Motion Artifact Removal for Functional Near Infrared Spectroscopy: A Comparison of Methods

Abstract

Near infrared spectroscopy (NIRS) is rapidly gaining popularity for functional brain imaging. It is well suited to studies of patients or children; however, in these populations particularly, motion artifacts can present a problem. Here, we propose the use of imaging channels with negligible distance between light source and detector to detect subject motion, without the need for an additional motion sensor. Datasets containing deliberate motion artifacts were obtained from three subjects. Motion artifacts could be detected in the signal from the co-located channels with a minimum sensitivity of 0.75 and specificity of 0.98. Five techniques for removing motion artifact from the functional signals were compared, namely two-input recursive least squares (RLS) adaptive filtering, wavelet-based filtering, independent component analysis (ICA), and two-channel and multiple-channel regression. In most datasets, the median change in SNR across all channels was the greatest using ICA or multiple-channel regression. RLS adaptive filtering produced the smallest increase in SNR. Where sharp spikes were present, wavelet filtering produced the largest SNR increase. ICA and multiple-channel regression are promising ways to reduce motion artifact in functional NIRS without requiring time-consuming manual techniques.