Fig. 1: The flow chart of the analysis.

Time series for each subject are extracted from rsfMRI according to the AAL atlas. The Pearson correlation is used to get the functional connectivity matrix. Next, the upper triangle of the functional connectivity matrix is stretched into a one-dimensional vector to get each subject’s feature vector. The feature vectors are concatenated to obtain input data matrices. A brain age prediction model is trained and five-fold cross-validated on the training set. Finally, brain-PAD scores are obtained by applying the model in the hold-out validation set and test set.