BRIDGING THE GAP: IMPLEMENTING LOW-COST AI RADIOLOGY SOLUTIONS IN RESOURCE-LIMITED SETTINGS

Abstract

Radiology service provision in low- and middle-income countries (LMICs) is hampered by chronic shortages of trained radiologists and inadequate digital infrastructure, most notably in rural healthcare facilities. This research prospectively assessed the feasibility, diagnostic performance, and operational robustness of a low-cost, offline-enabled artificial intelligence (AI) system for chest X-ray interpretation in resource-limited settings. An open-source deep learning model based on CheXNet was implemented in five hospitals in northern India and incorporated into regular clinical practice without the need for real-time internet connectivity. The system analyzed 1,830 radiographs over a period of six months, out of which 1,500 were employed for diagnostic assessment. The AI system attained 86.5% accuracy, 88% sensitivity, 85% specificity, and an area under the receiver operating characteristic (AUC) of 0.91, similar to board-certified radiologists. The system was 97% available and generated diagnostic reports in 5–10 seconds per image. 23 clinical and IT personnel provided structured feedback with high usability (mean score 4.4/5), faith in AI results (4.1/5), and willingness to implement the system (4.6/5). Unlike previous retrospective or cloud-based research, this study illustrates real-time clinical integration of an independent, edge-deployed AI system within LMIC primary care clinics. The results highlight the promise of AI-supported diagnostic tools to increase access to imaging services, assist frontline health workers, and minimize diagnostic delays. This paper provides necessary implementation evidence and concurs with global health priorities aimed at universal access to diagnostics. Subsequent studies ought to assess longitudinal effect, cross-modality extension, and national-scale integration in digital health systems.