Research map / Label-efficient and robust learning / Self-supervised and contrastive
Multistain Pretraining for Slide Representation Learning in Pathology
Guillaume Jaume, Anurag J Vaidya, Andrew Zhang, Andrew Song, Richard J Chen, Sharifa Sahai, Dandan Mo, Emilio Madrigal, Long P Le, Faisal Mahmood
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add global cross-stain InfoNCE alignment (ablation) | Madeleine · Loss ablation (4 tasks avg) | average score: 75.5 -> 81.5 | Improved |
| add local GOT alignment on top of InfoNCE (ablation) | Madeleine · Loss ablation (4 tasks avg) | average score: 80.1 -> 81.5 | Improved |
| GOT local loss alone (ablation) | Madeleine · Loss ablation (4 tasks avg) | average score: 75.7 -> 75.5 | No clear effect |
| replace InfoNCE with cross-modal MSE (ablation) | Madeleine · Loss ablation (4 tasks avg) | average score: 80.1 -> 70.8 | Hurt |
| add intra-modal loss on top of InfoNCE + GOT (ablation) | Madeleine · Loss ablation (4 tasks avg) | average score: 81.5 -> 81.4 | No clear effect |
| use multiple attention heads (four) instead of one (ablation) | Madeleine · Breast tasks | average AUC: +2.1% | Improved |
| add learnable stain encoding (ablation) | Madeleine · Breast subtyping vs molecular subtyping | AUC | Mixed |
| replace ABMIL backbone with TransMIL (ablation) | Madeleine · Breast tasks | average AUC: -2.9% vs ABMIL, -5.0% vs MH-ABMIL | Hurt |
| use CONCH patch encoder instead of CTransPath (ablation) | Madeleine · Breast tasks | AUC: +10.5% AUC | Improved |
| fine-tune Madeleine encoder versus random initialization | Madeleine · MGH ER 3-class quantification | performance: +7.5% | Improved |
| GigaPath slide encoder versus plain mean of its patch embeddings | GigaPath · TCGA Breast subtyping (k=1) | macro-AUC: 59.7 -> 58.7 | Hurt |
Closest papers by meaning
- Unsupervised Foundation Model-Agnostic Slide-Level Representation Learning (CVPR 2025 · similarity 0.92)
- Transcriptomics-guided Slide Representation Learning in Computational Pathology (CVPR 2024 · similarity 0.90)
- PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization (NeurIPS 2025 · similarity 0.90)
- Morphological Prototyping for Unsupervised Slide Representation Learning in Computational Pathology (CVPR 2024 · similarity 0.89)
- MLLM-HWSI: A Multimodal Large Language Model for Hierarchical Whole Slide Image Understanding (CVPR 2026 · similarity 0.89)
- SlideChat: A Large Vision-Language Assistant for Whole-Slide Pathology Image Understanding (CVPR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
