ESTRO 2024 - Abstract Book

S4237

Physics - Intra-fraction motion management and real-time adaptive radiotherapy

ESTRO 2024

1206

Poster Discussion

Better motion-compensated daily imaging using 4D-MRI for MR-linac gating workflows

Katrinus Keijnemans 1 , Tim Schakel 1 , Pim T.S. Borman 1 , Bas W. Raaymakers 1 , Eric S. Paulson 2 , Martin F. Fast 1

1 University Medical Center Utrecht, Department of Radiotherapy, Utrecht, Netherlands. 2 Medical College of Wisconsin, Department of Radiation Oncology, Milwaukee, Wisconsin, USA

Purpose/Objective:

Breathing induced tumor motion is one of the main uncertainties in abdominal and thoracic stereotactic body radiotherapy. This uncertainty is typically passively mitigated by creating an internal target volume treatment plan, which covers the expected amount of motion during free-breathing deliveries. On the MR-linac, real-time motion monitoring (MM) using (orthogonal) 2D cine MR images during treatment can facilitate the delivery of a gated treatment plan with smaller treatment margins. The vendor-provided 3D reference images for daily plan adaptation are typically acquired in free-breathing without motion compensation, or in end-exhale during a breath-hold or triggered acquisition. These 3D imaging strategies suffer from (residual) motion blurring and/or unpredictable acquisition times. This study explores the suitability of deriving improved motion-compensated 3D reference MRIs in a user-selected respiratory phase for daily plan adaptation and real-time MM based on 4D-MRIs acquired under free breathing conditions. MRI data were acquired in 3 healthy volunteers on a 1.5 T Unity MR-linac (Elekta AB, Sweden). Vendor-provided T 2 weighted 3D-MRI sequences for abdomen with and without respiratory triggering were acquired along with an axial T 2 -weighted simultaneous multi-slice (SMS) accelerated 4D-MRI sequence [1] and a T 2 /T 1 -weighted fat-suppressed pseudo-golden-angle radial stack-of-stars (SOS) 4D-MRI sequence [2]. The SMS-4D-MRI acquisition was interleaved with a 1D respiratory navigator (1D-RNAV) for sorting. The acquired 3D-MRI data (2x2x2.4 mm 3 voxel size) were reconstructed at 0.78x0.78x1.2 mm 3 , whereas the acquired 4D-MRI data (2x2x4 mm 3 ) were reconstructed at 1.64x1.64x4 mm 3 (SMS-4D-MRI) and 1.64x1.64x2 mm 3 (SOS-4D-MRI). Data were sorted into 10 phases that accounted for inhalation/exhalation hysteresis and corrected for gradient non-linearities. Optical flow was used for deformable image registration. High-definition images were obtained by warping the respiratory phases onto the anatomy of interest (end-exhale, mid-position, end-inhale) and calculating the time-weighted median voxel intensity [1]. The spatial resolution of the high-definition images in cranial-caudal direction was increased to 1.2 mm to match the 3D MRI resolution. The MM research package (MMRP) version 4.5.0 (Elekta AB, Sweden) was used to simulate real-time MM based on orthogonal cine MR images. The vendor-provided orthogonal T 2 /T 1 -weighted 2D cine images used for real-time MM were acquired in coronal and sagittal orientation for 3:59 min:s (1560 images). A sub-liver contour at the top of the right lobe was used as structure and manually translated to align with the different reference volumes. The MMRP used the first 60 cine MR images to generate a template, which was matched to the 3D reference volume for alignment. Afterwards, 2D-to-2D registrations were performed between incoming cine frames and the template frame [3]. Three registration modes were used that each created a different template: the exhale mode using end- Material/Methods: