Parallel Factor Analysis for multidimensional decomposition of fNIRS data - A validation study
AbstractSignificanceFunctional near-infrared spectroscopy (fNIRS) is a neuroimaging technique that uses near-infrared lights to estimate cerebral hemodynamic response, based on concentration changes in oxygenated and deoxygenated hemoglobin. A multi-dimensional decomposition technique, parallel factor (PARAFAC) analysis, has been validated for the identification of artifacts and cerebral activation patterns in electroencephalography and neuroimaging.AimWe aimed at introducing and validating the use of the PARAFAC model for fNIRS data analysis, which is inherently multidimensional (time, space, wavelength).ApproachEighteen healthy adults underwent fNIRS acquisition during a verbal fluency task. The signal-to-noise ratio and Pearson’s correlation were used to evaluate the efficacy of PARAFAC for motion artifact correction. Temporal, spatial and hemodynamic characteristics of the PARAFAC component allowed to identify task-related cerebral activations.ResultsMotion artifact correction with PARAFAC led to significant improvements in data quality and other advantages as compared to traditional 2D approaches (ICA, tPCA). Although PARAFAC revealed a slightly more distributed functional network, temporal and spatial characteristics of the task-related brain activation identified with PARAFAC mostly overlapped with those obtained with commonly used approaches.ConclusionThis study describes the first implementation of PARAFAC in fNIRS and supports it as a promising data-driven alternative for multi-dimensional data analyses in fNIRS.