Yulong Dou, Han Wu, Changjian Li, Chen Wang, Tong Yang, Min Zhu, Dinggang Shen, and Zhiming Cui.

This is the official repository of our paper CLIK-Diffusion: Clinical Knowledge-informed Diffusion Model for Tooth Alignment in Medical Image Analysis (MedIA) 2025. In this work, we formulate the complex problem of tooth alignment as a more manageable landmark transformation problem, which is further refined into a landmark coordinate generation task via diffusion model. To further encourage the integration of essential clinical knowledge, we design hierarchical constraints from three perspectives: dental-arch level, inter-tooth level, and individual-tooth level.

• [2026-01-26] Inference code for crown-only meshes is provided!
• [2025-08-05] Version of Record of our paper is available online, see Paper!
• [2025-07-25] Dataset is released!
• [2025-07-25] Inference code for full tooth meshes is released!
• [2025-07-22] Our paper is accepted by MedIA 2025!

First clone this repo. And then set up an environment and install required packages.

git clone https://github.com/ShanghaiTech-IMPACT/CLIK-Diffusion.git

conda create -n tooth python=3.8
conda activate tooth
pip install torch==1.10.0+cu113 torchvision==0.11.1+cu113
pip install -r requirements.txt

We now only provide case-by-case code for testing. We support two modes: Full Tooth (as presented in the paper) and Crown-only (including contact surfaces between adjacent teeth, which differ from intra-oral scans). To test our model ——

• First download the checkpoints from https://drive.google.com/drive/folders/1o9tVJ6p8Jbad3gu0ZUkX0tE5dp7Vkh9g?usp=sharing.

  • For Full Tooth, download both the diffusion model checkpoint and the standard landmark detection checkpoints.
  • For Crown-only, download the diffusion model checkpoint and the '[Crown]' version of the landmark detection checkpoints.

• Then put all downloaded checkpoints into ./Code/checkpoint.

• We have prepared some data samples for you to start a quick test.

• Run the following command based on your needs:

  1. For Full Tooth (as in paper): The pre-orthodontic tooth meshes are stored in ./Data/275.

  cd CLIK-Diffusion
  python Code/infer.py -i "./Data/275" -o "./Output"
  # add argument "-v" if you want see the results of landmark detection network

  2. For Crown-only: The pre-orthodontic crown meshes are stored in ./Data/275_crown.

  cd CLIK-Diffusion
  python Code/infer_crown.py -i "./Data/275_crown" -o "./Output"
  # add argument "-v" if you want see the results of landmark detection network

• You will find the post-orthodontic meshes which are predicted by our method in ./Output.

• Note that if you want to try your own data, it is important to maintain a similar scale and orientation to our dental data.

★ Our dataset is available for reserach purpose only. To apply for our CLIK-Diffusion dataset, please fill out the Data Access Agreement and send the signed e-copy to Yulong Dou (email: [email protected]) and Zhiming Cui (email: [email protected]), as well as CC your supervisor. We will send you the dataset link and password when recieving the data access form.

If you find this code or dataset useful, please cite our paper:

@article{DOU2025103746,
title = {CLIK-Diffusion: Clinical Knowledge-informed Diffusion Model for Tooth Alignment},
journal = {Medical Image Analysis},
volume = {106},
pages = {103746},
year = {2025},
issn = {1361-8415},
doi = {https://doi.org/10.1016/j.media.2025.103746},
url = {https://www.sciencedirect.com/science/article/pii/S1361841525002932},
author = {Yulong Dou and Han Wu and Changjian Li and Chen Wang and Tong Yang and Min Zhu and Dinggang Shen and Zhiming Cui}
}