Our team, in collaboration with researchers from Alfaisal University, Sorbonne Université and Sensoria Analytics, has contributed to a new study published in Scientific Reports journal that introduces a novel way to estimate pulse wave velocity (PWV).
This study proposes a novel data-driven approach for PWV estimation using features derived from Limited Penetrable Weighted Visibility Graphs (LPWVGs) constructed from photoplethysmography (PPG) waveforms and their first and second derivatives. By generating multiple LPWVGs with diverse weighting strategies, we capture the PPG signal’s rich temporal and morphological characteristics. A wide range of features was extracted, including descriptors from two-dimensional Semi-Classical Signal Analysis (SCSA), frequency-domain features, and morphological shape and local variation metrics. These were used to train an Explainable Boosting Machine (EBM), a glass-box machine learning model combining strong predictive power and interpretability.The proposed method was evaluated using positive and negative testing on real multicycle PPG datasets. The results demonstrate high accuracy and robustness,with an R2 = 0.91 and RM SE = 0.34 in the positive test and a RMSE = 1.49 for the negative test. These results support the feasibility of this approach for non-invasive PWV estimation in clinical and ambulatory settings, with potential applications in cardiovascular disease screening, risk stratification, and aging research.
For more information about the article, you can enter to the following link:
Scientific Reports article
