This project continues part A project, where we created a large database of PPG signals from various sources, together with demographic data of the subjects. The goal of the current project is to find relations between the shape of the PPG signal and cardiovascular diseases using the above-mentioned data. Visualization of these relations was achieved by extracting several features from the signal that could indicate various disorders in the cardiovascular system and locating the subjects on a “Health Map” according to these features using two different quasi-clustering methods: Gath-Geva and FCM. We also used the t-SNE technique for reducing the map dimensions and its visualization in 2D.
We decided to show on the map three diseases that are strongly related to various distortions in the shape of the PPG pulse: hypertension, airway obstruction and heart failure.
Resulting health maps did not show good disease separation because the data was very noisy, but manually selecting small groups of data points provided a fairly accurate clustering to the given clusters.