KEY POINTS
- The retrospective study included 34 patients with single liver metastases treated with five-fraction MR-guided SBRT on a 0.35-T MR-Linac. In total, 204 MRI scans were analyzed, including simulation and pre-fraction imaging.
- Fractionation ranged from 5 × 5 Gy to 5 × 9 Gy, with 17/34 patients receiving 5 × 9 Gy and another 10 receiving 5 × 8 Gy. Radiomic features were extracted from the GTV and five concentric 2-cm rings of surrounding healthy liver.
- Of 160 GTV features, 55 remained after removing highly correlated features and changes smaller than baseline imaging variability. Temporal trajectories identified three patient groups containing 4, 10 and 20 patients.
- Intensity-based energy was the principal GTV feature separating these groups (adjusted p=0.001). In the two larger groups, its longitudinal behavior was predominantly only weakly correlated with changes in tumor volume, suggesting that the imaging signal was not simply measuring macroscopic shrinkage.
- Longitudinal changes were also detectable in normal liver. In the innermost 2-cm ring around the PTV, 49 features remained for analysis and separated patients into groups of 19 and 15. Intensity-based energy and mean intensity differed with adjusted p<0.001, while maximum intensity differed with adjusted p=0.005.
- These normal-liver imaging changes were predominantly weakly correlated with mean cumulative radiation dose, and significant temporal features were identified even in more distant liver rings. This raises the possibility that longitudinal MRI captures tissue responses not fully described by physical dose alone.
- The study did not link radiomic trajectories to tumor control, progression-free survival or clinically relevant toxicity. No toxicity above grade 2 occurred, preventing meaningful correlation between normal-liver imaging changes and radiation-induced injury.
- The cohort was small and biologically heterogeneous, and all imaging came from a 0.35-T TRUFI sequence specific to the MRIdian platform. The identified clusters and intensity-based features therefore require independent validation before they can be used for response prediction or adaptive decision-making.
CLINICAL TAKEAWAY
Repeated MR-Linac imaging may contain measurable treatment-response information beyond conventional tumor size and dose metrics. Longitudinal radiomics could eventually support biologically informed adaptation during liver SBRT, but these trajectories remain exploratory imaging phenotypes rather than validated response or toxicity biomarkers.