Fig. 1.

A deep learning framework for computing personalized functional networks. (a) Schematic diagram of the self-supervised deep learning model for identifying personalized functional networks (FNs), consisting of a time-invariant representation learning module (green) and a functional network learning module (yellow). (b) Network architecture of the time-invariant representation learning module to account for temporal misalignment of resting-state fMRI data across different scans. (c) Network architecture of the functional network learning module for the prediction of personalized FNs. The numbers underneath convolutional (c1, c2, c3, c4, c5, and c6, green blocks) and deconvolutional (d1, d2, and d3, yellow blocks) layers indicate their corresponding numbers of kernels with a stride of 1 or 2. The kernel size in all layers is set to 3×3×3. Different colormaps are used to differentiate the input fMRI and feature maps learned by the deep learning model. The colors are for illustrations only.