Photoplethysmogram based vascular aging assessment using the deep convolutional neural network

Abstract

AbstractArterial stiffness due to vascular aging is a major indicator during the assessment of cardiovascular risk. In this study, we propose a method for age estimation by applying deep learning to a photoplethysmogram (PPG) for the non-invasive assessment of the vascular age. The proposed deep learning-based age estimation model consists of three convolutional layers and two fully connected layers, and was developed as an explainable artificial intelligence model with Grad-Cam to explain the contribution of the PPG waveform characteristic to vascular age estimation. The deep learning model was developed using a segmented PPG by pulse from a total of 752 adults aged 20–89 years, and the performance was quantitatively evaluated using the mean absolute error, root-mean-squared-error, Pearson’s correlation coefficient, and coefficient of determination between the actual and estimated ages. As a result, a mean absolute error of 8.1 years, root mean squared error of 10.0 years, correlation coefficient of 0.61, and coefficient of determination of 0.37, were obtained. A Grad-Cam, used to determine the weight that the input signal contributes to the result, was employed to verify the contribution to the age estimation of the PPG segment, which was high around the systolic peak. The results of this study suggest that a convolutional-neural-network-based explainable artificial intelligence model outperforms existing models without an additional feature detection process. Moreover, it can provide a rationale for PPG-based vascular aging assessment.