Open Access

ARTICLE

Identification of Tuberculosis and Coronavirus Patients Using Hybrid Deep Learning Models

Mohammed A. Al Ghamdi^*

Computer Science Department, Umm Al-Qura University, Makkah City, 24243, Saudi Arabia

* Corresponding Author: Mohammed A. Al Ghamdi. Email:

Computers, Materials & Continua 2023, 76(1), 881-894. https://doi.org/10.32604/cmc.2023.037826

Received 17 November 2022; Accepted 23 April 2023; Issue published 08 June 2023

Abstract

Considerable resources, technology, and efforts are being utilized worldwide to eradicate the coronavirus. Although certain measures taken to prevent the further spread of the disease have been successful, efforts to completely wipe out the coronavirus have been insufficient. Coronavirus patients have symptoms similar to those of chest Tuberculosis (TB) or pneumonia patients. Chest tuberculosis and coronavirus are similar because both diseases affect the lungs, cause coughing and produce an irregular respiratory system. Both diseases can be confirmed through X-ray imaging. It is a difficult task to diagnose COVID-19, as coronavirus testing kits are neither excessively available nor very reliable. In addition, specially trained staff and specialized equipment in medical laboratories are needed to carry out a coronavirus test. However, most of the staff is not fully trained, and several laboratories do not have special equipment to perform a coronavirus test. Therefore, hospitals and medical staff are under stress to meet necessary workloads. Most of the time, these staffs confuse the tuberculosis or pneumonia patient with a coronavirus patient, as these patients present similar symptoms. To meet the above challenges, a comprehensive solution based on a deep learning model has been proposed to distinguish COVID-19 patients from either tuberculosis patients or healthy people. The framework contains a fusion of Visual Geometry Group from Oxford (VGG16) and Residual Network (ResNet18) algorithms as VGG16 contains robust convolutional layers, and Resnet18 is a good classifier. The proposed model outperforms other machine learning and deep learning models as more than 94% accuracy for multiclass identification has been achieved.

---

Keywords

---