HER2 status scoring on H&E and IHC pathology images (test)
94AccuracyViT-based tumor and stain classifier
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ViT-based tumor and stain classifierReference=This study, Method description=ViT-based tumor and stain classifier, and stain segmentation, Evaluation protocol=four score classification (0, 1+, 2 + and 3 +)2025.12 | 94 | 98.2 | 100 | — | — | — | — | — | 0.991 | 1 | — | — | |
| Corr-A-NetReference=[31], Method description=Feature Learning framework based on Correlational Attention Neural Network for HER2-scoring2025.12 | 93 | — | — | 98 | — | — | — | — | — | — | — | 0.85 | |
| DenseNet201/GoogleNet/MobileNet_v2/ViTReference=[43], Method description=classification and random forest for HER2 score prediction2025.12 | 91.15 | — | — | — | — | — | — | — | — | — | — | — | |
| HE-HER2NetReference=[28], Method description=A transfer learning approach with Grad-CAM for explainability2025.12 | 87 | 88 | 86 | 98 | — | — | — | — | — | — | — | — | |
| end-to-end ConvNeXtReference=[42], Method description=end-to-end ConvNeXt network utilizing low-resolution IHC images2025.12 | 83.56 | — | — | 91.79 | — | — | — | — | 0.8352 | — | — | — | |
| WSMCLReference=[27], Method description=weakly supervised multi-modal contrastive learning on H&E and IHC2025.12 | 78.3 | — | — | — | 0.941 | 0.945 | 0.762 | 0.781 | — | — | — | — | |
| 3-stage attention modelReference=[45], Method description=A 3-stage model for weakly localizing features, an attention module, and HER2 expression level proximity computation2025.12 | — | 92.2 | — | 92.02 | — | — | — | — | — | 0.876 | 0.959 | — | |
| ViT-based pipelineReference=[46], Method description=ViT-based pipeline for HER2 scoring using H&E2025.12 | — | — | — | 90 | — | — | — | — | — | — | — | — | |
| Weak-supervised MoCo-v2 modelReference=[40], Method description=A weak-supervised model with MoCo-v2 contrastive learning2025.12 | — | — | — | 85 | — | — | — | — | — | — | — | — |