Automated detection of severity of hypertension ECG signals using an optimal bi-orthogonal wavelet filter bank.

Hypertension (HPT) is a serious risk factor for cardiovascular disease and if not controlled in the early stage, can lead to serious complications. Long-standing HPT can induce heart muscle hypertrophy which will be reflected on electrocardiography (ECG). However, early stage of HPT may have no clinically discernible ECG perturbations, and is difficult to diagnose manually from the standard ECG. Hence, we propose an automated ECG based system that can automatically detect the ECG changes in the early stages of HPT. This work is based on ECG signals obtained from 139 HPT patients (SHAREE database) and 52 healthy subjects (PTB database). The ECG signal is non-stationary with relatively short duration, and rhythmic. Two-band optimal bi-orthogonal wavelet filter bank (BOWFB) and machine learning are used to automatically diagnose low, high-risk hypertension, and healthy control using ECG signals. Five-level wavelet decomposition is used to produce six sub-bands (SBs) from each ECG signal using BOWFB. Sample and wavelet entropy features are calculated for all six SBs. The features calculated SBs are fed to the k-nearest neighbor (KNN), support vector machine (SVM), and ensemble bagged trees (EBT) classifiers. In this work, we have obtained the highest average classification accuracy of 99.95% and area under the curve of 1.00 using EBT classifier in classifying healthy control (HC), low-risk hypertension (LRHPT) and high-risk hypertension (HRHPT) classes with ten-fold cross validation strategy. Hence the developed system can be used in clinics, or even in remote detection of HPT stages using ECG signals.