Real-Time Scalp-Hemodynamics Artifact Reduction Using a Sliding-Window General Linear Model: A Functional Near-Infrared Spectroscopy Study

Abstract

Functional near-infrared spectroscopy (fNIRS) measures temporal hemoglobin changes in gray matter, reflecting brain activity. The primary advantage of fNIRS is real-time estimation of brain activity, with applications such as neurofeedback training. However, task-related scalp-hemodynamics distributed across the whole head are superimposed onto cerebral activity, leading to false estimation of brain activity. To prevent this, we propose a real-time artifact rejection method using short distance probes, by applying a sliding-window general linear model (GLM) with a real-time updated design matrix via a global scalp-hemodynamics model (GSHM). To assess the performance of our proposed method, we performed simulation, assuming that fNIRS signals, consisting of local cerebral blood flow (CBF) and scalp-hemodynamics, had a spatially common pattern. Simulation results were compared with off-line analysis and previous on-line methods, with scalp-hemodynamics excluded from the design matrices. The proposed method showed significantly higher performance for estimating CBF.