Ilona M Bloem^1,2, Louis N Vinke^3,4, Sam Ling^5,6; ¹Donders Institute for Brain, Cognition, and Behaviour, Radboud University, Nijmegen, the Netherlands, ²Netherlands Institute for Neuroscience, KNAW, Amsterdam, the Netherlands, ³Department of Psychiatry, Massachusetts General Hospital, Boston, MA, USA, ⁴Harvard Medical School, Boston, MA, USA, ⁵Psychological and Brain Sciences, Boston University, Boston, MA, USA, ⁶Center for Systems Neuroscience, Boston University, Boston, MA, USA

Contrast determines how strongly visual stimuli drive neural responses and, in turn, how well we can see. This contrast-response relationship is nonlinear, with gain control mechanisms acting to compress responses to high contrast stimuli, effectively setting the contrast range to which neural responses are most sensitive. Yet, capturing these saturating responses noninvasively in human fMRI has proven challenging. We previously showed that sustained contrast adaptation can reveal saturating contrast response functions (CRF) with fMRI, but this requires long interleaved top-up adaptation periods that are inefficient and limit experimental design. Here, we introduce a model-based fMRI approach to more flexibly and efficiently estimate saturating contrast responses: population contrast response function (pCRF) mapping. First, we benchmarked pCRF mapping against traditional deconvolved BOLD responses for stimuli presented with a fast event-related design. Participants viewed stimuli at nine fixed contrast levels (2.7–96% Michelson Contrast, 2 s duration), interleaved with 16% contrast top-up periods. Our results revealed tight correspondence in parameter estimates between the deconvolution-derived BOLD CRF and pCRF estimates, validating the model-based approach. We then tested the durability of pCRF mapping with a nonstandard design that allows for more flexibility and efficiency, but does not lend itself to traditional fMRI analyses. Participants were scanned while viewing stimuli of various contrast levels sampled from a mixed uniform and Gaussian distribution centered on 16% Michelson contrast. Stimuli were shorter (0.5 s) and no top-up adaptation periods were used, resulting in a continuous sequence of contrast events. Sampling from this biased distribution preserves coverage of the full contrast range while maintaining adaptation near the adapter contrast. Voxel-wise pCRF estimates could still be recovered from this unconventional design, demonstrating that model-based pCRF mapping technique broadens available experimental designs while still capturing the saturating nonlinearities that are fundamental to neural population responses.

Acknowledgements: NIH-EY028163 and NSF-1625552