Reproduce “SAM-Med2D” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junlong Cheng, Jin Ye, Zhongying Deng, Jianpin Chen, Tianbin Li, Haoyu Wang, Yanzhou Su, Ziyan Huang and 7 more
SAM-Med2D adapts the Segment Anything Model (SAM) for 2D medical image segmentation through large-scale data curation and adapter-based fine-tuning, achieving substantial performance improvements across various modalities and anatomical structures. It demonstrates that a single click can yield segmentation quality comparable to original SAM's bounding box performance, improving efficiency for clinical workflows.
Key insights
- Adapter-based fine-tuning effectively bridges the domain gap between natural and medical images while preserving the original SAM's pre-trained knowledge.
- Achieving high-quality medical image segmentation with a single point prompt can significantly increase annotation efficiency for clinical workflows, potentially matching the performance of bounding box prompts.
- Curating a large-scale, diverse dataset spanning multiple modalities and anatomical structures is essential for successful domain adaptation of foundation models in medical imaging.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.