Quantifying intrafractional colon tumor motion on 1.5 T MR-linac cine-MRI and applying anisotropic residual-motion margins for SBRT

PURPOSE: To noninvasively quantify the intrafraction motion of unresectable locally advanced colon cancer (UNLACC) via 1.5 T magnetic resonance imaging and linear accelerator (MR-linac) cine-MRI and assess the preliminary feasibility of translating motion envelopes into individualized, anisotropic residual-motion margins for stereotactic body radiotherapy (SBRT) workflow. METHODS: This prospective, single‑centre, observational cross‑sectional study serves as a preliminary test to validate the workflow of a phase I single-arm clinical study and falls under the preliminary feasibility verification phase (Clinical Trials.gov NCT06244537; Initial Release Time: 01/25/2024). We first performed internal validation of a rolling-ball-based motion extraction (RBME) algorithm using a dynamic phantom on the same 1.5 T MR linac, which demonstrated negligible systematic error with a Bland‒Altman mean bias of 0.01 mm (standard deviation 0.49 mm) and 95% limits of agreement of -0.98~0.95 mm. We then prospectively accrued 24 patients for a simulated MR-linac workflow and applied RBME to cine-MRI to generate patient- and axis-specific motion envelopes. Intrafraction motion was analysed in a repeated-measures framework to test axis-by-site effects. Finally, we conducted a planning simulation prescribing 25 Gy in 5 fractions; no patients yet treated according to the proposed strategy. The simulated planning target volume (PTV) is the gross tumor volume (GTV) plus an RBME-derived anisotropic residual-motion margin and an additional 3 mm setup margin. RESULTS: Phantom testing supported the accuracy of RBME ( ~ ± 1 mm agreement). In patients, the mean excursions peak in the caudal/right/anterior directions and are smallest in the cranial/left/posterior directions, with the average of means centered at approximately 4 ~ 5 mm. Across all lesions, mean root-mean-square (RMS) displacement was 2.64 ± 1.32 mm cranio-caudal, 2.91 ± 1.05 mm anterior-posterior, and 2.75 ± 1.43 mm left-right. The corresponding 95th percentile (P95) excursions were 4.49 ± 2.43 mm, 4.97 ± 1.74 mm, and 4.76 ± 2.62 mm. Axis-by-segment interactions were significant for RMS (χ² 19.93, df 8, P = 0.011) and P95 (χ² 20.44, df 8, P = 0.008), indicating location-dependent anisotropy. CONCLUSION: A phantom-validated RBME algorithm was used to characterise irregular intrafraction motion for UNLACC on a 1.5T MR-linac. The resulting motion envelopes should be regarded as hypothesis-generating and intended to inform further method development and protocol design.