Divisive normalization unifies disparate response signatures throughout the human visual hierarchy

Abstract

Neural processing is hypothesized to apply the same mathematical operations in a variety of contexts, implementing so-called canonical neural computations. Divisive normalization (DN) is considered a prime candidate for a canonical computation. Here, we propose a population receptive field (pRF) model based on DN and evaluate it using ultra-high-field functional MRI (fMRI). The DN model parsimoniously captures seemingly disparate response signatures with a single computation, superseding existing pRF models in both performance and biological plausibility. We observe systematic variations in specific DN model parameters across the visual hierarchy and show how they relate to differences in response modulation and visuospatial information integration. The DN model delivers a unifying framework for visuospatial responses throughout the human visual hierarchy and provides insights into its underlying information-encoding computations. These findings extend the role of DN as a canonical computation to neuronal populations throughout the human visual hierarchy.

Attention drives human numerosity-selective responses

“…Previous visual-processing studies show that directing attention to a distracter outside the receptive field can reduce the response elicited by a preferred stimulus in the receptive field of a relevant neuron (Moore and Armstrong, 2003;Motter, 1993), consistent with divisive surround suppression. Surround suppression of visual spatial responses elicits negative fMRI responses in early visual cortex (Aqil et al, 2021;Harvey et al, 2013b;Zuiderbaan et al, 2012). In addition, the attention field acts as a gain field and that may have a suppressive surround (Puckett and Deyoe, 2015).…”

What has vision science taught us about functional MRI?

“…Since its inception and initial application, the pRF model has been used to understand topographic organisation in regards to other stimulus types and cortical regions: somatosensory cortex (Puckett et al, 2020;Schellekens et al, 2021;Wang et al, 2021), auditory cortex (Thomas et al, 2015), numerosity maps in parietal cortex (Harvey et al, 2013;Harvey & Dumoulin, 2017;van Dijk et al, 2021), sensory substitution (Hofstetter et al, 2021), and semantic space (Huth et al, 2012). Having a domain like vision science, in which some of the results are expected from prior knowledge, has provided a solid foundation for the extension of this approach to other domains; for example, the pRF model has been expanded to assess canonical computation of normalisation that is thought to occur throughout the brain (Aqil et al, 2021). This forward modelling approach provides an alternative to the subtraction approach (measuring contrast maps between stimuli, task, or groups) (Van Orden & Paap, 1997), affording greater generalisation and explanatory depth.…”

Intrasession and Intersession Reproducibility of Artificial Scotoma pRF Mapping Results at Ultra-High Fields

“…We tested for effects of nonlinear fitting on V1 pRF results and found that neither reproducibility nor scotoma size estimation benefitted from this approach. Since surround suppression effects are considered stronger in early visual areas, the implementation of novel pRF models, including both effects, as Aqil et al (2021) , could improve reproducibility for pRF parameters or scotoma estimation even further.…”

“…Further studies showed nonlinearities when only parts of a pRF are stimulated simultaneously ( Kay et al, 2013 ). Novel pRF models implemented by Aqil et al (2021) combine both surround suppression and spatial nonlinearities, with the promise of increased model power in pRF analyses. An extension to elliptical Gaussians is also possible, however, studies have shown that early visual cortex pRFs are almost circular ( Zeidman et al, 2018 ; Lerma-Usabiaga et al, 2021 ).…”

“…Further studies showed nonlinearities when only parts of a pRF are stimulated simultaneously (Kay et al, 2013). Novel pRF models implemented by Aqil et al (2021) combine both surround suppression and spatial nonlinearities, with the promise of increased model power in pRF analyses.…”