Research map / Detection and segmentation / Open-set and few-shot detection
SegMoTE: Token-Level Mixture of Experts for Medical Image Segmentation
Yujie Lu, Jingwen Li, Sibo Ju, Yanzhou Su, He Yao, Yisong Liu, Min Zhu, Junlong Cheng
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 Progressive Prompt Tokenization (auto prompt tokens) (ablation) | SegMoTE · ISLES | Dice: 59.00 -> 65.28 | Improved |
| add Progressive Prompt Tokenization (ablation) | SegMoTE · ISIC2018 | Dice: 84.87 -> 87.68 | Improved |
| add Progressive Prompt Tokenization (ablation) | SegMoTE · SZ-CXR | Dice: 75.20 -> 77.63 | Improved |
| add Progressive Prompt Tokenization (ablation) | SegMoTE · ISIC2017 | Dice: 76.66 -> 77.62 | Improved |
| vary number of PPT learnable queries (2, 4, 8) (ablation) | SegMoTE · ISIC2017/2018, SZ-CXR, ISLES | Dice: ISLES 65.28 / 64.97 / 65.00 | No clear effect |
| use 12 experts instead of 4 (4 modalities) (ablation) | SegMoTE MoTE · out-of-domain subset | Dice | Hurt |
| use 8 or 12 experts instead of 4 (7 modalities) (ablation) | SegMoTE MoTE · out-of-domain subset | Dice | Hurt |
| unfreeze SAM mask decoder and train jointly with MoTE and PPT (ablation) | SegMoTE · ISLES | Dice: 77.30 -> 78.05 | Improved |
| unfreeze SAM mask decoder and train jointly with MoTE and PPT (ablation) | SegMoTE · SegThor | Dice: 83.39 -> 83.91 | Improved |
| SegMoTE vs IMIS on out-of-domain box prompts | IMIS · ISLES | Dice: 71.24 -> 77.30 | Improved |
| SegMoTE with 17M trainable params on 0.15M masks vs larger-data baselines | SAM-Med2D / IMIS · SegThor | Dice: 80.52 -> 83.39 | Improved |
Closest papers by meaning
- SegVol: Universal and Interactive Volumetric Medical Image Segmentation (NeurIPS 2024 · similarity 0.89)
- MedSAMix: A Training-Free Model Merging Approach for Medical Image Segmentation (AAAI 2026 · similarity 0.89)
- Simple-ViLMedSAM: Simple Text Prompts Meet Vision-Language Models for Medical Image Segmentation (CVPR 2026 · similarity 0.89)
- UniPro: Unified Multi-Mode Medical Image Segmentation from 2D Images to 3D Volumes via Propagation (NeurIPS 2026 · similarity 0.88)
- Enhancing SAM with Efficient Prompting and Preference Optimization for Semi-supervised Medical Image Segmentation (CVPR 2025 · similarity 0.88)
- Unleashing the Potential of SAM for Medical Adaptation via Hierarchical Decoding (CVPR 2024 · similarity 0.87)
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.
