Fig. 5: Ablation analysis and modality contribution in recurrence prediction. | npj Digital Medicine

Fig. 5: Ablation analysis and modality contribution in recurrence prediction.

From: Deep learning-enabled multiphoton microscopy predicts colorectal cancer recurrence from routine FFPE specimens

Fig. 5

a Attention weight distribution from Modality A (TPEF) and Modality B (SHG) in correctly and incorrectly predicted cases. b Schematic of the ablation setup, where the Modality A branch was removed to evaluate the independent contribution of SHG. c ROC curves comparing the full MPMRecNet model with single-modality variants. d PR curves for the same models. e UMAP-based dimensionality reduction of features from each model. f Sankey diagram comparing prediction outputs from Modality A, Modality B, and MPMRecNet with the ground truth labels.

Back to article page