Neural network models for predicting vascular age from PPG signals: A comparative study

Abstract

Cardiovascular diseases (CVDs) represent a significant global health issue, necessitating precise assessment methods. An important factor is vascular ageing, marked by a progressive decline in arterial elasticity, which impairs the ability of arteries to regulate blood flow effectively. Evaluating vascular age by comparing blood vessel health to chronological age offers valuable insights into arterial stiffness, aiding in the prevention of CVDs. This study employs four distinct neural network models to predict an individual's vascular age using photoplethysmography (PPG), a non-invasive, cost-effective, and reliable technique. PPG pulse waves from 4374 healthy adults, aged 25–75, grouped into six 10-year intervals from both radial and digital arteries, are used to explore age-related variations. The neural network models assessed include multilayer perceptron (MLP) and 1D convolutional neural network (CNN 1D) with raw signals, as well as 2D CNN and the pre-trained VGG-16 model with spectrograms as input. Results reveal that MLP achieved 95.3% accuracy for radial and 92.7% for digital arteries, CNN 1D achieved 99.3% for radial and 99.4% for digital arteries, and the 2D CNN model achieved 99.6% accuracy for both arteries. Notably, VGG-16 outperformed all models with an accuracy of 99.9% for radial and 99.8% for digital arteries. However, it is essential to consider that VGG-16's extended training time per epoch may pose limitations when dealing with large datasets and time constraints. In such scenarios, the more efficient 2D CNN, with appropriate hyperparameter tuning, may provide advantages in vascular age prediction. This predictive capability enhances the identification of cardiovascular ageing deviations and underlying disorders, improving assessment methods and proactive cardiovascular health management. By comparing blood vessel health to chronological age, this approach potentially enhances clinical practice, supports early intervention, and facilitates personalised treatment plans.