SU-E-J-05: A Dose-Based Metric to Assess the Accuracy of Deformable Image Registration




To examine the correlation between landmark-based spatial registration error and a novel metric for evaluating dose mapping error.


The dose mapping error was calculated from the difference between the dose distributions obtained using two different dose mapping techniques. One method calculates the dose on a deformed geometry while the other calculates the dose on the target geometry. In both cases the dose distributions are mapped in a consistent manner back to the reference geometry. For a realistic clinical lung plan, dose distributions were mapped to the Exhale phase for repeated deformable registrations with different average spatial registration accuracies as determined by distance to agreement of 300 manually identified landmarks. The dose mapping error and spatial registration error were compared at each landmark and the correlation was assessed.


No clear correlation was observed between spatial registration accuracy and dose mapping error, however, the average dose mapping error was found to increase as average spatial registration accuracy increased. The relationship between the dose mapping error and spatial error at each landmark point was found to be influenced by the dose gradient, density gradients and voxel size.


The dose mapping error was found to not have any direct relationship with spatial registration error but depends on a number of factors including: dose gradient, density gradient and voxel size. Our results indicate that landmark analysis on its own is not a sufficient predictor of the accuracy of dose mapping and that metrics that directly assess the dose mapping error should be used.

Natural Sciences and Engineering Research Council