MR Derived Cardiac Metabolism Changes in Patients With Obesity and Diabetes: Knowledge Discovery Via Bayesian Networks and Random Forest Classification.

Recent developments in magnetic resonance spectroscopy (MRS) techniques have demonstrated great potential for the quantification of myocardial energy metabolites, enabling an in-depth glimpse into the energetic state of the heart to better explain the onset and severity of disease states. However, more evidence is required to establish its clinical impact and explanatory power compared to other diagnostic variables. In this study, random forest classification was used on data from 195 subjects to discriminate between healthy vs. obese non-diabetic, healthy vs. diabetic, healthy vs. non-obese diabetic, obese non-diabetic vs. obese diabetic, and non-obese diabetic vs. obese diabetic patient subgroups using 31 P-MRS and 1 H-MRS measurements of cardiac energetics, along with MRI measures of cardiac function, fat volumes, blood pressure, and blood glucose and cholesterol levels. Achieving 78.12%, 88.57%, 85.19%, 90.48%, and 76.47% test accuracies, SHAPs (SHapley Additive exPlanations) feature importances indicate a higher predictive impact of metabolic metrics for classifying the diabetic heart compared to global function metrics, gained through most common imaging techniques. Bayesian networks generated through structure learning of the data further suggests a potential causal association of increased visceral fat, increased LVMass resulting in decreased PCr/ATP, and increased cardiac lipid levels attributed to these disease states. Through the results, we have been able to showcase the importance of MRS measurements, both as strong features separating the diseases with clinically-relevant accuracy and furthermore its causal connection to other cardiac parameters.