Code created for the final project of the AML lecture 2021
The COVID-19 pandemic continues to pose a challenge to our healthcare systems. Hospitals currently use PCR-tests to detect the presence of the coronavirus. While being accurate, these tests require hours at minimum before a result is available. To speed up diagnosis and consequently save time and lives, other additional methods may be of use. In this final project for the Advanced Machine Learning lecture a neuronal network capable of detecting and localizing COVID abnormalities in chest radiographs is developed. This procedure has the potential to help physicians decide on an appropriate medical treatment much faster than a PCR-test would allow while maintaining a sufficient accuracy. We hereby follow a machine learning challenge put out for tender on the online platform www.kaggle.com. Guidelines and - most importantly - a comprehensive data set are provided. Most of our later runs are visualized in depth on Weights & Biases https://wandb.ai/cov01.