Classification of HLHS Anomaly Images using ANN

congenital heart disease (CHD) is one of the severe structural abnormalities which cause neonatal mortality. The severity of CHD depends on the early stage, moderate stage, and advanced stage of diseases. According to the cardiological society of India (CSI), the variation of incidence is about 74/2000 live births present at the time of birth. Among the different CHD’s, Hypoplastic Left Heart Syndrome (HLHS) is considered a significant cyanotic heart disease. HLHS is a congenital disability that affects normal blood flow through the heart. The sonographic marker for HLHS identification is the underdevelopment of the left ventricle. Even though ultrasound devices are there to scan the fetus growth, it is still challenging due to the inherent speckle noise, continuous movement of the fetus and low quality of ultrasound devices used for scanning. So, an optimization methodology is still in demand for these kinds of problems. This paper combines different optimization techniques such as fast non-local mean denoising (FNLM) methodology for reducing the speckle noise, active geodesic contouring to segment the left ventricle for sonographic identification of HLHS and ANN with Bayesian Regularization for classifying the normality and abnormality present in the dataset. The dataset consists of 569 four-chamber view (4CV) ultra sound images for training and testing. The five different features used for evaluation are radius, texture, perimeter, smoothness, and symmetry. 80% of the data are used for training, and 20% for testing. The ANN Bayesian model numerically calculates the mean squared error (MSE) of 0.018 and classifies the abnormality with an accuracy of 97%. This model gives a second opinion for medical practitioners and gynecologists.