Is It Possible to Predict Cardiac Death?

Abstract

Cardiovascular diseases are the leading cause of death in all the world; despite having the knowledge of the main risk factors, they keep on being complicated pathologies to deal with. Cardiovascular management has introduced a lot of parameters as regards patients’ state of health; particularly, nuclear cardiology with Stress single-photon emission computed tomography myocardial perfusion imaging can carry out interesting parameters that have encouraged researchers to apply machine learning techniques to predict whether patients will die due to a cardiac event or not. The dataset consisted of 661 patients that were evaluated for suspected of known coronary artery disease at the Department of Advanced Biomedical Sciences of the University Hospital “Federico II” in Naples. Knime analytics platform was employed to implement a decision tree and Random forests. After a procedure of features reduction, 29 features were included, and the overall accuracy was 91.0%, while recall, precision, sensitivity and specificity overcame the value of 90.0%. This implementation shows the feasibility of machine learning combined with data coming from nuclear cardiology. Moreover, the possibility to predict cardiac death exploiting clinical data and parameters carried out from instrumental exams would help clinicians to provide patients with the best treatments and interventions.