Figure 5

Method overview (adapted from Zheng et al.²⁷). The WSI is first tiled into patches and feature extracted via self-supervised learning. The extracted features become the nodes of a graph network, which become the inputs to a graph-transformer classifier.