Adaptive hemodynamic response function to optimize differential temporal information of hemoglobin signals in functional near-infrared spectroscopy

It has been nearly twenty years since functional near-infrared spectroscopy (fNIRS) was first applied to assessing human brain functions. It has now become widely accepted as a common functional imaging modality with more than 100 publications of fNIRS-related scientific literature annually. However, universal analytical methods for fNIRS data have yet to be established. Although not frequently mentioned, temporal analysis of fNIRS data also poses a technical challenge: how oxygenated and deoxygenated hemoglobin (Hb) signals should be treated. With its analogy to fMRI, a general linear model (GLM) with regression to a canonical hemodynamic response function (HRF) has often been used. However, the Hb parameters do not necessarily follow the same behavior as the BOLD signal: rather, we often encounter different temporal profiles for the two Hb signals. Here we introduce adaptive methods to find the optimal HRF for temporal analysis of fNIRS data. Application of the GLM with regression to a temporally optimized HRF on the functional activation data during an overt confrontation naming task revealed different temporal structures for oxy-Hb and deoxy-Hb signals, with the latter having substantial temporal delay. However, when the temporally optimized HRF was used, the two parameters yielded reasonably compatible activation patterns including activation in classical language-related areas of the left hemisphere. These results suggest the potential use of the GLM with regression to an adaptive HRF to fully utilize temporal information of both Hb parameters.