Diagnosing COVID-19 using Chest X-Rays with fast.ai

In this article, I will be discussing how a Resnet architecture can be used to classify a dataset containing chest X-rays of patients who have COVID-19, pneumonia or are healthy.

The spread of coronavirus throughout the world has happened at an unprecedented rate, shutting down entire countries and crippling economies. Production of testing equipment is often very limited and the global community would undoubtedly benefit from having alternative ways to diagnosing COVID-19.

Currently, there is a very limited amount of data about Coronavirus cases, the main source I will be using for this article is Synthetaic which has collated data from COVID-NET and Image Data Collection. Due to the rapid pace which new COVID-19 data is being collected and the pace at which the Coronavirus literature is moving, I will be regularly updating my model and this article as more information becomes available.

Using standard Coronavirus testing methods the estimated false negative rate is around 10%, the aim for this project is to help provide a system for distinguishing between pneumonia and COVID-19 with a high rate of success.

Note: A link to my GitHub containing the code is at the bottom

Model

For this task, we will be using a Convolutional Neural Network(CNN) with the Resnet50 architecture, due to time constraints and a lack of data we will be using pre-trained weights gained from training this model on the…