Maternal and neonatal health remains one of the most pressing global health challenges, particularly in low- and middle-income countries (LMICs). Ethiopia continues to face a high burden, with maternal mortality estimated at 195 per 100,000 live births, neonatal mortality at 27 per 1,000 live births, and perinatal mortality rates ranging from 37‰ to 124‰ depending on the level of care. These outcomes remain substantially higher than the targets set under the United Nations Sustainable Development Goals (SDGs) for 2030.

The World Health Organization (WHO) recommends that all pregnant women receive at least one ultrasound scan before 24 weeks of gestation, yet nearly two-thirds of women worldwide-especially in LMICs-lack access to this service. Barriers include high costs of ultrasound machines, limited technical expertise, and shortages of skilled sonographers in rural primary care.

Artificial Intelligence-driven Point-of-Care Ultrasound (AI-POCUS) represents a promising innovation to expand prenatal imaging in resource-constrained settings by equipping frontline health workers with AI-supported diagnostic capabilities. This study, conducted under the Tsinghua University BRIGHT (Bringing Research to Impact for Global Health at Tsinghua) program, will evaluate the clinical effectiveness, feasibility, cost, and scalability of AI-POCUS in rural Ethiopia. A three-arm cluster randomized controlled trial will compare two AI-enabled ultrasound technologies-BabyChecker (Netherlands) and a China-developed AI-POCUS device-against standard antenatal care without ultrasound. Findings will generate robust clinical and policy-relevant evidence to guide the sustainable implementation of AI-enabled maternal health interventions in sub-Saharan Africa.

Maternal and neonatal morbidity and mortality remain unacceptably high in sub-Saharan Africa and continue to impede progress toward global health targets. In Ethiopia, recent estimates show maternal mortality at 195 per 100,000 live births and neonatal mortality at 27 per 1,000 live births. Perinatal mortality is also elevated, ranging between 66‰ and 124‰ in hospital-based settings and 37‰ to 52‰ in community-level health facilities. These figures surpass the Sustainable Development Goal (SDG) thresholds for 2030, underscoring the urgent need for innovative, scalable solutions.

Ultrasound imaging is a cornerstone of modern antenatal care. The WHO recommends at least one ultrasound before 24 weeks' gestation to assess gestational age, detect multiple pregnancies, identify fetal anomalies, and diagnose high-risk conditions such as preeclampsia, placenta previa, or growth restriction. However, nearly two-thirds of pregnant women worldwide still lack access to this basic diagnostic tool. In low-resource environments, the barriers include limited infrastructure, high equipment costs, technical complexity, and the scarcity of trained professionals capable of performing and interpreting scans. As a result, potentially preventable maternal and neonatal deaths remain common.

Artificial Intelligence-driven Point-of-Care Ultrasound (AI-POCUS) introduces a transformative opportunity to address these gaps. POCUS devices embedded with AI algorithms can guide non-specialist health workers in image acquisition and interpretation, reducing reliance on highly trained personnel and lowering barriers to integration within primary care. Such innovations may strengthen early detection of pregnancy complications, enable timely referral to higher-level care, and ultimately improve maternal and neonatal survival.

This study is embedded within the Bringing Research to Impact for Global Health at Tsinghua (BRIGHT) initiative. It will use a three-arm cluster randomized controlled trial (C-RCT) design to evaluate and compare: (1) BabyChecker, a portable AI-enabled ultrasound developed in the Netherlands, (2) A China-developed AI-POCUS device, and (3) Standard antenatal care (ANC) without ultrasound, reflecting current practice in many rural Ethiopian communities.

The study population will include pregnant women receiving antenatal care in rural Ethiopia, as well as primary health care providers delivering these services. Data will be collected at both the patient and facility level to capture maternal and neonatal health outcomes, health service utilization, and system-level performance indicators.

Evaluation will follow a multi-dimensional framework, addressing:

1. Clinical effectiveness: improved detection of high-risk pregnancies, reduced maternal and neonatal complications, and mortality.
2. Implementation feasibility and acceptability: user experience among health workers and pregnant women, integration into routine workflows, and perceived trust in AI-assisted care.
3. Economic evaluation: cost and cost-effectiveness of AI-POCUS compared to standard ANC, including resource utilization, referral patterns, and potential savings from earlier detection.
4. Scalability and policy relevance: analysis of barriers and enablers for broader adoption in Ethiopia and similar LMIC contexts, with direct input from policymakers and health system stakeholders.

The study aims to provide rigorous clinical evidence and practical implementation guidance on how AI-POCUS technologies can be sustainably scaled in resource-constrained settings. Findings are expected to inform national health policies, guide investment decisions, and offer a replicable model for expanding maternal health technologies across sub-Saharan Africa and other LMICs.

Inclusion Criteria:

1. Aged 15-49 years;
2. Gestational age less than 24 weeks at the first ANC visit;
3. No history of severe pregnancy complications (e.g., placenta previa, preeclampsia, etc.);
4. Signed informed consent and agreed to participate in the study.

Exclusion Criteria:

1. Pregnant women with cognitive impairments or unable to communicate effectively;
2. Failure to complete antenatal care within the specified timeframe;
3. Incomplete or unavailable records of antenatal care and delivery.

Undecided

Other parties masked in this trial include the data analysts, who will remain blinded to group assignments during statistical analyses to minimize bias in outcome assessment.