| Issue/Objective |
Adverse maternal outcomes remain a leading cause of mortality and morbidity in low-resource settings, primarily due to delayed risk detection, poor referral systems, and limited healthcare infrastructure. The integration of machine learning (ML) in maternal healthcare presents an innovative opportunity to bridge these gaps. This synthesis aims to explore how machine learning has been used to predict high-risk maternal conditions and proposes a conceptual framework tailored for resource-constrained environments. |
| Methodology/Approach |
A systematic review of peer-reviewed studies published between 2018 and 2024 was conducted using Google Scholar. Search strings combined the terms ("machine learning" OR "artificial intelligence") AND ("maternal health") AND ("low-resource settings"). Out of 80 initially retrieved papers, 10 met the inclusion criteria based on methodological quality and relevance to real-world implementation. Key elements such as algorithm type, dataset characteristics, performance metrics, and deployment feasibility were extracted and synthesized. A predictive framework suitable for low-resource settings was then developed based on recurring themes and best practices. |
| Results |
Multiple studies demonstrated the strong predictive capability of machine learning models. For instance, Willis et al. (2024) reported 89% accuracy using blind-sweep ultrasound to estimate gestational age. Boehmer et al. (2024) achieved 82-90% effectiveness in vital sign monitoring through reinforcement learning. Vousden et al. (2022) validated wearable diagnostics in clinical trials with high diagnostic accuracy. Ensemble models such as XGBoost consistently outperformed traditional approaches. However, challenges included poor generalizability across geographies, limited availability of localized training data, and integration issues within under-resourced health systems. |
| Discussion/Conclusion |
Machine learning has significant potential to revolutionize maternal healthcare in low-resource environments by enabling early detection and risk stratification. However, real-world application requires more than algorithmic accuracy-it depends on localized data generation, user-friendly interfaces, and policy support. The proposed framework emphasizes adaptable, scalable, and community-integrated machine learning tools. Pilot implementation is recommended in countries like Nigeria and Ethiopia to assess operational feasibility and impact. This work aligns strongly with the conference themes of global health equity, innovation, and localized impact. |
| Presenters and affiliations |
Ramon Ibraheem Obafemi Awolowo University Ile Ife |
