MedSAM3 v1: Delving into Segment Anything with Medical Concepts (LoRA Weights)

This repository contains the v1 LoRA weights for MedSAM3.

📊 Model & Dataset Statistics

We constructed a large-scale dataset uniformly sampled to ensure diversity and robustness. The model covers diverse medical modalities:

• Radiology: CT, MRI, PET, X-ray
• Optical/Microscopic: Microscopy, Histopathology, Dermoscopy, OCT, Cell
• Video/Procedure: Ultrasound, Endoscopy, Surgery video

Dataset Scale:

• 658,094 Images
• 2,863,974 Instance Annotations
• 330 Unique Medical Text IDs (Concepts)

⚠️ Usage Instructions

These are not standalone weights. To use this model, you must load these LoRA weights in combination with the base SAM3 model. Please refer to our official GitHub repository for detailed instructions on environment setup, weight loading, and inference.

• GitHub Repository: MedSAM3 on GitHub
• Paper: ArXiv

🖊️ Citation

@misc{liu2025medsam3delvingsegmentmedical,
      title={MedSAM3: Delving into Segment Anything with Medical Concepts}, 
      author={Anglin Liu and Rundong Xue and Xu R. Cao and Yifan Shen and Yi Lu and Xiang Li and Qianqian Chen and Jintai Chen},
      year={2025},
      eprint={2511.19046},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={[https://arxiv.org/abs/2511.19046](https://arxiv.org/abs/2511.19046)}, 
}