A support system for automatic classification of hypertension using BCG signals

Abstract

Hypertension (HPT) is a lethal medical disorder in which the blood vessels unusually have high pressure for an extended period. It may raise the risk of various health complications. Ballistocardiography (BCG) is an emerging tool to diagnose various heart-related diseases and is used to depict the repetitive vibrations in the human body induced by the sudden evacuation of blood into the major arteries with each heartbeat. This article presents the multi-resolution analysis of BCG signals for the screening of HPT patients using integrated tunable Q-factor wavelet transform (ITQWT). The TQWT decomposes an input BCG signal into various sub-bands (SBs). Specifying a predetermined accurate basis function for optimal decomposition utilizing TQWT is a difficult task. Therefore, in this study, for the first time, a multi-verse optimization (MVO) algorithm is integrated with TQWT for selecting optimum tuning parameters to decompose the input BCG signals into more representative SBs. To detect hypertensive BCG signals eleven statistical features are evaluated from each SB. Among them, a set of seven statistically-significant features are selected by applying the Kruskal–Wallis test and fed to a K-nearest neighbor (K-NN) classifier with six different kernels using a 10-fold validation scheme. The highest classification accuracy of 92.21%, sensitivity of 92.96%, and specificity of 91.60% are achieved using a weighted K-NN classifier. This paper presents, a non-parameterized approach for the optimal decomposition of BCG data to detect HPT more accurately. The primary benefit of the proposed support system is that it can detect HPT patients with high accuracy by reducing the clinician’s workload.