Performance Analysis of AI-Assisted Chest Radiography for COVID-19 Pneumonia Diagnosis in Resource-Limited Settings

Abstract

Introduction: Chest radiography (CXR) is commonly used for diagnosing lung and cardiothoracic disorders, including coronavirus disease (COVID-19) pneumonia. However, its diagnostic accuracy during the early COVID-19 stages was limited. Artificial intelligence (AI) can enhance CXR analysis and diagnostic accuracy. Objective: To evaluate AI in X-ray diagnostics for COVID-19 patients in Kyrgyzstan. Methods: Three radiologists reviewed CXR reports of pneumonia patients and healthy individuals. An AI system with the MedVit deep learning model identified COVID-19 pneumonia, and its reports were compared to radiologists’ interpretations to evaluate diagnostic accuracy. Results: AI’s performance in detecting pneumonia matched that of radiologists, with 88.31% sensitivity and 96.67% specificity. High Youden index values indicated quality. AI can enhance X-ray accuracy, especially in resource-limited settings, though challenges like data quality, standardization, and ethics must be addressed for widespread adoption. Conclusion: Collaboration between radiologists and AI can enhance radiological reports for patients with COVID-19 pneumonia, particularly in rural areas with staff shortages.