the cluster organizations. Our method is tested on a set of experiments with the purpose of enhancing late positive ERPs elicited by emotional picture stimuli. Results suggest that the best way to look for physiologically plausible late positive potential (LPP) sources is to explore in depth the tightness of those clusters that, taken together, best resemble the template. According to our results, after brain sources clustering, LPPs are always identified more accurately than from ensemble-averaged raw data. Since the late components of an ERP involve the same associative areas, regardless of the modality of stimulation or specific tasks administered, the proposed method can be simply adapted to other ERP studies, and extended from psychophysiological studies to pathological or sport training evaluation support.