Accelerated 3D bSSFP imaging for treatment planning on an MRI-guided radiotherapy system

Purpose: The purpose of this study was to introduce a compressed sensing and parallel imaging-combined technique to reduce the acquisition time of planning MRI for MR-guided radiotherapy (MRgRT) systems. Methods and materials: A variable-density Poisson-Disk (VDPD) undersampling acquisition along with compressed sensing reconstruction technique was developed and compared with the current planning MR protocol, which uses an optimized balanced steady-state free precession sequence with 7.5-fold (7.5x) acceleration achieved by GRAPPA and partial Fourier. The image quality of GRAPPA and VDPD with 7.5x and 15x acceleration was compared with fully sampled images on a phantom. Two volunteers were recruited to compare the in vivo imaging performance. Ten patients with abdominal tumors were scanned using the conventional GRAPPA 7.5x (25 s) and the proposed VDPD 15x (12.5 s) sequences. Three readers scored the two approaches in terms of the quality for organ and tumor delineation. The gross tumor volume (GTV) and two kidneys were contoured. Differences in centroid location and contour volumes, Dice coefficients, and mean distance-to-agreement (MDA) between contours draw on the two techniques were calculated. All studies were performed on a 0.35 T MRgRT system. Results: In the phantom study, VDPD with 15x acceleration rate had lower noise level than GRAPPA with 7.5x acceleration. In both the phantom and volunteer study, noise amplification was apparent when the acceleration rate was increased from 7.5x to 15x in the GRAPPA acquisition, whereas it was minimally increased using the VDPD approach. In the patient study, no significant difference was found for the scoring and contouring statistics between the two techniques, whereas VDPD only took half the scan time as GRAPPA. Volume difference for the GTV and two kidneys between GRAPPA 7.5x and VDPD 15x was around 7.6%, 1.3%, and 2.8%, respectively; while the Dice index was approximately 0.85, 0.92, and 0.90, respectively. Conclusion: The proposed technique reduced the acquisition time by half and provided comparable or improved image quality than the standard planning MRI protocol. (c) 2018 American Association of Physicists in Medicine