The Attentional Field Revealed by Single-Voxel Modeling of fMRI Time Courses

Abstract

The spatial topography of visual attention is a distinguishing and critical feature of many theoretical models of visuospatial attention. Previous fMRI-based measurements of the topography of attention have typically been too crude to adequately test the predictions of different competing models. This study demonstrates a new technique to make detailed measurements of the topography of visuospatial attention from single-voxel, fMRI time courses. Briefly, this technique involves first estimating a voxel's population receptive field (pRF) and then "drifting" attention through the pRF such that the modulation of the voxel's fMRI time course reflects the spatial topography of attention. The topography of the attentional field (AF) is then estimated using a time-course modeling procedure. Notably, we are able to make these measurements in many visual areas including smaller, higher order areas, thus enabling a more comprehensive comparison of attentional mechanisms throughout the full hierarchy of human visual cortex. Using this technique, we show that the AF scales with eccentricity and varies across visual areas. We also show that voxels in multiple visual areas exhibit suppressive attentional effects that are well modeled by an AF having an enhancing Gaussian center with a suppressive surround. These findings provide extensive, quantitative neurophysiological data for use in modeling the psychological effects of visuospatial attention.

Spatial sampling in human visual cortex is modulated by both spatial and feature-based attention

“…Second, it extends the utility of such attentional gain field models to situations where attention is dynamically deployed over space and time during the mapping of the pRF (Kay et al, 2015). In agreement with earlier reports (Klein et al, 2014; Puckett and DeYoe, 2015), we found that the best-fitting model implemented comparable attentional gain field sizes across visual regions. This strongly points to spatial attention being implemented as a global influence across visual cortex.…”

“…Two recent studies showed that in early visual areas, spatial attention shifted pRFs away from the attended location, but toward the attended location in higher visual areas (de Haas et al, 2014; Vo et al, 2017). Other studies showed that in precisely these visual regions, both the pRF and the attentional gain field are composed of a suppressive surround in addition to their positive peak (Zuiderbaan et al, 2012; Puckett and DeYoe, 2015). We leave the question whether these suppressive surrounds could explain such repulsive shifts in lower visual cortex for future research.…”

Extensive tonotopic mapping across auditory cortex is recapitulated by spectrally-directed attention, and systematically related to cortical myeloarchitecture

“…In future research it will be informative to examine interactions with several frontal regions whose potential analogues are known to have direct feedforward and feedback connections in macaque monkeys (Romanski and Goldman-Rakic, 2002), and where in ferret there are clear modulatory influences on primary and non-primary auditory cortex during learning (Atiani et al, 2014;Shamma and Fritz, 2014). Similar to recent work in vision (Klein et al, 2014;Puckett and DeYoe, 2015), it will also be useful to establish the shape of the attentional population receptive field, and how this varies across auditory areas and relates to stimulus-driven auditory population receptive field size (Thomas et al, 2015). Finally, following on our own pilot work, it will be exciting to explore whether higher-level auditory regionalization may follow along some of the 'fault lines' revealed by shared local tonotopic and myelin gradients, and whether or not more sophisticated and finegrained spectral attentional manipulations may reveal a relationship between the degree of attentional malleability and underlying cortical architecture and circuitry.…”

Bayesian population receptive field modeling in human somatosensory cortex

“…center. Because the DoG model has additional parameters compared to the Gaussian 465 model (i.e., is fundamentally more complex), testing for the most appropriate model 466 type often involves applying some sort of information criterion after the pRF analysis 467 (e.g., Akaike's information criterion) (Akaike, 1974;Puckett and DeYoe, 2015); 468 however, one of the strengths of the Bayesian pRF modeling approach is that the 469 estimation procedure directly provides an approximation of the model evidence -the 470 negative variational free energy (F). The free energy term increases with model 471 accuracy and decreases with model complexity, and can hence be used to compare 472 pRF models in order to determine the most accurate, least complex explanation of the 473 data.…”