An examination of application technique of independent component analysis for fNIRS signals

Abstract

A measurement technique of brain activity using fNIRS is drawing attention. The contamination of skin blood flow (SBF) is often interpreted as brain activities in fNIRS measurements, which makes the reliability of fNIRS studies uncertain. The independent component analysis (ICA) is frequently applied to decrease the SBF artifacts. However, there is not an appropriate criterion to determine which independent components contain meaningful features. Here, we proposed a criterion based on a cross-correlation coefficient (CCC) and a coefficient of spatial uniformity (CSU). fNIRS signals were measured while subjects utter vowels rhythmically while measuring SBF simultaneously. The result indicates that many channels contain SBF artifacts and show strong correlation with uttering. We applied an ICA to pseudo fNIRS signals, which generated by using SBF signals measured in an uttering task. The results suggest that brain activity is properly extracted by using CCC and CSU, unless SBF has a correlation with brain activities.