Purpose: To quantify the accuracy of clinical deformable image registration (DIR) software following AAPM TG Report 132 guidelines for CT images of various pediatric anatomical sites.
Methods: Five pediatric radiotherapy patients (ages 4-16 years) who represent five major tumor sites, including brain, head & neck (H&N), mediastinum, abdomen, and pelvis, were selected in this study. Planning CT images and normal tissue contours were exported to an independent validation software. Deformations with known amounts that represent the extreme scenarios were applied to these cases in the validation software. The deformations were set to 10Â° roll for brain, 10Â° pitch and 2cm tissue increase for H&N, 2cm respiratory motion for mediastinum, 2cm tissue increase for abdomen, 50% increase in bladder volume for pelvis. For each site, the ground truth deformed vector field (DVF), synthetic CT images, and structures were acquired in the validation software. The clinical DIR software under evaluation also generated DVF by morphing original to synthetic CT images without the knowledge of true DVF. Comparisons were then made to the ground truth in the validation software.
Results: All of 27 structures in 5 patients met the recommended TG-132 tolerances for the Mean Distance to Agreement (MDA<3mm), Target Registration Error (TRE<3mm), and Dice Similarity Coefficient (DSC>0.8). The DVF error in 95% of the voxels [d(95%)] for each structure were able to stay below <2mm. Average MDA (mm), TRE (mm), DSC, and d(95%) (mm) were [1.8, 2.0, 2.1, 2.2, 2.2], [0.6, 0.6, 0.8, 0.8, 0.8], [0.95, 0.93, 0.96, 0.97, 0.95], and [0.87, 0.22, 0.25, 0.25, 0.30] for brain, H&N, mediastinum, abdomen, and pelvis, respectively.
Conclusion: The evaluated clinical software performed well for selected pediatric patients on the DIR accuracy test based on quantitative metrics recommended by TG-132. The validation software facilitated these quality assurance tests for various scenarios in pediatric radiation therapy.