ClinicalCode

Checkpoints for Animating the Clinical Source Code of Disease Progression for Medical World Models.

Project page · Code and commands

Checkpoints

File	Function	Source
difftumor_hcc_fold0.pt	CT tumor segmentation within a supplied liver mask	HCC fold-0 DynUNet used in our experiments, based on DiffTumor / MeWM
ct_decoder.freeze.json	CT treatment-action decoding	ClinicalCode frozen Lingshu-I-8B decoder; identical to the decoder bundled with the public CT case
Raidionics-MRI_TumorCE_Postop-v13.zip	MRI enhancing-tumor segmentation	Unmodified Raidionics v1.3 archive
Raidionics-MRI_Cavity-v13.zip	MRI resection-cavity segmentation	Unmodified Raidionics v1.3 archive

Download

python -m pip install huggingface_hub
hf download qicq1c/ClinicalCode difftumor_hcc_fold0.pt --local-dir models/difftumor
hf download qicq1c/ClinicalCode ct_decoder.freeze.json --local-dir models/decoder
hf download qicq1c/ClinicalCode \
  Raidionics-MRI_TumorCE_Postop-v13.zip Raidionics-MRI_Cavity-v13.zip \
  --local-dir models/raidionics/archives
unzip -q models/raidionics/archives/Raidionics-MRI_TumorCE_Postop-v13.zip -d models/raidionics
unzip -q models/raidionics/archives/Raidionics-MRI_Cavity-v13.zip -d models/raidionics

File hashes are recorded in SHA256SUMS. The frozen language models are available from their original Hugging Face repositories, including Lingshu-I-8B.

Licenses

• ClinicalCode decoder: MIT.
• HCC DynUNet checkpoint: retained with the MeWM CC BY-NC 4.0 license. Original MeWM license.
• Raidionics model archives: BSD-2-Clause, copyright David Bouget. Original model repository.

Citation

@misc{chen2026clinicalcode,
  title  = {Animating the Clinical Source Code of Disease Progression for Medical World Models},
  author = {Chen, Qi and Yuille, Alan L. and Chen, Jieneng},
  year   = {2026},
  url    = {https://github.com/qic999/ClinicalCode}
}