Echofocus-CHD: Autonomous Detection and Stratification of Critical Congenital Heart Disease (CHD) in Prenatal Ultrasound

Critical Congenital Heart Disease (CHD) is one of the most important causes of neonatal morbidity and mortality and early prenatal diagnosis is essential for effective treatment and better clinical outcomes. In this study, ECHOFOCUS-CHD, an AI-inspired autonomous framework for the diagnosis of Critical Congenital Heart Disease from ultrasound images of fetuses is proposed. A quantitative experimental research design was used and 200 prenatal ultrasound scans acquired from hospitals and fetal echocardiography databases are used. A Convolutional Neural Network (CNN) was created to classify fetal cardiac conditions into normal, mild, moderate and critical CHD categories. The models are enhanced by using image preprocessing methods like noise reduction, normalization, contrast enhancement, and data augmentation. An overall accuracy of 94.5%, sensitivity of 92.8%, specificity of 95.6% with an AUC value of 0.96% was obtained, which shows an excellent diagnostic capability in the proposed system. The reliability of the framework was also verified using confusion matrix and ROC curve analyses. The system was also able to automatically classify the heart defects of foals using images from their prenatal scans, aiding clinical decisions during pregnancy and early treatment planning. The results show that AI-powered prenatal ultrasound analysis has the potential to greatly improve the effectiveness of early detection of CHD and prenatal screening.