5th International Conference on Gynecology and Obstetrics

Scarlet Barker has completed her MSc in the Control of Infectious Diseases at the London School of Hygiene and Tropical Medicine and holds a BA in Global Health and Social Medicine from King’s College London, United Kingdom. She is a research associate in maternal health, working on the CRADLE-SMART, based in Freetown, Sierra Leone.

Background: Accurate gestational age (GA) estimation is essential for antenatal care (ANC); however, access to ultrasound is limited in many low-resource settings. AI-enabled point-of-care ultrasound (AI-POCUS) may allow for GA assessment to be task-shared to non-specialist users.

Objectives: We evaluated agreement between AI-POCUS and conventional ultrasound, alongside feasibility, acceptability, and implementation in Sierra Leone.

Methods: In this prospective mixed-methods cohort study at Aberdeen Women’s Centre, Freetown, women with viable singleton pregnancies (<28 weeks’ GA) underwent same-day AI-POCUS by data collectors without sonography training, and ultrasound biometry by a fetal medicine-trained consultant. Agreement was defined as AI-POCUS minus specialist-ultrasound GA; bias and limits of agreement were estimated using quantile regression. The proportion within ±7 days was examined. Focus-group discussions (N=20) with women with an AI-POCUS scan and healthcare workers (HCWs) explored acceptability and implementation.

Results: Of 486 AI-POCUS scans, 17 (3.5%) failed to generate an estimate. Agreement varied systematically with GA. At 10 weeks, the median difference was +1.7 days (95% CI −0.1 to 3.5 days); by 28 weeks, it was -7.4 days (95% CI -9.0 to -5.8). 79% (95% CI 75–83%) of estimates were within ±7 days of specialist ultrasound.

Women valued AI-POCUS for birth preparedness and expressed a strong trust in its GA estimates. HCWs perceived potential for task-sharing to increase GA estimation skills and reduce care delays. Participants supported community rollout and identified equipment availability and reliable electricity as barriers to sustainability.

Conclusion: Non-specialist users successfully obtained AI-generated GA estimates in 96.5% of scans, with 79% within ±7 days of specialist ultrasound. Women and HCWs valued its potential to expand access to pregnancy dating. However, agreement diverged systematically with advancing GA. High perceived trust alongside gestation-dependent disagreement highlights the need for validation against independently established early-pregnancy dating before wider implementation.