Study on the prediction of congenital cardiac abnormalities using various Machine learning models

Introduction Congenital heart disease (CHD) involves structural heart defects present from birth. Ventricular septal defects (VSDs) are among the most common types. Early diagnosis is important and can be done using fetal echocardiography at 12–14 weeks of gestation. However, detection rates depend on the quality of diagnostic tools and expertise. Machine learning (ML) can enhance detection through various diagnostic modalities, including electrocardiogram (ECG) and ultrasonography (US). Aim and Objectives This study aims to improve CHD detection by integrating fetal echocardiography with machine learning techniques. Method The study explores methods for detecting CHD using an online dataset, employing preprocessing, feature extraction, and deep learning classification. Results There was notable variability in model performance metrics. The Decision Support System for Early Prediction (DSSEP) had the highest sensitivity (80.11%) but a lower positive predictive value (PPV) and specificity compared to the Heart Deep Learning model (CDLM), which showed the highest specificity (88.25%) and PPV (91.31%). The Predictive Analysis of Congenital Heart Defects (PACHD) model had the lowest sensitivity (59.78%) and PPV (56.45%), while the Machine Learning-Based Discharge Prediction (MLBDP) model had the lowest specificity (59.78%) and the highest miss rate (40.22%). These findings highlight the importance of selecting appropriate models based on performance metrics. Conclusion The DSSEP model demonstrated higher sensitivity and lower miss rates, making it strong for early detection, whereas the CDLM model offered higher specificity and PPV, reducing false positives.