Performance Analysis of Entropy Measures in Assessment of Mitral Regurgitation Using Cardio-mechanical Signals

In this paper, we investigated the effectiveness of entropy features in the classification of mitral regurgitation (MR), based on the cardio-mechanical signals. These entropy features were extracted from the original data and data after continuous wavelet transform (CWT) expansion. XGBoost was selected for the classifier. The performance of different entropy features and different training-test set classification methods were compared. The contribution of different entropy features to the classification results is compared by SHapley Additive exPlanations (SHAP) values. It is concluded that the entropy features extracted from the signal after CWT expansion perform better in classification results. Approximate entropy (ApEn), permutation entropy (PEn), and sample entropy (SampEn) demonstrated greater contributions to the classification outcomes. The entropy features extracted from the Y-axis (along the head-to-foot direction) channels are more important than those from other channels. However, the sensitivity decreases significantly when dividing the training-testing set in terms of subjects. These results provide valuable insights for future research on utilizing entropy measures for assessing MR.