Fig. 2.

Dimensionality reduction with Principle Component Analysis (PCA). PCA allows for the reconstruction of each subject's (subj) imaging data through a linear combination of principle components (PC), which reflect representative patterns over the studied population. The coefficients of each PC used to reconstruct each subject's imaging data become the features for the classifier. For illustrative purposes, shown here is the PCA analysis for AD asymmetry maps (LI) over the whole cohort, whereas independent PCAs are performed for each cross-validation split.