Abstract
Many animals, including humans, are better at detecting and discriminating cardinal orientations (vertical and horizontal) than oblique orientations. Studies of non-human primates have suggested that the physiological loci of these so-called oblique effects are present in early visual areas. However, in studies of humans the neural signatures of the oblique effect have been elusive. Some recent fMRI studies have failed to find differences in mean BOLD responses to different orientations, while others have demonstrated a reverse oblique effect, finding that mean BOLD responses are greater for oblique than for cardinal orientations. Here, we present a novel, non-parametric approach to fMRI pattern analysis that allows us to measure the precision of feature representations in cortical activity patterns. Using this approach, we find robust oblique effects in activity patterns in early visual areas (V1-V3), with more sharply tuned pattern responses around cardinal orientations. This advantage in tuning precision for cardinal orientations corresponded well with behavioral discrimination thresholds, and was observed even when mean BOLD responses revealed a reverse oblique effect. Computational modeling of orientation-tuned cells suggests that behavioral discrimination thresholds, the oblique effect in pattern activity, and the reverse oblique effect in mean BOLD can be accounted for simultaneously. The results of these simulations rule out several potential explanations for the observed fMRI effects, such as differences in the gain or number of neurons across the orientation space, and support a recently proposed efficient encoding model in which neurons tuned to oblique orientations have wider tuning bandwidths than those tuned to cardinal orientations. These results help reconcile previous conflicting findings in fMRI, and support the notion that the neural locus of the oblique effect includes early visual cortex. Additionally, our study demonstrates how pattern similarity analysis provides a general, powerful approach for understanding neural representations of continuous feature spaces such as orientation.
Meeting abstract presented at VSS 2016