Machine learning-based predictive model for the diagnosis of PH pulmonary hypertension diagnosis integrating cardiac magnetic resonance imaging and clinical biomarkers

Figures

• 

  Fig. 1. Flow chart of data analysis in this study

• 

  Fig. 2. Development and validation of the MRI index for PH prediction. (A) Key MRI features selected by recursive feature elimination (RFE) to construct the MRI index. (B) Comparative performance of machine learning models for PH prediction. (C) Confusion matrix of the MRI index for PH prediction in the training set. (D) Scatter plot of correctly and incorrectly classified samples by the MRI index in the training set. (E) Confusion matrix of the MRI index for PH prediction in the internal validation set. (F) Scatter plot of correctly and incorrectly classified samples by the MRI index in the internal validation set. (G) Receiver operating characteristic (ROC) curve of the MRI index in the training set. (H) Precision-recall curve of the MRI index in the training set. (I) Specificity-sensitivity trade-off curve of the MRI index in the training set. (J) Accuracy-threshold relationship of the MRI index in the training set. (K) ROC curve of the MRI index in the internal validation set. (L) Precision-recall curve of the MRI index in the internal validation set. (M) Specificity-sensitivity trade-off curve of the MRI index in the internal validation set. (N) Accuracy-threshold relationship of…

• 

  Fig. 3. SHAP analysis of the PH prediction model. (A) Feature importance shows PA, RVEF, and REDV as top predictors. (B) Beeswarm plot reveals how feature values influence predictions (red=high, blue = low). (C) Heatmap clusters samples by similar SHAP patterns, highlighting key features like PA and REDV. (D) Waterfall plot explains a high-risk case: PA dilation (+ 1.589) and increased REDV (+ 1.889) raised PH risk, while higher RVEF (-0.037) was protective

• 

  Fig. 4. Development of the integrated PH prediction model combining laboratory biomarkers with MRI index and its clinical presentation. (A) Random forest selection of key predictive biomarkers from comprehensive laboratory tests. (B) Decision curve analysis demonstrating the clinical net benefit of biomarker-based PH prediction. (C) Enhanced predictive performance shown by decision curve analysis of the combined MRI-biomarker model. (D) Nomogram visualization of the final integrated prediction model incorporating both MRI parameters and selected biomarkers for point-of-care risk assessment

• 

  Fig. 5. Performance of the machine learning-based predictive model for PH. (A) Calibration curve of the nomogram in the training set, (B) Calibration curve of the nomogram in the internal validation set, (C) Calibration curve of the nomogram in external validation set, (D) ROC curve of the nomogram in the training set, (E) ROC curve of the nomogram in the internal validation set, (F) ROC curve of the nomogram in the external validation set

• 

  Fig. 6. Web-based calculator for the nomogram-derived PH prediction model. (A) Interface screenshot displaying the calculated PH probability based on MRI index parameters, accompanied by SHAP value interpretation of key imaging features. (B) Interactive nomogram implementation showing real-time risk calculation with detailed visualization of model components (MRI parameters and laboratory biomarkers) and their respective point contributions