Poster — Thur Eve — 33: A Dose‐Based Metric for Evaluation of Image Registration Accuracy

Current approaches for the valuation of image registration accuracy rely on comparison of calculated point displacements with measured motion of manually identified anatomical landmarks or contours. In the context of using image registration to map dose distributions, the interpretation of a landmark analysis in terms of a dose error is not readily obvious. In this work we propose a new method to evaluate image registration accuracy based on a dose mapping error. The dose error is calculated by comparing two different dose mapping approaches which use the same deformation vectors but one scores the energy deposited on the target geometry while the other scores energy deposition on a deformed reference geometry. Any error in the deformation vectors will lead to a discrepancy between these geometries and a difference in the warped dose distribution. The dose warping error was evaluated on a set of inhale and exhale images for a lung patient. Multiple image registrations were performed with different levels of registration accuracy. At each landmark point the dose warping error was computed as well as a simplified dose error calculated from the landmark error and the dose gradient. A comparison of dose mapping error and landmark error revealed that the latter is not sufficient to predict the accuracy of the dose warping and factors such as the smoothness of the deformations must be considered. The proposed dose mapping error can be readily applied to determine the accuracy of image registrations used in treatment planning applications.