Room: AAPM ePoster Library
Purpose: We compare deformable image registration (DIR) techniques based upon contours (C-DIR), image features (I-DIR), and contours and images combined (CI-DIR), and determine which techniques are expected to provide reliable region-of-interest (ROI) dose summation for adaptive re-planning.
Methods: Analyses were performed using original-plan and re-plan kVCT images from previously-treated pelvis cases with clinician-drawn ROIs on both sets. All DIRs were performed with preliminary rigid registration (RR) using a commercial DIR tool. For C-DIR and CI-DIR, the constraining ROIs were the bladder, colon, rectum, and small bowel. DIR quality assurance (QA) metrics included ROI Jacobian determinant (JD) distributions, along with changes in mean distance-to-agreement (?MDA) and Dice similarity coefficient (?DSC) relative to RR alone. ?MDA < 0 and ?DSC > 0 indicate improvement in DIR over RR alone. Lower standard deviation of JD (s(J)) within the ROI suggests more reliable dose summation interior to the ROI. MDA and DSC measure registration accuracy along the ROI boundary, and thus indicate the reliability of ROI maximum dose calculation.
Results: C-DIR exhibits large JD variations (high s(J)) throughout the imaged volume; s(J) is smaller for I-DIR and CI-DIR. This suggests that I-DIR and CI-DIR exhibit less severe deformations within the ROIs and are thus expected to lead to better ROI dose summation. ?MDA < 0 for only the constraining ROIs, mesorectum, and sigmoid for C-DIR, whereas for I-DIR and CI-DIR ?MDA < 0 for all ROIs considered. For C-DIR, ?DSC > 0 for only the constraining ROIs and mesorectum, whereas for I-DIR and CI-DIR ?DSC > 0 for all ROIs. This suggests that I-DIR and CI-DIR provide better ROI surface propagation.
Conclusion: Based upon the JD, ?MDA, and ?DSC DIR QA metrics, I-DIR and CI-DIR are expected to provide more reliable dose summations than C-DIR, and thus are recommended for adaptive re-planning.