Mapping sequences can bias population receptive field estimates

Abstract

11Population receptive field (pRF) modelling is a common technique for estimating the stimulus-12 selectivity of populations of neurons using neuroimaging. Here, we aimed to address if pRF 13properties estimated with this method depend on the spatio-temporal structure and the 14 predictability of the mapping stimulus. We mapped the polar angle preference and tuning width of 15 voxels in visual cortex (V1-V4) of healthy, adult volunteers. We compared sequences orderly 16 sweeping through the visual field or jumping from location to location employing stimuli of 17 different width (45° vs 6°) and cycles of variable duration (8s vs 60s). While we did not observe 18 any systematic influence of stimulus predictability, the temporal structure of the sequences 19 significantly affected tuning width estimates. Ordered designs with large wedges and short cycles 20 produced systematically smaller estimates than random sequences. Interestingly, when we used 21 small wedges and long cycles, we obtained larger tuning width estimates for ordered than random 22 sequences. We suggest that, ordered and random mapping protocols show different susceptibility 23 to other design choices such as stimulus type and duration of the mapping cycle and can produce 24 significantly different pRF results. 25 26 27 28 adaptively redistributed to optimize the sampling of the visual stimulus according to task 60

Highly accurate retinotopic maps of the physiological blind spot in human visual cortex

“…Some of these used analysis approaches (e.g. using ridge regression) that account for the considerable spatiotemporal correlation in conventional pRF stimulation paradigms (Alvarez et al, 2015; Infanti & Schwarzkopf, 2020). We decided against this, as these modelling approaches drastically increase processing time.…”

“…While such fits must be treated with caution, this can nevertheless be a great advantage. Forward‐modelling is probably also less susceptible to spatiotemporal correlations in the stimulus sequence (although see [Alvarez et al, 2015; Infanti & Schwarzkopf, 2020; Linhardt et al, 2021]). Moreover, forward‐modelling is based on theoretical models for the shape and function of pRFs.…”

Highly accurate retinotopic maps of the physiological blind spot in human visual cortex

“…Some of these used analysis approaches (e.g. using ridge regression) that account for the considerable spatiotemporal correlation in conventional pRF stimulation paradigms (12, 37). We decided against this, as these modelling approaches drastically increase processing time.…”

“…when studying subtle changes in pRF parameters around the blind spot. Randomized stimulation sequences that break spatiotemporal correlations might further enhance the accuracy of these maps, although they are also associated with poorer signal-to-noise ratios (23, 37, 38).…”

Cortical Depth-Dependent Modeling of Visual Hemodynamic Responses

“…At that time, fMRI resolutions were not fine enough to resolve cortical depths reliably, so averaging over the different depths not only made the resulting calculations much more tractable but also did not affect the accuracy of the predictions of the model compared to experimental data. Moreover, it was able to produce key findings such as astrocyte-induced hemodynamic time delays (Pang et al, 2017), origins of resting-state fMRI spectrum (Pang and Robinson, 2019), techniques for imaging ocular dominance and orientation preference maps (De Oliveira et al, 2019), and improvements in population receptive field estimation (Infanti and Schwarzkopf, 2020).…”