Purpose: Geometric accuracy is vital for delivering high quality MR-guided radiotherapy (MRgRT). MR images acquired on 0.35T MR-linac systems are distorted when acquired at different gantry angles, requiring repositioning of the gantry for acquiring new volumetric images. This work investigates correction of distorted images acquired at different gantry positions using deformation vector fields (DVFs) from reference images.
Methods: Two phantoms were imaged on a 0.35T MR-linac using a clinical 3D TRUFI sequence. The phantoms included a cylindrical phantom, part of the QUASARTM MRID3D geometric distortion analysis system, and a QUASARTM 3D grid phantom. Acquisitions were performed at 30° increments, from 0° to 330°, with TR: 3ms, TE: 1ms, flip angle: 60°, FoV: 450x450x360mm³, imaging matrix: 300x300x240. Acquisitions were performed with the onboard Siemens distortion correction function turned on and off (DstOn and DstOff). The MRID3D phantom images were imported into the distortion analysis software to produce DVFs for each gantry angle with DstOn and DstOff. These DVFs were then applied to the 3D grid phantom images in the proprietary System Distortion Resampler software from ModusQA. Resampled images were compared to the original and resampled G0 distortion corrected images visually and with the structural similarity index measure (SSIM).
Results: Significant distortion was present in 3D grid phantom images with DstOff, and was partially corrected with DstOn. After resampling with DVFs the phantom appeared visually identical to the reference image. SSIM increased from a mean (±SD) of 0.26±0.001 and 0.06±0.001 before resampling to 0.70±0.004 and 0.69±0.005 after resampling for DstOn and DstOff, respectively.
Conclusion: The resampling software was successfully able to apply DVFs from one phantom to a second to correct for geometric distortion artifacts generated by imaging at different gantry angles on an MR-linac system. This gives the possibility of applying DVFs to patient images for accurate reconstruction during MRgRT.