Purpose: Deformable image registration (DIR) in abdomen is challenging due to large organ deformation and motion. Using several DIR quality assurance (QA) metrics, we investigate which DIR technique provides the most accurate DIR-based dose accumulation (DDA) results in MRI-guided abdominal adaptive radiation therapy (MRgART).
Methods: MRI-to-MRI registration techniques investigated included rigid registration (RR), image-based DIR (IB-DIR) which minimizes intensity differences between image pairs, multi-modality DIR (MM-DIR) which incorporates diffusion regularization and feature-similarity scoring, and contour-based DIR (CB-DIR) which minimizes distances between surfaces for corresponding contours in each image. QA metrics included Jacobian determinant (JD) distributions, mean distance-to-agreement (MDA), Dice similarity coefficient (DSC), and a DVH overlay technique. For the latter, the source-image DVH calculated using source-image dose and warped target-image contours is overlaid with the target-image DVH calculated using warped source-image dose and target-image contours. The registrations and QA were applied to daily MRIs acquired during MRgART for pancreatic cancer cases on a 1.5T MR-Linac.
Results: Among the DIR techniques, MM-DIR demonstrated the most uniform JD distributions within the duodenum and stomach, indicating less extreme local deformations. CB-DIR yielded the most favorable registrations along the contour boundaries, with MDA < 1 mm and DSC > 0.8 for both duodenum and stomach. MDA and DSC results for MM-DIR were more favorable than those obtained from RR or IB-DIR. DVH overlay results were best for RR, although for MM-DIR the overlap was favorable for the stomach. The DVH overlay was least favorable for CB-DIR.
Conclusion: The accuracy of image registration between abdominal MRIs and its effect on DDA in MRgART were evaluated for four different registration techniques. For tracking the accumulative maximum point dose along a contour boundary proximal to the PTV, CB-DIR may be the most accurate technique. For the most accurate DDA throughout the whole contour, MM-DIR may be preferred.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the MCW Fotsch Foundation and by Manteia Med.