Skip to content

Latest commit

 

History

History
91 lines (77 loc) · 2.53 KB

README.md

File metadata and controls

91 lines (77 loc) · 2.53 KB

Texture-Preserving Diffusion Model for CBCT-to-CT Synthesis

flow chart

Introduction

This repository implements a texture-preserving diffusion model for CBCT-to-CT synthesis, designed to improve the quality and accuracy of medical image synthesis.

Dataset

You should structure your aligned dataset in the following way:

data_path/
  2PA092
    ├── CT
    │   ├── 0.npy
    │   ├── 1.npy
    │   ├── ...
    ├── CBCT
    │   ├── 0.npy
    │   ├── 1.npy
    │   ├── ...
  2PA093
    ├── CT
    │   ├── 0.npy
    │   ├── 1.npy
    │   ├── ...
    ├── CBCT
    │   ├── 0.npy
    │   ├── 1.npy
    │   ├── ...
  ...

The dataset links used in this study are as follows:

  1. SynthRAD2023
  2. Pelvic-ReferenceData

Before using the dataset, convert the nii.gz data to npy format using reg/nii2npy.py, and perform registration using syn. All the data is standardized using the mean and standard deviation of the raw data. The data processing is handled in dataloader.py

Dependencies

albumentations==1.3.0
einops==0.6.1
matplotlib==3.5.2
numpy==1.21.5
opencv_contrib_python==4.7.0.72
Pillow==9.5.0
scikit_image==0.19.2
scikit_learn==1.2.2
SimpleITK==2.2.1
timm==0.6.12
torch==2.0.0
tqdm==4.64.1

Installation

  • Clone this repo:

git clone https://github.com/zyj15416/TPDM-CBCT2CT.git
cd TPDM-CBCT2CT

Train

python train.py, the hyperparameter settings are handled in `utils/utils.py`.

Predict

You can use `predict.py` to load the weights and test the dataset, and you can modify the number of samples to improve the inference time.

Pretrained Models

We have released pretrained weights for cbct->ct in SynthRAD2023 challenge dataset(https://pan.baidu.com/s/1ATAeiGnNeaF1G0E1z974KQ?pwd=k7m4 ). You can save these weights in relevant checkpoints folder and perform inference.
After downloading the pretrained weights, place them in the corresponding folder, such as `./weights/weights.pth`. Then, replace the weight path in `predict.py` accordingly to perform model inference.

  • sample_test01

 A limited set of preprocessed CBCT and CT samples is provided in `sample_test01` to facilitate validation of the model's performance.

Citation

If you find this repository useful for your research, please use the following.
{

}