Classification of heart sounds using a multilayer feed-forward neural network

Cardiovascular diseases are among chronic diseases that seriously threaten human health. Medical experts sometimes listen to heart sounds through auscultation system, either from stethoscope or PCG (phonocardiogram), as one of the ways in diagnosing cardiovascular diseases. The auscultation process is subjective and largely depends on the experience, skill, knowledge or hearing ability of the physician to interpret heart conditions. This paper examines heart valve-related diseases using a multilayer feed-forward neural network (MFNN). The heart sounds were digitally filtered with moving average filter, segmented with sliding window and then reduced to an arbitrary frame of 64 points. After that, it is pre-processed by fast Fourier transform prior to feeding into the neural network for classification. To check for network robustness, the heart sound samples were injected with different level of noise. A total of 55 types of sample data were generated and used for classification. The study produces a 100% correct classification of eleven heart-valve diseases.