P238] Dosimetric evaluation of deformable image registration error using 4DCT-MRI datasets

Abstract

Purpose Deformable image registration (DIR) is irreplaceable for 4D dose calculation (4DDC). Any geometric errors from estimated deformable vector fields (DVFs) directly lead to error in accumulated dose distribution, influencing clinical decision. Quantifying DIR error is imperative but difficult due to the lack of DVF ground-truth. This study aims to evaluate the dosimetric impact of DIR error by utilizing the advantage of 4DCT-MRI images whose ground-truth motions (DVF) are pre-defined. Methods 6 synthetic 4DCT-MRI datasets were generated by respectively warping the reference CT of 3 liver patients (CTV volume of 122, 264 and 403 cc) with 2 deformable motion fields (as ground-truth DVF) extracted from 4DMRI (spatial-mean-peak-amplitude of (a) 7.82 and (b) 20.61 mm). The Velocity™ B-splines DIR was applied to each 4DCT-MRI datasets for estimating motion between phases using binary-mask encompassing the full patient or just liver. Geometric error was quantified by the differences of estimated DVFs and corresponding ground-truth DVFs. Dosimetric assessment was performed by analyzing the differences between estimated plans and ground truth 4D plans using estimated or ground-truth DVFs as input for 4DDC. Comparisons of such effects were conducted for single- and multi-field scanned proton therapy treatment plans using either single or re-scanned dose delivery. Results For all cases and scenarios, substantial dose differences were observed between the estimated and ground-truth DVFs and 4D plans. DVF-error magnitude were found to be 1.3±2.1 mm and 3.2±3.4 mm inside liver for (a) small and (b) large motion respectively (averaged over phases and patients). For single-field-no-rescan scenario, interplay effects (D5-D95-CTV) were (a) 2.3±3.3% and (b) 6.0±3.8% under-estimated. The volumes receiving point-to-point absolute dose difference over 10% mainly dominate outside target, being (a) 10.8±2.1% and (b) 24.4±4.4% of the 10 mm-extended-CTV. Applying either multi-field planning or rescanning is helpful for mitigating DIR-error. For multi-field-rescan scenario, difference in D5-D95 can reduce to 0.1±0.7/1.5±1.1% for (a)/(b) motion cases with max error within 3% being achieved for all investigated cases, and less than (a) 6.0±2.5% and (b) 4.8±2.8% volume in the extended-CTV can receive absolute dose difference over 10%. Conclusions The DIR error of Velocity™ was quantitatively estimated, whose dosimetric impacts were shown to be significant for local dose distribution but limited on the plan statistical quantities.