It can be seen that these predictive patterns consist of small subregions in which activity increases and decreases with larger angles. Specifically, some voxels have

higher responses for orientations >45° (yellow), whereas other voxels show higher responses for orientations <45° (blue). We compared our multivariate results to a more conventional univariate whole-brain analysis searching for correlations between stimulus orientation and the BOLD signal in each voxel BMS-354825 in vivo by using a parametric approach (Büchel et al., 1998). This analysis did not reveal any significant voxels (p < 0.0001, uncorrected, k = 5). Furthermore, a region of interest (ROI) analysis at a more liberal threshold of p < 0.05 revealed no univariate correlations with stimulus orientation in the early visual cortex (t = 1.29, p = 0.21), the lateral parietal cortex (t = 1.34, p = 0.20), and the medial frontal gyrus (t = 0.56, p = 0.58) as identified by our multivariate analysis (see above). This suggests that the results of the multivariate analysis are above and beyond what could have been obtained through univariate approaches. Our results so far suggest that information about the physical properties of the stimulus, i.e., its orientation,

is encoded in the early visual cortex as well as in higher brain regions such as the putative LIP. However, our model suggests that the orientation of the Gabor is not used directly to make the perceptual decision. Torin 1 datasheet What is used to make the choice is the decision variable DV. Thus, activity patterns in brain regions that are directly involved in perceptual decision-making should correlate with DV. We identified such brain regions by applying the same local information mapping procedure described above, but this time searching for representations of DV rather than orientation. We found significant information (p < 0.0001, k = 20, corrected for about multiple comparisons

at the cluster level, p < 0.001) about the model-derived decision variable in the left putative LIP (BA 7 [-24, −63, 48], t = 5.98, Figure 5A), the ACC (BA 32 [-3, 45, 24], t = 9.01, Figure 5C) and the precuneus (BA 23 [0, −39, 39], t = 6.57) but not the early visual cortex (see Figure S2 and Table S2 for complete results). In these regions distributed patterns of activity can be used to make linear predictions about the decision variable derived from the reinforcement learning model ( Figures 5A and 5C, right). Again, a univariate whole-brain analysis searching for correlations with DV revealed no significant voxels (p < 0.0001, uncorrected, k = 5). Furthermore, an ROI analysis revealed no significant (p < 0.05) univariate correlations with DV in the lateral parietal cortex (t = 0.64, p = 0.53) or the ACC (t = 0.75, p = 0.46) as identified by our multivariate analysis (see above). The physical stimulus orientation is correlated with the decision variable (DV) that is used by the model to make perceptual choices.