Abstract
Purpose: Although modern MRI scanners have built-in distortion corrections, residual distortion from gradient non-linearity (GNL) may still degrade treatment planning accuracy for large field of views (FOVs). We characterized residual 3D GNL distortion for large FOVs using a novel 3D distortion phantom and developed a correction scheme for patient images. Methods: A 465×350×168mm3 3D distortion phantom with ∼4700 known landmarks was scanned with a 1.0T open MR-SIM (TE/TR/α =5.54ms/30ms/28°, voxel size =1×1×2mm 3). Reverse gradient techniques (identical parameters except read gradient polarities of ±4.48 mT/m) were used to isolate GNL. To increase the FOV in the superior-inferior direction to ∼38cm, an automated script translated the phantom to two off-axis positions yielding ∼14,000 landmarks for evaluation. Centroid analysis was performed between measured and phantom schematic landmark locations. Data were fitted to a sixth-degree polynomial to generate a 3D correction map and applied to images. To correct patient MR-SIM images, distortion maps were resampled to match the acquisition grid and registered. Using distortion maps as warping functions, images were corrected via inverse warping. Pre-and post-correction images were assessed via difference maps and visual evaluation with treatment planning contours. Results: Within a 7.5cm radial distance of isocenter, no pixels had distortions >1mm. However, for radial distances between 7.5 and 15cm from isocenter, ∼35% of pixel were distorted>1mm and 4% were distorted>2mm. Over the full sampled FOV, ∼66% were distorted>1mm, ∼28%>2mm, and 10%>3mm. Applying additional in-house corrections reduced distortions to 2mm. Patient scans showed negligible distortions near isocenter for targets and organs at risk. Conclusion: Distortion magnitudes resulting from GNL were non-negligible for large FOVs away from the magnet isocenter and necessitate additional corrections. Although corrections may be negligible near the treatment planning isocenter, effects of distortion on peripheral structures warrant further study. Research supported in part by a grant from Philips HealthCare (Best, Netherlands), and by NIH grant R01 EB016079
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have