Differentiating BOLD and non-BOLD signals in fMRI time series using cross-cortical depth delay patterns.

Over the past two decades, rapid advancements in magnetic resonance technology have significantly enhanced the imaging resolution of functional Magnetic Resonance Imaging (fMRI), far surpassing its initial capabilities. Beyond mapping brain functional architecture at unprecedented scales, high-spatial-resolution acquisitions have also inspired and enabled several novel analytical strategies that can potentially improve the sensitivity and neuronal specificity of fMRI. With small voxels, one can sample from different levels of the vascular hierarchy within the cerebral cortex and resolve the temporal progression of hemodynamic changes from parenchymal to pial vessels. We propose that this characteristic pattern of temporal progression across cortical depths can aid in distinguishing neurogenic blood-oxygenation-level-dependent (BOLD) signals from typical nuisance factors arising from non-BOLD origins, such as head motion and pulsatility. In this study, we examine the feasibility of applying cross-cortical depth temporal lag (CortiLag) patterns to automatically categorize BOLD and non-BOLD signal components in modern-resolution BOLD-fMRI data. We construct an independent component analysis (ICA)-based framework for fMRI de-noising, analogous to previously proposed multi-echo (ME) ICA, except that here we explore the across-depth instead of across-echo dependence to distinguish BOLD and non-BOLD components. The efficacy of this framework is demonstrated using visual task data at three graded spatiotemporal resolutions (voxel sizes = 1.1, 1.5, and 2.0 mm isotropic at temporal intervals = 1700, 1120, and 928 ms). The proposed CortiLag-ICA framework leverages prior knowledge of the spatiotemporal properties of BOLD-fMRI and serves as an alternative to ME-ICA for cleaning moderate- and high-spatial-resolution fMRI data when multi-echo acquisitions are not available.