Purpose: Deformable dose-accumulation accuracy depends both on local spatial dose gradients of each treatment fraction and the accuracy of the underlying image registration. Conventional deformable registration evaluation methods focus only on the geometrical target registration error (TRE) without considering the dose-distribution characteristics. In this study, we propose a general procedure to incorporate dose distribution features in the registration evaluation process. It can support the selection of registration methods for patient-specific treatment planning and the estimation of dose accumulation uncertainty from the chosen registration methods.
Methods: Our method generates a TRE tolerance map based on the dose distributions and the user-selected dose-error tolerance level in 4 steps: 1) Region of interest (ROI) are selected to excludes low-dose areas. 2) In a 15 mm radius neighborhood around each voxel in the ROI , the neighborhood voxels with doses that deviate from the center voxel dose by more than the dose-error tolerance (5% of the prescribed dose) are labeled. 3) For each voxel, the TRE tolerance vector is created that points to the nearest labelled voxel in its neighbor and has a magnitude equal to the distance between center and nearest labelled voxels. 4) The TRE tolerance map was composed by the TRE tolerance vectors for all voxels in the ROI.
Results: We applied the proposed method to generate the dose-based TRE tolerance map for a SBRT case, and evaluated a Demons image registration method based on landmarks and the tolerance maps. The average landmark TRE was 2.7 mm and >90% of landmark pairs passed the TRE tolerance test.
Conclusion: A procedure was developed to generate the dose-based TRE tolerance map, which could assist the selection of image registration method for subject-specific cases or the production of dose accumulation uncertainty from the registration methods.
Funding Support, Disclosures, and Conflict of Interest: The project described was partially supported by the Agency for Healthcare Research and Quality (AHRQ) grant number R01-HS022888, National Institute of Biomedical Imaging and Bioengineering (NIBIB) grant R03-EB028427 and National Heart, Lung, and Blood Institute (NHLBI) grant R01-HL148210.
Dose, Registration, Image Processing