This prospective study aims to investigate the anatomical and biomechanical characteristics of MVP using RT3DEE and advanced artificial intelligence (AI)-based image analysis techniques. The goal is to develop an automated framework for the segmentation and morphological assessment of the MV apparatus, and to conduct FE modeling to evaluate the biomechanical implications of degenerative MVP and associated surgical repairs.
Imaging Protocol Intraoperative RT3DE will be performed in the operating room as part of the standard imaging protocol during surgical MVP repair. A Vivid S70N ultrasound system (GE Healthcare) equipped with a 6VT 4D multiplane RT3DE probe will be used. Standard mid-esophageal RT3DE views will be acquired, including full left ventricular (LV) chamber volumes and zoomed, gated 3D views of the MV complex (annulus, leaflets, and papillary muscles). To enhance image quality, zoomed acquisitions will be limited to the smallest pyramidal volume capturing the entire mitral complex, ensuring frame rates ≥20 Hz. All imaging datasets will be anonymized prior to analysis.
Surgical Protocol MVP surgical repair will be conducted under general anesthesia through a right mini-thoracotomy. Surgical techniques will include: i) leaflet resection (removal of excess leaflet tissue); ii) neochordal implantation (placement of expanded polytetrafluoroethylene, ePTFE, artificial chordae); annuloplasty (implantation of an annuloplasty ring or pericardial band for annular stabilization).
In cases where significant leaflet tissue is excised (≥10×10 mm), samples will be collected and stored in the institutional BioCor biobank for histological and morphometric analysis. Samples will be decontaminated, cryopreserved, and stored at -80°C. After histological evaluation, all biological material will be destroyed.
Clinical Data Collection Demographic (e.g., age, sex) and clinical data (e.g., diagnosis, medications, comorbidities) will be obtained from medical records and anonymized.
Neural Network Training for Image Segmentation Anonymized RT3DE datasets will be manually segmented by expert operators using advanced image analysis software (e.g., 3D Slicer). A minimum of 150 RT3DE acquisitions will be processed to generate binary masks of the MV annulus, leaflets, and papillary muscles.
These segmented datasets will be used to train a convolutional neural network (CNN), likely based on the 3D U-Net architecture; 70% of the datasets will be used as training data and 30% will be reserved for testing and validation.
The CNN will learn to automatically segment MV structures and generate 3D surface models. Key anatomical features-such as annulus contour, leaflet free margin, commissures, and papillary muscle positions-will be extracted automatically.
Quantitative geometric parameters will include annular area, perimeter, anteroposterior diameter, commissural width, annular height and ellipticity, leaflet dimensions, coaptation zones and billowing extent.
FE Analysis Patient-specific 3D models of the mitral valve will be reconstructed from the segmented RT3DE data. These models will undergo smoothing and remeshing to generate high-quality meshes for structural FE analysis. Chordae tendineae will be modeled based on established anatomical templates and tuned to match physiological lengths. Tissue material properties will be assigned based on published experimental data.
FE simulations will be used to quantify stress distribution on MV leaflets, assess load transfer between leaflets and papillary muscles and evaluate mechanical effects of surgical interventions (e.g., neochordal implantation, annuloplasty) These simulations are retrospective and exploratory; they will not influence surgical decision-making.
Follow-up No post-operative follow-up is required beyond standard clinical care. The study is focused on intraoperative imaging and tissue collection, followed by offline image analysis and computational modeling.
