Transfer Learning-Based Benchmarking Study for Diagnosis of COVID-19 from Lung CT Scans

Abstract

The virus known as Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) or Coronavirus Disease 2019 (COVID-19), which emerged from the city of Wuhan in the People’s Republic of China, has affected the whole world. This disease, which is categorized as an epidemic disease, continues to increase despite the various measures taken. It is aimed to reduce death and infected people rates with vaccination studies, inspection and early diagnosis. On the other hand, new types of coronavirus cases are emerging and people are kept under surveillance to prevent the spread of the virus. By keeping the infected people under quarantine, the transmission of the epidemic to more people is prevented. For this reason, early diagnosis kits and tests are vital. Today, various abnormalities are detected by specialists thanks to medical imaging tools. On the other hand, this process is performed on medical images using image processing techniques. Thanks to methods such as image classification, image segmentation, image quantification and various operations such as object detection, localization and quantitative analysis on the object are performed. In this study, it is aimed to detect COVID-19 on lung CT scan images with deep learning methods. CNN-based state-of-art deep learning models, which were pre-trained with millions of images and applied transfer learning method for a similar problem, were used in this study. This process was performed by choosing VGG19, ResNet152 and MobileNetV2 models and the results were compared. According to the performance criteria, validation accuracy of 93.53%, 95% and 87.28% was obtained from VGG19, ResNet152 and MobileNetV2 models, respectively. These results show that these models give good results for the detection of COVID-19 from lung CT scan images.