Real-time Motion Artifacts and Low-Frequency Drift Correction for Functional Near-infrared Spectroscopy

Abstract

The paper investigates a real-time filtering technique for low-frequency drifts and motion artifacts (MAs) correction. The optical intensities of two wavelengths are generated by imitating brain activations using a balloon model. Two types of MAs (spike-like and step-like) and low-frequency drifts are added to the generated brain signals, forming the final synthetic brain activityies. A novel method, differential median filter (DMF), is adopted to recover the uncontaminated signals. Evaluation metrics, d1, d2, d∞, and baseline-correction ratio (BCR), are used to find out the best window sizes (8.75 s for the first median filter and 5 s for the second). The proposed method is compared with a wavelet-based MA correction method using artifact power attenuation (APA) and normalized mean-squared error (NMSE). The results show that the proposed method outperforms the wavelet-based method both in terms of the attenuation of two types of MAs and of signal distortion.