Research map / Label-efficient and robust learning / Self-supervised and contrastive
AnatoMask: Enhancing Medical Image Segmentation with Reconstruction-guided Self-masking
Yuheng Li, Tianyu Luan, Yizhou Wu, Shaoyan Pan, Yenho Chen, Xiaofeng Yang
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 |
|---|---|---|---|
| replace random masking with reconstruction-guided AnatoMask | STU-Net-B SSL pretraining · TotalSegmentator | DSC (%): 83.6 -> 84.3 | Improved |
| use hierarchical decoder instead of simple decoder (ablation) | AnatoMask · TotalSegmentator | DSC (%): 83.4 -> 84.3 | Improved |
| AnatoMask masking on a simple decoder vs random masking (ablation) | simple decoder MIM · TotalSegmentator | DSC (%): 82.9 -> 83.4 | Improved |
| add self-distillation (EMA teacher) (ablation) | AnatoMask · TotalSegmentator | DSC (%): 83.7 -> 84.3 | Improved |
| mask high-loss regions instead of low-loss regions (ablation) | AnatoMask · TotalSegmentator | DSC (%): 83.6 -> 84.3 | Improved |
| use easy-to-hard instead of hard-to-easy masking dynamics (ablation) | AnatoMask · TotalSegmentator | DSC (%): 84.0 -> 84.3 | Improved |
| increase masking ratio from 0.6 to 0.8 or 0.9 (ablation) | AnatoMask · TotalSegmentator | DSC (%): 84.3 -> 84.1 (0.8), 84.2 (0.9) | Hurt |
| scale backbone from STU-Net-B to STU-Net-L and H | AnatoMask · TotalSegmentator | DSC (%): 84.3 -> 87.8 -> 89.0 | Improved |
| transfer AnatoMask pretraining vs SparK | STU-Net-B · AMOS22 (transfer) | DSC (%): 88.5 -> 89.1 | Improved |
| transfer AnatoMask pretraining vs SparK | STU-Net-B · FLARE22 (transfer) | DSC (%): 93.8 -> 94.0 | 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.
- UniMiSS — UniMiSS: Universal Medical Self-Supervised Learning via Breaking Dimensionality Barrier (ECCV 2022) · 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
- Learning Generalizable 3D Medical Image Representations from Mask-Guided Self-Supervision (CVPR 2026 · similarity 0.90)
- Revisiting MAE Pre-training for 3D Medical Image Segmentation (CVPR 2025 · similarity 0.88)
- Mine yOur owN Anatomy: Revisiting Medical Image Segmentation With Extremely Limited Labels (TPAMI 2024 · similarity 0.86)
- Masked-Diffusion Autoencoders for 3D Medical Vision Representation Learning (CVPR 2026 · similarity 0.86)
- Why Invariance is Not Enough for Biomedical Domain Generalization and How to Fix It (NeurIPS 2026 · similarity 0.85)
- Unleashing the Potential of SAM for Medical Adaptation via Hierarchical Decoding (CVPR 2024 · similarity 0.84)
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.
