Automatic component rejection based on fuzzy clustering for noise reduction in electroencephalographic signals

Among the techniques for measuring brain response, Electroencephalography (EEG) remains as the most popular for acquiring electrical brain activity over time. Currently, Event Related Potentials (ERPs) are used to detect electrophysiological responses of the brain due to a stimulus. They are usually measured by EEG due to its high temporal resolution and minimal invasiveness of the procedure. Nevertheless, EEG is well known for its high noise levels. Artifact noise (Muscular movement) is the most common and non-desired source of noise in EEG. There are basically two different ways of reducing it: (i) Suppressing the time windows with artifacts (manually). (ii) Using noise estimation algorithms to remove non-deterministic components produced by artifacts in the signal. Typically; those algorithms are based on Independent Component Analysis (ICA). However, ICA requires performing component rejection by qualified medical personnel using visual inspection, i.e., they are dependent of trained personnel for performing visual artifact or component rejection. In this manuscript a new approach for automatic component rejection based on fuzzy clustering is proposed. It considers the contributions of all components in order to remove those with non-desired elements. The proposed approach supports the decision-making procedure imitating the human learning process. It does not require the number of classes as input parameter as most of the based fuzzy clustering classification methodologies, and estimates the similarity among data leading to a non-iterative process.