fNIRS: Non-stationary preprocessing methods

• MVE is able to accurately detect (97.56 %) noisy channels in fNIRS data. • CCFA filtering is able to produce a higher SNR than other conventional methods. • Choosing correct filtering window can improve SNR of a specific HRF amplitude range. In this paper we present algorithms for preprocessing of functional Near Infrared Spectroscopy (fNIRS) data. We propose a statistical method that provides an automatic identification of noisy channels and a non-stationary filtering procedure for both detrending and removal of high frequency contamination sources. A recently published Cumulative Curve Fitting Approximation (CCFA) algorithm was used for the filtration of the signals to reduce distortion effects due to the non-stationarity of the fNIRS data. The output was compared to Discrete Cosine Transform (DCT) based filtering, followed by Low Pass Filtering (LPF) and to Band Pass Filtering (BPF) methods. The results demonstrate that CCFA based filtering can produce a greater Signal to Noise Ratio (SNR) improvement in comparison to the commonly/conventionally used methods.