This white paper introduces HeartAssist™, a deep learning–based tool designed to automate classification and measurement in fetal heart ultrasound imaging. Developed by Samsung Medison and evaluated at Asan Medical Center, the tool supports standardized assessment of the fetal heart across ten key ultrasound views.
A total of 4,000 fetal echocardiograms between 20 and 40 weeks of gestation were analyzed. HeartAssist™ automatically classified views such as the four-chamber view, LVOT, RVOT, three-vessel views, aortic arch, and ductal arch. From these views, 44 parameters were measured, including diameters, areas, circumferences, and Z-scores for fetal cardiac growth comparison.
The study found a 95.5% classification accuracy, 100% annotation accuracy, and a dice similarity coefficient (DSC) of 84.8% for segmentation performance. HeartAssist™ significantly reduced processing time by 87% compared to manual measurements. It also performed reliably on unclear or ambiguous images.
HeartAssist™ was implemented on ultrasound systems such as WS80A and HERA W10, supporting clinical workflows with its automatic view recognition and real-time measurement capabilities. The tool helps address challenges like inter-observer variability, fetal motion, and image artifacts—common barriers in fetal echocardiography.
By standardizing fetal heart image interpretation, HeartAssist™ improves diagnostic consistency and may enhance detection rates for congenital heart disease (CHD). The system offers clinical support for both specialists and non-experts performing fetal echocardiograms.
To learn how HeartAssist™ for Fetus streamlines fetal heart ultrasound with automated classification and measurement, watch the HeartAssist™ for Fetus workflow guide video .
