A Wearable Sensor for Arterial Stiffness Monitoring Based on Machine Learning Algorithms

Abstract

Arterial stiffness is strongly associated with cardiovascular events. Existing devices for evaluating arterial stiffness based on ultrasound or pulse wave velocity suffer a lot from complexity and inconvenience in home-care settings. This paper proposed a wearable sensor for arterial stiffness monitoring via machine learning techniques. The proposed sensor is comprised of one electrocardiogram (ECG) and one photoplethysmogram (PPG) module. The ECG and PPG signals were first simultaneously collected by the wearable sensor, and 21 features were extracted from two signals for arterial stiffness evaluation. A genetic algorithm-based feature selection method was then used to select the important indicators. Multivariate linear regression (MLR), decision tree, and back propagation (BP) neural network were employed to develop the model. Vascular age and 10-year cardiovascular disease risk from OMRON arteriosclerosis instrument were deemed as the gold standard to evaluate arterial stiffness. Experimental results based on 501 diverse subjects showed that the MLR approach exhibited the best accuracy in vascular age estimation (correlation coefficient, 0.89; mean of the residual, 0.2136; and standard deviation of the residual, 6.2432). While the BP neural networks-based approach was best in cardiovascular disease risk estimation (correlation coefficient, 0.9488; mean of the residual, - 0.3579%; and standard deviation of the residual, 3.7131%). The results indicate that the proposed learning-based sensor has great potential in arterial stiffness monitoring in home-care settings.