Statistical valuation of cognitive load level hemodynamics from functional near-infrared spectroscopy signals

Abstract

Human cognitive load level assessment is a challenging issue in the field of functional brain imaging. This work aims to study different cognitive load levels statistically from brain hemodynamics. Since the functional brain activities can be evaluated by functional near-infrared spectroscopy (fNIRS), a renowned fNIRS dataset is considered for this work. The dataset contains fNIRS data of three types of n-back tasks (0-back, 2-back, and 3-back) of twenty-six healthy volunteers. The fNIRS signals were pre-processed and separated according to the tasks and trials. The mean changes of oxygenated hemoglobin (HbO₂) and deoxygenated hemoglobin (dHb) are calculated from each trial corresponding to the tasks and tested for significant inference among three levels utilizing analysis of variance (ANOVA). From the outcomes of the ANOVA ($p<0.005$), two significant channels (AF7 (frontal) and C3h (motor)) were figured out. The significance of these two channels was further justified using the property consistency test by three different time intervals of hemodynamics inside the total task period. The latter result also explored the functional pattern of the hemodynamics of AF7 and C3h positions. Moreover, two-level cognitive load (due to easy i.e., 0-back test and hard i.e., 2-back and 3-back task) is classified using support vector machine and found classification accuracy in average 73.40%±0.076 for HbO₂ data and 71.48%±0.061 for dHb data. The study signposts the collective role played by both fNIRS signals and statistical valuation of functioning cognitive load efficacy to use fNIRS as a cognitive load assessment biomarker.