MaternalU5Triage AI Technical White Paper

Description MaternalU5Triage: An AI-Powered Clinical Decision Support System for Maternal, Neonatal and Under-Five Emergency Triage is a comprehensive technical white paper describing the design, development, implementation, clinical validation, and future evolution of an artificial intelligence (AI)-enabled Clinical Decision Support System (CDSS) for maternal, neonatal, and under-five emergency care. The platform was developed to strengthen emergency triage, risk stratification, referral decision-making, and continuity of care by providing healthcare professionals with transparent, evidence-based, and explainable AI-assisted clinical recommendations. MaternalU5Triage integrates deterministic clinical rules, supervised machine learning, Retrieval-Augmented Generation (RAG), Large Language Models (LLMs), knowledge graphs, and Explainable Artificial Intelligence (XAI) within a human-in-the-loop decision-support framework that preserves clinician oversight and accountability. The current implementation comprises a functional Android-based software prototype with implemented clinical workflows, evidence-based decision pathways, and AI-assisted reasoning capabilities. The platform has undergone multi-site proof-of-concept clinical validation at the University of Nigeria Teaching Hospital (UNTH) Parklane, Enugu, and Community Primary Health Care Centre, Abor. The evaluation involved 50 healthcare professionals and 35 representative clinical cases/patient encounters using structured clinical validation and usability assessment methods to evaluate clinical decision-support performance, workflow integration, usability, acceptability, and implementation feasibility. Evaluation findings demonstrated technical feasibility, clinical relevance, and strong user acceptance. The platform achieved 70% agreement with clinician decisions, 79% correct triage classification, an average triage completion time of 90 seconds per patient, a System Usability Scale (SUS) score of 84.2/100, 85% overall user satisfaction, and 74% intention among participants to continue using the platform. Feedback obtained during the evaluation informed iterative improvements to system functionality, user interface design, clinical workflows, and AI-assisted decision-support capabilities. This white paper documents the complete technical architecture of MaternalU5Triage, including system architecture, software architecture, AI architecture, hybrid clinical reasoning, training data strategy, Retrieval-Augmented Generation (RAG), Explainable Artificial Intelligence (XAI), interoperability, cybersecurity, privacy and data governance, responsible AI governance, validation methodology, implementation strategy, deployment considerations, monitoring, and future development roadmap. The platform is designed to support interoperability with established healthcare information standards, including FHIR, SNOMED CT, ICD-11, and LOINC. The document is intended for researchers, clinicians, software engineers, digital health practitioners, health informaticians, policymakers, funding organizations, and implementation partners interested in trustworthy AI-enabled clinical decision support systems for maternal, newborn, and child health. It serves as both a technical reference and an implementation resource for advancing safe, explainable, and evidence-based AI applications in healthcare, particularly within low-resource and resource-constrained settings.