The CHS-Net is trained and evaluated on the synthesized dataset generated using publicly available COVID-19 CT segmentation datasets: COVID-19 CT segmentation nr. 2 [1] and COVID-19 CT lung and infection segmentation [2]. These two datasets are merged to form an aggregated dataset that addresses the problem of limited availability of the COVID-19 data. The fused dataset consists of 3561 CT slices with dimensions as 256 x 256 x 1, each having associated lungs mask and COVID-19 infection mask. The ground truth label is a binary mask with 0 value being the background region and 1 value being the target region. The complete dataset is available here.

The source code will be shared soon.

@article{punn2020chs,
  title={CHS-Net: A Deep learning approach for hierarchical segmentation of COVID-19 infected CT images},
  author={Punn, Narinder Singh and Agarwal, Sonali},
  journal={arXiv preprint arXiv:2012.07079},
  year={2020}
}