Figure 1. Overview of the proposed method.

After brain surface information is extracted, resampling is performed and noise is removed. Z-scores are then computed for each subject’s cortical thickness with respect to the cognitively normal (CN) subjects as their individual ‘cortical atrophy’. The similarity of any pair of subjects is defined using a correlation coefficient between cortical atrophy levels of the subjects. Similarity is therefore more sensitive to the shape of the cortical atrophy patterns, rather than overall levels (corr_AB > corr_AC). Modular organization of subjects was extracted using the defined similarity. Note: cortical atrophy plots in the second and third boxes of the overall pipeline are depicted as an example for illustration purposes.