The Classification of Independent Components for Biomedical Signal Denoising: Two Case Studies

Abstract

This chapter presents two experiences on the recovery of biomedical signals of interest from noisy datasets, i.e., the extraction of the fetal phonocardiogram from the single-channel abdominal phonogram and the recovery of the Long Latency Auditory Evoked Potential from the multichannel EEG (in children with a cochlear implant). These by implementing denoising strategies based on (1) the separation of components statistically independent by using Independent Component Analysis (ICA) and, of especial interest in this chapter, (2) the classification of the components of interest by taking advantage of properties such as temporal structure, frequency content, or temporal and spatial location. Results of these two case studies are presented on real datasets, where either focused (1) on rhythmic physiological events such as the fetal heart sounds or (2) on spatially localized events like the cochlear implant artifact, the classification stage has been fundamental on the performance of the denoising process and thus, on the quality of the retrieved signals.