Click here to

Session: Machine Intelligence in Image Processing and Motion Correction I [Return to Session]

MRIGRT - 4D Motion Modeling and Real-Time Online 3D Motion Tracking

y zhang1, X Wu1*, H Gach1, J Ginn2, D Yang2, (1) Washington University, Saint Louis, MO, (2) Duke University, Chapel Hill, NC


TH-E-BRC-4 (Thursday, 7/14/2022) 1:00 PM - 2:00 PM [Eastern Time (GMT-4)]

Ballroom C

Purpose: Real-time monitoring, estimation, and prediction of the respiratory movement of tumors and nearby OARs are very challenging for treating thoracic and upper abdominal cancers. The latest MRI-guided radiotherapy system can acquire 2D Cine MRIs at 8Hz during beam delivery and perform real-time 2D image-based tumor motion tracking and beam gating. However, 2D imaging is often inadequate to accurately track 3D motion of tumors and OARs.

Methods: A novel motion modeling and estimation approach named Low-Resolution-Motion-Modeling (LRMM) was developed to enable 3D motion tracking with two stages: 1) pre-delivery motion measurement and modeling; 2) 3D motion estimation during treatment delivery. Forty low-resolution (4.5x4.5x7 mm3) free-breathing 3D-Cine MRIs (2 volumes per second) were added after the regular breath-hold 3D MRI patient setup scan. A groupwise deformable registration AI model was developed and applied to compute the 4D motion on the 3D Cine volumes before beam delivery. To compute 3D motion during beam delivery, each frame of the continuously acquired 2D-Cine MRIs was approximated from the central slices of the 3D Cines. The corresponding 3D motion field for each 2D-Cine frame was estimated from the weighted sum of the determined weights with corresponding deformation fields calculated in the modeling stage.

Results: Real-time 3D motion computation (0.012 second per frame) during beam delivery was accomplished without computationally expensive iterative optimization. Synthetic experiments and healthy subject studies were conducted to demonstrate and evaluate the proposed approach. An average 0.5±0.4 mm 3D motion tracking accuracy was obtained after registration and motion modeling on XCAT digital motion phantom simulations. Data from three healthy human subjects were processed and visually evaluated.

Conclusion: A novel 4D motion management procedure was successfully developed. It allows tracking tumor and OAR motions in 3D and real-time, which is an important advancement over the current 2D tracking and gating.

Funding Support, Disclosures, and Conflict of Interest: Supported by NHLBI R01HL148210


MRI, Registration, Modeling


IM/TH- Image Registration Techniques: Modality: MRI

Contact Email