ESTRO 38 Abstract book

S350 ESTRO 38

Netherlands) using a dedicated eye coil and sequences to reduce eye-motion artefacts. UM and OARs were automatically segmented, and subsequently, a 3D model of the eye was created (Figure 1). A TPS was developed in-house using a semi-analytical broad beam algorithm for passive scattering, in which the 3D eye-model was integrated. Human eye motion was mimicked through an Euler-sequence of rotations around the optical axis (abduction/adduction, elevation/depression and possible torsion). These rotations were used to simulate various clinically feasible gazing angles. For tumor coverage a safety margin was set by adjusting the range and modulation width of the Spread-Out-Bragg Peak proximal and distal edges. Lateral conformation was achieved by a collimator adjusted to the tumor contour. To optimize dose to OARs, a weighted-sum objective function was computed for each gazing angle, which also included an objective to penalize extreme gazing angles. Weights could be adjusted to prioritize sparing of specific OARs. Results The novel treatment planning approach has been tested on 5 patients. Tumor coverage was reached for all cases (D 95% > 95%). Figure 2 shows an example of the weighted- sum objective function for one patient for the clinical feasible gazing angles. Fig 2.A shows the map if optic nerve is prioritized. It shows that the half right’s patient side should be avoided, while for the left half most angles are optimal. Fig 2.B shows the objective values for equally weighted sparing of OARs. This tool has great prospects for improving decision-making.

Conclusion The GAN model was able to generate high-quality synthetic CTs from MR images, which can directly be used for MRI-only treatment planning. Given the sensibility of scanned proton beam dose distributions on small differences along the beam path, our model shows a superior dosimetric accuracy that competes very closely to more conventional methods, such as atlas-based approaches, but requires no hand-crafted features and beats them in terms of computational speed. Our next task involves a complete evaluation of the plans robustness against multiple source of errors. OC-669 Development of a novel MRI-only treatment planning approach for ocular proton therapy E. Fleury 1,2 , P. Trnková 1,2 , E. Erdal 1 , K. Hassan 3 , J. Beenakker 4 , J. Herault 5 , J. Pignol 6 , M. Hoogeman 1,2 1 Erasmus Medical Center, Radiation Oncology, Rotterdam, The Netherlands ; 2 Holland Proton Therapy Center, Radiation Oncology, Delft, The Netherlands ; 3 Leiden University Medical Center, Division of Image Processing, Leiden, The Netherlands; 4 Leiden University Medical Center, Department of Ophtalmology, Leiden, The Netherlands; 5 Centre Antoine Lacassagne, Radiation Oncology, Nice, France ; 6 Dalhousie University, Radiation Oncology, Halifax, Canada Purpose or Objective Proton therapy (PT) for Uveal Melanomas (UM) is often employed to lower the risk of irreversible side-effects such as vision impairment, loss or even enucleation. However, a main shortcoming is that high-resolution 3D image data for target volume definition and treatment planning is not used in clinical practice. Instead, the target volume is defined by 2D fundus photography or ultrasound and demarcated by tantalum clips stitched to the sclera. Moreover, the sole commercial treatment planning system (TPS) uses a simplistic eye model, and fails to account for detailed 3D information to accurately characterize the tumor and organs-at-risk (OARs). Besides, the optimal gazing angle for each patient needs to be manually defined and is not objectively optimized. To overcome these limitations, we aim to develop an MRI-only workflow for PT of UM in a multidisciplinary and multicenter approach. This project is to develop an MRI-based planning method with dose-based gaze-angle optimization. Material and Methods 5 patients referred to a single academic center for UM treatment were prospectively included into the study after signing an informed consent. High-resolution ocular images were acquired on a 7T Philips Achieva MRI (Best,