303. Image registration – precise quality assessment of radiotherapy without necessity of showing corresponding points in simulation and portal images

PurposeEnabling the quality assessment of radiotherapy to be made in daily practice, using the new software tool to analyze the simulation and portal images.MethodIn the registration of the anatomical structures as well as the irradiation fields, the features used as landmarks are the edges. The significant edge fragments must be chosen manually, but without showing any specific corresponding points. Field edges marked with wires in the simulation image are found fully automatically with the original combination of a dedicated line edge detector and a version of hierarchical, combined Hough transform. The registration is guided by the robust accuracy criterion using the modified Hausdorff distance measure. The only parameter of the measure – quantile rank, or share of data used in comparison – is not fixed, but evolves from 1 to 0 during the optimization of the accuracy. This has two advantages. 1: The user can choose the result found for the share corresponding to the actual share of erroneous data in the images, which can be seen only after the results for all the possible ranks are known. 2: The algorithm can avoid the local minima. The registration takes few seconds on a typical PC. The method has been implemented in a software tool which supports the complete process of measurement, and has been tested in clinical triais with positive result.ConclusionsThe modified Hausdorff distance measure with evolving rank is a good and efficient registration accuracy measure for quality assessment of radiotherapy based on the comparison of portal and simulation images