22nd CiNet Monthly Seminar: Kendrick Kay, “Ultra-high-resolution fMRI: the problem of veins and a potential solution”

CiNet Monthly Seminar

June 14, 2018
14:30 ~ 15:30
CiNet 1F Conference Room

“Ultra-high-resolution fMRI: the problem of veins and a potential solution”

Kendrick Kay
Assistant Professor
University of Minnesota

Host : Hiromasa Takemura (Kida Group)

Advances in MR hardware, pulse sequences, and reconstruction techniques have made it possible to measure fMRI signals at sub-millimeter resolution with extensive spatial coverage and relatively high signal-to-noise ratio. However, it is not clear whether ultra-high-resolution fMRI is a practical and robust tool for neuroscientists, given the challenge of stable in vivo imaging and the corrupting influence of veins. Here we acquire ultra-high-resolution fMRI data in human occipital, temporal, and parietal cortex during simple controlled visual experiments (7 T, T2*-weighted gradient-echo EPI, 0.8-mm isotropic, 2.2-s TR, 84 slices), and develop analysis tools to maintain the high spatial resolution of the data and to produce results that are interpretable with respect to quantitative physical units (e.g. mm). We demonstrate that BOLD responses can be measured at a fine scale with high accuracy and reliability. However, simple inspection of T2*-weighted intensities reveals that these responses, though measurable, are corrupted by a complex network of cortical veins. Venous effects are widespread, heterogeneously distributed, tend to be found in outer cortical depths, and are more prevalent in gyri than sulci. Moreover, we find clear relationships between veins and large BOLD response amplitudes and between veins and increased variability of BOLD response amplitudes; these effects make it challenging to draw valid inferences regarding local neural activity. We then move on to our efforts to develop a solution for the problem of veins. We show preliminary results of a data-driven analysis technique in which the signatures of veins are deduced and removed from the data. Applied to datasets in which the pattern of expected neural activity is known, the technique successfully removes a number of artifacts. The technique is simple, robust, and can be retrospectively applied to existing fMRI data. Thus, we believe that the technique may prove useful to fMRI studies in which spatial accuracy is of critical importance.