# Interactive-Automatic Segmentation and Modelling of the Mitral Valve

## TL;DR

This paper presents a semi-automated, user-in-the-loop segmentation method for creating accurate, patient-specific 3D models of the mitral valve from ultrasound images, aiding surgical planning and training.

## Contribution

It introduces a novel 3D active contour-based segmentation approach that efficiently captures mitral valve geometry with high accuracy from ultrasound data.

## Key findings

- Achieves mean surface distance of 1.40 mm in patient group
- Attains 1.01 mm accuracy near ultrasound probe
- Demonstrates potential to improve pre-surgical modeling workflow

## Abstract

Mitral valve regurgitation is the most common valvular disease, affecting 10% of the population over 75 years old. Left untreated, patients with mitral valve regurgitation can suffer declining cardiac health until cardiac failure and death. Mitral valve repair is generally preferred over valve replacement. However, there is a direct correlation between the volume of cases performed and surgical outcomes, therefore there is a demand for the ability of surgeons to practice repairs on patient specific models in advance of surgery. This work demonstrates a semi-automated segmentation method to enable fast and accurate modelling of the mitral valve that captures patient-specific valve geometry. This modelling approach utilizes 3D active contours in a user-in-the-loop system which segments first the atrial blood pool, then the mitral leaflets. In a group of 15 mitral valve repair patients, valve segmentation and modelling attains an overall accuracy (mean absolute surface distance) of 1.40+-0.26 mm, and an accuracy of 1.01+-0.13 mm when only comparing the extracted leaflet surface proximal to the ultrasound probe. Thus this image-based segmentation tool has the potential to improve the workflow for extracting patient-specific mitral valve geometry for 3D modelling of the valve.