Segment anything model (SAM) has revolutionized natural image segmentation, but its performance on medical images is limited. This work presents MedSAM, the first attempt at extending the success of SAM to medical images, with the goal of creating a universal tool for the segmentation of various medical targets. Specifically, we first curate a large-scale medical image dataset, encompassing over 200,000 masks across 11 different modalities. Then, we develop a simple fine-tuning method to adapt SAM to general medical image segmentation. Comprehensive experiments on 21 3D segmentation tasks and 9 2D segmentation tasks demonstrate that MedSAM outperforms the default SAM model with an average Dice Similarity Coefficient (DSC) of 22.5% and 17.6% on 3D and 2D segmentation tasks, respectively. The code and trained model are publicly available at https://github.com/bowang-lab/MedSAM.