Sebastian Waz¹ (), Yalin Wang², Zhong-Lin Lu¹³; ¹New York University, ²Arizona State University, ³NYU Shanghai

In human subjects, visual neuroscience often depends, critically, on the estimation of retinotopic maps from BOLD fMRI. Intepreting brain activity as it relates to basic visual perception requires knowledge of the boundaries of (e.g.) areas V1, V2, and V3 on the cortex and the mapping of retinal eccentricity and polar angle onto these areas. Any approach to drawing the areas’ boundaries and the mappings therein is limited in accuracy by the resolution of the fMRI scanner and is susceptible to noise. These issues can be addressed using hierarchical Bayesian modeling, wherein information is pooled across subjects, with potential to enhance estimates in both precision and unbiasedness. We introduce a novel hierarchical Bayesian Gaussian process model (HGP) to estimate retinotopic maps for 162 subjects of the Human Connectome Project (Van Essen et al., 2013; Uğurbil et al., 2013), taking population receptive field model estimates (Dumoulin and Wandell, 2008) on these data as input to the HGP. Subjects were functionally aligned using optimally chosen wedges from their cortical ROIs. We found that the standard deviation across subjects of the estimated V1 ventral-dorsal boundary was 0.042 wedge-arcs on average, ranging from 0.036 to 0.049 wedge-arcs along the length of this boundary. For the contour line at 4° eccentricity, the standard deviation was 0.056 wedge-radii on average, ranging from 0.044 to 0.070 wedge-radii along this contour. At the population level, a systematic distortion of the eccentricity map was observed along the ventral V1/V2 boundary. Future work will be done to assess how the hierarchical framework in HGP mitigates noise-induced bias when generalized to a new data set. Since the HGP was fitted to relatively high-resolution 7T scanner data, it may prove especially valuable as a prior for 3T scanner data which are commonly used but require additional constraints to accurately estimate retinotopic maps.

Acknowledgements: Supported by National Eye Institute Grant R01 EY032125