Ensemble Learning of Blood Metabolic Biomarkers: A Novel CAD System of Heart Failure

Acute myocardial infarction (MI) is complex, and a variety of pathologies can result in this clinical diagnosis. Current guidelines recognize two classes of MI: thrombotic (Type 1) and non-thrombotic (Type 2), which have similar prevalence but require different treatment. Unfortunately, diagnostic criteria for differentiating between the two do not exist. This results in inefficient and sub-optimal care of patients suspected of MI. This paper introduces a novel CAD system for categorizing MI by analyzing the association between blood metabolites with the formation of thrombosis as a cause of MI. The data collection stage involves a new non-targeted approach that detects the quantities of both know and unknown metabolites from blood samples. Our system uses a randomized ensemble-learning scheme to increase the model accuracy. This scheme includes a recursive feature elimination (RFE) algorithm to discard weakly correlated metabolites and a multi-classifier algorithm that votes between five different classifiers in order to enhance accuracy. The output of our system includes suggested clusters of metabolites that act as biomarkers to differentiate between Type 1 and Type 2 MI. Our experiment achieves diagnostic accuracy of %94, sensitivity of %90 and specificity of %81.