Abstract
One of our primary goals in using functional magnetic resonance imaging (fMRI) is to be able to infer neuronal response properties from the fMRI signal. The mismatch in spatial scale makes this challenging: our best imaging resolution element contains hundreds of thousands of neurons. Here, we use size tuning in V1 to study the interaction between imaging resolution and inference of the underlying neural population response. We used 7 Tesla fMRI with 1 millimeter (isotropic) resolution to measure responses to sinusoidal luminance gratings with 0.25, 0.5, 1.0, 2.0 or 4.0 degree diameters centered at 3 degrees eccentricity. Stimuli were presented in a simple block design (8 cycles 12 sec stimulus/12 sec rest per scan) while subjects performed a contrast discrimination task; each scan contained one stimulus size. Seeds for computing regions of interest were located at the center of mass of the response to the smallest stimulus. Average response was then calculated for ROIs of increasing diameter; ROI diameter was computed on the cortical surface and translated back to the functional data. For ROIs less than 3 mm in diameter, the size-tuning function matched neural response predictions – the strongest response was to stimuli 1 degree in diameter, falling off as stimulus size increased until the response to a 4-degree (diameter) stimulus was 60% of the maximum (1-degree) response. For larger regions of interest the peak response occurred for larger stimuli: 4- and 6-millimeter (diameter) ROIs responded most strongly to 2-degree stimuli, for example. Predicting this result from spatial pooling by fMRI of the underlying neuronal population responses is straight-forward. However, it nicely illustrates the difficulty of inferring neuronal response properties from fMRI response: without an accurate encoding model with which to interpret the data, a low-resolution experiment would conclude that V1 neurons prefer larger stimuli than they actually do.
Meeting abstract presented at VSS 2015