Congenital Heart Disease Detection Using Spectral Analysis and CNN

Congenital Heart Disease (CHD) encompasses a broad range of birth defects that affect the functioning of a normal heart. In this paper, a novel approach has been proposed for the classification of three abnormalities/diseases of CHD, namely Atrial Septal Defect, Ventricular Septal Defect and Atrioventricular Septal Defect when compared to a normal heartbeat. These classes are analyzed by processing an audio file as input into a spectrogram which in turn is an input to the Convolutional Neural Network. Classification accuracy of 97.12% was achieved and can potentially be used in the development of methodology for the diagnosis of CHD.