Significance: Isolating task-evoked brain signals from background physiological noise (e.g., cardiac, respiratory, and blood pressure fluctuations) poses a major challenge for the analysis of functional near-infrared spectroscopy (fNIRS) data. Aim: The performance of several analytic methods to separate background physiological noise from brain activity including spatial and temporal filtering, regression, component analysis, and the use of short-separation (SS) measurements were quantitatively compared. Approach: Using experimentally recorded background signals (breath-hold task), receiver operating characteristics simulations were performed by adding various levels of additive synthetic "brain" responses in order to examine the sensitivity and specificity of several previously proposed analytic approaches. Results: We found that the use of SS fNIRS channels as regressors of no-interest within a linear regression model was the best performing approach examined. Furthermore, we found that the addition of all available SS data, including all recorded channels and both hemoglobin species, improved the method performance despite the additional degrees-of-freedom of the models. When SS data were not available, we found that principal component filtering using a separate baseline scan was the best alternative. Conclusions: The use of multiple SS measurements as regressors of no interest implemented in a robust, iteratively prewhitened, general linear model has the best performance of the tested existing methods.
Keywords: functional near-infrared spectroscopy; noise removal; sensitivity–specificity; short-separation measurements; systemic physiological noise.
© The Authors. Published by SPIE under a Creative Commons Attribution 4.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.