Estimation of Carotid-Femoral Pulse Wave Velocity based on Radial-Tibial Pulse Analysis: a Multivariable Approach

Cardiovascular diseases (CVDs) represent the leading cause of death, accounting for around 32% of fatalities globally. Furthermore, CVD prevalence and mortality rates are expected to increase in the coming years due to the population's average age growth. This trend carries substantial implications for clinical practices and healthcare expenditures. Monitoring symptoms becomes essential to mitigate the negative impacts of these pathologies. Arterial stiffness is one of the highly predictive parameters in this regard, as the arteries naturally become stiffer with age and in the presence of specific diseases. One effective method to access arterial elasticity is by measuring Pulse Wave Velocity (PWV), which indicates the blood pulse speed to travel through two sites of the cardiovascular system. It is widely acknowledged that carotid-femoral PWV (cfPWV) is considered the gold-standard measurement of arterial stiffness. However, cfPWV measuring presents several limitations as it is operator-dependent, technically challenging, and requires extensive training for accurate measurement. This thesis project explores the correlation between peripheral (radio-tibial) and systemic (carotid-femoral) PWV. In particular, this research focuses on developing a model for estimating cfPWV based on the information extracted from these features (derived from radial-tibial PWV and radial pulse wave morphology) in conjunction with data from parameters strongly associated with cfPWV, such as age and blood pressure. A trial conducted at "A. O. U. Citt¿ della Salute e della Scienza" in Turin enrolled 91 healthy volunteers (46 males; age 51 ¿ 16 years; height 168 ¿ 9 cm; weight 69 ¿ 14 kg), leading to a dataset of clinical reports containing all the desired parameters (age, systolic pressure, diastolic pressure, PWV, etc.) and waveforms (ECG and Pulse). The measurements are accessed with SphygmoCor, the gold-standard device for this evaluation. A Python script is developed to extract high-resolution waveform images from this dataset and, concurrently, the values documented in the collected data from the SphygmoCor reports. Consequently, the images are digitized using a specialized MATLAB routine, and the waveform data are given as input to an algorithm to calculate PWV and pulse transit time (PTT) values. Secondly, feature extraction algorithms are implemented in MATLAB, ranging from typical metrics used for PWV calculation to new parameters related to the morphology of the radial pulse wave, such as dicrotic notch and diastolic peak amplitude. After features extraction, a correlation analysis is conducted in MATLAB and RStudio to identify the metrics most correlated with the cfPWV. Subsequently, a routine is implemented to build a multivariate linear regression model that computes all feature combinations and selects the best one based on adjusted R-squared, RMSE, and p-value parameters. Finally, the model is compared to a neural network using the same feature combination as input. The best-selected model reported a Pearson coefficient of 0.81, demonstrating a high correlation with the cfPWV value. This model could be used to develop wearable devices enabling real-time monitoring of PWV.