A Machine Learning Approach to Predict Successful Trans-Ventricular Off-Pump Micro-Invasive Mitral Valve Repair

Area of Science:

• Cardiovascular Surgery
• Medical Imaging
• Machine Learning in Medicine

Background:

• The NeoChord procedure is a beating-heart technique for degenerative mitral regurgitation (MR).
• It addresses MR caused by leaflet prolapse or flail using a trans-ventricular approach.
• Echocardiography guides the procedure.

Purpose of the Study:

• To develop a machine learning (ML) scoring system to predict NeoChord procedure success.
• Success is defined as achieving less than moderate MR at follow-up.
• The system utilizes pre-procedural clinical and echocardiographic data.

Main Methods:

• Eighty patients undergoing the NeoChord procedure were analyzed.
• Three-dimensional transesophageal echocardiography assessed preoperative mitral valve anatomy.
• Supervised ML models (random forest, decision tree) were trained and validated.

Main Results:

• The random forest model achieved a cross-validated area under the curve of 0.79 (mean) and 0.83 (median).
• The decision tree model achieved a cross-validated area under the curve of 0.72 (mean) and 0.77 (median).
• Key predictors included age, flail gap, early systolic mitral valve area, and indexed left atrial volume.

Conclusions:

• Machine learning effectively predicts mid-term success for the NeoChord technique.
• Integrating clinical and 3D echocardiographic data enhances prediction accuracy.
• This approach may aid in preoperative patient selection for the NeoChord procedure.