Deformable Model-Based Segmentation of 3D CT by Matching Distributions
R.J. Radke, D. Freedman, T. Zhang, Y. Jeong, and G.T.Y. Chen
Medical Physics, 2004

A rate-limiting step in tomographic image guided therapy or 4D CT-based treatment planning is the significant amount of time and human intervention required to delineate the tumor and nearby critical structures on each scan. A radiation oncologist can often take 30-45 minutes to outline all of the structures of interest in every axial slice. Performing this operation each of the 30-40 times a patient is treated is tedious. Using an uncompiled MATLAB implementation on a modest hardware platform (1.67 GHz AMD machine with 448 MB RAM), we present an algorithm that performs the same procedure in less than five minutes. Our early results show in-plane agreement with manual segmentation of multiple organs in the male pelvis on the order of 2- 3mm.