Spectrogram Image-based Machine Learning Model for Carotid-to-Femoral Pulse Wave Velocity Estimation Using PPG Signal

Carotid-to-femoral pulse wave velocity (cf-PWV) is a critical biomarker for evaluating arterial stiffness and cardiovascular risk. Monitoring cf-PWV is essential for cardiovascular disease diagnosis and prediction. However, the complexity during the measurement process of cf-PWV makes it prone to present errors and inaccuracies. For this reason, a learning-based non-invasive measurement of cf-PWV using peripheral signals could overcome some of the difficulties presented in the classical measurement process and improve the quality of the estimation. In this paper, a spectrogram-based machine learning model obtained from the photoplethysmogram (PPG) waveform is proposed for the estimation of the cf-PWV. For this purpose, two machine learning models have been developed using three different types of features. The first category is based on an adaptive signal processing method called Semi-Classical Signal Analysis (SCSA) that relies on the spectral problem of the Schrodinger operator; the second type proposed is energy texture-based, and the third is the statistical texture representation. Finally, the training and testing datasets were extracted from in-silico, publicly available pulse waves and hemodynamics data. The obtained results provide evidence for the feasibility and robustness of the spectrogram to transform the signals into an image and machine learning method as a tool for estimating the cf-PWV.