Automated Liver Segmentation for Cone Beam CT Dataset by Probabilistic Atlas Construction

Abstract

Cone beam CT based image guided radiation therapy can be used to measure and correct positional errors for target and critical structures immediately prior to or during the treatment delivery. Data correlation between Planning CT images and daily CBCT images is the key issue for adaptive radiation therapy, including image registration and segmentation processing. In this paper, aiming for getting accurate liver contour structures automatically in daily CBCT images which is very low-contrast comparing the planning CT, probabilistic atlas is constructed from 50 high contrast planning CT images with manual delineation by oncologist. The incoming CBCT images are registered with the atlas using the deformable registration algorithm, and the liver contour structures are generated automatically by using the deformation map. The experiment results demonstrate the efficiency of our algorithm.