Machine learning-aided screening of COVID-19 in Lung Ultrasound images.

Abstract

In this project, the transfer learning technique was exploited on several deep learning algorithms to classify COVID-19 in Lung Ultrasound images. Training and preliminary testing of the algorithms was performed with the use of a data set of 792 images that contained several distinct features that are indicative of either of the classes of interest i.e. COVID-19 and healthy. Particularly the VGG-16 framework provided outstanding optimal results with a remarkable accuracy of 97.5 percent and a recall of 95.7 percent. The astounding success rate makes the model a very useful advisory screening approach in this field of medical imaging to aid further diagnosis of COVID-19.