MedSAM

MedSAM is the foundation model for universal medical image segmentation, delivering accurate, promptable segmentation across diverse modalities (CT, MRI, X-ray) and pathologies.

This is a high-performance, generalized segmentation model, a direct adaptation of the Segment Anything Model (SAM) for clinical use. MedSAM was fine-tuned on a massive dataset: over 1.5 million image-mask pairs, spanning 10 imaging modalities and 30+ cancer types. This extensive training enables it to bypass the traditional need for modality-specific models, offering robust, zero-shot segmentation across diverse anatomies. The core value proposition is efficiency: clinicians provide a simple prompt (like a bounding box or text concept), and MedSAM generates a precise mask, significantly expediting diagnosis and treatment planning workflows.

https://github.com/bowang-lab/MedSAM