Decoding Vascular Age from Brachial Waveform Morphology via Machine Learning and Model-Based Data Augmentation

Abstract

Vascular aging is increasingly recognized as a critical marker of cardiovascular health, yet no gold-standard method exists for its quantification. In this study, we present a machine learning (ML)–based framework for decoding vascular age (VasAge) from brachial pressure waveforms, under the assumption that vascular and chronological age are equivalent in healthy individuals. The ML model was trained on Fourier-based harmonic features extracted from high-resolution brachial waveforms obtained using a laboratory-developed cuff device. We tested our method on a clinical dataset of 111 subjects (45 women, mean age 65 years), consisting of 16 healthy and 95 unhealthy individuals. To develop the ML model, waveforms from the healthy subgroup were augmented using a synthetic waveform database generated by a physiologically relevant pulse wave computational model. For model development in the healthy population where chronological and vascular age are equivalent, the model's predictions correlated with measured age (r = 0.91), with marginal systemic bias (bounded to 5% of the mean). We then applied this model to the population of unhealthy individuals, finding higher vascular age than chronological age in this group (mean difference of 9.5 years), with sensitivity to elevated aortic stiffness and systolic blood pressure (p < 0.05). These results demonstrate that vascular age can be accurately estimated from the morphology of a single brachial pressure waveform using a Fourier-based ML approach, even with a moderately sized sample. This method offers a cost-effective, non-invasive strategy for personalized cardiovascular risk assessment and holds promise for clinical translation.

Files

Additional details