Helping Healthcare Providers to Differentiate COVID-19 Pneumonia by Analyzing Digital Chest X-Rays: Role of Artificial Intelligence in Healthcare Practice

ABSTRACT

Background: Detecting COVID-19 pneumonia and differentiating it from community acquired pneumonia (CAP) has been a challenging task for healthcare providers since the pandemic began. We therefore aim to develop and evaluate a simple, non-invasive tool to accurately detect COVID-19 by using digital chest X-ray (CXR). Methods and Results: We performed a retrospective, multi-center study in which deep learning frameworks were used to develop the system architecture of the diagnostic tool. The tool was trained and validated by using data from the GitHub database and two hospitals in Bangladesh. Python programming was used to calculate all statistical estimates. Our study revealed that the artificial intelligence (AI)-based diagnostic tool was able to detect COVID-19 accurately by examining chest X-ray (CXR). During the testing phase, the tool could interpret CXR with precision of 0.98, recall/sensitivity of 0.97 and F1 score of 0.97 for COVID-19. The evaluation results showed high sensitivity (90%) and specificity (92%) in detecting COVID-19. The AUC values for COVID-19 and pneumonia were 0.91 and 0.87, respectively. Conclusion: The developed AI-based diagnostic tool can offer the healthcare providers an effective means of detecting and differentiating COVID-19 from other types of pneumonia, thus contributing to reducing the long-term impact of this deadly disease. © 2022, International Medical Research and Development Corporation. All rights reserved.