EEG Data Analysis for Neurofeedback Applications

This thesis deals with signal processing methods within the context of EEG neurofeedback applications. Special attention is reserved to the use of exploratory data analysis (EDA) methods like independent component analysis (ICA) and common spatial patterns (CSP) to decompose a set of multichannel data into more informative sources, by projecting them onto opportunely designed spatial filters. ICA relies on the assumption of statistical independence of some latent sources, assumed to be linearly mixed at the EEG electrodes. This work present a method to evaluate residual dependencies between sources estimated by ICA to be used in a hierarchical clustering procedure. The properties of every group of components are evaluated at each level of the hierarchical tree by two indices that aim at assessing both cluster tightness and physiological reliability through a template matching process. Regarding CSP, a method is introduced to overcome the order indeterminacy directly in the electrode space in order to have a physiological confirmation of the decomposition. These methods are tested, in combination with a supervised classification stage, to discriminate between two brain state associated to affective picture processing. Results show that both methods, and especially ICA, are able to enhance the presence of late positive potentials elicited by emotionally arousing pictures with respect to neutral ones.