Machine learning for pacemaker implantation prediction after TAVI using multimodal imaging data.

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Pacemaker implantation (PMI) is a frequent complication following transcatheter aortic valve implantation (TAVI).
• Computed tomography (CT) scan data are recognized predictors of PMI, but integrated models are lacking.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting PMI after TAVI.
• To assess the specific contribution of CT imaging data in conjunction with clinical, ECG, and transthoracic echocardiography (TTE) data.

Main Methods:

• Retrospective analysis of 520 TAVI patients.
• Utilized recursive feature elimination with SHAP values for variable selection.
• Trained and evaluated six ML models, including Support Vector Machines (SVM).

Main Results:

• The best ML model achieved an AUC-ROC of 92.1%, F1 score of 71.8%, and accuracy of 87.9%.
• The model incorporated 22 variables, with 9 derived from CT imaging.
• Critical predictors included membranous septum measurements and their dynamic variations.

Conclusions:

• The developed ML model offers robust prediction of PMI post-TAVI.
• CT imaging data significantly contribute to the model's predictive performance.
• The model facilitates personalized risk assessment and is available online for clinical application.