Oral-3D

This is the official implementation of paper:

Oral-3D: Reconstructing the 3D Structure of Oral Cavity from Panoramic X-Ray

Model Overview

Oral-3d is a GAN-based model that learns to generate 3D oral structure from panoramic X-ray image

Set up environemnt

 conda create -n Oral_3D
 conda activate Oral_3D
 pip install -r requirements.txt

Data Format

We assume that all the required data is saved in .mat file. An example case be seen seen in ./data/mat A mat file is expected to contain below fields:

# Given the CBCT in [256, 288, 160]
CBCT: [256, 288, 160] # CBCT data after pre-processing
MPR: [80, 160, 576] # Flattened 3D image used for training
Case_ID: [1] # ID of the case for saving generation image.
PriorShape: [577, 2] # Shape of the dental arch
Ideal_PX: [160, 576] # The Panoramix X-ray Image

Train

We assume all the required .mat files are store under {data_root}. If there is no split file under the directory, the dataset will split the data in a ratio of 3:1:1 for training, validation, and test and save this in split.csv under {data_root}. Or else the dataset will load the training, validation, and test data according to existing split.csv.

To train the model, the user could run by:

python main.py --mode train --data_root {data_root}

Test

To test the model, we assume the data is saved as in the same format as in Data Format with the split.csv file.

 python main.py --mode test

The user could add --test_only option to skip loading train and val data

License and Citation

@inproceedings{song2021oral,
  title={Oral-3d: Reconstructing the 3d structure of oral cavity from panoramic x-ray},
  author={Song, Weinan and Liang, Yuan and Yang, Jiawei and Wang, Kun and He, Lei},
  booktitle={Proceedings of the AAAI conference on artificial intelligence},
  volume={35},
  number={1},
  pages={566--573},
  year={2021}
}