Limitations of 2-dimensional line-scan MRI for directly measuring neural activity

A 2D-line-scan MRI sequence has been reported to directly measure neural responses to stimuli (the "DIANA response"). Attempts to replicate the DIANA response have failed, even with higher field strength and more repetitions. Part of this discrepancy is likely due to a limited understanding of how physiological noise manifests in 2D-line-scan acquisition sequences. Specifically, it is unclear what the consequences are of breaking the assumption that the imaging substrate remains constant between each line acquisition. To answer this question, we collected 2D-line-scan data at 3T from human subjects viewing a blank screen. We found temporal fluctuations in the reconstructed time series that could easily be confused with neural responses to stimuli. These fluctuations were present both in the head and in the surrounding empty volume along the span of the phase-encoding direction from the head. The timing of these fluctuations varied systematically and smoothly along the phase-encoding direction. These artifacts are similar to well-known phase-encoding artifacts in EPI and GRE images, but are exacerbated due to longer acquisition times in the 2D-line-scan sequence (seconds vs. milliseconds). We explain these artifacts with a model that accounts for the acquisition sequence and incorporates time-varying contrast fluctuations and movement in the imaging substrate. Using the model, we quantify the amount of cortical- and scan-averaging one might need to reliably distinguish a DIANA response from noise, and show that navigator echoes might help in reducing phase-encode noise in the 2D-line-scan sequence.