Deep Learning-Based Detection of Abnormal Axial Cross-Sectional Planes in Fetal Echocardiography

Early diagnosis of congenital heart diseases (CHDs) via fetal echocardiogram ultrasonography (US) is critical for minimizing infant mortality and enabling timely medical interventions. In this research, we proposed a deep learnin method that examines many fetal heart cross-sections at once, namely the four-chamber-view (4CV) and the three-vessel-and-trachea-view (3VTV). Three deep learning models: DenseNet169, MobileNetV2, and Xception—were trained using our dataset, which included 1,781 pictures, to categorize fetal heart images into three classifiers: normal, 4CV abnormal, and 3VTV abnormal. High classification performance was shown by the experimental findings, with individual models obtaining accuracies ranging from 91% to 94%. Our method presents a more thorough evaluation of embryonic heart problems by utilizing various ultrasound planes, which may enhance early detection, direct clinical judgments, and ultimately improve mother and fetal health outcomes.