S1894
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
guide automatic plan optimization. Material/Methods: Dosimetric and geometric data from 335 single brain lesion clinical plans were collected by 8 CK centers. The body volumes included in the 100%, 85%, 65%, 50%, 40% and 30% isodose lines were gathered and the corresponding isodose equivalent radius estimated as Reff = (3Volume/4p)1/3. Conformity index (CI) and Dose Gradient Index (DGI) were assessed together with DGI 10thpercentile values (Minimal). Linear regression models (based on a subset of 225 plans satisfying CI<1.2 and DGI ≥ Minimal) relating the PTV equivalent radius with the Reff were generated for the selected isodose values, aiming to predict the dose fall-off outside the PTV for new patients. Based on this information, a plan template was created and tuned to automate planning optimization. The dose prescription was set to cover 100% of the PTV with 100% of the prescribed dose, limiting the maximum dose to 125%. The isodose shells predicted by the KB model were used as the only optimization constraints. The KB model was validated both with an internal (48 pts- 6 Inst) and external (85 pts- 7 Inst) setting, re-optimizing previous clinical plans. Automatic plans were compared against manual clinical plans, both in terms of dosimetric and technical parameters. Results: R2 values for the regression between Reff and PTV radius were >0.98, showing an increasing inter-center variability for decreasing isodose values. Automatic plans always yielded CI <1.20 and DGI values above Minimal. Both for internal and external patients, automated KB plans showed better CI (p<0.05) and significantly (p<0.0001) steeper dose fall-off in terms of DGI, irradiated healthy brain, and Reff at 85%, 65%, 50%, 40% isodoses, when compared to clinical manual plans. Moreover, automatic plans were more efficient, with a significant (p<0.0001) reduction of beams and nodes compared to the manual ones. Results are summarized in Table 1 for the external setting. The gain obtained by the KB predictive model was not uniform among centers in terms of DGI (fig.2); excepting Inst 6, all Institutes showed a significant benefit in terms of CI/DGI, with Inst 1 showing the largest improvement.
Conclusion: Inter-center differences enhanced the advantages of a multi-institute approach. Predictive models for dose fall-off in CK brain SRS/ SRT planning are feasible and can be exploited to automate plan optimization, avoiding suboptimal plans. Keywords: KB model, Dose gradient prediction, Cyberknife
Digital Poster 2701
Dosimetric evaluation of Magnetic Resonance for Calculating Attenuation (MRCAT) and CT based radiotherapy plans Pak-Hang Nam, Chi-To Yung, Dejun Zhou, Chi-Wah Kong, Kin-Yin Cheung, Siu-Ki Yu, Bin Yang Medical Physics Department, Hong Kong Sanatorium & Hospital, Hong Kong, Hong Kong
Purpose/Objective: Magnetic-resonance-guided online-adaptive radiotherapy utilizes daily MR images for online-
Made with FlippingBook - Share PDF online