Interpretable machine learning models to support differential diagnosis between Ischemic Heart Disease and Dilated Cardiomyopathy

The differential diagnosis between Ischemic Heart Disease (IHD) and Dilated Cardiomyopathy (DCM), particularly in the early stages of the diseases, can often be difficult. Left ventricular ejection fraction (LVEF) and heart rate variability (HRV) analysis are shown to be helpful tools for diagnosing several cardiac diseases. There is a growing interest in application of machine learning techniques to guide the diagnosis. However, often black-box machine learning models create dissatisfaction among clinicians due to the lack of a model interpretability. The aim of our study was to compare the classification performance of interpretable and clinically plausible models applied for early differential diagnosis between DCM and IHD (NYHA = 1) based on LVEF and HRV features. The study encompassed 196 IHD and 117 DCM subjects. The models were produced by classification tree, logistic regression and naïve Bayes algorithms considering the set of selected HRV and LVEF features, chosen with the information gain method. The results showed that the most informative features for classification between IHD and DCM were LVEF, LF, NN50, pNN50, and meanRR. The naive Bayes model with classification accuracy of 73.5% outperformed classification tree and logistic regression models with 67.4% and 67.1% accuracies, respectively. We also demonstrated that the produced models together with nomograms allow probabilistic interpretation of the classification output between IHD and DCM, which is an important factor to guide the clinical decision making in differential diagnosis.