<![CDATA[This article introduces a hybrid cardio-ankle vascular index–machine learning method (CAVI-ML) that combines the classical non-invasive CAVI with a random forest (RF) model to reduce CAVI’s residual dependence on systolic blood pressure (SBP) and the pressure gradient. The key change is replacing the standard coefficients a and b with RF-predicted apred and bpred. Using a synthetic pulse wave database (PWDB) of 4,374 virtual pulse waves (ages 25–75), classical CAVI was computed via the Bramwell–Hill equation, and 11 non-invasive hemodynamic features per subject were used to train an RF regressor (100 trees; 63.2% bootstrap samples; 3 variables per split). We then used tree-averaged apred and bpred to calculate normalised CAVIDyn. We compared performance with classical CAVI and the gold standard carotid-femoral pulse wave velocity (cfPWV) using the metrics: R², root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE). CAVI-ML reduced correlation with SBP by 59% (r = 0.613 → 0.251) and with diastolic pressure by 73% (r = 0.735 → 0.201), and was 62% less blood-pressure-dependent than cfPWV (r = 0.662). CAVI-ML improves independence from instantaneous blood pressure and shows promise for non-invasive arterial stiffness estimation and cardiovascular risk prediction, but requires validation on clinical data.]]>
Copyrights © 2026
