Estimating pulse wave velocity from the radial pressure wave using machine learning algorithms.

Area of Science:

• Cardiovascular research
• Biomedical engineering
• Machine learning applications

Background:

• Vascular ageing is a key risk factor for cardiovascular disease.
• Carotid-femoral pulse wave velocity (cfPWV) is a gold standard for measuring vascular ageing but requires two measurement sites and expert operation.
• A simpler, single-site method for assessing vascular ageing is needed.

Purpose of the Study:

• To develop and validate machine learning pipelines for estimating cfPWV from a single peripheral pulse wave measurement.
• To assess the feasibility of using radial pressure waves for vascular ageing assessment.
• To compare the performance of Gaussian process regression and recurrent neural networks (RNNs) for this task.

Main Methods:

• Two machine learning pipelines were developed: one using Gaussian process regression with pulse wave analysis features, and another using an RNN on the entire radial pressure waveform.
• The pipelines were trained and tested on data from the Twins UK cohort (3,082 subjects) and a virtual subject database (4,374 subjects).
• Performance was evaluated using mean difference and limits of agreement (LOA) on a hold-out test set.

Main Results:

• The RNN pipeline achieved a mean difference of 0.05 m/s and LOA of 3.21 m/s & -3.11 m/s on human subjects.
• The Gaussian process regression pipeline showed a mean difference of 0.2 m/s and LOA of 3.75 m/s & -3.34 m/s.
• The RNN method demonstrated robustness to noise, with less than 2% error increase with 20% random noise.

Conclusions:

• Machine learning models, particularly RNNs, can accurately estimate cfPWV from single-site radial pressure waves.
• This approach offers a potential non-invasive method for assessing vascular ageing and cardiovascular risk.
• The developed pipelines provide a more accessible alternative to traditional cfPWV measurements.