Session Item

Orthogonal cine MRI has been clinically used for motion monitoring/gating in MR-guided-radiotherapy (MRgRT), but might misestimate motion due to the lack of volumetric motion information. We aim to develop a fast 3D volumetric dynamic MRI for potential respiratory motion monitoring/gating in MRgRT.

7 healthy volunteers underwent free-breathing abdominal scans on a 1.5T MR-sim in RT position with an 18-channel body matrix coil and a spine coil. 3D gradient-echo sequence VIBE (Volumetric Interpolated Breath-hold Examination) was accelerated by 4-fold CAIPIRINHA (Controlled aliasing in parallel imaging results in higher acceleration). View sharing technique TWIST was used (20% and 25% undersampling rate for central and peripheral K-space) to further accelerate acquisition. This imaging protocol (Axial, TE/TR=0.6/1.5ms, flip angle=5^o, partial-Fourier factor=0.75, 56 slices/volume, voxel size=3x3x4.5mm³, temporal resolution= ~0.33s) yielded a frame rate of ~3 frame-per-second (fps) to cover the entire abdominal motion. 144 time frames (8064 slice images) were acquired with a scan duration of ~49s. Dynamic images were retrospectively reconstructed and reformatted. Abdominal organ motions with respiration in 3D were demonstrated by rigid and deformable registration using 3D slicer and Elastix.

All scans were successful. The high 3fps frame rate was fast enough to divide a normal respiratory cycle of ~3-4s into ten phases directly on volumetric images, without relying on any external/internal respiration surrogate. It completely avoided the volume discontinuity (missing, overlap, zig-zag boundary etc) problems associated with slice-by-slice MRI acquisition and complicated respiration phase or magnitude sorting algorithms. The reformatted CAIPIRINHA-TWIST-VIBE images of a volunteer were illustrated (Fig. 1), along with the voxel-wise dynamic displacement vector field (DVF) calculated by deformable registration. The respiration induced blurring was minor at diaphragm in SI direction. The cardiac motion effect on image quality was unnoticeable. Slight ghosting artifact along AP was observed due to the highly undersampled K-space acquisition, which might not much affect image segmentation and registration for clinical purpose, but need to be further validated. Other limitations in image quality included relatively low spatial resolution and limited image contrast due to the extremely short TE/TR. The translational displacements of liver dome and kidneys (to the first frame) with respiration in a subject were also illustrated (Fig. 2). Future work of reconstruction latency reduction, motion prediction and ghosting artifact removal using deep learning is on-going.

CAIPRINHA-TWIST-VIBE was developed and achieved 3fps frame rate to capture volumetric motion information in the abdomen. It has potentials for motion monitoring, gating and tracking of future MRgRT.