Kathi Eickhoff^1,2,3, Arjan Hillebrand⁴, Maartje C. de Jong^1,2,5, Serge O. Dumoulin^1,2,3,6; ¹Spinoza Centre for Neuroimaging, Amsterdam, the Netherlands, ²Netherlands Institute for Neuroscience, Amsterdam, the Netherlands, ³Vrije Universiteit, Amsterdam, the Netherlands, ⁴Department of Clinical Neurophysiology and Magnetoencephalography Centre, Amsterdam UMC location Vrije Universiteit Amsterdam, the Netherlands, ⁵University of Amsterdam, the Netherlands , ⁶Utrecht University, the Netherlands

The visual system is organized retinotopically. In humans, this organization can be studied non-invasively by estimating the receptive fields of populations of neurons (population receptive fields; pRFs) with functional magnetic resonance imaging (fMRI). However, fMRI is too slow to capture the temporal dynamics of visual processing that operate on the scale of milliseconds. Other non-invasive techniques such as magnetoencephalography (MEG) provide this temporal resolution while lacking the spatial resolution to disentangle the precise locations of pRFs in the cortex. Here, we introduce a forward modeling approach that combines fMRI’s spatial- and MEG’s temporal resolution, enabling us to estimate pRFs on the neuronal timescale. Using fMRI, we estimated the participants pRFs using conventional pRF-modeling. With MEG, we measured event-related field (ERF) responses while the participants (N=5) viewed briefly presented (100ms) contrast-defined bar and circle shapes. Next, we combined the pRF models with a forward model that maps the pRFs’ electrical brain activity to the sensor level, to predict MEG’s sensor-level responses to the stimuli. We computed the goodness of fit between the predicted and measured ERF responses at each time point using cross-validated variance explained. To evaluate whether the recorded ERFs were optimally fitted by the pRFs, we moved the pRF positions away from the fMRI-estimated positions and refitted the new predictions to the experimental ERF data. We found that the fMRI-estimated pRFs explained up to 90% of the variance in individual MEG sensor’s ERF responses. The variance explained varied over time, but the pRF model accurately captured the ERF responses between 80ms to 250ms after stimulus presentation. Perturbing the pRF positions decreased the explained variance, suggesting that the ERF responses were driven by the pRFs. In conclusion, pRF models accurately capture event-related MEG responses, enabling routine investigation of the spatiotemporal dynamics of human pRFs with millisecond resolution.