# MedSAM / SAM-Med3D (segment anything for medicine)

Source: https://onco.cc/technologies/medsam/  
OnCo record `medsam` (Technology). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

Adaptations of Meta's Segment Anything model that outline tumours and organs on any scan with a click.

## Summary

MedSAM is a promptable segmentation transformer created by fine-tuning Meta's Segment Anything model on medical images, so a user can click or draw a box and receive an outline of a tumour or organ. The Nature Communications 2024 paper fine-tuned SAM on 1.5M image-mask pairs across imaging modalities, and SAM-Med3D extends the approach from 2D slices to volumes. It is used for radiotherapy contouring and for measuring treatment response, where hand outlining is slow and variable between readers. The models need prompts rather than working fully automatically, and boundary errors on small lesions remain a known weakness that matters for early or metastatic disease. For a newcomer: MedSAM lets a clinician outline a tumour on almost any scan with a click, but a person still has to point it in the right place.

## Fields

- Kind: Technology
- Status: emerging
- Last checked: 2026-09-08
- Tags: foundation-model; radiology
- Principle: MedSAM is a promptable segmentation transformer fine-tuned on medical masks.
- Since: 2024
- Strengths: Interactive, general
- Limitations: Needs prompts; boundary errors on small lesions

## Sources

- MedSAM, Nature Communications 2024: https://doi.org/10.1038/s41467-024-44824-z

## Connected records

- fronts: [AI & Computation](https://onco.cc/fronts/ai-computation/), [Imaging](https://onco.cc/fronts/imaging/)
- technologies: [AI in radiology](https://onco.cc/technologies/radiology-ai-screening/), [IMRT / IGRT (modern external beam)](https://onco.cc/technologies/imrt-igrt/)
- key papers: [Segment anything in medical images](https://onco.cc/key-papers/paper-ma-nat-commun/)

---
JSON: https://onco.cc/api/v1/entities/medsam.json