ESTRO 38 Abstract book

S1053 ESTRO 38

Table 1: Summary of classification results from both data groups testing using T2 images Conclusion These results are promising, a larger data set is required to further develop these approaches. However, in the absence of this, there are many areas of improvement still to be explored. EP-1934 A study of RBE and NTCP uncertainties underlying model-based patient selection to proton therapy S.N. Fly 1 , J. Pedersen 1 , J.B. Petersen 1 , C.H. Stokkevåg 2 , L.P. Muren 1 1 Aarhus University Hospital, Department of Medical Physics, Aarhus, Denmark ; 2 Haukeland University Hospital, Deptartment of Oncology and Medical Physics, Bergen, Norway Purpose or Objective Normal tissue complication probability (NTCP) models have been proposed for patient selection to proton therapy (PT). NTCP model parameters from different (mostly photon-based) studies often show large variability with wide confidence intervals. PT dose distributions should also be corrected for the higher and variable relative biological effectiveness (RBE) of protons. A number of different models for variable RBE are available, with their inherent model parameter dependencies. The aim of this study was to investigate how the choice of RBE model and their uncertainties relate to the uncertainties of current NTCP model parameters. Material and Methods Three published variable RBE models (McNamara et al., Wedenberg et al., and Carabe et al.) - as well as the generic (constant RBE = 1.1) alternative - were applied when analysing spot scanning PT plans of six prostate cancer patients (prescription doses of 78 Gy (RBE = 1.1)). The RBE corrected dose volume histograms of the rectum and bladder were adjusted for fractionation effects using the linear-quadratic model (EQD 2Gy ) before entered into the Lyman-Kutcher-Burman (LKB) NTCP model. Nine different published LKB parameter sets (including the QUANTEC parameters, Michalski et al) were used for the rectum and three sets for the bladder. The effect of varying α/β-ratio within the range of published values was investigated. Uncertainties in the NTCP values were calculated from the confidence intervals of the published NTCP parameters using both error propagation and Monte Carlo simulations. Results The choice of NTCP model parameter sets had a much larger influence on the resulting NTCP predictions than the choice of variable RBE model (Fig. 1). The value of the α/β-ratio had a considerable effect on the resulting NTCP estimates for the variable RBE models (Fig. 1). The McNamara model resulted in higher NTCP values than the Wedenberg and Carabe models (Fig. 1) while all three variable RBE models resulted in higher NTCPs than the generic constant RBE alternative, with the differences increasing for lower α/β-ratios. The uncertainty in the NTCP parameters had a large impact on the NTCP values, with differences up to 5% for the rectum (with the QUANTEC parameters) and more than 10% for the bladder (Fig. 2).

Conclusion The choice of NTCP model parameters has a much larger influence on resulting risk predictions than the choice of variable RBE model, even for the rectum where there are more published morbidity studies than for most other organs. The three variable RBE models resulted in higher NTCPs than the generic constant RBE, with the differences increasing for lower α/β-ratios. Even without considering implications of potential differences in volume effects between photons and protons (due to the reduced dose bath of PT), this study clearly shows that model-based patient selection to PT should be performed with caution, taking current model and parameter uncertainties into account. EP-1935 Delta radiomics Features Analysis in GLIoblastoma multifome GLI.F.A. Project. A multi- centric study S. Chiesa 1 , F. Beghella Bartoli 1 , I. Palumbo 2 , R. Barone 1 , M. Lupatelli 2 , C. Masciocchi 1 , R. Tarducci 2 , A. Rongoni 2 , D. Cusumano 1 , R. Russo 3 , P. Floridi 2 , S. Longo 1 , N. Dinapoli 3 , M. Balducci 1 , V. Valentini 1 , C. Aristei 2