We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
COVID-19 diagnosis from chest CT scan images using deep learning.
- Authors
ALASSIRI, Raghad; ABUKHODAIR, Felwa; KALKATAWI, Manal; KHASHOGGI, Khalid; ALOTAIBI, Reem
- Abstract
Coronavirus disease 2019 (COVID-19) has caused nearly 600 million individual infections worldwide and more than 6 million deaths were reported. With recent advancements in deep learning techniques, there have been significant efforts to detect and diagnose COVID-19 from computerized tomography (CT) scan medical images using deep learning. A retrospective study to detect COVID-19 using deep learning algorithms is conducted in this paper. It aims to improve training results of pre-trained models using transfer learning and data augmentation The performance of different models was measured and the difference in performance with and without using data augmentation was computed. Also, a Convolutional Neural Network (CNN) model was proposed and data augmentation was used to achieve high accuracy ratios. Finally, designed a website that uses the trained models where doctors can upload CT scan images and get COVID-19 classification (https://covid-e46e8.web.app/) was designed. The highest results from pre-trained models without using data augmentation were for DenseNet121, which was equal to 81.4%, and the highest accuracy after using the data augmentation was for MobileNet, which was equal to 83.4%. The rate of accuracy improvement percentage after using data augmentation was about 3%. The conclusion was that data augmentation could improve the accuracy of COVID-19 detection models as it increases the number of samples used to train these models.
- Subjects
COVID-19; DEEP learning; COMPUTED tomography; COVID-19 testing; DATA augmentation; CONVOLUTIONAL neural networks; MACHINE learning
- Publication
Romanian Journal of Information Technology & Automatic Control / Revista Română de Informatică și Automatică, 2022, Vol 32, Issue 3, p65
- ISSN
1220-1758
- Publication type
Article
- DOI
10.33436/v32i3y202205