Direct imaging of neuronal activity: a multivariate perspective on DIANA data in human subjects

Functional magnetic resonance imaging (fMRI) relies upon the blood-oxygenation-level-dependent (BOLD) signal as a proxy for neuronal activity. Whole-brain coverage and millimetric resolution have propelled this modality as a workhorse for human neuroscience. However, the signal fluctuations induced by the hemodynamic response are slow at the timescale of seconds. Therefore, a direct imaging contrast of neuronal activity at the electrophysiological timescale of milliseconds would be a major discovery and breakthrough. Recent work termed DIANA has demonstrated such a candidate mechanism in rodent fMRI, but subsequent replication studies have mostly failed both in animal and human experiments. In this work, we propose to re-analyze the dataset publicly available in human subjects at 7T. Given the low signal-to-noise ratio, we apply a multivariate analysis to the voxels in the region-of-interest of the visual cortex where activity can be expected. We report a weak correlate of possible activity in the visual cortex in one subject and one paradigm. Several guidelines are formulated for future experiments in order to improve the data quality to validate or invalidate the DIANA contrast mechanism.