Fig. 3
From: A graph neural network framework for mapping histological topology in oral mucosal tissue
Training procedure for a GraphSage model to identify the edges crossing the BM. Ground truth annotations for nuclei locations, cell class and BM localisation were integrated with a d-distance limited Delaunay triangulation to create a node and edge labelled ground truth cell-graph. Nodes were labelled according to cell class and edges were labelled as crossing (red edges) or not crossing (black edges) the BM. Subsequently the ground truth cell-graphs were employed for training GraphSage for edge classification