Hemodynamic modelling improves population receptive field estimates

Population receptive field (pRF) modelling is a ubiquitous tool in sensory neuroscience for estimating the functional architecture of the human brain. Most pRF models assume a canonical hemodynamic response function (HRF) to account for neurovascular effects. But how does this assumption affect results in practice? Here, we concurrently fit the HRF within the pRF model. Using simulations, we demonstrate that this algorithm accurately estimated different ground truth HRFs used for synthesizing data. Moreover, concurrent fitting improves the accuracy of pRF estimates, especially for the complex Difference-of-Gaussian model. Next, we reanalyzed empirical datasets with different stimulus paradigms and scanning parameters. Concurrent fitting substantially reduced the proportion of implausibly small pRFs. Importantly, the best-fitting HRF differed substantially from canonical HRFs or from those measured independently. HRFs also peaked earlier in higher than early visual cortex, suggesting response nonlinearities. This variability need not only be hemodynamic but could result from neural factors. Critically, all these differences could produce spurious pRF estimates and alter the interpretation of reported findings.