Research map / Label-efficient and robust learning / Test-time adaptation
You Point, I Learn: Online Adaptation of Interactive Segmentation Models for Handling Distribution Shifts in Medical Imaging
Wentian Xu, Ziyun Liang, Harry Anthony, Yasin Ibrahim, Felix Cohen, Guang Yang, Konstantinos Kamnitsas
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 online adaptation (PI+MI) to a frozen interactive model | ICNN* · BRATS T1 (Flair to T1), 10 clicks | Dice (%): 62.5 -> 88.0 | Improved |
| add Post-Interaction adaptation alone versus TSCA | online adaptation · ATLAS T1, 10 clicks | Dice (%): 64.0 -> 81.8 | Improved |
| add Mid-Interaction adaptation on top of Post-Interaction (ablation) | PI · BRATS T1, 10 clicks | Dice (%): 78.9 -> 88.0 | Improved |
| add CCG loss to Dice-Focal loss in Mid-Interaction (ablation) | PI+MI · BRATS T1, 10 clicks | Dice (%): 48.9 -> 88.0 | Improved |
| add Dice-Focal loss alongside CCG loss in Mid-Interaction (ablation) | PI+MI · BRATS T1, 10 clicks | Dice (%): 86.8 -> 88.0 | Improved |
| add Post-Interaction adaptation when MI is already used, 5-click budget (ablation) | MI · BRATS T2, 5 clicks | Dice (%): 89.8 -> 90.3 | Improved |
| add Post-Interaction when MI is already used, 10-click budget (ablation) | MI · brain MRI, final click | Dice (%) | Mixed |
| add class-limited mechanism to CCG loss (ablation) | PI+MI · ATLAS | Dice (%): 80.7 -> 86.0 | Improved |
| use Gaussian kernel instead of uniform kernel in CCG loss (ablation) | PI+MI · ATLAS | Dice (%): 79.8 -> 86.0 | Improved |
| add CCG loss during pretraining of interactive model (ablation) | interactive model · in-distribution and cross-distribution | Dice (%) | Improved |
| increase click budget per image from 10 to 50 (ablation) | TSCA · BRATS T1, early clicks | Dice (%): 34.4 -> 28.9 at click 1 | Hurt |
| use SAM-based Med-SA (frozen) instead of adapted U-Net | Med-SA · BRATS T1, 10 clicks | Dice (%): 62.5 -> 35.7 versus ICNN* | Hurt |
Closest papers by meaning
- Continuous Adaptation for Interactive Object Segmentation by Learning from Corrections (ECCV 2020 · similarity 0.87)
- Deep Interactive Segmentation of Medical Images: A Systematic Review and Taxonomy (TPAMI 2024 · similarity 0.87)
- Each Test Image Deserves A Specific Prompt: Continual Test-Time Adaptation for 2D Medical Image Segmentation (CVPR 2024 · similarity 0.86)
- TETRIS: Towards Exploring the Robustness of Interactive Segmentation (AAAI 2024 · similarity 0.85)
- Interactive Medical Image Segmentation: A Benchmark Dataset and Baseline (CVPR 2025 · similarity 0.85)
- UniPro: Unified Multi-Mode Medical Image Segmentation from 2D Images to 3D Volumes via Propagation (NeurIPS 2026 · 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.
