Synergistic Foundation Models for Semi-Supervised Fetal Cardiac Ultrasound Analysis: SAM-Med2D Boundary Refinement and DINOv3 Semantic Enhancement

Researchers introduced a semi-supervised fetal cardiac ultrasound framework that combines SAM-Med2D and DINOv3.

The system uses SAM-Med2D for sharper segmentation boundaries and DINOv3 to generate stronger pseudo-labels.

A two-stage training strategy helps preserve segmentation gains while recovering classification accuracy.

The method delivered strong results on the FETUS 2026 benchmark, supporting earlier congenital heart disease screening.