ABSTRACT
Lung diseases affect many populations around the world and their symptoms may range from common cough to chronic lung infections caused by unhygienic living conditions, unhealthy habits(smoking) and often inter-species virus/bacterial transmission. Moreover, the death toll and individuals affected by lung infections have skyrocketed after the contagious COVID-19 outbreak in 2019 December in Wuhan China. The Big Data revolution has increased the number of labelled and analyzed x-ray image data in the medical field, which has triggered more solutions for preventive and early diagnostics measures in the area. However contagious nature of COVID-19 makes it unsafe for medical practitioners despite the use of preventive gear and the varying examination skills of radiologists generates a biased result with different x-rays. Employing Deep Neural Network-based methodologies would help overcome the current issue. In this paper, we have compared the performance of pre-trained models Resnet18, Resnet50 and the fusion of the two Resnet models using transfer learning. We have performed cross-validation of 5 folds with 25 epochs for each fold to obtain the optimal metrics performance for all three models. Average accuracy, precision, f1-score and recall of 88.75%, 89.89%, 88.75% and 88.66% was reported for resnet18 respectively while Resnet50 yield 90.25%, 90.26%, 90.25% and 90.24% for the same. The proposed fusion model gave increased performance metrics with an accuracy of 95.75%, precision of 95.89%, recall of 95.75% and an f-1 score of 95.75%. © 2022 IEEE.