Pediatric CHD Detection

This study proposes a deep learning method using seven standard echocardiographic views to improve the detection of congenital heart disease (CHD) in children, addressing limitations of prior approaches that relied on fewer views. Using echocardiographic data from 1,411 pediatric patients, the model achieved strong performance in the test set, with an AUC of 0.91 and an accuracy of 92.3%, and remained robust even under shear transformation interference when appropriate images were provided. The results demonstrate that incorporating multiple anatomical views enhances diagnostic accuracy and robustness, highlighting the model’s practical clinical value.