Figure 3.

Classifier importance maps representing voxels that accurately distinguish internal mental states. (A) Subject-level importance maps showing individualized brain patterns representing voxels that are important for distinguishing neural signatures of attention to the Breath, MW, and Self (X = 0). For each task condition, importance values were computed by multiplying each voxel’s classifier weight for predicting the condition and the average activation during the condition (McDuff et al., 2009). The maps were thresholded at ±2 SD and displayed on the MNI152 template to identify the most important voxels for each participant. Orange importance voxel indicates positive z-scored average activation values, and blue importance voxels indicate negative z-scored average activation values. (B) To initially characterize the distribution of voxels that supported accurate classification, group importance frequency maps indicate the number of participants for which the voxel accurately distinguished each mental state. All important voxels were summed, irrespective of average positive or negative z-scored activation. Frequency maps were also computed that independently summed positive (Supplementary Figure S2A) and negative (Supplementary Figure S2B) z-scored activation voxels, as well as histograms of frequency counts (Supplementary Figures S2C–E). Note that the maximum frequency for any importance map was 10/14.