Research map / Detection and segmentation / Open-set and few-shot detection
Atlas is Your Perfect Context: One-Shot Customization for Generalizable Foundational Medical Image Segmentation
Ziyu Zhang, Yi Yu, Simeng Zhu, Ahmed Aly, Yunhe Gao, Ning Gu, Yuan Xue
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 |
|---|---|---|---|
| derive FM prompts from a registered atlas instead of oracle clicks (ablation) | nnInteractive · HaN-Seg | Dice: 42.42 -> 61.74 | Improved |
| add atlas-derived prompts to nnInteractive (ablation) | nnInteractive · Pengwin | Dice: 72.09 -> 94.56 | Improved |
| add adaptive fusion of atlas and FM predictions (ablation) | AtlasSegFM · Pengwin | Dice: 94.56 -> 95.50 | Improved |
| add adaptive fusion of atlas and FM predictions (ablation) | AtlasSegFM · HaN-Seg | Dice: 61.74 -> 63.71 | Improved |
| add fusion on top of vesselFM alone (ablation) | vesselFM · Fe-MRA | Dice: 55.28 -> 84.42 | Improved |
| add rigid pre-registration to atlas registration (ablation) | atlas registration · Pengwin | Dice: 63.24 -> 70.55 | Improved |
| add affine pre-registration after rigid (ablation) | atlas registration · Pengwin | Dice: 70.55 -> 82.75 | Improved |
| use mask prompt instead of box prompt from atlas (ablation) | AtlasSegFM · Pengwin | Dice: 88.96 -> 95.77 | Improved |
| use AtlasSegFM instead of Iris on organs-at-risk | one-shot segmentation · BrainRT organs-at-risk | Dice: 10.45 -> 77.07 | Improved |
| AtlasSegFM versus 5-click nnInteractive on AVT vessels | AtlasSegFM · AVT | Dice: 83.38 -> 81.34 | Hurt |
| AtlasSegFM (MedSAM2-based) versus 5-click nnInteractive on HaN-Seg | AtlasSegFM · HaN-Seg | Dice: 55.29 -> 52.61 | 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.
- UniverSeg — UniverSeg: Universal Medical Image Segmentation (ICCV 2023) · 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
- Training Like a Medical Resident: Context-Prior Learning Toward Universal Medical Image Segmentation (CVPR 2024 · similarity 0.86)
- SegVol: Universal and Interactive Volumetric Medical Image Segmentation (NeurIPS 2024 · similarity 0.84)
- Tyche: Stochastic In-Context Learning for Medical Image Segmentation (CVPR 2024 · similarity 0.84)
- Domain Generalized Medical Landmark Detection via Robust Boundary-Aware Pre-Training (AAAI 2025 · similarity 0.84)
- VISTA3D: A Unified Segmentation Foundation Model For 3D Medical Imaging (CVPR 2025 · similarity 0.84)
- Interactive Medical Image Segmentation: A Benchmark Dataset and Baseline (CVPR 2025 · 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.
