work on COVID-19 detection from X-ray images by fine-tuning popular convolutional networks (such as ResNet, SqueezeNet, and DenseNet) here: https://arxiv.org/pdf/2004.09363.pdf

first re-labeled the publicly available COVID-19 images (collected by Joseph Paul Cohen PhD), with the help of a board-certified radiologist, and created COVID-Xray-5k dataset for binary classification of COVID-19 (we made this dataset publicly available via our Github).
We then trained multiple models on this dataset, and evaluated their sensitivity, specificity, ROC curve, AUC, and confusion matrix.

The PyTorch code for training and inference on our model are available here:
https://github.com/shervinmin/DeepCovid

Although the result looks promising, this is a first version of these models, and more experiments will be done once a larger dataset of cleanly labeled X-ray and CT images become available for COVID-19, for more concrete evaluation.

🗣 @AI_Python_arXiv
✴️ @AI_Python_EN
❇️ @AI_Python