Skip to content

Codes of "Knee Cartilage Defect Assessment by Graph Representation and Surface Convolution"

License

Notifications You must be signed in to change notification settings

zixuzhuang/CSNet

Repository files navigation

Cartilage Surface Network (CSNet)

This is an official implementation of the paper Knee Cartilage Defect Assessment by Graph Representation and Surface Convolution.

Abstract

Knee osteoarthritis (OA) is the most common osteoarthritis and a leading cause of disability. Cartilage defects are regarded as major manifestations of knee OA, which are visible by magnetic resonance imaging (MRI). However, the physiologic characteristics of the cartilage may hinder such efforts: the cartilage is a thin curved layer, implying that only a small portion of voxels in knee MRI can contribute to the cartilage defect assessment; heterogeneous scanning protocols further challenge the feasibility of the CNNs in clinical practice; the CNN-based knee cartilage evaluation results lack interpretability.

To address these challenges, we model the cartilages structure and appearance from knee MRI into a Graph Representation. Then, guided by the cartilage graph representation, we design a non-Euclidean deep learning network with the self-attention mechanism, named Cartilage Surface Network (CSNet), to extract cartilage features in the local and global, and to derive the final assessment with a visualized result.

Framework

Teaser image

Visualized Diagnostic Results

Teaser image

Catalog

  • Knee Graph Representation code
  • Cartilage Surface Network code
  • Visualization code
  • Reproduced Methods for Comparisons
  • Our Segmentation Tool

Installation

This code requires PyTorch 1.1+ and DGL 0.6+. Clone this repo and install the dependencies by

git clone https://github.com/zixuzhuang/CSNet.git
cd CSNet
pip install dgl-cu113 dglgo -f https://data.dgl.ai/wheels/repo.html
pip install -r requirements.txt

Please note: install the dgl library that corresponds to your CUDA version.

Usage

In order to run our framework successfully, the code need to be executed in the following order:

1. Transform the knee MRI into the graph representation for computation and the knee mesh for visualization

This step requires knee bone segmentation. We recommend using our segmenation tool to obtain them.

python graph_construction.py -org <mri-data> -seg <seg-data> \
 -grade <grade> -graph <graph-save-path> -mesh <mesh-save-path>

It can be run directly without any parameters because this demo already comes with them.

python graph_construction.py

2. Train the CSNet

python train_csnet.py -c <config-file> -f <fold-idx> -t <test-mode>

It also can be run without parameters.

python train_csnet.py

Then the corresponding experiment log and model file can be found in results.

3. Evaluate the result and show the visualization

python eval_csnet.py -graph <graph-file> -mesh <mesh-file> \
 -net <model-weight-file> -img <visualized-result-save-path>

If no parameters are added, it will give results based on the untrained CSNet.

python eval_csnet.py

The visualized result is in in data/visualization

Citation

@article{zhuang2022knee,
  title={Knee Cartilage Defect Assessment by Graph Representation and Surface Convolution},
  author={Zhuang, Zixu and Si, Liping and Wang, Sheng and Xuan, Kai and Ouyang, Xi and Zhan, Yiqiang and Xue, Zhong and Zhang, Lichi and Shen, Dinggang and Yao, Weiwu and Wang, Qian},
  journal={arXiv preprint arXiv:2201.04318},
  year={2022}
}

About

Codes of "Knee Cartilage Defect Assessment by Graph Representation and Surface Convolution"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages