Room: Track 2
Purpose: MRI-guided Online adaptive replanning (MRgOLAR), enabled with MR-Linac, can substantially improve radiation therapy, however, it is labor-intensive and time-consuming. As OLAR is not necessary for all situations, being able to quickly identify indicators for plan adaption is desirable. Previously reported indicators based on image intensity variations are not reliable for MRgOLAR due to the inconsistent nature of MRI intensities. This work aims to investigate whether the structural similarity index (SSIM) between planning and fractional MRIs can be an effective indicator for OLAR.
Methods: Five representative pancreatic cancer patients underwent MRgOART on a 1.5T MR-Linac were utilized. For each fractional image, both repositioned and re-optimized plans were generated. The superior plan was determined based on the commonly used dose-volume criteria. A high-dose gradient region (HDGR) was defined on the planning image as the bounding box that fully contains the 30 Gy isodose line. Fractional MRIs were rigidly registered to the planning MRI and slice-by-slice SSIM was performed in the pre-defined HDGR. In addition, the 2D volume difference between the two images for the most critical organ around pancreatic head, i.e. duodenum was also calculated. A threshold was determined for SSIM based on the correlation between the SSIM, volume difference and the optimal plan decision.
Results: SSIM values of > 0.5 resulted in mean 2D volume differences of 3±1%, indicating a high degree of structural similarity within HDGR for the slice pairs. Values of SSIM < 0.5 in two or more consecutive slices indicated sizable organ deformation (mean diff: 16±11%), which would require OLAR.
Conclusion: The threshold of SSIM < 0.5 in two or more consecutive MRI slices indicated meaningful anatomic changes and can be automatically obtained and used to quickly and objectively determine the need for online adaptive replanning during MR-guided RT.
Not Applicable / None Entered.