SU-E-J-57: Dosimetric Impact of Segmentation Differences in the Context of Brain IMRT




To gauge the dosimetric impact of segmentation differences amongst experts and an automated system for intracranial organs at risk in the presence of large space-occupying lesions being treated with IMRT.


Eight expert raters and a single automated system were used to segment normal tissues of the brain using MR and CT for 20 patients. The expert segmentations were used to generate ground truth estimations. One hundred-eighty IMRT plans were generated to test impact of segmentation differences on three dosimetric end points: target coverage, dose to ground truth, and dose reported versus dose to ground truth. Several dosimetric figures of merit with guidance from QUANTEC were used quantitate these differences: mean/maximum dose, V45, V54, V59, V64 and D1mL for the normal tissues; min, mean, maximum, and V95 for target coverage.


We found target coverage was not affected by segmentation differences. Inverse-planning was also generally robust to segmentation differences and not result in marked over doses to the simulated ground truth structures. Dose reporting discrepancies, however, were commonplace and could be large; that is, the dose reported by the plan compared to that which was the best estimate of true dose, were over or under reported by as much as 30 Gy; one third of maximum doses were under- or over-reported by 2 Gy or more.


This work indicates that inverse-planning in the context of large brain lesions may be generally robust to segmentation differences in terms of target coverage and true dose to the normal tissues. Discrepancies in dose reporting of maximum dose were large, could impact treatment decisions, and may contribute to lack of consensus regarding toxicity, especially relating to the brainstem and optic pathway. These results indicate using figures of merit for planning other than the maximum dose, such as the mean and volume doses.