An empirical investigation of deep transfer learning approach to automatic detection of coronavirus disease (COVID-19) in CT images

Abstract

The recently discovered COVID-19 coronavirus pneumonia is highly contagious, pathogenic, and incurable without the use of an antiviral medication or vaccination that has received clinical approval. The commonest COVID-19 symptoms include a dry cough, fever, and sore throat. A severe type of pneumonia with life-threatening difficulties, such as infected tremor, multi-organ failure, pulmonary edema, and acute respirational distress syndrome (ARDS), can improve as a consequence of these symptoms. While computer-aided diagnosis (CAD) methods may aid in the early discovery of COVID-19 anomalies, it may also aid in monitoring the evolution of the illness, thereby lowering death rates. For automatic COVID-19 categorization, we compare well-known deep learning-technique feature extraction structures in this study. Deep convolutional neural networks were drawn from a source, VGG16, Xception, and GoogleNet were proposed to create the most precise and precise attribute, which is a crucial part of learning. The extracted features were then used to categorize images as either COVID-19 cases or controls using a variation of machine learning classifiers. To enable a stronger ability to generalize for unidentified data, our approach shunned data pre-processing for certain tasks techniques. A COVID-19 dataset of and CT scan images was used to validate the performance of the suggested technique. With 90% classification accuracy, the Xception feature extractor and CNN delivered the best performance.