Enhancing the Quality of Chest X-Ray Images for Detection of Covid-19 through Data Augmentation Technique

Abstract

Corona (Covid-19) is a rapidly transmitting deadly disease caused by the severe acute respiratory syndrome Coronavirus-2 (SARS-CoV-2). Covid19 disease proliferation has caused adverse effects on the world economy, academics, and majorly on the health of individuals. Due to the recent outbreak of the covid19 pandemic worldwide, utilizing computer-aided diagnosis techniques plays an important role in classification and predominantly reduces the burden of clinicians. Chest X-ray images (CXR) are an effective tools assisting for rapid diagnosis of disease in its early stages by using intellectual deep learning techniques for image classification. In this research, Convolutional neural network (CNN) ImageNet pre-trained Resnet50 model was trained and evaluated using the transfer learning technique. The procured expertise of model is then transformed and finely tuned can better generalize and improve performance for classifying chest X-rays as the COVID-19 or normal or viral pneumonia by finding viral abnormalities. To accomplish this, CNN requires a substantial amount of data for training. It is strenuous to glean a substantial percentage of radiographic images in a short time. Hence, we opted for data augmentation techniques which increased the data size and image quality that yielded 97 percent accuracy. We envisage more impactful radiology systems will be built using these techniques for screening COVID-19 using chest radiographs.