S2131
Physics - Inter-fraction motion management and daily adaptive radiotherapy
ESTRO 2026
1 Department of Radiotherapy, University Medical Center Utrecht, Utrecht, Netherlands. 2 Department of Biomedical Engineering, Eindhoven University of Technology, Eindhoven, Netherlands Purpose/Objective: Online adaptive MR-guided radiotherapy enables high- precision treatment delivery. To facilitate long-course online adaptive treatments in esophageal cancer, automated online target definition without the need for manual adjustments is essential. Current deformable image registration (DIR) algorithms for contour propagation typically rely solely on the planning and fraction MR images. This study investigated whether incorporating masks of organs adjacent to the esophageal CTV improves target propagation. Material/Methods: A 3D deep learning model [1] was trained on planning MR scans and contours from 23 esophageal cancer patients treated on a 1.5T MR-linac. For 14 of these patients, the model segmented the heart, lungs, aorta, trachea, and bronchi on the planning and all fraction MRs. These segmentations served as input for an in- house developed DIR algorithm to propagate all contours from the planning to the fraction MRs. Propagated CTVs from contour-guided DIR were qualitatively evaluated and compared with clinically used software (Monaco) and the same DIR algorithm without contour guidance (default DIR). For each patient, a radiation oncologist specialized in esophageal cancer blindly assessed the three fractions with the highest 98th-percentile Hausdorff distance between contour-guided DIR and Monaco, using a Likert scale based on clinical acceptability.For five patients, treatment plans were optimized for each fraction using unadjusted contours from contour- guided DIR with a 1 mm PTV margin. These plans were accumulated on the planning MR using an independent DIR method, and planning CTV coverage was evaluated. Results: Contour-guided DIR significantly outperformed both Monaco (p<0.0001) and default DIR (p=0.0082) (Figure 1 and 2). CTVs from contour-guided DIR required no major edits, indicating substantial timesaving compared with Monaco, where 11 of 42 CTVs required major edits. Moreover, the CTV was clinically acceptable without adjustments in more fractions for contour-guided DIR (n=31) than for Monaco (n=13) and default DIR (n=27).
the initial treatment plan projected and recalculated on the daily anatomy; and (3) the most recent ART plan projected and recalculated on the current anatomy. Dosimetric parameters among different plans were compared using a linear mixed-effects model (LMM). Results: In the DLAS-based ART workflow, automatic segmentation was completed in 1.38 ± 0.90 minutes, with manual modification of targets and organs of interest (OOIs) taking 7.75 ± 0.40 minutes, leading to a total treatment time of 22.49 ± 2.33 minutes. In 44 oART fractions, the ART plan achieved the most favorable dosimetric outcomes, providing superior target coverage and OOIs sparing, followed by the Prior ART plan, while the Primary plan performed least favorably. The fixed-effect LMM analysis confirmed significantly improved dosimetric outcomes for the ART plan compared with the Primary plan. CTV_D95%, CTV_D98%, and CTV_D100% increased by 1.05, 0.86, and 11.57 Gy, respectively, with a 2.53% rise in CTV_V95%. Correspondingly, PTV_D95%, PTV_D98%, and PTV_D100% improved by 4.77, 8.14, and 11.57 Gy. Furthermore, OOIs revealed variable dose-response patterns. Bladder V40Gy and Dmean increased by 8.28% and 2.56 Gy, respectively; whereas rectal V40Gy, Dmean, and D2cc decreased by 14.15%, 4.27 Gy, and 0.79 Gy. For the small intestine, Dmean and V20Gy decreased by 0.91 Gy and 5.87%, with no significant difference in V40Gy. Colon D2cc decreased by 0.82 Gy, while V40Gy and V20Gy remained unchanged. Pelvic bone V10Gy decreased by 1.82%, and femoral head (R/L) D5% by 2.01 Gy and 1.69 Gy, respectively. A significant positive correlation was found between bladder Dmean change and bladder volume. In addition, the random effects of the model indicated inter-patient heterogeneity, suggesting that the dosimetric benefits of ART varied among patients. Conclusion: The DLAS-based ART workflow was time-efficient. While oART enhanced target coverage without imposing an additional treatment burden, it also revealed variability in OOIs dose responses. Future work should focus on optimizing the dosimetric trade- offs between target coverage and OAR sparing to advance personalized radiotherapy. Keywords: online adaptive radiotherapy; endometrial cancer;
Digital Poster 1709
Contour-guided deformable image registration improves online target definition in MR-guided radiotherapy for esophageal cancer Koen M. Kuijer 1 , Ivo W. Jutte 1,2 , Lando S. Bosma 1 , Cornel Zachiu 1 , Stella Mook 1 , Gert J. Meijer 1
Made with FlippingBook - Share PDF online