Echocardiography-based machine learning algorithm for distinguishing ischemic cardiomyopathy from dilated cardiomyopathy

Figures

• 

  Fig. 1. Machine learning workflow for differentiation of ICM and DCM. A machine learning model was developed to differentiate between ICM and DCM. The machine method included feature selection using a statistical method and 10-fold cross-validation, and 4 different algorithms were compared. ICM: ischemic cardiomyopathy; DCM: dilated cardiomyopathy

• 

  Fig. 2. Analysis of correlations between variables in the dataset. The inclusion of highly correlated variables can lead to numerical instability, obscure the interactions between different features, affect the interpretability of machine learning models, and also lead to overfitting. Therefore, it is best to exclude one of the two relevant variables. The correlation graph shows that the dataset did not contain many relevant variables, indicating that the model is simple and stable

• 

  Fig. 3. Ranking of feature importance. The importance graph for the variables was obtained after training with tuning parameters on the training data set. The most important variables in the study cohort for the differentiation of ICM and DCM are ranked. (A) random forest, (B) logistic regression, (C) extreme gradient boosting, and (D) artificial neural network

• 

  Fig. 4. Receiver operating characteristic curves for prediction of ICM and DCM in the test sample. XGboost model presented a higher AUC for distinguishing ICM and DCM than all other models (LR, RAN and NNET). AUC: area under the curve; LR: logistic regression; NNET: neural network algorithm; RAN: random forest; XGboost: extreme gradient boosting

• 

  Fig. 5. Receiver operating characteristic curves for prediction of ICM and DCM in the external validation. XGboost model presented a higher AUC for distinguishing ICM and DCM than all other models (LR, RAN and NNET). AUC: area under the curve; LR: logistic regression; NNET: neural network algorithm; RAN: random forest; XGboost: extreme gradient boosting