We present SMILE (Super Modality Image Learning and Enhancement), an anatomy-aware diffusion model for clinically reliable CT contrast enhancement. SMILE achieves significant improvements: +14.2% SSIM, +20.6% PSNR, +50% FID, and enables cancer detection from non-contrast CT scans with +10% F1 score improvement.
Click the image above to watch the video on YouTube.
See More, Change Less: Anatomy-Aware Diffusion for Contrast Enhancement
Junqi Liu, Zejun Wu, Pedro R. A. S. Bassi, Xinze Zhou, Wenxuan Li, Ibrahim E. Hamamci, Sezgin Er, Tianyu Lin, Yi Luo, Szymon Płotka, Bjoern Menze, Daguang Xu, Kai Ding, Kang Wang, Yang Yang, Yucheng Tang, Alan Yuille, Zongwei Zhou★
Johns Hopkins University
To set up environment, see INSTALL.md for details.
git clone https://github.com/MrGiovanni/SMILE.git
cd SMILE
while read p; do pip install "$p" || echo "Failed to install $p, skipping..."; done < requirements.txtbash download_ckpts.shbash download_demo_dataset.sh
bash inference.shThe enhancement results are in the ./out folder.
First, save your data folder (e.g., PanTS) in ./data using the same format as our demo data.
Second, modify the parameters in the inference.sh:
-
Dataset_Name: name of your own dataset name (e.g.,PanTS).data └──PanTS/ ├── PanTS_000001/ │ └── ct.nii.gz ├── PanTS_000002/ │ └── ct.nii.gz -
TARGETS(optional): enhancement targets. Default as("arterial" "venous" "delayed"). -
GUIDE_CSV(optional): .csv file to guide model to inference on specific cases. An example:
Inference ID
BDMAP_xxxx01
BDMAP_xxxx02
BDMAP_xxxx03
BDMAP_xxxx04
...Note
We are calling for more baseline methods.
| model | paper | github | SSIM | PSNR | FID | Intensity Correlation | Download |
|---|---|---|---|---|---|---|---|
| Pix2Pix |  |
 |
60.7 | 18.8 | 299.7 | 0.26 | Download |
| CycleGAN |  |
 |
71.9 | 18.2 | 271.1 | 0.09 | Download |
| DALL-E |  |
 |
51.4 | 16.3 | 423.7 | 0.71 | Download |
| Stable Diffusion |  |
 |
64.6 | 16.0 | 406.3 | 0.45 | Download |
| CUT |  |
 |
75.4 | 21.4 | 269.5 | 0.06 | Download |
| SMILE |  |
 |
86.1 | 25.8 | 133.4 | 0.95 | Download |
Our work further includes CTVerse, a large-scale multi-phase CT dataset containing 477 patients from 112 hospitals, with all four contrast phases (non-contrast, arterial, venous, and delay).
# This work is currently under peer review, but early access is available!
# To request the CTVerse dataset files, please email Zongwei Zhou at [email protected]@article{liu2025see,
title={See More, Change Less: Anatomy-Aware Diffusion for Contrast Enhancement},
author={Liu, Junqi and Wu, Zejun and Bassi, Pedro RAS and Zhou, Xinze and Li, Wenxuan and Hamamci, Ibrahim E and Er, Sezgin and Lin, Tianyu and Luo, Yi and Płotka, Szymon and others},
journal={arXiv preprint arXiv:https://www.arxiv.org/abs/2512.07251},
year={2025},
url={https://github.com/MrGiovanni/SMILE}
}
This work was supported by the Lustgarten Foundation for Pancreatic Cancer Research, the Patrick J. McGovern Foundation Award, and the National Institutes of Health (NIH) under Award Number R01EB037669. We would like to thank the Johns Hopkins Research IT team in IT@JH for their support and infrastructure resources where some of these analyses were conducted; especially DISCOVERY HPC. Paper content is covered by patents pending.