Purchase this article with an account.
Daniel Drucker, Geoffrey Aguirre; Integral versus separable perceptual dimensional pairs are reflected in conjoint versus independent neural populations. Journal of Vision 2009;9(8):805. doi: 10.1167/9.8.805.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Some visual properties of shapes are apprehended separately, others as a composite; these are termed separable and integral dimensions. We hypothesize that integral dimensions are represented by populations of neurons representing dimensions conjointly, while separable dimensions are represented by independent populations.
Using a carry-over fMRI design (Aguirre, 2007) we measured recovery from neural adaptation associated with shape changes along a single perceptual dimension or combined across two. Changes along both dimensions produce recovery from adaptation that is the sum of the recovery for each dimension in the case of separate representation, but is subadditive in the conjoint case. We studied two sets of shapes that varied along two parameterized dimensions. For a set of ineffable “popcorn” shapes, the two dimensions were behaviorally integral; while for “lens” shapes the dimensions of curvature and thickness were separable (replicating Op-de-Beeck, et al. 2001, 2003).
Five participants were studied using fMRI while they viewed a continuous stream of these shapes. Two covariates modeled linear recovery from adaptation proportional to shape changes within the two stimulus spaces. We identified voxels in ventral temporal cortex in each subject that demonstrated a linear recovery from adaptation along both of the “popcorn” dimensions and both of the “lens” dimensions. A “Euclidean Contraction” covariate modeled variance attributable to sub-additive recovery from adaptation for combined changes in the shape space. The “lens” shape space had substantially reduced loading upon this covariate as compared to the “popcorn”, indicating that the shape space defined by behaviorally separable dimensions was associated with a more “feature” based neural representation within ventral extrastriate cortex. This suggests that integral and separable dimensions are represented by neural populations which differ in their tuning for stimulus properties. More generally, this approach may be used to characterize tuning of neural populations for stimulus features using fMRI.
This PDF is available to Subscribers Only