Feasibility of cardiovascular risk assessment through non-invasive measurements

The present work is a first step in building a wearable system to monitor the heart functionality of a patient and assess the cardiovascular risk by means of non-invasive measurements, such as electrocardiogram (ECG), heart rate, blood oxygenation, and body temperature. Also clinic data obtained by means of a patient interview are taken into account. In this feasibility study, measures from a pre-existing dataset are exploited. They are processed with a machine learning algorithm. Features are first extracted from the measures collected with the wearable sensors. Then, these features are employed together with clinic data to classify the patients health status. A Random Forest classifier was employed and the algorithm was characterized considering different setups. The best accuracy resulted equal to 78.6% in distinguishing three classes of patients, namely healthy, unhealthy non-critical, and unhealthy critical patients.