USF-MAE

This study evaluated a self-supervised ultrasound foundation model (USF-MAE) for automated classification of fetal renal anomalies using 969 prenatal ultrasound images. Compared with a DenseNet-169 baseline, the model achieved consistent performance improvements in both binary and multi-class classification, with particularly large gains in multi-class settings. Interpretability analysis showed that predictions were guided by clinically relevant renal structures, supporting the potential of ultrasound-specific foundation models for improving prenatal detection of kidney anomalies.