Independent component analysis (ICA) is a blind source separation technique that allows the separation of linear mixtures of signals into maximal statistically independent sources, normally called independent components (ICs). This technique relies on several mathematical assumptions which need to be met by the signals of interest.

In the field of neurophysiologic signals ICA has been shown to be successful in disentangling multi-channel electroencephalogram (EEG) recordings into a number of artifacts and brain-related ICs. Thus this technique allows attenuating artifacts from EEG data by computing the back-projection of all ICs but those identified as representing artifact related activity. The categorization of ICs relies mainly on visual inspection, which makes it subjective and time consuming.

In this talk I will discuss the evaluation and optimization of EEG decompositions by means of ICA. I will also present semi-automatic procedures which improve the identification of ICs representing biological artifacts, and consequently facilitate the attenuation of these same artifacts from EEG recordings.