Back to 2026 Abstracts
Ai-applied Machine Learning In Preoperative Planning Of Minimally Invasive Mitral Valve Repair
Rosa Giusti1, Rafik Margaryan2, Giacomo Bianchi3, Marco Solinas4.
1Ospedale del Cuore, Massa, Italy, 2Rafik Margaryan, Massa, Italy, 3Giacomo Bianchi, Massa, Italy, 4Marco Solinas, Massa, Italy.
BACKGROUND: Mitral valve disease, particularly mitral regurgitation due to mitral valve prolapse, is a common reason for cardiac surgery. Surgical repair, preferred over replacement, provides better long-term outcomes. Successful repair relies on accurate preoperative assessment, including measurements of mitral annular dimensions, chordal length, and leaflet morphology. While imaging techniques like echocardiography and cardiac CT are essential for surgical planning, they can be affected by operator variability. Machine learning (ML) can enhance preoperative assessments by analyzing multimodal imaging data to predict optimal surgical strategies and implant sizes, such as annuloplasty rings and neochordae. This study aimed to validate a CT-based protocol and develop an ML model to predict implant sizes for personalized surgical planning.
METHODS: This prospective observational study included 32 patients undergoing minimally invasive mitral valve repair for posterior or bileaflet prolapse between January 2022 and November 2025. Only patients receiving prosthetic annuloplasty rings and pre-measured neochordae were included. Preoperative imaging, including cardiac CT with acquisition during the cardiac cycle, was performed. Data on mitral annular dimensions and chordae lengths were used to train ML models to predict implant sizes. Data were divided into training (80%) and test (20%) sets, with cross-validation to identify the best model.
RESULTS: The ML model successfully predicted the optimal sizes for annuloplasty rings and neochordae based on preoperative imaging data. The BaggingClassifier was the top-performing model, with an area under the curve (AUC) of 0.81 for the training set. Predicted implant sizes closely matched intraoperative measurements, validating both the CT-based protocol and the ML model. However, additional data is needed to further refine the model"s performance on test data and improve predictive accuracy.
CONCLUSIONS: The ML model reliably predicted annuloplasty ring and neochordae sizes, aiding preoperative planning. By integrating preoperative imaging, the model helps surgeons fully plan procedures, particularly benefiting less experienced surgeons and improving outcomes while reducing complications.
Back to 2026 Abstracts
