Assessment of Arterial Stiffness Using Radial Artery Diameter Variation Signals

This work presents a proof of concept for the development of a predictive model of arterial stiffness evolution with age, using arterial diameter variation (VDA) records as a non-invasive biomarker, acquired in the radial artery. A record storage system is proposed that considers the extraction of images of the wave of each cardiac cycle and consults a deep learning model based on convolutional neural networks (CNN). A model was trained to estimate age from the images of the cardiac cycles and its performance was evaluated on training datasets (randomly selected records) and test (records of other patients). Heatmaps were used to visualize the regions of the cardiac cycles that the model considered relevant for prediction. The results showed an average absolute error, considering the estimations of the cycles) of 7.78 years in the estimation of age for the complete records. The cause of this error is analyzed. The limitations of the study, such as the size and selection of the sample, are discussed and future improvements are proposed, such as training with more cases and patient selection. The potential of the system to assess cardiovascular health and estimate other relevant descriptors is highlighted.