Automated quantification of mitral valve morphology and dynamics using real-time 3D echocardiography

Mitral valve repair has become the surgical treatment of choice in patients with mitral regurgitation (MR). However, recent long-term studies have indicated that the recurrence rate of significant MR after repair is much higher than previously thought. Disappointing clinical outcomes likely result from inadequate mitral valve assessment, a necessary step in predicting which patients will benefit from mitral valve repair over replacement and in identifying repair strategies that target patient-specific distortions in valve geometry. Physiological models of the mitral valve constructed from real-time 3D echocardiographic (rt-3DE) images can provide visual and quantitative information that allows for an optimized patient-specific approach to valve surgery. Rt-3DE is an ideal imaging modality for surgical guidance and overcomes many limitations of traditional two-dimensional echocardiography. While the current commercial rt-3DE platforms can acquire high-quality 3D images, these platforms do not provide fully automated quantification or spatially dense representations of valve geometry for interactive visualization. They require off-line user interaction and have limited potential for image-based surgical guidance. To overcome this limitation, this dissertation presents and validates both semi- and fully automated 3D mitral valve segmentation algorithms. The former method is based on user-initialized active contour evolution, while the latter makes use of fully automated multi-atlas label fusion. A unique feature of both algorithms is the use of template-based medial modeling to represent the mitral leaflets as structures with locally varying thickness. In the deformable medial modeling framework, each instance of the valve is mapped to a common shape-based coordinate system, which enables landmark identification and automated morphometry. The utility of the algorithms is demonstrated in three applications: statistical analysis of normal mitral annular geometry, 3D printing of image-derived mitral valve models, and estimation of leaflet stress distributions using image-based models as input to finite element analysis. Furthermore, the dissertation extends 3D mitral leaflet segmentation to 4D and proposes methods for improved spatiotemporal analysis of rt-3DE image data. The work is a step towards providing surgeons a practical tool for interactively visualizing and quantitatively assessing mitral valve morphology and dynamics in the peri-operative period, with the goal of improving long-term outcomes of surgical treatment for MR