Abstract:

Objective: This study aimed to analyze the value of artificial intelligence (AI)-assisted imaging diagnostic technology in improving the detection rate of pulmonary tuberculosis in general hospitals. Methods: This prospective and retrospective study, conducted from June to December 2024, selected 75786 patients who underwent chest CT scans at the outpatient clinics of two general hospitals, Jiangsu Provincial People’s Hospital and the Second Affiliated Hospital of Nanjing Medical University, for AI-assisted screening for pulmonary tuberculosis. Subsequently, radiologists reviewed the AI-suspected images, excluding those without suspected pulmonary tuberculosis, before a panel of experts confirmed the diagnosis. A total of 259 patients were ultimately diagnosed with pulmonary tuberculosis. Simultaneously, we used binary logistic regression to analyze the impact of clinical characteristics and CT findings on the effectiveness of AI analysis, aiming to demonstrate the effectiveness of AI in detecting pulmonary tuberculosis in actual clinical practice. Results: Among 75786 CT scans from two general hospitals, AI (AI-guided diagnostic tool) used a threshold of 0.56 to identify suspected pulmonary tuberculosis, AI indicated 8574 scans as potentially indicating pulmonary tuberculosis, ultimately confirming 259 cases. Among these, 196 were clinically diagnosed, accounting for 75.68% (196/259), representing a 102.34% ((259-128)/128) increase in detection rate compared to the number of reported cases during the same period (χ²=41.800, P<0.001). The median (interquartile range) AI threshold for confirmed pulmonary tuberculosis was 0.65 (0.59, 0.69), while the median threshold for excluded patients was 0.62 (0.59, 0.70), with no statistically significant difference (U=1783.500, P=0.738). Calcification and fibrosis were more prevalent in non-pulmonary tuberculosis cases, with adjusted odds of 0.52 (95%CI: 0.32-0.84, P=0.008) and 0.52 (95%CI: 0.31-0.86, P=0.011), respectively. Conclusion: AI-assisted CT image reading systems can significantly improve the detection efficiency of pulmonary tuberculosis patients and reduce the missed diagnosis rate; however, the diagnosis of pulmonary tuberculosis still relies on comprehensive clinical diagnosis. Calcification and fibrosis imaging features are not helpful in the diagnosis of pulmonary tuberculosis. Therefore, AI-assisted CT image reading has good application potential in general hospitals.

Key words: Tuberculosis, pulmonary, Diagnosis, differential, Artificial intelligence, Tomography, X-ray computed, Hospitals