Application of artificial intelligence in reading X-ray images for pulmonary

Abstract

Research on the implementation of Artificial Intelligence (AI) in analyzing chest X-ray images to support pulmonary tuberculosis screening has been implemented at the VISA Department, Cho Ray Hospital. The objective of this study is to evaluate the effectiveness of AI in diagnosing pulmonary tuberculosis. The research methodology employed AI software based on two deep learning algorithms implemented through convolutional neural networks: automatic lung nodule detection and automatic pulmonary tuberculosis warning. These algorithms were trained on large datasets from Seoul National University Hospital and other medical institutions. The study results demonstrated that the diagnostic performance of AI surpassed that of many radiologists, achieving AUROC (area under the alternative free-response of receiving operating curves) scores of 0.91 (lung nodule detection) and 0.988 (pulmonary tuberculosis warning), with a specificity of 95.2 % and a sensitivity of 80.7 %. The AI system processed 9660 chest X-rays and supported the detection of 46 suspected pulmonary tuberculosis cases. The novel contributions of this study highlight the remarkable efficiency of AI in practical applications within high-frequency X-ray environments. AI effectively supported the detection of small, hard-to-identify lesions, enhanced diagnostic quality, reduced errors, and minimized additional procedures such as sputum tests or computed tomography scans. Consequently, it reduced costs, time, and radiation exposure for patients. This study opens up significant potential for optimizing the pulmonary tuberculosis screening process in Viet Nam.