Abstract
In visual cortex, divisive normalization is modulated by features, such as orientation similarity: surround suppression is more potent when the surrounding orientation content is collinear with the orientation content of the center, compared to when the surround is orthogonal to the center. Correspondingly, neural responses to orthogonal stimuli are stronger than responses to stimuli in a collinear arrangement. Here, we provide a closer characterization of this feature-tuned aspect of normalization in human fMRI. Using full-field stimuli composed of alternating wedges filled with band-pass filtered noise enabled us to vary orientation differences between neighboring components in 15° increments between 0° and 90°, and examine whether orientation-tuned suppression strength changes as a function of orientation similarity. In agreement with previous research, we demonstrate that strongest BOLD response is seen for orthogonal and weakest BOLD response for collinear configurations, gradually decreasing as orientation similarity increases. We quantified the bandwidth of tuned normalization by fitting a Gaussian function to the tuned normalization curves for each subject. Having established this measure, we next asked how attention modulates the bandwidth of tuned normalization. The normalization model of attention posits that divisive normalization facilitates attentional modulation, and prior work has shown that stronger divisive normalization produces stronger attentional modulation. Would directing attention to the stimulus also affect the bandwidth of tuned normalization? In a second experiment, participants viewed similar stimuli while either performing a task at fixation (drawing attention away from the stimulus) or detecting color changes within the oriented stimulus (directing attention to the stimulus). We replicated our previous results – BOLD activation gradually increased from collinear to orthogonal. Our results demonstrated larger attentional effects for the collinear configuration, consistent with a previous finding from our lab. However, we find that tuned normalization bandwidth appears to remain unchanged with attention.