PCAMNet

Code for paper "Position-Prior Clustering-based Self-attention module for Knee Cartilage Segmentation".

The framework of PCAM is shown below:

Requirements

Python 3.6.2

Pytorch 1.7

CUDA 11.2

Numpy 1.19.1

SimpleITK 2.1.1.2

Data process

The model was trained and evaluated on OAI-ZIB Datasets. This public dataset includes 507 3D DESS MR data with 81120 slices. The pixel spacing is $0.3645mm\times0.3645mm$ and the slice thickness is 0.7mm for all volumetric data. For each volumetric data, it contains 160 slices in $384 \times384$.

For data preprocess, each volume data is cropped into $256\times256$ in center position along the height and width dimension and is cropped randomly along depth (slice) dimension into $32$. The finally data size is $256\times256\times32$.

Our segmentation model is evaluated by four evaluation metrics, which are Dice score, Volumetirc Overlap Error (VOE), Average Symmetric Surface Distance (ASSD). We performed experiments to evaluate the performance of the proposed model. Please refer to the original paper for more details.

##result

Cite

Please consider citing this project in your publications if it helps your research. The following is a BibTeX reference. The BibTeX entry requires the url LaTeX package.

@article{liang2022pcam,
  title={Position-Prior Clustering-based Self-attention module for Knee Cartilage Segmentation},
  author={Dong Liang, Jun Liu, Kuanquan Wang, Gongning Luo, Wei Wang, Shuo Li},
  Conference={MICCAI2022},
  publisher={Springer}
}

Acknowledgment

Thansks OAI (https://nda.nih.gov/oai/) for providing public dataset and ZIB(Zuse Institute Berlin) for providing corresponding masks.