Artificial intelligence-based prediction of procedural success in transcatheter mitral valve edge-to-edge repair

Patient selection for transcatheter mitral valve edge-to-edge repair (MTEER) remains challenging, particularly in individuals with complex mitral valve anatomy. Conventional risk scores incorporate limited clinical variables and do not adequately account for detailed echocardiographic features, resulting in suboptimal prediction of procedural success.

To develop and evaluate machine learning (ML) and deep learning (DL) models integrating clinical and echocardiographic data to predict procedural success following MTEER.

Consecutive patients undergoing MTEER at a single tertiary center between 2014 and 2022 were retrospectively analyzed. Procedural success was defined as residual mitral regurgitation ≤ mild and mean mitral valve gradient < 5 mmHg at the end of the procedure. Pre-procedural transesophageal echocardiographic (TEE) videoclips were used to generate image datasets for ML and DL model training. Machine Learning algorithms (Decision Tree, Random Forrest, Gradient Boosting, k-Nearest Neighbors, Support Vector Machines) were used to predict the primary outcome from patient demographics and baseline TEE measurements. Inception, Xception, and MobileNet architectures were evaluated using 10-fold cross-validation with strict patient-level data separation. Model performance was assessed using accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC).

Among 467 included patients, procedural success was achieved in 90.7%. Deep learning models demonstrated superior performance compared with logistic regression and ML. The Xception architecture achieved the highest discriminatory ability (AUC 0.76), with favorable accuracy and sensitivity for predicting procedural success.

Artificial intelligence has the potential to improve prediction of procedural outcomes following MTEER. Deep learning models that integrate echocardiographic imaging and clinical data can accurately predict procedural success and may enhance pre-procedural risk stratification, patient selection, and Heart Team decision-making.