Optimal prenatal diagnosis model for fetal heart ventricular septal defect detection using hybrid deep learning

Ventricular septal defect (VSD) is congenital cardiac defect that accounts for 26% of all congenital heart disorders (CHDs) and has an incidence rate of 3.5 per 1,000 live births. It results from chromosomal abnormalities, sequence variants, and euploidy, among other genetic pre-dispositions, as well as environmental variables. Also, 20%–30% of CHDs are caused by genetic factors, and 36.8% of VSD cases have inherited causes. Although surgical procedures often result in positive outcomes, the prognosis for VSDs linked to genetic disorders is frequently less encouraging. Together with ultrasound screening, early prenatal genetic examination improves diagnostic precision, aids in parental decision-making, maximizes prenatal treatment, and lowers newborn mortality from CHD. Obstacles like high demand for fetal cardiologists and the great number of screening cases restrict the effectiveness of detection. To increase the precision of fetal cardiac VSD identification; this work presents a sophisticated prenatal diagnostic model that makes use of hybrid deep learning (DL). Over the course of four years (2021–2024), a dataset of 1,350 2D ultrasound fetal heart pictures was gathered from online sources and a private scanning centre, guaranteeing adherence to the Helsinki Declaration and World Medical Association ethical criteria. The proposed model follows a structured pipeline, including speckle noise removal, optimal ROI segmentation, feature extraction using attention mechanism, optimal feature selection, and a DL based diagnostic system for VSD detection and classification. According to experimental results, the suggested model outperforms state-of-the-art models by 6.4%, achieved excellent prediction accuracy of 98.5%. These findings suggest that the model improves diagnostic precision and speeds up doctors’ decision-making when diagnosing VSD.