Multicenter Implementation Evaluation of AI-Enabled Chest X-Ray Screening for Tuberculosis and Lung Nodules in Ethiopia
Year: 2026
Type: article
Abstract: Introduction: Tuberculosis (TB) and lung cancer remain major public health burdens in low- and middle-income countries, including Ethiopia, where limited diagnostic and radiological capacity constrains early detection. Integrating artificial intelligence (AI) into chest X-ray (CXR) screening offers a scalable approach to enhance diagnostic accuracy, supporting timely patient management, and improving clinical outcomes. This study evaluated the prospective implementation of an AI decision-support tool (qXR) across three referral hospitals using the Consolidated Framework for Implementation Research (CFIR 2.0) and Proctor’s implementation outcomes framework. Methods: This prospective multicenter implementation study was conducted in three major referral hospitals in Addis Ababa, selected for their established TB and oncology services. Implementation determinants were mapped to CFIR 2.0 domains (Intervention Characteristics, Inner Setting, Outer Setting, Individuals, and Implementation Process). Implementation outcomes (Feasibility, Adoption/Concordance, Fidelity, Penetration, and Operational Challenges) were evaluated using descriptive statistics in SPSS version 25. Presumptive cases flagged by AI tool were reviewed by radiologists; concordant cases proceeded to bacteriological confirmation (GeneXpert) for TB or low-dose CT (LDCT) and Lung-RADS® evaluation for pulmonary nodules Results: Out of the 23,916 collected chest radiographs, 21,692 were validly processed (Feasibility). The AI flagged 3,872 presumptive TB cases and 2,576 pulmonary nodules (392 high-risk, 2,184 low-risk). Radiologists reviewed 3,652 TB-flagged and 1,523 nodule-flagged cases, demonstrating low radiologist-AI concordance (Adoption): 12.1% (n=443) for TB and 11.3% (n=172) for nodules. Diagnostic cascade fidelity declined significantly during follow-up (Fidelity/Penetration): only 268 concordant TB cases underwent GeneXpert testing, yielding 24 confirmed TB cases, while 77 nodule cases underwent LDCT, yielding two confirmed primary lung cancers. Systemic determinants within the Inner and Outer Setting, including server downtime, X-ray equipment failure, GeneXpert cartridge shortages, and staff turnover, substantially curtailed downstream detection yield. Conclusion: AI-enabled CXR is a practical screening tool, but sustainable impact requires human oversight, robust infrastructure, and strengthened health system support. Crucially, this necessitates seamless workflow integration, defined as the harmonization of the AI solution with existing digital radiography, DICOM protocols, Picture Archiving and Communication Systems (PACS), and routine radiologist reporting procedures. [Ethiop. J. Health Dev. 2026; 40(2)] Keywords: Artificial Intelligence; Tuberculosis; Lung nodule; Qure.AI; Low- and middle-income countries
Authors: Amanuel Getachew, Eyob Woldeyohannes, Amsalu Bekele, Rahel Argaw Kebede, Aschalew Worku, Hanan Yusuf, Amsalu Bitew, Azmera Gissila, Hanna Damtew, Hawi Farris, Yonas Dendea, Wondwosen Ergete, Nathan Estifanos, Mekonnen Birhanu, Solomon Tessema, Selamawit Tilahun, Million Mola, Eyob Kebede Etissa
Language: English
Subfield: Infectious Diseases
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7222369715
