A Novel Metric to Evaluate Dose Deformation Error for Deformable Image Registration Algorithms
M Surucu*, J Roeske, Loyola University Medical Center, Maywood, ILSU-E-J-110 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To propose a novel metric to estimate the dosimetric error associated with deforming dose cloud using the mapping generated by deformable image registration (DIR).
Methods: The increasing use of the DIR techniques in daily clinical practice requires extensive quality assurance. Deforming dose using the deformation map generated by deforming one CT image to the other is controversial. We propose here a novel method that can serve as a quantitative QA metric for DIR. First, various isodoses in the original dose are converted into structures. Then a deformation map is generated by deforming the original CT to new CT scan. The subsequent mapping is then used to deform the isodose level structures and dose into the new CT (Dose/structure/deform set). The same isodose levels in the deformed dose are converted into structures (Dose/deform/structure set). Then, differential dose surface histograms (DDSH) for each contour are generated using the deformed dose. The difference between the full-width-half-maximum (FWHM) of the DDSHs at the corresponding isodose levels in both sets is defined as the dose deformation factor (DDF). For demonstration purposes, DDF was calculated at 14 different isodose levels for a representative head & neck and representative lung cancer patient that were rescanned during treatment.
Results: The mean DDF was 0.38 Gy (range: 0.06 to 1.28Gy) for lung case and was 0.14 Gy (range: 0.02 to 0.85Gy) for the HN case. The maximum DDF values occurred at the 50% and 30% isodose levels for lung and HN cases, respectively.
Conclusion: The proposed DDF can be used to evaluate the dose deformation error at different dose levels, independent of the deformation algorithm. The unit of the DDF is Gy and is useful in understanding the clinical effect of dose deformation at different isodose levels. This metric can be used to evaluate commercially available DIR software.