A combined simulation and machine learning approach to classify severity of infarction patients

Knowledge-driven and data-driven strategies have been widely used to address many bioengineering and clinical open questions. However, little attention has been paid to the potential advantages the integration of such strategies could open up. To this aim, in this paper, we describe a sequential simulation and machine learning (ML) framework. Firstly, an ad-hoc mathematical model, based on differential equations, was used to simulate - starting from real data - cardiac troponin concentration curves of 27 patients (with Acute Myocardial Infarction and ST-segment elevation) in a 200h time frame; later, the curves were analyzed to extract 4 time-domain features which, fed to 3 tree-based ML algorithms, allowed to successfully classify - ML scores > 75% for Gradient Boosted Tree - patients in two risk classes according to Thrombolysis in Myocardial Infarction risk index. These promising results could stimulate researchers to consider combined knowledge-driven and data-driven strategies to address other cardiovascular and/or clinical research questions.