Detection of Covid-19 from Computed Tomography Images with DenseNet Based Deep Learning Models


Ala A., POLAT Ö.

29th Signal Processing and Communications Applications Conference (SIU2021), İstanbul, Turkey, 9 - 11 June 2021 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu53274.2021.9477885
  • City: İstanbul
  • Country: Turkey
  • Keywords: covid-19, classification, deep transfer learning, densenet architectures
  • Sivas Cumhuriyet University Affiliated: Yes

Abstract

Covid-19 infection caused by the new type of coronavirus (SARS-CoV-2), which emerged in Wuhan city of China in December 2019, has become a deadly pandemic in the world. Fast and accurate detection of Covid-19 will allow treatment to be started without wasting time, and thus the patient's chance of survival will increase. In this study, DenseNet121, DenseNet169 and DenseNet201 architectures, which are deep transfer learning models, are proposed for the detection of Covid-19 from computed tomography (CT) images. In order to train and test the models used, a publicly available dataset consisting of 2482 CT images, 1252 with Covid-19 and 1230 without Covid-19, was used. Each model was tested 10 times and the performance of the models is given in terms of precision, recall, f1-score and accuracy. Looking at the average of 10 tests, the best model is DenseNet201 with 97.72% success. The best classification performance was obtained by using DenseNet169 model with 98.66% accuracy. With the use of other DenseNet models, very good results have been yielded. Therefore, the proposed model is promising in medical sciences and can help radiologists make quick and correct decisions.