Research map / Vision-language and multimodal / CLIP and zero-shot
MUSE: Harnessing Precise and Diverse Semantics for Few-Shot Whole Slide Image Classification
Jiahao Xu, Sheng Huang, Xin Zhang, Zhixiong Nan, Jiajun Dong, Nankun Mu
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 SMMO stochastic multi-view optimization (ablation) | MUSE (BM+TI) · CAMELYON 4-shot | ACC: 65.13 -> 71.18 | Improved |
| add SFSE fine-grained semantic enhancement (ablation) | BM + TI · CAMELYON 4-shot | ACC: 65.13 -> 65.16 | No clear effect |
| add SFSE on top of TI + SMMO (ablation) | MUSE · CAMELYON 4-shot | ACC: 71.18 -> 74.86 | Improved |
| add traditional text-patch interaction to base MIL (ablation) | Base MIL · CAMELYON 16-shot | ACC: 82.37 -> 86.61 | Improved |
| use cosine retrieval instead of random selection (ablation) | SMMO · CAMELYON 4-shot | ACC: 72.23 -> 74.86 | Improved |
| use cosine instead of L2 or random retrieval (ablation) | SMMO · CAMELYON 8-shot | ACC: 84.20/84.49 -> 84.01 | Mixed |
| use stochastic optimization instead of multi-mean (ablation) | SMMO · CAMELYON 4-shot | ACC: 70.33 -> 74.86 | Improved |
| use stochastic instead of multi-mean optimization (ablation) | SMMO · CAMELYON 8-shot | ACC: 84.08 -> 84.01 | No clear effect |
| build knowledge base with Qwen2-7B instead of Llama-3.2-1B (ablation) | MUSE · CAMELYON 4-shot | ACC: 70.44 -> 74.86 | Improved |
| build knowledge base with Qwen2-7B (ablation) | MUSE · CAMELYON 8-shot | ACC: 84.94 -> 84.01 | Mixed |
| use MUSE instead of FOCUS | MUSE vs FOCUS · CAMELYON 4-shot | ACC: 68.13 -> 74.86 | Improved |
| use MUSE instead of FOCUS | MUSE vs FOCUS · TCGA-NSCLC 4-shot | ACC: 79.14 -> 79.90 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ViLa-MIL — ViLa-MIL: Dual-scale Vision-Language Multiple Instance Learning for Whole Slide Image Classification (CVPR 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification (CVPR 2025 · similarity 0.91)
- Few-Shot Learning from Gigapixel Images via Hierarchical Vision-Language Alignment and Modeling (NeurIPS 2025 · similarity 0.91)
- Generalizable Whole Slide Image Classification with Fine-Grained Visual-Semantic Interaction (CVPR 2024 · similarity 0.90)
- MAPLE: Multi-scale Attribute-enhanced Prompt Learning for Few-shot Whole Slide Image Classification (NeurIPS 2025 · similarity 0.89)
- SlideChat: A Large Vision-Language Assistant for Whole-Slide Pathology Image Understanding (CVPR 2025 · similarity 0.89)
- PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization (NeurIPS 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.
