CT-FM (whole-body CT foundation model)

CT-FM is a whole-body CT foundation model built with 3D self-supervised pretraining, learning from unlabelled volumes so that downstream tasks need fewer annotations. The 2025 arXiv paper describes pretraining on 148,000 CT scans and applies the model to organ segmentation, triage of findings and image retrieval. It is intended for imaging researchers who want a shared 3D backbone across CT tasks, including tumour segmentation and follow-up in oncology. It remains a research release, and prospective clinical evaluation and regulatory review have not been reported, so its performance in routine radiology is unknown. For a newcomer: CT-FM is a general-purpose model that learned the anatomy of the whole body from a large pile of CT scans and can be adapted to specific jobs.