SU‐E‐J‐36: Accuracy of MRI/CT Registration for Spinal SBRT Target Definition

Purpose: To quantify the uncertainties in imaging registration for treatment of spinal SBRT by tomotherapy. The uncertainties are quantified for kilovoltage computed tomography/magnetic resonance imaging (kVCT/MRI) and then kVCT/megavotage CT (MVCT) in a two‐step procedure. Methods: Spinal MRI and kVCT and MVCT images for 11 patients were imported to VelocityAIS software, which can perform automated and manual rigid registration between different imaging modalities. Gross tumor volume (GTV) and spinal cord were contoured on MRI.Image registration of MRI and kVCT favoring GTV was repeated for 2 times for each patient. GTV and cord were then mapped to kVCT, giving GTVmap1, GTVmap2, cordmap1, and cordmap2. GTVmap1 is then expanded isotropically by 1mm 2mm 3mm and 4mm until the expanded volume encompasses GTVmap2. Similarity of the contours is defined by Dice index, ratio of overlapping volume of the contours from two registrations to the original volume. kVCT/MVCT registrations were then performed. Contours were mapped to MVCT. Dice index was then calculated.Results: For repetitive MRI/kVCT registrations, manual registration was always required to improve the accuracy, resulting in average Dice index of 0.81±0.08 for cord and 0.91±0.04 for GTV between different trials. Most patients require 2mm expansion in order to compensate for the difference between the two repetitive MRI/KVCT registrations, indicating the reproducibility of the registration has a 2mm uncertainty for GTV. Compared to the kVCT/MRI registration, kVCT/MCVT registration can be performed reproducibly with automated registration, resulting in close to 100% Dice index and 0 uncertainties Conclusions: An intrinsic 2mm error was observed for MRI/KVCT registration but not between the kVCT and MVCT. The results indicate the necessity of taking the 2 mm uncertainties into consideration in contour transfer from MRI to CT.