ESTRO 2022 - Abstract Book

S11

Abstract book

ESTRO 2022

Conclusion We present a generic DL-based proton dose engine that can be applied to arbitrary geometries and nominal beamlet energies. Offering MC accuracy 1000 times faster, our model allows several steps of the treatment workflow to benefit from drastic speed improvements: from treatment planning to dose recalculation for robustness analysis, or adaptive proton treatments. Straightforward extensions to other heavy ions could offer similar benefits for helium or carbon treatments and enable real-time adaptive treatment delivery.

OC-0039 Improving 4D optimized Pencil Beam Scanned proton plan robustness using motion guided dose delivery

Y. Zhang 1 , N. Vatterodt 1,2 , A. Duetschler 1,3 , S. Safai 1 , D. Weber 1,4,5 , A. Lomax 1,1

1 Paul Scherrer Institut, Center for Proton Therapy, Villigen-PSI, Switzerland; 2 Martin-Luther-Universität Halle-Wittenberg, Institut fuer Physik, Halle, Germany; 3 ETH Zürich, Department of Physics, Zürich, Switzerland; 4 Department of Radiation Oncology, University Hospital of Zürich, Zürich, Switzerland; 5 Department of Radiation Oncology, University of Bern, Bern, Switzerland Purpose or Objective By optimizing beam weights according to a pre-treatment motion model, together with the delivery timeline, 4D optimized (4DO) Pencil Beam Scanned (PBS) proton plans can inherently mitigate the detrimental effects of organ motion. However, such plans are sensitive to changes between the nominal motion for plan optimization (PM) and the actual motion during delivery (DM). We propose and validate in silico , motion guided 4D optimized plan delivery approach (MG4dOPD) to improve robustness, by controlling motion variability within an uncertainty band around the PM. Materials and Methods The two-field 4DO plan were calculated (PTV) for 10 lung cancer cases, using 4DCT(MRI) datasets [1] from 5 patient geometries, each modulated by 2 deformable motions (fig2A). Based on PM using the tumour isocenter as surrogate, uncertainty bands for limiting DM were generated using an adaptive temporal motion model (TMM) (fig1A). To simulate DMs with controlled variabilities (fig1B), 1D surrogate motions with irregular patterns were generated within uncertainty bands of three widths (±2/4/6mm), each for 10 scenarios. A subject-specific spatial motion model (SMM) was established by correlating surrogate motions with deforming anatomy using Principle Component Analysis [2], for the purpose of PBS proton 4D dose calculations. SMM is then used to estimate the 3D deforming motion and associated 4DCTs from each simulated DMs. Moreover, 4DO plans with/without rescanning in the optimization stage were also calculated. The impact of different uncertainty band widths was investigated and compared to the uncontrolled DM without limiting variabilities. Results were then compared to conventional 3D optimized plans (3DO), calculated on a geometric ITV (gITV: encompassing PTVs of all PM phases) based on averaged or inhalation CT’s and with/without volumetric rescanning (VS). All plans were quantified by DVH and V95 in CTV.