Fetal growth restriction affects up to 10 percent of pregnancies and is a significant contributor to perinatal mortality. Monitoring fetal abdominal circumference (AC) is a standard method for assessing fetal development, but this requires access to ultrasound imaging and skilled personnel. In low-resource settings, such access is often limited.
The ACOUSLIC-AI (Abdominal Circumference Operator-agnostic UltraSound measurement in Low-Income Countries) challenge examined whether artificial intelligence can support AC measurement using ultrasound images collected by novice operators with minimal training. The challenge focused on a method that does not require the user to interpret images in real-time.
Study design and method
The study used blind-sweep ultrasound data from healthcare workers in Sierra Leone, Tanzania, and the Netherlands. Operators had received one hour of training and performed standardized free-hand sweeps without viewing the ultrasound images during acquisition. The equipment used was a low-cost portable probe connected to a mobile device.
Participants in the challenge developed algorithms to identify suitable frames for fetal AC measurement and generate segmentation masks of the fetal abdomen. The datasets used for training and evaluation included more than 500 cases, annotated by experienced clinical readers. Evaluation was conducted using established image analysis metrics.
Results
Three algorithms showed performance levels consistent with clinical standards. One model achieved limits of agreement comparable to those reported for interobserver variability in clinical AC measurements. These results suggest that automated measurements are viable using data from blind-sweep acquisitions, even without expert sonographers.
The study also showed that consistent performance could be achieved across different regions and healthcare settings. This supports the feasibility of implementing similar AI tools in environments with limited technical capacity.
Implications for field-based maternal care
The approach tested in the ACOUSLIC-AI challenge corresponds with the operational conditions of many maternal care programs in low-income countries. Eliminating the need for real-time operator judgment enables more consistent data collection and potential scale-up of fetal growth monitoring.
This model of care may support antenatal screening programs that rely on community health workers or non-specialist staff. It also provides a reference framework for incorporating AI-based fetal biometry into broader maternal health strategies.
Conclusion
The ACOUSLIC-AI challenge presents a validated method for estimating fetal abdominal circumference using AI and blind-sweep ultrasound. It offers a practical solution for extending access to fetal measurements where standard sonography is unavailable. The study contributes to the ongoing development of appropriate diagnostic tools for resource-constrained environments.
