Performance Analysis of Deep Transfer Learning for Manifestation of COVID-19 Using Chest X-ray

Lungs can be affected by various bacterial and viral infections other than novel coronavirus, popular as COVID-19. Any machine learning technique should be capable to differentiate among these infections and classify an image to generate inference matching with actual cause of disease. In this paper, we have analyzed the performance of VGG19 for diagnosis of COVID-19 using X-ray images of lungs infected by bacterial and viral pneumonia. The visual clarity of X-ray images is very low compared to CT scan. However, the accuracy obtained by supports our claim of using VGG19 as a low coast and easily accessible automated alternate to CT scan based diagnosis. Datasets having images of 3-class (including normal, viral pneumonia and COVID-19) and 4-class (including normal, bacterial, viral pneumonia and COVID-19) categories were used to analyze the performance of VGG19 Deep Transfer Learning Model for accurate diagnosis of COVID-19. Sensitivity and accuracy of VGG19 were compared with AlexNet and ResNet19 models. VGG19 produced an accuracy of 98.2% with 3-class dataset and 94.4% with 4-class dataset.