Prediction of Ascending Aortic Dilation and Analysis of Influencing Factors in Bicuspid Aortic Valve Patients Using an Explainable Machine Learning Model.

Objective: Developing a machine learning (ML) model to predict the risk of ascending aortic dilation in patients with a bicuspid aortic valve (BAV). Using SHapley Additive exPlanations (SHAP) to interpret and visualize the model. Methods: This study enrolled 102 BAV patients, who were divided into two subgroups based on ascending aorta diameter (dilated and nondilated). All participants underwent routine echocardiography, clinical baseline data collection, and measurement of plasma matrix metalloproteinases (MMPs) and their tissue inhibitors (TIMPs). Feature selection was performed using univariate analysis followed by the least absolute shrinkage and selection operator (LASSO)-logistic regression (LR) method. Five common ML prediction models were developed: support vector machine (SVM), LR, gradient-boosting machine (GBM), neural network (NNET), and Naïve Bayes (NB) classifier. To identify the best-performing predictive model for ascending aortic dilation in BAV patients, an evaluation of predictive efficacy was carried out by employing ROC curves, calibration curves, and DCA curves. Finally, the optimal model's predictions were interpreted using SHAP. Results: Random allocation of the entire patient population resulted in a training set ( Conclusions: The GBM model offers a valuable tool for predicting ascending aortic dilation in BAV patients. Moreover, SHAP analysis enhances the model's utility by providing clear, actionable insights for clinical management.