ESTRO 38 Abstract book

S538 ESTRO 38

using the independent dose calculation system, CheckTomo. Total delivered dose was determined by accumulating daily dose at each finite element (or, voxel). 3D voxel-wise statistical analysis was performed for dose differences between patients with and without rectal bleeding (RB) ≥Gr2 at 2 years (prospectively collected; CTCAEv4.03; n = 15 ,training; n = 6, validation), for planned and delivered dose. Subregions were identified where p<0.05 (SRR 0.05 ) and p<0.01 (SRR 0.01 ) (Figure 1), and the equivalent uniform dose (EUD) was calculated for each SRR, as well as the full rectal wall (RW). Univariate analyses were performed to identify the dose metrics and clinical prognostic factors to be included in a multivariable logistic regression model. NTCP was calculated as: NTCP = 1/(1+ exp ( -S )), where S is dependent upon the outcome of the logistic regression (Schaake et al., 2017).

Conclusion When evaluating delivered dose on the daily anatomy, the relatively simple 2D in-plane geometric distortion correction closely approximates the 3D correction for axial 4D-MRIs. Further research is needed to assess the impact of image distortions on contouring and plan re- optimisation (‘adapt-to-shape’). PO-0983 Accumulating delivered dose to the rectum using finite element analysis improves toxicity prediction L. Shelley 1,2 , D. Noble 3 , M. Romanchikova 4 , K. Harrison 5 , A. Bates 3 , M. Sutcliffe 1 , S. Thomas 2 , N. Burnet 6 , R. Jena 3 1 University of Cambridge, Department of Engineering, Cambridge, United Kingdom ; 2 Cambridge University Hospitals, Department of Medical Physics & Clinical Engineering, Cambridge, United Kingdom ; 3 Cambridge University Hospitals, Department of Oncology, Cambridge, United Kingdom ; 4 National Physical Laboratory, Data Science, Teddington, United Kingdom ; 5 University of Cambridge, Cavendish Laboratory, Cambridge, United Kingdom ; 6 University of Manchester, Manchester Academic Health Science Centre, Manchester, United Kingdom Purpose or Objective In prostate cancer radiotherapy (RT), the rectum is a dose- limiting organ. Existing NTCP models are based on planned DVH data from the static pre-treatment CT scan. However, delivered dose to the rectum differs from planned due to interfraction motion. Here we present a system for accumulating motion- inclusive delivered dose to the rectal wall, using biomechanical modelling and finite element analysis (FEA) to build voxel-by-voxel dose histories, and use this data to construct NTCP models incorporating spatial features. Material and Methods 185 prostate cancer patients treated with helical IGRT were split into training (n = 139) and validation groups (n = 46). Dose prescriptions were 74 Gy/37# and 60 Gy/20#, combined by converting to EQD in 37 fractions. Rectal position during treatment was identified from daily MVCTs using a locally developed autosegmentation algorithm. The auto-contours were used to generate a simulated rectal model in the FEA environment, Abaqus (Dassault Systèmes®). For each patient, the model was biomechanically grown and deformed to the planned and daily contours. Daily dose to the rectal wall was calculated

Results Univariate analysis revealed that EUD of SRR 0.01 and RW, from delivered dose, were discriminative of RB. No significant associations were observed for corresponding planned dose metrics. Pre-treatment status of hypertension (HTN) was associated with RB. A NTCP model was generated for each dose metric. For the training set, all NTCP models were predictive of RB for both planned and delivered dose metrics (Figure 2), the strongest being SRR 0.01 (delivered)+HTN, with AUC 0.801 [95%CI: 0.697, 0.905]. The only model predictive of RB upon testing of the validation cohort was SRR 0.01 (delivered)+HTN, AUC 0.829 [0.693, 0.966]. , SRR 0.05 ,