Fig. 7 | Scientific Reports

Fig. 7

From: Pan-cancer frozen section classification using a soft mixture of experts vision transformer under weak supervision

Fig. 7The alternative text for this image may have been generated using AI.

Calibration curves of three baseline models on the test set. Calibration curves showing the relationship between predicted malignancy probability and the observed proportion of malignant cases for (A) Soft MoE-ViT, (B) Standard ViT, and (C) ResNet50. The dashed diagonal line represents perfect calibration. Among the three models, Soft MoE-ViT exhibits the closest alignment to the ideal calibration curve and achieves the lowest Brier Score (0.167), indicating more reliable probability estimates. These results highlight the importance of well-calibrated confidence outputs for intraoperative frozen-section decision-making.

Back to article page