Evaluation of deformable image coregistration in adaptive dose painting by numbers for head-and-neck cancer

Abstract

Purpose: To assess the accuracy of contour deformation and feasibility of dose summation applying deformable image coregistration in adaptive dose painting by numbers (DPBN) for head and neck cancer.

Methods and materials: Data of 12 head-and-neck-cancer patients treated within a Phase I trial on adaptive (18)F-FDG positron emission tomography (PET)-guided DPBN were used. Each patient had two DPBN treatment plans: the initial plan was based on a pretreatment PET/CT scan; the second adapted plan was based on a PET/CT scan acquired after 8 fractions. The median prescription dose to the dose-painted volume was 30 Gy for both DPBN plans. To obtain deformed contours and dose distributions, pretreatment CT was deformed to per-treatment CT using deformable image coregistration. Deformed contours of regions of interest (ROI(def)) were visually inspected and, if necessary, adjusted (ROI(def_ad)) and both compared with manually redrawn ROIs (ROI(m)) using Jaccard (JI) and overlap indices (OI). Dose summation was done on the ROI(m), ROI(def_ad), or their unions with the ROI(def).

Results: Almost all deformed ROIs were adjusted. The largest adjustment was made in patients with substantially regressing tumors: ROI(def) = 11.8 ± 10.9 cm(3) vs. ROI(def_ad) = 5.9 ± 7.8 cm(3) vs. ROI(m) = 7.7 ± 7.2 cm(3) (p = 0.57). The swallowing structures were the most frequently adjusted ROIs with the lowest indices for the upper esophageal sphincter: JI = 0.3 (ROI(def)) and 0.4 (ROI(def_ad)); OI = 0.5 (both ROIs). The mandible needed the least adjustment with the highest indices: JI = 0.8 (both ROIs), OI = 0.9 (ROI(def)), and 1.0 (ROI(def_ad)). Summed doses differed non-significantly. There was a trend of higher doses in the targets and lower doses in the spinal cord when doses were summed on unions.

Conclusion: Visual inspection and adjustment were necessary for most ROIs. Fast automatic ROI propagation followed by user-driven adjustment appears to be more efficient than labor-intensive de novo drawing. Dose summation using deformable image coregistration was feasible. Biological uncertainties of dose summation strategies warrant further investigation.