Interactive-Automatic Segmentation and Modelling of the Mitral Valve
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
URL with Digital Object Identifier
© 2019, Springer Nature Switzerland AG. 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±0.26mm, and an accuracy of 1.01±0.13mm 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.