P232] Deformable image registration for CT-images of the thorax region

Abstract

Purpose Evaluate the performance of three commercial deformable image registration (DIR) algorithms for computed tomography (CT) images of the thorax. Methods Deformable registrations were performed on CT-images of a thorax phantom with tumor inserts and on six 4-dimensional patient CT-series of the thorax. For each DIR the shape and position of the deformed structures and the corresponding structures in the target image that the DIRs were set to reproduce, were compared for shifts between 0–53 mm. Two quantitative measures were used, the center of mass shift (CMS) and the Dice Similarity Coefficient (DSC) as well as visual examination of the registered images. Dose calculations on the deformed patient images were compared to calculations on the un-deformed target images for the gross tumor volume (GTV) (mean dose and D98%), lung (V20Gy, V12Gy), heart and spinal cord (D2%). Results In phantoms, moving structures with constant volume and shifts ⩽30 mm were reproduced with DSC⩾0.91 and CMS⩽0.9 mm by all the algorithms. For shifts >30 mm two algorithms failed. Deformations including volume changes were less accurate with 3/12 and 9/12 DIRs considered visually unacceptable, corresponding to CMS>2.0 mm and DSC Conclusions Visually acceptable DIRs were achievable for moving structures with constant volume and shifts ⩽30 mm in phantoms and shifts⩽16 mm in patients. Visually acceptable DIRs implied dose deviations ⩽3.6 percentage units on patient CT-images of the thorax region.