Certain investigations on semantic segmentation and performance analysis of deep convolutional neural networks for covid19 using chest x ray images

Abstract

The Coronavirus disease-19 (COVID-19) is the most recently newlinediscovered infectious disease, and the ongoing pandemic witnessed a high newlinedeath rate globally due to severe acute respiratory syndrome (SARS)- newlineCoronavirus-2 (CoV-2). The global pandemic is in progress currently as a newlineresult of the COVID-19 outbreak. The source of the outbreak, the newlinecoronavirus-2 (CoV-2) produces a severe acute respiratory infection that newlineaffects the human respiratory system. It has been reported that the COVID-19 newlinevirus can rapidly mutate and cause lung damage before infected persons newlinereceive specific medicine. COVID-19 coexists with other chest illnesses, newlinemaking the diagnosis more difficult. If the disease is detected at the newlinepreliminary stage, it could minimize the impact on human health and death newlinerate. The most precise gold-standard molecular laboratory technique for newlineCOVID-19 diagnosis is a reverse transcription-polymerase chain reaction newline(RT-PCR). Unfortunately, it is a labour and time-intensive method. newlineComputer Tomography (CT) and Chest X-Ray (CXR) are two newlineexamples of radiographic imaging techniques that have become a quotsuccessful newlineadditionquot to RT-PCR. Even though PCR sputum testing is the gold-standard, newlinediagnosis of COVID-19 using CXRs is faster. CXR screening is one of the newlinemost studied radiological imaging modalities due to its minimal radiation newlinedose and ease of use and availability. Due to the need for large CXR datasets newlinewith various input image sizes collected from COVID-19 individuals, the newlinecurrent Deep Learning (DL)-based CNN models need to be more accurate in newlinerecognizing COVID-19 from CXRs. Although most studies have reported newlinehigh sensitivity, specificity, and accuracy values, these results tend to be newlinebiased when cross-validated with different datasets. newline newline