ESTRO 2023 - Abstract Book

S2112

Digital Posters

ESTRO 2023

8) reconstructed delivered dose to the bladder and bladder wall and investigated the relation with acute patient reported urinary toxicity. They found that absolute bladder wall dose was most predictive for treatment related toxicity, i.e. V25Gy > 9cc. We therefore compared the dose to the bladder wall for the sub-fractionation workflow with CTV-PTV margins of 2, 2 and 3 mm for LR, SI and AP directions of 25 IMRT pre-treatment plans with the dose in the bladder wall of 25 IMRT pre- treatment plans with an isotropic CTV-PTV margin of 5 mm. To solely focus on the impact of margin reduction, the V25Gy constraint was not yet applied in these plans. The bladder was defined as the entire volume circumscribed by the outer bladder wall, including the bladder content. For practical reasons, the bladder wall was created by 4 mm contraction of the outer bladder wall (Figure 1), following the experience from Willigenburg et al (2022) and Maggio et al (PMB 2013, 58:N115).

Results The average V25Gy dose in the bladder wall for the 5 mm margin plans was 8.0±2.3cc (range 4-12.6cc) whereas for the 2- 3 mm plans this was 5.9±1.5cc (range 4.2-9.7cc). This was a statistically significant difference (p < 0.01). The corresponding numbers of patients for which the V25Gy < 9cc constraint was violated were 7 (28%) and 2 (8%) respectively. Conclusion The clinically relevant bladder wall V25Gy was significantly higher in the 5 mm margin group than in the 2-3 mm group, even when this constraint was not explicitly applied during treatment planning. We therefore expect a lower GU acute toxicity rate in the sub-fractionation group and will present early results of patient reported toxicity in this group. T. Roque 1 , M.S. Agarwal 2 , G. Klausner 3 , S. Guihard 4 , H.M. Parenica 2 , S.N. Andersen 2 , T.A. Cuthbert 2 , E. Maani 2 , C. Anderson 2 , A. Oumani 5 , S. Kandiban 1 , P. Chailloleau 1 , O. Teboul 5 , N. Paragios 6 , W.E.I. Jones 7 , S. Stathakis 7 1 TheraPanacea, Clinical Affairs, Paris, France; 2 UT Health San Antonio, Department of Radiation Oncology, San Antonio, USA; 3 Hôpital Universitaire de Genève, Department of Radiation Oncology, Geneve, Switzerland; 4 Institut de cancérologie Strasbourg Europe, Department of Radiation Oncology, Strasbourg, France; 5 TheraPanacea, AI Engineering, Paris, France; 6 Centrale Supelec, University of Paris-Saclay, Department of Mathematics, Gif-sur-Yvette, France; 7 UT Health San Antonio, Department of Radiation Oncology, San Antonio, France Purpose or Objective The efficacy of traditional radiotherapy planning is hampered by changes in patient anatomy over the course of treatment. Adaptive radiotherapy – involving organs at risk-annotation, dose accumulation, and re-planning - is a pivotal transformation in radiation oncology towards better outcomes, reduced toxicity and decreased lifelong side effects and sequels. Its implementation could be effective and feasible within the clinical workflow by harnessing daily fraction images (e.g. cone- beam CTs) acquired for patient positioning. However, the resolution, contrast, and field of view of CBCTs are often insufficient for use in segmentation or dose computation. In this study, an artificial intelligence-based synthetic augmented CT is proposed and clinically evaluated in terms of its ability to generate automatic contours to overcome these challenges and unlock the potential of using CBCT images for pelvic offline adaptive radiotherapy. Materials and Methods ART-Net®, a CE-marked, FDA-cleared anatomically preserving deep-learning ensemble architecture for automatic contouring (AC) of OAR, was applied to synthetic CTs generated from CBCTs. For a total of 40 patients, automatic annotations of 8 OARs (anal canal, bladder, femoral heads R/L, penile bulb, prostate, rectum and seminal vesicles) were performed and compared against the inter-expert variability between nine expert annotators (divided into 3 groups). The Dice score coefficients (DSCs) are computed for each contour and the average of these DSC across all images is computed to generate the mean interexpert DSC used to evaluate the quality of the auto-contours. Results Average DSCs for ART-Net® on the test set for all OARs are presented in Tab. 1. Auto-contouring DSCs were above the inter- expert variability for all organs, except for the prostate where auto-contouring performed at the level of inter-expert variability. In terms of DSC, the best-performing organs were the femoral heads (90%) and bladder (89%) for the AI solution and the bladder (91%) and femoral heads (89%/88%) for the clinical experts. On the other hand, worst-performing organs were anal canal (36% vs 28%) and penile bulb (44% vs 40%) for the AI solution and for the clinical experts, respectively. PO-2348 AI-based delineation for CBCT offline adaptive radiotherapy: an interexpert variability evaluation