engleski

Comparison of different metods for atlas construction

Atlas-based image segmentation is an important method for analysis of cardiac Doppler ultrasound images. Various methodologies for atlas construction and optimal atlas selection have been proposed in the literature. In this article, we propose novel methods for atlas construction and atlas-based segmentation. The atlas-based segmentation method utilizes a specially formulated image registration procedure. Four atlas construction methods are compared with respect to accuracy of the obtained atlas-based segmentation methods. The shape-based atlas constructed using a distance transform achieves the minimal error. Finally, we propose a new method for reference atlas segmentation by mapping and averaging manual segmentations of all images to the atlas. Experimental validation was done on an image set of 140 Doppler ultrasound images of cardiac outflow velocity profiles from both healthy volunteers and patients. The proposed methods resulted in higher accuracy of cardiac ultrasound atlas-based segmentation.

atlas; construction; evaluation; comparison; distant transformation; registration; segmentation

nije evidentirano