DefeatCovid19-Net, a deep convolutional neural network design tailored for the detection of COVID-19 cases from chest radiography images.
The workflow takes as input a '.png', '.jpeg' or '.jpg' chest radiography image and outputs the most likely diagnosis: Non-COVID-19 or COVID-19. The workflow produces a report showing the input image as well as the prediction.
The network of choice is ResNet34, provided by torchvision and pretrained on Imagenet. The net is first trained on the Kaggle Chest X-Ray Pneumonia dataset (5856 images) and then on the COVID-19 Chest X-Ray dataset (123 usable images). Axial and lateral images were removed from the latter dataset. COVID-19 diagnoses were labelled 1, 0 otherwise (SARS/ARDS/Pneumocystis/Streptococcus/No finding).
Example of the output report generated by the workflow:
Required inputs
- Chest X-Ray: A Chest X-Ray image. Must be tagged as 'png' or 'jpeg'.
Output container files
- report.pdf: PDF report.
- report.txt: txt with prediction
- report/
- online_summary_report.html: Online report, visible in the second tab of "Show results".
- input image
References
- DefeatCovid19-Net: GitHub
- Paul Mooney, Chest X-Ray Images (Pneumonia), Kaggle dataset, https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia, 2018
- Joseph Paul Cohen, COVID-19 image data collection, https://github.com/ieee8023/covid-chestxray-dataset, 2020
Create free account now!
