IEEE International Conference on Signal Processing and Communication ( ICSPC 17)

HEART SOUNDS SEGMENTATION AND CLASSIFICATION USING ADAPTIVE LEARNING NEURAL NETWORKS

Heart diseases are a cause for a large number of deaths across the globe and are particularly greater in developing and third world nations. It is of prime importance to tackle the problem. Our motivation is to help doctors screen such diseases early on to prevent unnecessary deaths. In the given paper we describe a novel approach towards classifying heart sounds based on whether they belong to the normal or murmur category of the given dataset. Using a dataset of audio signals collected from hospital trials using Digital stethoscopes, we propose a two-step process to detect heartbeat anomaly. First, an algorithm based on peak detection using Shannon energy envelope calculation and feature construction for Systoles (S1 or lub) and Diastole (S2 or dub) segments of a heart sound is implemented. Next, the classification of these heartbeats is done using neural networks. The accuracy is further increased by the use of backpropagation neural networks with adaptive learning technique.