Room: Exhibit Hall | Forum 2
Purpose: Tumor motion variability causes radiotherapy uncertainty for thoracic and upper abdominal cancer due to respiratory irregularities. This study evaluated a novel approach to reconstruct time-resolved (TR) 4DMRI images through library matching between online sparsely-sampled images and an ultra-fine respiratory-correlated (RC) 4DMRI library using a golden-angle radial acquisition (GRA) with compressed sensing (CS) reconstruction.
Methods: A library-matching approach was evaluated for TR-4DMRI reconstruction via simulation. An ultra-fine RC-4DMRI library was built by resorting simulated radial spokes to 50 respiratory states (âˆ†=1-2mm at the diaphragm, 50 spokes/state) using an internal navigator after continuous GRA with stack-of-stars sampling. Three sparsely-sampled 2D slices with low-resolution/quality at 1-2cm interval were used to find a match in the high-resolution library and a range of 6-15 spokes (4ms/spoke) were tested using a T1-weighted 4D digital motion phantom with two diaphragmatic motions of 2.5cm and 4.0cm and a 4s period. Residual motion during acquisitions was simulated and factored into library building and sparse sampling. Both diaphragm position matching and rigid image matching were applied and compared. The image quality of 6-15 spokes was evaluated visually and quantitatively, while the accuracy of library matching was evaluated using the ground truth.
Results: The 50-state RC-4DMRI library is reconstructed with high-fidelity, high-resolution (2x2x4mm3), and minimal artifacts, compared to the reference 4D digital phantom. The 10 spokes/slice sampling provides a frame rate of 8Hz and an optimal matching result, with a good balance between the number of spokes and residual motion error. Library matching error is 0.7Â±0.5 state (2.5cm/4s) and 0.7Â±1.2 state (4.0cm/4s), where a state represents a 2.0mm misalignment at the diaphragm, leading to ~2.0mm uncertainty in TR-4DMRI reconstruction.
Conclusion: This study has demonstrated the feasibility of library-matching reconstruction of TR-4DMRI with high spatiotemporal resolution using GRA/CS. Further investigation is on-going to acquire patient data to reconstruct TR-4DMRI images.