S2773
RTT - RTT contouring, target definition, and treatment planning
ESTRO 2026
Roel Bouwmans, Lieke T.C. Meijers, Jonna K. van Vulpen, Thomas Willigenburg, Astrid L.H.M.W. van Lier, Cornel Zachiu, Gijs H. Bol, Martijn P.W. Intven Radiotherapy, UMC Utrecht, Utrecht, Netherlands Purpose/Objective: Online MR-guided adaptive radiotherapy (MRgRT) using a 1.5T MR-Linac (Unity, Elekta) requires daily contouring to generate a new treatment plan. For abdominal patients, this process is time-consuming due to suboptimal deformation of anatomical structures and the need for extensive online contour adaptation. While the clinical use of an in-house developed deformable image registration algorithm (Evolution)1and deep-learning (DL)networks2has already reduced delineation time in prostate treatments, their performance for abdominal targets remains unclear. The purpose of this study was to evaluate the feasibility and accuracy of Evolution for gross tumor volume (GTV) contouring in abdominal MRgRT. Material/Methods: Fourteen abdominal MRgRT fractions were analyzed (5 pancreas, 4 liver, 1 abdominal lymph node). All patients underwent a full adaptive treatment (ATS) workflow, including daily recontouring and replanning based on triggered T2-weighted Multi-Vane XD (MVXD) MR scans. For each fraction, two GTVs were generated: one using Monaco deformable registration (v6.2.1.0, Elekta) and one using the Evolution algorithm. Both were compared to the clinically used GTV, generated online by Monaco and manually corrected and approved by a radiation oncologist, using Dice Similarity Coefficient (DSC), absolute Hausdorff Distance (HD), and Mean Distance to Agreement (MDA). Additionally, three radiation oncologists indicated which contour they preferred for clinical use (42 assessments total) (Figure 1).
Conclusion: The open-source model is ready for clinical use in reference scans, where no baseline contours exist. In online adaptation, its performance is comparable to DIR and does not significantly affect editing time. Future work could focus on patient-specific models or targeted deployment for selected patients or specific (e.g. non-vessel) organs that showed favorable performance in reference scans. References: [1] Kraus AC, Iqbal Z, Cardan RA, Popple RA, Stanley DN, Shen S, et al. Prospective Evaluation of Automated Contouring for CT-Based Brachytherapy for Gynecologic Malignancies. Adv Radiat Oncol. 2024;9:101417. Keywords: AI, MRI-linac, Locally advanced pancreatic cancer
Proffered Paper 1576
Evaluation of Deformable Image Registration- Based Auto-Contouring for GTV Delineation in MR- Guided Adaptive Radiotherapy of Upper Abdominal Tumors
Made with FlippingBook - Share PDF online