Fig. 2: In-depth evaluation and analysis of HGTs: model comparison and component analysis. | npj Precision Oncology

Fig. 2: In-depth evaluation and analysis of HGTs: model comparison and component analysis.

From: Cell graph analysis in hepatocellular carcinoma: predicting local recurrence and identifying spatial relationship biomarkers

Fig. 2: In-depth evaluation and analysis of HGTs: model comparison and component analysis.

A A comparison of HGTs with other state-of-the-art models, including DeepGraphSurv, Patch-GCN, and TEA-graph. B An analysis of the impact of various node features within HGTs, encompassing the Nucleus/Cell area ratio (Nucleus/Cell), Coordinates (Coord), a combination of Coordinates and Morphological Features (Coord + Morph Features), and a fusion of Coordinates with the Nucleus/Cell area ratio (Coord + Nucleus/Cell). C A study exploring the effects of varying the number of GNN layers in HGTs and their impact on model performance. D Comparison of the effects of incorporating PairNorm and Residual Connections into HGTs, evaluating their impact on model performance. E An analysis of different graph construction methods used in HGTs and their respective impacts on the model’s effectiveness. F Experiments evaluating different feature fusion strategies in the second stage, specifically comparing GAT, HyperGraph, and Transformer.

Back to article page