Fig. 5.

Personalized functional networks (FNs) identified by the DL model are with high reproducibility and associated with behavior. (a) Personal FNs including DMN and FPN of three randomly selected individuals using HCP REST1 and REST2 rsfMRI data. The isoline of value 0.15 in each FN is demonstrated by the black contour to facilitate the visual comparison across sessions and subjects. (b) FN based identification procedure. Given a query set of FNs from one target subject (sbj), we computed the similarity $r_i$ between its FNs and all the sets of FNs in the Database. The matched identity $I{D}^{\ast }$ is the one with the highest similarity. (c) The identification rate when 17 FNs (All FNs) or combined fronto-parietal FNs (FP FNs) were used for the identification. (d) Prediction accuracy of 13 cognitive measures based on personalized FNs computed by the proposed DL model. Violin plots show the distribution of prediction accuracy of 100 repetitions for each measure. Asterisks denote the cognitive measures that are significantly associated with personalized FNs.