S2772
RTT - RTT contouring, target definition, and treatment planning
ESTRO 2026
University Hospital, Odense, Denmark. 3 Department of Clinical Research, University of Southern Denmark, Odensen, Denmark Purpose/Objective: MRI-guided ablative radiotherapy shows promise for locally advanced pancreatic cancer (LAPC) owing to superior soft-tissue contrast and on-couch plan adaptation. However, adaptation depends on accurate organ segmentation, which is often achieved through time-consuming manual editing of contours propagated from reference scans. To address this, we developed an open-source nnU-Net model (https://zenodo.org/records/17569867) for AI-based MRI segmentation of abdominal organs. The current study aimed to (a) evaluate the model during reference planning and (b) compare its performance with contours propagated via deformable image registration (DIR) in simulated online adaptive Training data comprised 108 scans from 32 patients and included both T2w and T1vane image sequences acquired for reference planning on a 1.5 T MRI scanner or for online adaptive planning on a 1.5 T MRI- linac. Eighteen organs (Table 1) were contoured per national guidelines. Testing used T2w and T1vane reference and adaptive scans from 10 independent consecutive LAPC patients. To simulate adaptive workflows, curated reference contours were propagated to adaptive scans using the MRI-linac treatment planning system’s DIR algorithm. Two RTTs and two oncologists rated AI and DIR contours on a 5- point Likert scale [1] ranging from 5 (use as-is) to 1 (unusable) for whole volumes in reference scans and within a 2-cm ring around the PTV for adaptive scans. Time for contour corrections within the ring was recorded by an experienced RTT and compared between AI and DIR using paired t-tests. Results: AI performance was high in reference scans and slightly lower in adaptive scans (Table 1). Median ratings were ≥ 3 for all organs and ≥ 4 for all organs except for the BiliaryTract, Duodenum, Stomach, and V_Portal. Major edits were only required in some scans for varios vessels and the BiliaryTract. In adaptive scans, AI and DIR performed comparably for most organs. Median (range) correction times for T2w scans were 2.8 (0.6–4.5) min for the AI vs 3.3 (1.4–4.7) min for DIR contours (p=0.36). For T1vane scans, correction times were 4.4 (0.9–5.8) min for AI vs 3.2 (1.3–5.0) min for DIR contours (p=0.11) (Figure 1). workflows for LAPC. Material/Methods:
volume (PTV). Dosimetric parameters assessed included PTV coverage (V95%, V90%), maximum dose (Dmax), conformity index, and homogeneity index. Additionally, doses to OARs—such as the ipsilateral lung, thyroid, humeral head, contralateral breast, spinal cord, esophagus, and posterior skin—were analyzed. Statistical significance was determined using independent t-tests and Mann–Whitney U tests, with p < 0.05 considered significant. Results: Results demonstrated that the AP/Lat technique significantly improved PTV coverage (V95% = 79.96 ± 1.61%) compared to the AP/PA setup (66.82 ± 10.66%; p = 0.02). Other dosimetric indices (Dmax, conformity, and homogeneity) were similar between the two techniques. While the AP/Lat approach yielded a slightly lower mean dose to the ipsilateral lung (8.50 ± 0.99 Gy vs. 9.74 ± 1.55 Gy; p = 0.17), it showed a modest increase in thyroid dose (23.48 ± 3.10 Gy vs. 19.68 ± 3.06 Gy; p = 0.08), though neither difference reached statistical significance. Importantly, the posterior skin dose was significantly reduced with AP/Lat (11.36 ± 4.72 Gy) compared to AP/PA (23.50 ± 10.58 Gy; p = 0.04), a key consideration for patients undergoing latissimus dorsi flap reconstruction. No significant differences were observed in doses to the contralateral breast, spinal cord, or esophagus. Conclusion: In conclusion, the AP/Lat beams arrangement offers superior target coverage and significantly reduces posterior skin dose without increasing exposure to critical OARs. These findings support the AP/Lat technique as a potentially preferred approach for SCV irradiation, especially in patients at risk for reconstructive complications. References: Firouzjah RA, Banaei A, Farhood B, Bakhshandeh M. Dosimetric comparison of four different techniques for supraclavicular irradiation in 3D-conformal radiotherapy of breast cancer. Health Phys. 2019 May;116(5):631–6. doi:10.1097/HP.0000000000000991. Keywords: Lateral SCV, Lung dose, beam arrangment Automatic organ segmentation to streamline MRI- guided adaptive radiotherapy for locally advanced pancreatic cancer Elisabeth Kildegaard 1 , Bahar Celik 1 , Rana Bahij 2 , Anders Smedegaard Bertelsen 1 , Ebbe Laugaard Lorenzen 1 , Carsten Brink 1,3 , Tine Schytte 2,3 , Uffe Bernchou 1,3 1 Laboratory of Radiation Physics, Department of Oncology, Odense University Hospital, Odense, Denmark. 2 Department of Oncology, Odense Proffered Paper 1518
Made with FlippingBook - Share PDF online