Prenatal CHD Screening

This study evaluated a previously trained deep learning model on second-trimester fetal ultrasound images from a low-risk population to detect congenital heart disease (CHD). The model achieved 91% sensitivity and 93% specificity, significantly outperforming both real-time clinical screening and blinded human experts, who showed much lower sensitivity. The findings demonstrate that deep learning can substantially improve prenatal CHD detection, even for lesions not previously seen during training.