Detecting the novel coronavirus pneumonia in frontal chest X-ray images using transfer learning of CheXNet with a focus on Grad-CAM visualiztions.

• Preprint available on arXiv: COVID-CXNet
• Quick look at the final best results on paperswithcode: leaderboard
• Article about detailed steps and challenges toward developing COVID-CXNet on Medium: Thoughts on COVID-19 Pneumonia Detection in Chest X-ray Images

Chest x-ray images of patients with (mostly) PCR-positive COVID-19 are collected from different publicly available sources, such as SIRM. Please cite the associated paper if you are using CXR images. If this repo helped you with your research stuff, you can star it.

@article{haghanifar2020covidcxnet,
  title={COVID-CXNet: Detecting COVID-19 in Frontal Chest X-ray Images using Deep Learning},
  author={Arman Haghanifar, Mahdiyar Molahasani Majdabadi, Seokbum Ko},
  url={https://github.com/armiro/COVID-CXNet},
  year={2020}
}

If you are merging COVID-19 CXR images into your own datasets, please attribute the authors in any publications (DOI: 10.6084/m9.figshare.12580328. You may include the version of the dataset found on the figshare webpage for reproducibility.

• View COVID-19 images in the directory: chest-xray-images/covid19
• Download COVID-19 images as a single ZIP file: FigShare (755 images, next update on 900 images)
• Download the complete dataset from Kaggle: coming soon

There are currently ~820 images with different sizes and formats, and the data will be updated regularly. Metadata will be added soon. Normal CXRs are collected from different datasets, without a pediatric image bias. Note that a - sign at the end of image name indicates that CXR did not reveal any abnormalities, but the patient had CT/PCR-proven COVID-19 infection (probably patient is in early stages of disease progression). Besides, a p letter at the end of image name means that the image is taken from pediatric patient.