Covidense: Providing a Suitable Solution for Diagnosing Covid-19 Lung Infection Based on Deep Learning from Chest X-Ray Images of Patients

Amir   Sorayaie Azar; Ali   Ghafari; Mohammad   Ostadi Najar; Samin   Babaei Rikan; Reza  Ghafari; Maryam Farajpouri Khamene; Peyman   Sheikhzadeh

doi:10.18502/fbt.v8i2.6517

Amir Sorayaie Azar Department of Computer Engineering, Urmia University, Urmia, Iran
Ali Ghafari Department of Medical Physics and Biomedical Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran
Mohammad Ostadi Najar Faculty of Information Technology and Computer Engineering, Urmia University of Technology, Urmia, Iran
Samin Babaei Rikan Department of Computer Engineering, Urmia University, Urmia, Iran
Reza Ghafari Pharmacy Faculty, Urmia University of Medical Sciences, Urmia, Iran
Maryam Farajpouri Khamene Faculty of New Sciences and Technologies, University of Tehran, Tehran, Iran
Peyman Sheikhzadeh Department of Nuclear Medicine, Imam Khomeini Hospital Complex, Tehran University of Medical Sciences, Tehran, Iran

DOI: https://doi.org/10.18502/fbt.v8i2.6517

Keywords: Covid-19; Deep Learning; Convolutional Neural Network; Transfer Learning; Chest X-ray Images

Abstract

Purpose: Coronavirus disease 2019 (Covid-19), first reported in December 2019 in Wuhan, China, has become a pandemic. Chest imaging is used for the diagnosis of Covid-19 patients and can address problems concerning Reverse Transcription-Polymerase Chain Reaction (RT-PCR) shortcomings. Chest X-ray images can act as an appropriate alternative to Computed Tomography (CT) for diagnosing Covid-19. The purpose of this study is to use a Deep Learning method for diagnosing Covid-19 cases using chest X-ray images. Thus, we propose Covidense based on the pre-trained Densenet-201 model and is trained on a dataset comprising chest X-ray images of Covid-19, normal, bacterial pneumonia, and viral pneumonia cases.

Materials and Methods: In this study, a total number of 1280 chest X-ray images of Covid-19, normal, bacterial and viral pneumonia cases were collected from open access repositories. Covidense, a convolutional neural network model, is based on the pre-trained DenseNet-201 architecture, and after pre-processing the images, it has been trained and tested on the images using the 5-fold cross-validation method.

Results: The accuracy of different classifications including classification of two classes (Covid-19, normal), three classes 1 (Covid-19, normal and bacterial pneumonia), three classes 2 (Covid-19, normal and viral pneumonia), and four classes (Covid-19, normal, bacterial pneumonia and viral pneumonia) are 99.46%, 92.86%, 93.91 %, and 91.01% respectively.

Conclusion: This model can differentiate pneumonia caused by Covid-19 from other types of pneumonia, including bacterial and viral. The proposed model offers high accuracy and can be of great help for effective screening. Thus, reducing the rate of infection spread. Also, it can act as a complementary tool for the detection and diagnosis of Covid-19.