Purpose: There are technical limits to temporal acquisition of 3D MR data sets. In radiation therapy, MR-linac utilizes a â€œ2D+1â€? model: 2D cine imaging based on single user-selected plane is superimposed on corresponding plane extracted from 3D data set prior to treatment. We propose a more robust model which takes advantage of expanded 3 simultaneous planes of cine imaging available on an Elekta Unity MR-linac to improve targeting accuracy.
Methods: MR imaging was obtained from Philips Ingenia Ambition 1.5T that utilizes all imaging exam cards available on Unity. T2-weighted abdominal imaging along with subsequent 3 planes of cine imaging at frame rate of 2/sec were acquired, and a pancreas target contoured. The 3 planes were combined into a single volume, â€œ2.5Dâ€? representation of target volume. This 2.5D object was 1) superimposed on 3D data for contour comparison based on Dice coefficient; and 2) rigidly registered with initial 3D data. This allowed for 2.5D+1 monitoring of target volume during irradiation, no longer requiring user selection of a single plane for gating.
Results: Beyond using threshold Dice coefficient to trigger gating based on overlap of single cine imaging plane onto the 3D pre-treatment data, Unity allows this to be expanded to 3 orthogonal planes, which for the target had contour boundary accuracy of 2.5mm in 3 planes. Alternatively, the perpendicular distances from the contours defined on cine images to the 3D target can be used to estimate target motion. Rigid registration based on solids allows for position accuracy of <2mm and can identify target deformation.
Conclusion: Our 2.5D+1 model demonstrates that accurate registration can be used to guide MR-linac gating and offers improved accuracy and robustness over 2D+1 model. Clinical implementation will require computation hardware optimization so that gating reflects temporal changes in target position in near real-time.