Comparison of segmentation methods for analysis of endocardial wall motion with real-time three-dimensional ultrasound

Abstract

This paper presents a new methodology for validation of endocardial surface segmentation with real-time three-dimensional (RT3D) ultrasound via analysis of ventricular anatomical shape and deformations. When comparing manual tracing and deformable model segmentation methods, we observe high correlation for volume quantification while 3D shapes show significant differences when directly compared by point matching. In the absence of real three-dimensional ground truth for screening of ventricular anatomy, this study aims to define new tests to compare segmented shapes and analyze their accuracy in the context of wall motion analysis. Endocardial surfaces are fitted with finite element modeling in spheroidal prolate coordinates and analysis is performed via construction of node parameter maps in time. Comparison of parameter maps for healthy volunteers and patients with abnormal wall motion are reported.