Figure 5.

Analysis pipeline of the SGCP model. Contrastive learning-based graph feature embedding for the input image is performed at stage I (left). With the embedded features, brain region parcellation can be achieved by voxel-wise classification via supervised learning at stage II (right).