S1825
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
(2006).[3] V. Malka, S. Fritzler, E. Lefebvre, et al., “Electron acceleration by a wake field forced by an intense ultrashort laser pulse,” Science, 298(5598), 1596–1600 (2002). Keywords: VHEE, laser plasma acceleration Digital Poster 1251 Physics-informed deep learning for direct prediction of deliverable VMAT treatment plans Simon Glatzer 1,2 , Matthias Kronsteiner 1,2 , Gerd Heilemann 1,2 , Minh Vu 3 , Josef A Lundman 3 , Joakim Jonsson 3 , Jörgen Olofsson 3 , Kristina Sandgren 3 , Anders Garpebring 3 , Tommy Löfstedt 4 , Tufve Nyholm 3 , Wolfgang Lechner 1 , Dietmar Georg 1 , Attila Simkó 3 1 Department of Radiation Oncology, Medical University of Vienna, Vienna, Austria. 2 Comprehensive Center for AI in Medicine, Medical University of Vienna, Vienna, Austria. 3 Department of Diagnostics and Intervention, Umeå University, Umeå, Sweden. 4 Department of Computing Science, Umeå University, Umeå, Sweden Purpose/Objective: Auto-planning using AI has shown great potential to accelerate high-quality radiotherapy treatment planning. Previously proposed models rely on surrogate losses that do not directly reflect the clinical dose distribution and the physical relationship between fluence and dose delivery. We propose a physics-informed model that integrates a differentiable dose engine into the training, allowing for more accurate dosimetric optimization immediately during the learning process. This approach represents a key step toward physically meaningful and thus consistent, deep learning based
Digital Poster 1235
optimization of a laser-plasma accelerator for high-quality, very-high-energy electron beams Lidan Grishko, Arnaud Courvoisier, Eyal Kroupp, Anton Golovanov, Sheroy Tata, Noam Ben Pazi, Tomer Friling, Victor Malka Department of Physics of Complex Systems, Weizmann Institute of Science, Rehovot, Israel Purpose/Objective: With the growing incidence of cancer worldwide, there is a continuous need for innovative and cost-effective treatment modalities. Radiation therapy accounts for approximately half of all cancer treatments. While photon-based therapies continue to improve, they still face limitations in dose conformity and can cause unnecessary exposure to healthy tissues. Proton therapy offers superior dose localization through the Bragg peak, but its systems are complex, costly, and require long treatment times.Very high-energy electron (VHEE) therapy has recently emerged as a promising alternative, offering the possibility to treat deeper tumors with improved dose conformity, when combined with scanning delivery [1]. Laser-plasma accelerators (LPAs) provide a compact and potentially affordable source for producing high-quality, high- energy electron beams required for VHEE therapy [2]. Material/Methods: Ultrashort, high-intensity laser pulses are focused on a tailored gas-jet target to drive plasma wakefields. The interaction between the laser and the plasma leads to electric field gradients on the order of tens of GV/m, enabling the acceleration of injected electrons to very high energies [3]. A novel three-dimensional, large- field dosimetry method was developed to assess the dose deposition in a water-equivalent phantom for each laser shot. Results: Electron beams with energies up to 200 MeV were produced. The three-dimensional dose reconstruction method enabled precise characterization of the beam and demonstrated promising, clinically relevant performance. Conclusion: These results highlight the potential of compact LPA- based systems for future clinical applications in radiation therapy. References: [1] C. DesRosiers, V. Moskvin, A. F. Bielajew, and L. Papiez, “150–250 MeV electron beams in radiation therapy,” Physics in Medicine & Biology, 45(7), 1781 (2000).[2] Y. Glinec, J. Faure, V. Malka, T. Fuchs, H. Szymanowski, and U. Oelfke, “Radiotherapy with laser- plasma accelerators: Monte Carlo simulation of dose deposited by an experimental quasimonoenergetic electron beam,” Medical Physics, 33(1), 155–162
adaptive planning. Material/Methods:
This study included 258 patients treated for prostate cancer using VMAT plans with a prescription dose of 60 Gy in 20 fractions. The dataset was expanded to 418 cases through data augmentation. The model is based on a previously established [1][2] encoder– decoder architecture predicting multileaf collimator (MLC) trajectories. This work builds upon a differentiable, physics-informed dose engine (Simkó et al., submitted), which enables gradient-based optimization through the physical dose calculation using a pencil beam kernel [3]. Here, we demonstrate its integration into a machine-parameter prediction model for deliverable VMAT plans. This allows the network to optimize towards clinically relevant dose objectives in contrast to conventional parameter- based regression losses. The model was trained on 299 treatment plans, validated on 100, and tested on an independent set of 19 plans. The AI generated
Made with FlippingBook - Share PDF online