Validation of a Model-Based Segmentation Approach to Propagating Normal Anatomic Regions of Interest Through the 10 Phases of Respiration

Abstract

To validate a model-based segmentation (MBS) algorithm in a commercial radiation treatment planning system for use in propagating the contours of normal anatomic regions of interest (ROIs) through the respiratory phases that constitute a four-dimensional (4D) computed tomography (CT) image data set. The 4D CT data sets for 12 patients treated for non-small-cell lung cancer were acquired. Five ROIs were selected for delineation: right and left lungs, spinal cord, heart, and esophagus. These ROIs were manually delineated on the CT data set corresponding to the end-inspiration respiratory phase (0%). An MBS algorithm implemented on the treatment planning system propagated the ROIs sequentially through the respiratory phases that constituted the 4D CT data sets, concluding with the 0% phase data set, which was propagated from the 90% phase data set. The propagated ROIs on the 0% phase were compared with the original ROIs on that phase by using visual assessment and a quantitative measure of coincidence. Acceptable propagation accuracy within 1 mm of uncertainty was achieved for lungs and spinal cord. Propagation of the heart produced slightly larger contours that were similar to interphysician variations in contouring the heart. The esophagus was poorly propagated because of lack of tissue contrast and definitive shape. The MBS propagation is a promising tool for efficiently propagating contours through the different phases of respiration. However, propagating the esophagus through this technique may be difficult because of the lack of definitive shape and clearer boundaries from surrounding tissue.