Purpose: Free-breathing 4D-MRI is a promising alternative to 4D-CT for respiratory motion management on MRI-guided linear accelerator (MRI-LINAC) but requires substantial scan time (normally 5-7 minutes). Unmet clinical needs are 1) 4D-MRI with improved image quality 2) 4D-MRI with sufficient image quality but acquired in significantly shortened scan time, and 3) motion-resolved 3D-MRI with sufficient SNR for visualization and diagnosis. Thus, we developed and tested a respiratory motion detection and deep-learning based reconstruction procedure for a 0.35T MRI-LINAC system.
Methods: We evaluated the proposed procedure using both physical motion phantom and human subject studies. Our proposed procedure has three steps: 1) A T1-weighted self-navigated respiratory motion detection method (CAPTURE) was employed to scan subjects for 2000 spokes (5-7 minutes). 2) A deep learning motion-resolved 4D-MRI reconstruction (Phase2Phase, P2P) was employed to reconstruct 4D-MRIs. 3) Deformable motion vector fields (MVFs) were computed from the 4D-MRIs and then incorporated directly using a motion integrated reconstruction (MOTIF) for 3D-MRIs using 2000 or 200 spokes (30-40 seconds).
Results: Evaluation on the respiratory motion phantom experiment showed that the proposed procedure could reverse the effects of motion blurring and restore edge sharpness. In the human study on 12 healthy volunteers, a blinded radiological review by two radiologists found that images reconstructed by MOTIF on 2000 spokes exhibited superb image quality for sharpness, contrast, and artifact-freeness (scoring>8 out of 10), and images reconstructed by MOTIF on 200 spokes significantly improved sharpness, contrast, and artifact-freeness, comparing to results by non-motion-corrected MRI.
Conclusion: Our procedure produced high-quality, respiratory motion-resolved 4D-MRIs and 3D-MRIs from short scans in an MRI-LINAC system. The proposed method involves a motion correction technique (CAPTURE), and a motion integrated reconstruction (MOTIF), with incorporation of the deep-learning P2P reconstruction for motion estimation. The MOTIF 3D-MRIs could provide improved diagnostic and monitoring information for MRI-guided radiotherapy.