C106-02 Beyond a Screening Tool: How Artificial Intelligence Computer-aided Chest X-ray Screening Could Impact TB Management

Abstract Rationale TB remains the infectious disease leading cause of death annually. Rapid screening tools for timely treatment and containment is critical. In January 2025, the WHO endorsed six computer-aided detection (CAD) platforms to assist with rapid diagnosis using chest x-rays (CXR), including Qure AI (qXR - Qure.ai, Mumbai, India), but questions remain regarding optimal implementation of these new artificial intelligence (AI) tools. As patient characteristics affect TB disease severity and lung damage, we explored if baseline qXR TB scores could be utilized for more nuanced interpretation than just binary TB positive or negative; and further, the scores utility predicting risk for post-TB lung disease. Methods We conducted a retrospective study of clinical data from patients seen at a TB clinic in South Korea. We assessed patient demographics, and serologic and diagnostic data of 151 patients diagnosed and treated for TB. CXR’s taken prior to treatment (0 months) and at treatment completion (6 months) were evaluated by qXR software and analyzed. Approval was obtained from the Institutional Review Board of Guro Hospital, Korea University. Results The average age was 56 years-old (Standard deviation (SD) = 15.8), was predominantly male (74%), and the mean BMI was 22.1 (SD = 2.7). Forty percent (61/151) were current smokers, and the majority (92%) had no previous TB history (140/151). No one had HIV and 16% (24/151) had diabetes. Definitive TB diagnosis was based on bacterial confirmation for 68% (102/151) of patients, and the remaining 33% (49/151) were diagnosed clinically by a pulmonologist. When evaluated by qXR, 86% of the patients screened positive at 0 months (median TB score 0.920, IQR 0.69, 0.97) and 69% remained positive at 6 months (median TB score 0.728, IQR 0.378, 0.940). Stratifying by BMI, smoking status, and diabetes, indicated a statistically significant difference in median TB scores at both enrollment and treatment completion, whereas age stratification only showed a significant difference at 6 months (Table 1).Seventy (46%) patients completed pulmonary function tests at 6 months. Low DLCO (80%) was associated with a positive reticulonodular and nodule score at 6 months. (Table?1). Conclusion We demonstrated that patient characteristics are associated with diverse qXR scores, post treatment parenchymal damage, and propose that qXR score components could be utilized for predicting post-treatment lung damage. Novel interpretation of qXR scores beyond detecting a binary TB status could substantially improve our understanding of TB and its long-term impacts from a single CXR. This abstract is funded by: None