Noise removal of functional near infrared spectroscopy signals using emperical mode decomposition and independent component analysis

Abstract

Currently, researchers are getting more interests in discovering brain activities by non-invasive methods of using functional near-infrared spectroscopy (fNIRS). However, fNIRS collected signals usually contain noises which significantly affect the measurement of fNIRS experiments. There have been available methods proposed to remove artifacts of fNIRS signals. Among those approaches, adaptive filers are effective to mitigate physiological noises measured by extra sensors. However, the use of sensors attached on the human subjects during fNIRS measurement is uncomfortable for a user and is complicated for setup. Therefore, the method to extract the physiological signals automatically from fNIRS signals without the needs of other sensors is getting more attention from research community. In this work, we propose the combination of emperical mode decomposition (EMD) method and independent component analysis (ICA) to extract the heart rate signal. EMD is the fundamental part of Hilbert–Huang transform which is used to decompose signal into intrinsic mode functions that are not set analytically and are instead determined by an analyzed sequence alone. ICA uses Hyvarinen’s fixed-point algorithm to estimate the independent components from given multidimensional signals. Our proposed approach is able to extract the heart rate signal from multiple fNIRS channels with the accuracy of 80–90% compared with the one measured from the real device. Our further work will integrate this result with noise attenuation using adaptive filters to mitigate the global inference of physiological activities to fNIRS measurement.