Analyzing human brain functional near-infrared spectroscopy data with transformer model
OA Version
Citation
Abstract
Functional Near-infrared Spectroscopy(fNIRS) is an optical neuroimaging technology measuring local cortical concentration changes of oxygenated and deoxygenated hemoglobin, which are associated with brain activities. For the robust estimation of hemodynamic brain responses, the current best practice is General Linear Model (GLM) with temporally embedded Canonical Correlation Analysis (tCCA), which has been tested to improve the performance by reducing nuisance signals from systemic physiology and motion. However, some challenging confounding signals from motions, including optode shifts or non-linear correlation between motion and physiological responses, are hard to be reduced or eliminated with traditional methods including tCCA-GLM. This study proposes a transformer-based model aiming to remove local and systemic physiological confounding signals and increase the detection accuracy of hemodynamic responses.