Photoplethysmography acquired through the camera and the diode of a smartphone is a cheap and non invasive method that can be used to esti- mate the vascular aging of a patient. In this study we would like to apply a β−Convolutional Variational Autoencoder to perform this task. In this report are shown the best parameters and the best structure of the Neural Network with the best reconstruction performances of the peak. The final results show that it is possible to separate partially in the latent space the PPG of the young people from one of the older ones. In particular, it is possible to recognize the main feature that allows us to distinguish a PPG belonging to a young person, in the presence of an accentuated concavity in the peak, identified with the dicrotic notch. The results coming from the application of Deep Learning to PPG could lead to the diffusion of PPG as a screening tool.
The complete report is contained in the report directory