COVID-19 Detection using Convolutional Neural Network Architectures based upon Chest X-rays Images

Abstract

The novel Coronavirus has really been the unexpected Catastrophe, which no one could have even thought of. It has emerged as a pandemic and has raised questions on the health infrastructure and facilities available as it is required to get a large number of people tested. RT-PCR is the standard diagnostic test being used. But there are several issues with RT-PCR testing like relying only on one approach even when it is getting difficult for most countries to procure the required amount of testing kits, also there have been some cases of false positive results. All these factors definitely call for a search of alternative testing methodologies, so it not reqired to rely on just one approach. Apart from RT-PCR testing, Chest X-rays can be a great tool to know about the status and severity of COVID-19. Once obtaining the chest X-rays of a person a call has to be made on the COVID-19 status and that needs very high accuracy and expertise and again there will be a shortage of the expertise. So, a solution has been proposed to this problem by developing a COVID-19 detection system, which can assist the medical experts regarding the report and then the experts can make a final call. As observed, DeepLearning techniques, especially Convolutional Neural Network (CNN) have proven to be outstanding in medical image classification and analysis, and as our task is similar to medical image classification, CNN is a good choice for our use case. So, we analyzed four different CNN architectures on Chest X-ray for Covid-19 Diagnosis. The models used are pretrained on ImageNet datasets. Transfer Learning has been used to generate results. A comparative study of the results obtained with different architecture reveals that structures based on CNN have great potential for Covid-19 diagnosis and detection.