The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Download
Citation
Articles | Volume XLVI-4/W5-2021
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-4/W5-2021, 57–63, 2021
https://doi.org/10.5194/isprs-archives-XLVI-4-W5-2021-57-2021
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-4/W5-2021, 57–63, 2021
https://doi.org/10.5194/isprs-archives-XLVI-4-W5-2021-57-2021

  23 Dec 2021

23 Dec 2021

AI_COVID: AUTOMATIC DIAGNOSIS OF COVID-19 USING FRONTAL CHEST X-RAY IMAGE

I. Allaouzi1, B. Benamrou2, A. Allaouzi3, M. Ouardouz2, and M. Ben Ahmed1 I. Allaouzi et al.
  • 1LIST/Faculty of Sciences and Techniques, Abdelmalek Essaadi University, Tangier, Morocco
  • 2MMC/Faculty of Sciences and Techniques, Abdelmalek Essaadi University, Tangier, Morocco
  • 3LISAC/Faculty of Science Dhar El Mahraz, Sidi Mohamed Ben Abdellah University, Fez, Morocco

Keywords: COVID-19, Chest X-ray, Artificial Intelligence, DenseNet-121, SVM, Transfer learning

Abstract. With the continued growth of confirmed cases of COVID-19, a highly infectious disease caused by a newly discovered coronavirus called Severe Acute Respiratory Syndrome Coronavirus 2, or SARS-CoV-2, there is an urgent need to find ways to help clinicians fight the virus by reducing the workload and speeding up the diagnosis of COVID-19. In this work, we propose an artificial intelligence solution “AI_COVID” which can help radiologists to know if the lungs are infected with the virus in just a few seconds.

AI_COVID is based on a pre-trained DenseNet-121 model that detects subtle changes in the lungs and an SVM classifier that decides whether these changes are caused by COVID-19 or other diseases. AI_COVID is trained on thousands of frontal chest x-rays of people who have contracted COVID-19, healthy people, and people with viral or bacterial pneumonia. The experimental study is tested on 781 chest x-rays from two publicly available chest x-ray datasets COVID-19 radiography database and COVIDx Dataset. The performance results showed that our proposed model (DenseNet-121 + SVM) demonstrated high performance and yielded excellent results compared to the current methods in the literature, with a total accuracy of 99.74% and 98.85% for binary classification (COVID-19 vs. No COVID-19) and multi-class classification (COVID-19 vs. Normal vs. Pneumonia), respectively.