MedSAM

MedSAM (Medical Segment Anything Model) is a foundation model for medical image segmentation, developed by the Bo Wang Lab and adapted from Meta’s Segment Anything Model (SAM).

Key Capabilities

• Universal Medical Segmentation: Fine-tuned on an unprecedented dataset of over 1.5 million image-mask pairs encompassing 10 imaging modalities (CT, MRI, ultrasound, pathology, endoscopy, etc.) across 30+ cancer types.
• Promptable Interface: Like the original SAM, it can generate highly accurate segmentation masks for anatomical structures or lesions based on simple prompts, such as a bounding box or a single click, dramatically reducing annotation time.
• Zero-Shot Generalization: Exhibits strong zero-shot transfer capabilities, accurately segmenting new, unseen medical targets that were not explicitly included in its training data.

Usage in Healthcare

MedSAM addresses the massive bottleneck of manual data annotation in medical imaging. Radiologists and researchers use MedSAM to semi-automatically contour tumors, organs, and cells, accelerating the creation of new specialized datasets and improving the efficiency of quantitative medical research and clinical radiotherapy planning.