Improving Tuberculosis Diagnosis Through Artificial Intelligence (CAD4TB) and Stool Xpert MTB/RIF Testing: A Prospective Study From Oromia, Ethiopia

Background Tuberculosis remains the leading cause of death by a single infectious agent globally, with Ethiopia among the highest tuberculosis- and human immunodeficiency virus/tuberculosis-burden countries. Diagnostic gaps-particularly among household contacts (HHCs) unable to expectorate-hinder early case detection. Computer-aided detection software for chest radiography and nonrespiratory molecular assays, such as stool-based Xpert MTB/RIF testing, represent promising strategies for scalable screening. Methods We conducted a prospective diagnostic accuracy study at St Luke Catholic Hospital, Oromia, Ethiopia, enrolling 478 participants (152 tuberculosis index patients and 326 HHCs). All HHCs ≥4 years underwent digital chest radiographic screening, with or without CAD4TB (Delft Imaging) software assistance, and provided stool and sputum samples for Xpert MTB/RIF testing. The accuracy of CAD4TB and stool Xpert testing was evaluated against sputum Xpert testing as the reference. Results CAD4TB showed strong diagnostic performance, with a sensitivity of 0.77 (95% confidence interval, .70-.83) and specificity of 0.93 (.90-.96). Performance was higher among adults (sensitivity and specificity, 0.79 and 0.94) than in children (0.64 and 0.92). Stool and sputum Xpert testing demonstrated high concordance (Cohen's κ = 0.76), with a sensitivity of 0.77 (95% confidence interval, .70-.84) and specificity of 0.97 (.93-.99). During the study, 10.6% of HHCs (34 of 321) were newly diagnosed microbiologically with tuberculosis. Conclusions The combined use of CAD4TB and stool Xpert testing significantly improves tuberculosis detection, particularly among HHCs in high-burden, low-resource settings. This strategy is especially valuable in children and adults unable to produce sputum and where radiological expertise is limited.