Hypertension assessment based on feature extraction using a photoplethysmography signal and its derivatives

Abstract

Objective. Long-term abnormal blood pressure (BP) can lead to various cardiovascular diseases; therefore, it is significant to assess BP status as a preventative measure. In this study, a feature-extraction-based approach is proposed and performed on an open clinical trial dataset. Approach. Firstly, a complete ensemble of empirical mode decomposition with an adaptive noise algorithm and wavelet threshold analysis is applied to eliminate the noise interference from an original photoplethysmography (PPG) signal compared to other signal filters. Considering the strong connection between hypertension and diabetes, an analysis of variance test with a 95% confidence interval is firstly carried out to select these leading extracted morphological features, which are uniquely related to hypertension, from the PPG signal and its derivatives. Subsequently a variety of classification models are evaluated at different BP levels and their performances are compared. Main results and Significance. The test results demonstrate that the support vector machine classification model achieves a greater performance compared to other explored models in this paper, with accuracy of 78%, 87% and 88% for cases including normal versus prehypertension subjects, normotension versus hypertension subjects and non-hypertension versus hypertension subjects, respectively, which further illustrates the great potential of the proposed method in hypertension assessment.