S1826
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
Orthopaedics and Trauma Surgery, University Medical Center, Mainz, Germany
plans were then evaluated using predefined clinical goals. Results: The network generates a complete, deliverable plan, with all machine parameters in less than 500 ms. All 19 predicted plans were scaled to achieve target coverage (D95% = 57 Gy). With this scaling 42 % satisfied the high-dose objective (D1% < 66 Gy). Bladder dose goals (V54Gy, V48Gy, V44Gy were met in 95% of plans, whereas rectal sparing remained more challenging (met in ≈ 70%). Femoral head goals were met in all cases. Conclusion: This study demonstrates, for the first time, that a physics-driven loss can directly guide the prediction of deliverable machine parameters. Although current performance is still limited, this approach establishes the foundation for preference and adaptation-driven re-optimization within milliseconds - an essential step toward faster adaptive radiotherapy. Explicitly incorporating dose calculation into the learning process improves interpretability, as the link between predicted fluence and resulting dose distributions becomes directly observable, paving the way for clinically responsive auto-planning. References: Gerd Heilemann et al. „Generating deliverable DICOM RT treatment plans for prostate VMAT by predicting MLC motion sequences with an encoder-decoder network“. In: Medical Physics 50 (Aug. 2023), S. 5088– 5094. issn: 24734209. doi: 10.1002/mp.16545Gerd Heilemann et al. „Ultra-fast, one-click radiotherapy treatment planning outside a treatment planning system“. In: Physics and Imaging in Radiation Oncology 33 (Jan. 2025). issn: 24056316. doi: 10.1016/j.phro.2025.100724Tufve Nyholm et al. „Photon pencil kernel parameterisation based on beam quality index“. In: Radiotherapy and Oncology 78 (Mar. 2006), S. 347–351. issn: 01678140. doi: 10.1016/j.radonc.2006.02.002 Keywords: AI, Automated Treatment Planning, Deep Learning Digital Poster 1284 Planning feasibility study for neoadjuvant hypofractionated radiotherapy with simultaneous integrated tumor dose escalation for soft tissue sarcomas Nina Gercek 1 , Elena Kyryschuk 1 , Justus Kaufmann 1 , Stephanie Göller 1 , Sophia Drabke 1 , Liv-Annebritt Weimer 1 , Anna Sabrina Schunn 1 , Frank Traub 2 , Heinz Schmidberger 1 1 Department of Radiation Oncology, University Medical Center, Mainz, Germany. 2 Department of
Purpose/Objective: Neoadjuvant radiotherapy (naRT) with simultaneous integrated boost (SIB) might increase immunogenicity of soft tissue sarcoma (STS). A retrospective planning study on patients with extremity STS evaluated the feasibility of isotoxic dose escalation in ultrahypofractionated naRT with SIB focusing on respecting dose guidance and achieving sufficient target coverage. Material/Methods: We included ten patients with extremity STS treated with naRT between 2021-2023. Four volumetric- modulated-arc-therapy-plans per patient were calculated to evaluate dose escalation in macroscopic tumor volume (PTV_Boost) within the extended volume (PTV_low). PTV_low received 25 Gy in 5 fractions with SIB dose levels 1 – 4 to PTV_Boost (6,0 × 5, 6,5 × 5, 7,0 × 5, 7,5 × 5 Gy), respectively. We assumed feasibility, if ≥ 80% of plans met goals for target coverage ( ≥ 80% coverage of PTV_Boost) and adhered to dose guidance (skin_max <38,5Gy). Treatment plan parameters including monitor units (MU), modulation factor (MF), gradient index of PTV_Boost (GI_ PTV_Boost), dose range in PTV_Boost, conformity index of PTV_low (CI_PTV_low), and dose to organs of interest, were assessed. Results: Minimum and maximum PTV_Boost doses ranged from 22,94-32,15 Gy in level 1 and 27,22-40,01 Gy in level 4. MU increased from 1285±300 to 1474±391 with a significant difference between level 1 and 4 (Wilcoxon-test; p<0,01). MF, and therefore plan complexity, remained stable across all levels (p>0,05). Target coverage and GI_PTV_Boost differed significantly across levels (Friedman-test, p<0,001). Pairwise Wilcoxon comparison showed a significant difference between levels 1 and 4 (p<0,01). Goals for mean PTV_Boost coverage were met in all plans. Cochran’s Q-test (4,714, df=3, p=0,194) indicated a trend toward greater dose deviation with increasing levels. CI_PTV_low remained high ( ≥ 0,95) in all levels, without significant differences. Skin maximal doses showed significant differences between levels 1 and 3, 1 and 4, and 2 and 4 (Wilcoxon-test, p<0,001). Skin dose maxima in level 1 – 3 were <38,5 Gy in all plans, while 70% of level 4 plans did not comply with skin dose guidance (Cochran’s Q-test, p<0,01). Post-hoc McNemar-test revealed no significant pairwise differences between dose levels. Conclusion: The results demonstrate that dose escalation within PTV_Boost is technically feasible in treatment planning. Target coverage remained stable across increasing dose levels, with only a minor trend toward
Made with FlippingBook - Share PDF online