Research map / Label-efficient and robust learning / Self-supervised and contrastive
MUSE: Multi-Scale Dense Self-Distillation for Nucleus Detection and Classification
Zijiang Yang, Hanqing Chao, Bokai Zhao, Yelin Yang, Yunshuo Zhang, Dongmei Fu, Junping Zhang, Le Lu, Ke Yan, Dakai Jin, Minfeng Xu, Yun Bian and 1 more
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 a decoder with multi-level hybrid nucleus representations to pretraining (ablation) | MUSE ViT-B · Fine-tuning overall | accuracy (%): 84.47 -> 86.26 | Improved |
| add multi-level context on top of the decoder (ablation) | MUSE ViT-B · Fine-tuning overall | accuracy (%): 86.19 -> 86.26 | Improved |
| add nucleus-level self-distillation loss to image-level loss (ablation) | MUSE ViT-B · Fine-tuning overall | accuracy (%): 82.34 -> 86.26 | Improved |
| use nucleus loss alone instead of both losses (ablation) | MUSE ViT-B · Fine-tuning overall | accuracy (%): 86.26 -> 84.19 | Hurt |
| pretrain with multi-scale 20x-40x patching instead of fixed 20x (ablation) | MUSE ViT-B · 20x evaluation fine-tuning | accuracy (%): 84.74 -> 86.21 | Improved |
| pretrain with multi-scale 20x-40x patching instead of fixed 40x (ablation) | MUSE ViT-B · 40x evaluation fine-tuning | accuracy (%): 85.47 -> 86.31 | Improved |
| pretrain at fixed 40x and evaluate at 20x (ablation) | MUSE ViT-B · 20x evaluation fine-tuning | accuracy (%): 86.21 -> 82.89 | Hurt |
| expand annotated patches to large-FoV samples during fine-tuning (ablation) | LFoV-MUSE ViT-B · OCELOT | average F1 (%): 69.00 -> 70.63 | Improved |
| add consistency regularization on unlabeled regions (ablation) | LFoV-MUSE ViT-B · OCELOT | average F1 (%): 70.63 -> 73.29 | Improved |
| pretrain with a larger field of view (LFoV-MUSE) | MUSE ViT-B · Fine-tuning overall | accuracy (%): 86.26 -> 87.51 | Improved |
| pretrain DINOv2 on TCGA_nu instead of LVD-142M for ViT-B | DINOv2 ViT-B · Fine-tuning overall | accuracy (%): 82.34 -> 81.92 | Hurt |
| pretrain DINOv2 on TCGA_nu instead of LVD-142M for ViT-S | DINOv2 ViT-S · Fine-tuning overall | accuracy (%): 80.76 -> 81.66 | Hurt |
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.
- SENC — SENC: Handling Self-collision in Neural Cloth Simulation (ECCV 2024) · better on 1 of 1 shared result
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
- UniCell: Universal Cell Nucleus Classification via Prompt Learning (AAAI 2024 · similarity 0.88)
- Towards Effective and Efficient Context-aware Nucleus Detection in Histopathology Whole Slide Images (AAAI 2026 · similarity 0.88)
- Benchmarking Self-Supervised Learning on Diverse Pathology Datasets (CVPR 2023 · similarity 0.88)
- Incremental Nuclei Segmentation from Histopathological Images via Future-class Awareness and Compatibility-inspired Distillation (CVPR 2024 · similarity 0.87)
- Sequential Attention-based Sampling for Histopathological Analysis (NeurIPS 2025 · similarity 0.85)
- Dynamic Entity-Masked Graph Diffusion Model for Histopathology Image Representation Learning (AAAI 2025 · similarity 0.85)
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.
