BoneFM is the skeleton-focused CT foundation backbone used by BoneCoT: Multi-center validation of a whole-body skeleton foundation model for bone metastases guided by clinician-derived chain of thought.
Metastatic Bone Tumor Clinical Center, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China
Institute of Diagnostic and Interventional Radiology, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China
Mailman School of Public Health, Columbia University, New York, NY, USA
Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, WA, USA
*These authors contributed equally: Hui Zhao, Ruipeng Zhang, Zhiyu Wang
BoneFM is a Vision Transformer backbone adapted from DINOv2-style self-supervised learning for skeleton-focused CT representation. BoneCoT uses BoneFM features and clinician-derived task dependencies for downstream bone metastasis and bone-related disease reasoning.
This repository hosts the public BoneFM backbone checkpoint:
File
Description
BoneFM.pth
BoneFM pretrained backbone checkpoint for the BoneCoT public code
BoneFM is intended for non-commercial research on skeletal CT representation learning and downstream bone-related disease modelling. It can be used as a feature backbone with the public BoneCoT code when users provide their own de-identified image data and clinically appropriate labels.
BoneFM and BoneCoT are not standalone clinical diagnostic devices. They should not be used for patient management without local validation, regulatory review, and qualified clinical oversight.
Input Convention
Prepare CT slices with the bone-window convention used by the public BoneCoT code:
Please cite the final Nature Biomedical Engineering record once it is live:
bibtex
1@article{bonecot2026,
2 title = {BoneCoT: Multi-center validation of a whole-body skeleton foundation model for bone metastases guided by clinician-derived chain of thought},
3 author = {Zhao, Hui and Zhang, Ruipeng and Wang, Zhiyu and Gu, Yifeng and Xu, Shengyuan and Wang, Sheng and Li, Yuehua},
4 journal = {Nature Biomedical Engineering},
5 year = {2026},
6 doi = {10.1038/s41551-026-01736-1}
7}
BoneFM builds on DINOv2-style self-supervised vision-transformer code. Please also cite the relevant DINOv2 work when using inherited implementation components.
License
The public BoneFM release is made available under CC BY-NC 4.0 for non-commercial research use, subject to any applicable third-party code licenses in the accompanying implementation.
