Unwrapping aortic valve dysfunction through complex network analysis: A biophysics approach

Abstract

The development of sensitive and low-cost techniques for identifying valve dysfunction has become inevitable in the context of increasing death due to cardiac diseases. The present work attempts to propose a novel technique for cardiac auscultation based on graph theory. The sixty heart sound signals from normal heart (NMH) and with aortic stenosis (ASH) are subjected to Fast Fourier Transform (FFT) and complex network analyses. The murmur signals, a time-series signal, carry information about the blood flow through the heart, which gets exposed in the graph constructed and its features. The finer details of the murmur signal from the defective aortic valve and the normal aortic valve are reflected as the increased number of frequency components in FFT and as interconnected clusters without uncorrelated nodes in the graph of ASH. The distinction in graph features forms the basis of classification based on machine learning techniques (MLTs). When the unsupervised MLT-principal component analysis gives 86.8% total variance, the supervised MLTs-K nearest neighbor (KNN), support vector machine, and KNN subspace ensemble classifiers give 100%, 95.6%, and 90.9% prediction accuracy, suggesting its potential in remote auscultation in rural health centers.