Edge-Based Graph Neural Networks for Cell-Graph Modeling and Prediction

Identification and classification of cell-graph features using graph-neural networks (GNNs) has been shown to be useful in digital pathology. In this work, we consider the role of edge labels in cell-graph modeling, including histological modeling techniques, edge aggregation in GNN architectures, and edge label prediction. We propose EAGNN (Edge Aggregated GNN), a new GNN model that aggregates both node and edge label information to take advantage of topological information about cellular data and facilitate edge label prediction. We introduce new edge label features that improve histological modeling and prediction. We evaluate our EAGNN model for the task of detecting the presence and location of the basement membrane in oral mucosal tissue, as a proof-of-concept application.