Investigating orientation-tuned normalization in human visual cortex
OA Version
Citation
Abstract
The brain’s ability to parse sensory information into coherent scenes is
underpinned by numerous mechanisms. One such mechanism is divisive normalization, a neural computation which divides neural responses to stimulation by the pooled activity of neurons responding to neighboring stimuli. Normalization strength is modulated by context; similar features (e.g., orientation) suppress each other more than dissimilar features. This feature-tuned aspect of normalization appears essential for scene segmentation, yet it has been largely understudied. Using functional magnetic resonance imaging, this work explored the feature-tuned aspect of normalization in human early visual cortex across three experiments.
Experiment 1 examined the degree to which suppression in visual cortex is sensitive to orientation differences, by presenting observers (n = 10) with a bandpass- filtered noise stimulus and parametrically varying the orientation difference between its components. Consistent with prior work, parallel stimuli caused strongest BOLD signal suppression, while orthogonal stimuli caused least suppression. Extending previous findings, I obtained a measure of the orientation-tuned suppression function across a spectrum of orientation differences and estimated its bandwidth, providing a link between orientation difference and suppression strength.
Experiment 2 tested the effects of visual attention on orientation-tuned suppression. I examined whether attention alters its bandwidth by measuring the orientation-tuned suppression function of 10 observers while they attended either to the oriented stimulus or away from it. Attention increased the BOLD signal magnitude across all orientation differences (consistent with prior findings), but did not change other aspects of the function. Therefore, the specificity of orientation-tuned normalization appears unaltered by attention.
Finally, Experiment 3 explored the orientation tuning of normalization by measuring contrast response functions (CRF) to a central oriented stimulus suppressed by either a parallel or orthogonal high-contrast annulus (n = 10), and quantifying the resulting CRF with a normalization model. Consistent with non-human animal research, the relative orientation of center and surround impacted the CRF, with signs of stronger suppression seen with parallel surround. Interestingly, these modulations were spatially specific and largely limited to measurements from the boundary between the central and surrounding stimulus. Together, these results help extend our understanding of contextual effects in the human visual cortex.