A hybrid deep transfer learning based approach for COVID-19 classification in chest X-ray images

Abstract

As a contagious disease originating from a novel coronavirus, COVID-19 leads to swollen air sacs in the lungs. It can be diagnosed using a chest X-ray (CXR) images, which is usually cheaper and less harmful than a CT scan and is always available in small or rural hospitals. X-ray machines, however, sometimes cannot diagnose COVID-19. Since the COVID-19 dataset is small and cannot be diagnosed from CXR, pre-trained neural networks can be employed for coronavirus diagnosis. This paper mainly aims to use pre-trained deep transfer learning (DTL) architectures and conventional machine learning (ML) models as an automated instrument to diagnose COVID-19 from CXRs. To overcome the lack of a large number of images, DTL is utilized to extract image features for better classification. Then, to optimize the decision-making level for infectious diseases similar to bacterial and viral pneumonia, the extracted features are selected and classified. Our proposed method was validated by creating a new CXR database from Vasei Hospital in Sabzevar, Iran. Our hybrid model achieved hit rates above 99% and outperformed for CXR of COVID-19 and similar pneumonia classification. Comparative analysis shows the superiority of the proposed COVID-19 classification model based on DTL over other competitive methods.