Purchase this article with an account.
Aidan P. Murphy, Hiroshi Ban, Andrew E. Welchman; Decoding disparity-defined surface curvature in the human brain. Journal of Vision 2011;11(11):332. doi: https://doi.org/10.1167/11.11.332.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Binocular disparity provides a powerful cue to depth structure that is important for both identifying and interacting with objects. Previous electrophysiological studies have located neurons tuned to disparity-defined surface curvature in both ventral (IT) and dorsal (IPS) processing pathways; however the cortical circuits mediating the perception of curvature in the human brain remain largely unknown. To test human cortical responses to curved 3D surfaces, we performed pattern classification analyses of human fMRI data obtained during stereoscopic presentation of random dot stereograms (RDS). Participants viewed RDS containing disparity-defined concave and convex hemi-cylindrical surfaces at two disparity magnitudes (±6 and ±12 arcmin). In addition, we presented control stimuli that were constructed by randomly shuffling the locations of disparities within the cylindrical stimuli. The fMRI data were used to train a support vector machine (SVM) classifier to predict the disparity sign (crossed vs. uncrossed) of stimuli based on activation in visual cortex. Comparing SVM classification accuracies for curved versus randomized stimuli indicated that intermediate regions of extrastriate cortex (especially V3A) encode information that is diagnostic of global 3D shape rather than just disparity content. Further, in dorsal visual areas (V3, V3B and V7) we observed increased classification accuracy with increased disparity magnitude for curved surfaces compared to randomized versions. In contrast, ventral area LO appeared insensitive to changes in the magnitude of convexity; specifically, the SVM's accuracy in predicting the categorical convexity sign (convex vs. concave) remained the same when disparity magnitude differed between training and test stimuli. These results support the proposal that dorsal areas respond to metric depth structure while ventral areas encode depth configurations.
This PDF is available to Subscribers Only