ABSTRACT
BACKGROUND & PURPOSE: To propose a novel mid-position (midP) workflow for MRI-guided liver SBRT and provide a validation of the required midP-MRI generation and registration steps. MATERIALS & METHODS: The first step of the midP workflow is the generation of a simulation midP-MRI from a 4D-MRI scan using deformable image registration. Next, a planning midP-CT is warped to the midP-MRI to enable planning in the midP-MRI anatomy. For daily MRI-guidance, three different registration methods to the simulation midP-MRI are proposed; (1) 4D rigid registration of all phases of the daily 4D-MRI, (2) 3D rigid registration of the daily midP-MRI, and (3) 3D deformable registration of the daily midP-MRI. The midP-MRI image quality was assessed with respect to 4D-MRI acquisition time, which is related to over-sampling of the data acquisition (i.e. number of dynamics). The deformable registration precision for the midP-MRI generation was validated using the distance discordance metric (DDM). The deformable CT-MRI and daily MRI-MRI registration accuracies were quantified using the 'full circle method'. RESULTS: The DDM was 1.5â¯mm (median) within the liver, independent of the number of dynamics. The root-mean-squared difference between midP-MRIs based on 10 and 60 dynamics was only 5.2%. The full circle CT-MRI deformable registration error had a median 3D vector length of 1.8â¯mm in the liver. The daily MRI-MRI registration error was submillimeter for all three evaluated methods. CONCLUSION: The feasibility of an MRI-guided mid-position workflow for liver SBRT is supported by the demonstrated high precision of all image processing and registration steps.
