SCR-Registration

Code Repository for the paper accepted at MICCAI 2023 "X-Ray to CT Rigid Registration Using Scene Coordinate Regression".
Currently, we provide the test dataset used for a single patient data (VOL-17-1882) from the paper, thus only the inference is supported. Due to the large size of the training dataset, we plan to release the scripts used for generating the simulated data instead. Thanks for your patience and feel free to open issue in case of any inquiry.

Introduction

SCRNet is a deep learning-based method for rigid registration of X-ray and CT images. The network is trained to regress the scene coordinates of the X-ray image from the CT image. The scene coordinates are then used to compute the correspondences between the X-ray and CT images. The correspondences are used to compute the rigid transformation between the X-ray and CT images. The paper can be found here.

Requirements and Installation

The code is tested on Ubuntu 20.04.6 with RTX 3090 (CUDA 12.0). The following packages are required to run the code:

• Python ^3.9
• Poetry ^1.7.1

We recommend using pyenv to manage the Python version and poetry to manage the dependencies.
Please refer to the official websites for the installation of pyenv and poetry. To install the dependencies, run the following command in the root directory of the repository:

poetry install

We provide option in the config to output registration overlays and point clouds. To enable this, pyvista needs to be configured to run in headless mode. Please refer to the documentation for more details.

Inference

To run the inference, please download the pretrained models and test dataset from here and put it under the weights and data directory respectively. The folder structure should look something like this.

data
└── VOL-17-1882_CIOS_FUSION
    ├── annotations
    ├── correspondences
    ├── images
    ├── pelvis_in_world_coordinate.stl
    └── test.json

and

weights
├── DFLNet
│   └── VOL-17-1882_CIOS_FUSION.ckpt
├── PoseNet
│   └── VOL-17-1882_CIOS_FUSION.ckpt
└── SCRNet
    └── VOL-17-1882_CIOS_FUSION.ckpt

Then, run the following command to test the SCRNet model (replace the config file with the one you want to test):

poetry run python run_inference.py --config configs/test/SCRNet/VOL-17-1882_CIOS_FUSION.yaml

also, check the configs/test/SCRNet/VOL-17-1882_CIOS_FUSION.yaml for the available options to ouput overlays and point clouds for each prediction. The results will be saved under the results directory.
To show the aggregated evaluation results, run the following command:

poetry run python eval_metrics.py --config results/SCRNet/VOL-17-1882_CIOS_FUSION

Acknowledgements

We thank the authors of Automatic annotation of hip anatomy in fluoroscopy for robust and efficient 2D/3D registration for their work on related work on registration and open-source code on DeepFluoroLabeling-IPCAI2020. We also thank the authors of DeepDRR -- A Catalyst for Machine Learning in Fluoroscopy-guided Procedures for their work on realistic X-ray simulation and open-source code on DeepDRR. Please refer to their repositories for more details.

Citation

If you find this repository useful, please cite our paper:

@InProceedings{10.1007/978-3-031-43999-5_74,
author="Shrestha, Pragyan
and Xie, Chun
and Shishido, Hidehiko
and Yoshii, Yuichi
and Kitahara, Itaru",
title="X-Ray to CT Rigid Registration Using Scene Coordinate Regression",
booktitle="Medical Image Computing and Computer Assisted Intervention -- MICCAI 2023",
year="2023",
publisher="Springer Nature Switzerland",
address="Cham",
pages="781--790",
isbn="978-3-031-43999-5"
}