Classification of heart sound signals for the detection of heart diseases

This paper proposed the classification of heart sound signals for the detection of heart diseases. The heart sound signals were acquired from pediatric patients of National Heart Institute, Kuala Lumpur. Each signal was characterized by applying Nonlinear ARX (NARX) model and weight parameters of each disease were estimated. Prior to classification, the spectrogram was applied to the modeled signal for feature extraction and selection. The obtained frequency pattern features were fed to the FFNN and trained using Resilient Backpropagation (RPROP) algorithm. With optimized learning parameter of 0.07, the network gave its best performance at 32-220-6. The accuracy of the network when validated with the diagnostic test was above 97% which suggested that the network performed well and was operating as gold standard. The classification of heart diseases was further improved to 100% when overall testing was performed