Interpretable machine learning models based on CMR radiomics for predicting left ventricular diastolic dysfunction in patients with metabolic‐associated steatotic liver disease and type 2 diabetes mellitus

Figures

• 

  Figure 1. Flowchart for selecting and distributing T2DM and MASLD patients in predictive model development and validation. T2DM, type 2 diabetes mellitus; MASLD, metabolic‐associated steatotic liver disease; LVEF, left ventricular ejection fraction; CMR, cardiac magnetic resonance; SHAP, Shapley additive explanations.

• 

  Figure 2. Acquisition and segmentation of CMR images. CMR, cardiac magnetic resonance.

• 

  Figure 3. Radiomics feature selection using LASSO regression. (a) Trajectory of each independent variable coefficient (colored curves). (b) Repeat 10‐fold cross‐validation 50 times to identify optimal lambda in LASSO, resulting in 7 nonzero coefficients. Red dots indicate mean values. LASSO, least absolute shrinkage and selection operator.

• 

  Figure 4. Comparison of ML classifiers (Logit, SVM, RF, XGBoost) using clinical data: (a) ROC curves, (b) calibration plots, and (c) DCA. ROC‐AUCs: 0.751, 0.722, 0.715, 0.727. ML, machine learning; ROC, receiver‐operating characteristic; AUC, area under the curve; DCA, decision curve analysis; Logit, logistic regression; SVM, support vector machine; RF, random forest; XGBoost, extreme gradient boosting.

• 

  Figure 5. Comparison of ML classifiers (Logit, SVM, RF, XGBoost) using radiomics features: (a) ROC Curves, (b) calibration plots, (c) DCA. ROC‐AUCs: 0.817, 0.831, 0.882, 0.875. ML, machine learning; ROC, receiver‐operating characteristic; AUC, area under the curve; DCA, decision curve analysis; Logit, logistic regression; SVM, support vector machine; RF, random forest; XGBoost, extreme gradient boosting.

• 

  Figure 6. Comparison of ML classifiers (Logit, SVM, RF, XGBoost) on clinical and radiomics data: (a) ROC curves, (b) calibration plots, (c) DCA. ROC‐AUCs: 0.876, 0.866, 0.909, 0.931. ML, machine learning; ROC, receiver‐operating characteristic; AUC, area under the curve; DCA, decision curve analysis; Logit, logistic regression; SVM, support vector machine; RF, random forest; XGBoost, extreme gradient boosting.

• 

  Figure 7. Evaluation of optimal ML model's performance with external verification cohort: (a) ROC curve (AUC = 0.865), (b) calibration curve, (c) DCA. ML, machine learning; ROC, receiver‐operating characteristic; AUC, area under the curve; DCA, decision curve analysis.

• 

  Figure 8. SHAP analysis of XGBoost model for LVDD risk prediction in T2DM and MASLD patients: (a) summary plot of cumulative influence, (b) feature significance ranking. SHAP, Shapley additive explanation; XGBoost, extreme gradient boosting; LVDD, left ventricular diastolic dysfunction; T2DM, type 2 diabetes mellitus; MASLD, metabolic‐associated steatotic liver disease.

• 

  Figure 9. SHAP force plots of individual predictions: (a) for LVDD+ patient, (b) for LVDD‐ patient. SHAP, Shapley additive explanations; LVDD, left ventricular diastolic dysfunction.