Diagnosis of COVID-19 with a Deep Learning Approach on Chest CT Slices


Creative Commons License

Yener F. M. , Oktay A. B.

2020 Medical Technologies Congress (TIPTEKNO), ELECTR NETWORK, 19 - 20 November 2020 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/tiptekno50054.2020.9299266
  • Country: ELECTR NETWORK
  • Keywords: deep learning, convolutional neural networks, transfer learning, COVID-19, computed tomography, chest, CONVOLUTIONAL NEURAL-NETWORK, CLASSIFICATION

Abstract

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) first broke out in Wuhan, China and COVID-19 disease spread throughout the world by its highly contagious nature. High death numbers have caused a massive panic across the globe. Fast and early diagnosis is the key for preventing the virus from spreading. Besides PCR test, computed tomography (CT) of lungs is also used for diagnosis of COVID-19. Since the amount of testing kits for the diagnosis is insufficient and the conventional diagnosis methods are slow, developing AI-based fast diagnosis tools is not only an alternative way but also an urgent requirement for such alarming situations as those people faced with today. In this study, we employed three popular CNN models, VGG16, VGG19, and Xception, to classify CT scans of suspected patient cases as COVID-19 infected and non-COVID-19. VGG16 achieved 93% accuracy with the best parameters on the test set.