Motion artifact correction in fNIRS signals based on spline interpolation and locally weighted regression

Abstract

Functional near-infrared brain imaging (fNIRS) is a novel, noninvasive technique for brain function testing. The relative motion between the light source detector and the scalp is the main source of motion artifacts, including spike artifacts, slow drift and baseline mutation, which can seriously affect the signal quality. In order to remove the effects of motion artifacts, this paper proposes a hybrid "Spline-Loess (SPLS)" correction algorithm, which effectively removes all types of artifacts in time series signals. First, the baseline drift and spike artifacts present in the signal are identified by the moving standard deviation (MSD) detection method. The drifting signal is then corrected by fitting through cubic spline interpolation, and finally the spikes are removed by smoothing the artifact signal region using a locally weighted regression algorithm. In this paper, an adaptive algorithm is designed for the selection of the parameters of the weighting function in the original local weighted regression: the bandwidth parameter h of the weighting function will be determined by the standard deviation of all the signals in the segment to be smoothed. Different types of artifacts are added to the resting-state signal, and then the "SPLS" algorithm is used to correct the signal, and the Pearson's coefficient (R), the absolute mean-square error (RMSE), and the peak error (E_p) are used to compare with the existing seven correction methods. After the experiment, it is found that the two indexes R=0.824 and RMSE=1.78 of the corrected signal using the "SPLS" algorithm, which achieve the best filtering effect compared with other methods.