AI-assisted Handheld Obstetrics Ultrasound Scanning by Non-Experts
This project is creating one of the most comprehensive datasets for training AI to support pregnancy care in low-resource settings. Covering 2,000 pregnancies across five countries, it captures a wide range of conditions, including multiple pregnancies, non-pregnant women, and rare complications. Scans are performed by both expert and novice operators to reflect real-world conditions. The data will be made publicly available to enable AI tools that support non-specialist health workers.
Pregnancy care in low-resource settings is often limited by lack of specialist health workers and diagnostic tools. Standard ultrasound requires trained professionals, which restricts access. This project addresses these gaps by collecting high-quality, standardized ultrasound scans across diverse populations.
To build a global, diverse ultrasound dataset that can train AI models to support pregnancy care delivered by non-specialists, making obstetric diagnostics more reliable and accessible.
The project has created a rich dataset using the 8-sweep “blind sweep” method, covering key pregnancy and reproductive details. The dataset is uploaded to a centralized cloud and will be publicly available, enabling AI tools to improve diagnostic reliability and expand access to quality care in underserved regions.