Segmentation of Right Ventricle in 3D Ultrasound Recordings

Abstract

This thesis presents segmentation of the right ventricle of the heart in real-time tracking of 3D ultrasound recordings. A simple deformable model for the right ventricle is developed based on statistical data from manual segmentations, and the model has been tested out in a set of 3D ultrasound recordings and compared to manually segmented right ventricular volumes. The manual segmentation method with volume approximation is also developed. The segmentation tests on the recordings are performed with an already present Kalman filter based real-time contour tracking framework. The ability of the models to fit to the shape of the right ventricle has been evaluated, and the resulting volume curves have been inspected. Deformable models of the right ventricle are constructed by placing nodes in an initial three-dimensional mesh, and subdivision schemes are applied to make smooth surfaces. There have also been experimented with models of different resolutions and initial positions. A background study of right ventricular anatomy, subdivision, model-based segmentation and Kalman filter theory is included, and clinical applications of volume measurements in real-time are suggested. The results of the segmentation are promising, and indicate that models adjust to the right ventricle during the heart beat.