Figure 4
Fitting the neural responses based on the extracted semantic components. (a) Prediction accuracy (measured as Pearson correlation between predicted and observed HFB responses) for various time shifts of the semantic components. Prediction accuracy was cross-validated over five folds. Testing different time shifts was done to account for a delay in neural processing of the high-level visual information. Top panel shows the mean prediction accuracy for all significant electrodes (\(p<1\times {10}^{-5}\), Bonferroni corrected for the total number of electrodes) per time shift. Shading represents standard error of the mean. Bottom panel shows the number of electrodes with a significant fit per time shift. The red bar indicates the model with top average prediction accuracy as well as highest number of electrodes with a significant fit, which corresponds to a time shift of 320 ms. (b) Cortical map of the prediction accuracy for the time shift of 320 ms (\(p<1\times {10}^{-5}\), Bonferroni corrected for the total number of electrodes). Left panel shows a volume-based plot. Each cross-validated prediction accuracy value was assigned to the center coordinate of the corresponding electrode and projected to the MNI common space. Individual electrode locations were normalized to the MNI space using subject-specific affine transformation matrices obtained with SPM8. For the visualization purposes a 2D Gaussian kernel (FWHM = 8 mm) was applied to the coordinate on the MNI brain volume corresponding to the center of the electrode, so that the projected values (e.g. prediction accuracy) faded out from the center of the electrode toward its borders. Right panel shows a surface-based plot of the cross-validated prediction accuracy at the time shift of 320 ms. It shows the same prediction accuracy values as the left panel but projected on the surface for a better display of the fit in lateroocipital and fusiform cortices. See “Methods” for more details about the projection to a uniform regular grid on the average surface.
