COVID-19 Detection Using Radiography Images Based on Transfer Learning with DenseNet

Abstract

Shubham, Abhinav Sinha, RashmiCOVID-19 spread has now nearly come to a halt, despite of daily increase in positive cases in India. It has deeply affected daily lives, public health, and the economy of the whole world. An important step in controlling COVID-19 spread is to identify the infected patients as soon as possible and treating them. There is a need for supplementary diagnostic tools apart from RT-PCR, which is easy to use and less contagious. Significant findings have proven that chest X-rays (CXR) in combination with deep learning algorithms for images, like pretrained CNNs are vital in finding features that are related to COVID-19. Using pretrained networks, so-called transfer learning can extract features from CXR images which can help detect COVID presence. In this work, CXR images were analyzed using one of the advanced CNN architectures, DenseNet201 using MATLAB. This architecture is 201 layers deep, capable to classify into 1000 classes. The last layers have been modified so that DenseNet201 can be used to properly predict COVID+VE and COVID-VE CXR images.