Automatic Real-time Left Ventricular Tracking in 3D Echocardiography: Using probabilistic contour tracking and sequential state estimation algorithms

Abstract

This thesis presents a new framework for automatic real-time left ventricular (LV) tracking in 3D+T echocardiography. The framework enables usage of existing biomechanical deformation models for the heart, with nonlinear modes of deformation, combined with edge models for the endocardium boundary. Tracking is performed in a sequential state estimation fashion, using an extended Kalman filter to recursively predict and update contour deformations in real-time. Contours are detected using normal-displacement measurements from points on the predicted contour, and are processed efficiently using an information-filter formulation of the Kalman filter. Promising results are shown for LV-tracking using a truncated ellipsoid contour model, with deformation parameters for translation, rotation, scaling and bending in all three dimensions. The tracking framework automatically detects LV position initially, even in situations where it is partially outside the volume. It also successfully tracks the dominant motion throughout the heart cycle, and correctly identifies the long-axis.