InvCoSS: Inversion-driven Continual Self-supervised Learning in Medical Multi-modal Image Pre-training

Overview

This repository contains the official implementation of InvCoSS, a novel inversion-driven continual self-supervised learning framework for medical multi-modal image pre-training.

Key Features:

• Continual self-supervised learning for medical imaging
• Inversion-driven approach for knowledge preservation and transfer
• Multi-modal support (CT, MRI, X-ray, etc.)
• Self-supervised pre-training

Code and Pre-trained Weights

Code and pre-trained weights will be released upon paper acceptance.

⭐ Star this repository to stay updated!

Citation

@article{luo2025invcoss,
  title={InvCoSS: Inversion-driven Continual Self-supervised Learning in Medical Multi-modal Image Pre-training},
  author={Luo, Zihao and Rui, Shaohao and Tang, Zhenyu and Wang, Guotai and Wang, Xiaosong},
  journal={arXiv preprint arXiv:2512.19213},
  year={2025}
}