A machine learning approach to detect aortic valve dysfunction through phase portrait feature extraction
Department of Optoelectronics, University of Kerala, 695581, Trivandrum, Kerala, India
Accepted: 30 October 2021
Published online: 14 November 2021
The paper reports a novel method of early detection of heart valve dysfunction, drawing phase portrait and power spectral features of normal (NM) and aortic stenosis (AS) murmur signals. The 40 signals of NM and AS are subjected to spectral, fractal, and nonlinear time series analyses. Using machine learning techniques, the signals are classified and predicted based on phase portrait and power spectral features.The appearance of multiple high-intense frequency components in the spectral analyses of AS, without separation between the lub and dub, reveals valve dysfunction and turbulent blood flow. The higher value of Lyapunov exponent, sample entropy, and fractal dimension of AS compared to NM reflect the valve dysfunction, which gets displayed through the phase portrait also. When the principal component analysis (PCA) of power spectral density data classifies NM and AS with a variance of 87.5%, the weighted K-nearest neighbour (KNN) gives 100% predictive accuracy. The classification based on phase portrait parameters using KNN and linear discriminant analysis is found to give 96% predictive accuracy. The proposed digital auscultation technique based on phase portrait and spectral features, to understand the valve dysfunction, suggests its possible application of practising in rural primary health centres.
© The Author(s), under exclusive licence to EDP Sciences, Springer-Verlag GmbH Germany, part of Springer Nature 2021