MedCRP-CL: Continual Medical Image Segmentation via Bayesian Nonparametric Semantic Modality Discovery

MedCRP-CL is a framework for online task structure discovery and structure-aware continual learning in medical image segmentation. Utilizing the Chinese Restaurant Process (CRP), the method dynamically infers task groupings (semantic modalities) from clinical text prompts as tasks arrive, without requiring predefined cluster counts or replay of raw patient data.

• Paper: MedCRP-CL: Continual Medical Image Segmentation via Bayesian Nonparametric Semantic Modality Discovery
• Repository: GitHub - zygao930/MedCRP-CL

Usage

To run evaluation using the official codebase and this checkpoint, clone the GitHub repository and run:

# Download checkpoint
pip install huggingface_hub
huggingface-cli download clg-g/MedCRP-CL --local-dir checkpoints/MedCRP-CL

# Run evaluation
python scripts/inference.py checkpoints/MedCRP-CL

Citation

@misc{gao2026medcrpclcontinualmedicalimage,
  title         = {MedCRP-CL: Continual Medical Image Segmentation via Bayesian Nonparametric Semantic Modality Discovery},
  author        = {Ziyuan Gao},
  year          = {2026},
  eprint        = {2605.20297},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2605.20297}
}