MRI-Derived Variables Combined with Machine Learning for Pulmonary Hypertension Risk Prediction: A Retrospective Analysis
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Pulmonary hypertension (PH) is a severe and progressive vascular disease for which early diagnosis and risk stratification are critical for improving patient outcomes. However, current diagnostic approaches exhibit limitations in achieving precise risk prediction. This study aimed to develop and validate a machine learning-based risk prediction model that integrates MRI-derived variables for PH risk assessment.
Methods
This retrospective study enrolled 210 participants who underwent MRI at Zhongnan Hospital of Wuhan University between January 2021 and December 2023, including 87 PH patients and 123 controls. Key MRI features were selected through recursive feature elimination (RFE) with a Random Forest algorithm. Multiple machine learning models, including XGBoost and logistic regression, were trained and evaluated employing 10-fold cross-validation. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). SHAP analysis was utilized to interpret the contribution of individual features, and a nomogram integrating MRI and clinical variables was developed for personalized risk prediction.
Results
Six key MRI-derived features were identified, among which pulmonary artery diameter and left atrial anterior-posterior diameter were the most significant predictors. The XGBoost model exhibited the best performance, achieving an area under the curve (AUC) values of 1.0 and 0.969 in the training and testing sets, respectively. Calibration curves demonstrated excellent agreement between predicted and observed outcomes. DCA revealed high net clinical benefits across a range of risk thresholds. The developed nomogram offers an intuitive tool for individualized PH risk prediction, demonstrating strong interpretability and clinical utility.
Conclusion
This study developed a highly accurate and reliable machine learning-based risk prediction model for PH based on MRI-derived features. By integrating SHAP analysis and a nomogram, the model provides a novel, non-invasive approach for early diagnosis and personalized risk stratification of PH, highlighting the significant potential for clinical application.
