Reproduce “Self-Prompt SAM: Medical Image Segmentation via Automatic Prompt SAM Adaptation” 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
Bin Xie, Hao Tang, Dawen Cai, Yan Yan, Gady Agam
Researchers developed Self-Prompt-SAM, a framework that adapts the Segment Anything Model (SAM) for automated 3D medical image segmentation without manual prompt input. This approach integrates an automatic prompt generation module and depth-fused adapters, enabling SAM to handle diverse medical imaging modalities and 3D volumetric data. The framework achieved performance improvements, for example, outperforming nnUNet by 2.3% in Dice Similarity Coefficient on the AMOS2022 dataset.
Key insights
- Automating prompt generation via a Multi-Scale Prompt Generator (MSPGenerator) is effective for adapting foundation models like SAM to medical imaging, eliminating the need for manual prompt engineering.
- Adapter-based modifications (MAdapter, DFusedAdapter) allow a 2D-centric model like SAM to effectively process 3D medical volumetric data and handle diverse imaging modalities.
- Combining automatically generated bounding box, point, and mask prompts yields optimal segmentation performance, demonstrating the benefit of multi-modal prompting strategies in this context.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.