Discrimination of vascular aging using the arterial pulse spectrum and machine-learning analysis.

Aging contributes to the progression of vascular dysfunction and is a major nonreversible risk factor for cardiovascular disease. The aim of this study was to determine the effectiveness of using arterial pulse-wave measurements, frequency-domain pulse analysis, and machine-learning analysis in distinguishing vascular aging. Radial pulse signals were measured noninvasively for 3 min in 280 subjects aged 40-80 years. The cardio-ankle vascular index (CAVI) was used to evaluate the arterial stiffness of the subjects. Forty frequency-domain pulse indices were used as features, comprising amplitude proportion (Cn), coefficient of variation of Cn, phase angle (Pn), and standard deviation of Pn (n = 1-10). Multilayer perceptron and random forest with supervised learning were used to classify the data. The detected differences were more prominent in the subjects aged 40-50 years. Several indices differed significantly between the non-vascular-aging group (aged 40-50 years; CAVI <9) and the vascular-aging group (aged 70-80 years). Fivefold cross-validation revealed an excellent ability to discriminate the two groups (the accuracy was >80%, and the AUC was >0.8). For subjects aged 50-60 and 60-70 years, the detection accuracies of the two trained algorithms were around 80%, with AUCs of >0.73 for both, which indicated acceptable discrimination. The present method of frequency-domain analysis may improve the index reliability for further machine-learning analyses of the pulse waveform. The present noninvasive and objective methodology may be meaningful for developing a wearable-device system to reduce the threat of vascular dysfunction induced by vascular aging.