ESTRO 2021 Abstract Book

S1543

ESTRO 2021

(Platform for Radiotherapy Outcomes Plan Evaluation and Learning). Chalkidou et al [1] found that BED may be correlated to local control. The purpose of this sub-study was to study this link in greater detail using the submitted DICOM data. Materials and Methods Patients with a single oligo-metastatic lesion were selected from the CtE database where the original primary site had been recorded. Each case was checked against the DICOM database and if found was selected for analysis after the PTV and GTV were identified. The D95% was calculated for both the GTV and the PTV. The BED (Biological Effective Dose) was calculated for a range of possible D95 values by interpolating between the GTV and PTV. Values for the α/β ratio were selected based on the review by [2]. Prostate and breast were assigned primary site specific α/β ratios, 2 Gy and 3 Gy respectively, whilst other sites were assigned 10 Gy. In [2] it was noted that the difference in α/β ratio appeared to be driven by a higher β value for prostate and breast, rather than a low α. If all primaries are assumed to have a common α then BED is proportional to total effect (TE) as BED = TE/α. Due to the wide range of doses present, the data were binned with equal probability bins. The analysis was performed at 6 months, 12 months, 18 months and 24 months post treatment. Results At all time-points and doses the reported local control was high (> 85%). A total of 378 patients were included in the 6 month follow-up, 305 at 12 months, 226 at 18 months and 165 at 24 months. At all four time-points there was a trend for local control to decrease with decreasing BED, as shown for the 18 month data. The significance of this trend was assessed with a Fisher’s exact test. At six months p=0.08, at 12 months p=0.42, p=0.09 at eighteen months and p=0.78 at 2 years. If the specific prostate and breast α/β ratios were removed and all primaries were assumed to have the same α/β ratio then no suggestion of a dose response was observed at any time point.

Conclusion The very high absolute local control rates make it challenging from a statistical standpoint to confirm the presence of a dose response, however some indication of a dose response is seen at all time points studies and the difference was borderline significant at 6 and 18 months if prostate and breast are assumed to have significantly lower α/β ratios than other primary types. Work is continuing to quantify how sensitive this result is to assumptions about α/β ratio and the uncertainty in dose by building models from the data to estimate what values for α and β are most consistent with the data. [1] Chalkidou, A. et al. (2020). Lancet Oncol. 22(1):98. [2] van Leeuwen, C. M. et al. (2018). Radiat Oncol . 13(1):96. PO-1814 Enhancing a radiomic-based model prediction of patient outcome in locally advanced rectal cancer B. Tang 1 , Q. Peng 1,1 , J. Lenkowicz 2 , L. Boldrini 2 , H. Qing 3 , N. Dinapoli 2 , V. Valentini 2 , L.C. Orlandini 1 1 Sichuan Cancer Hospital & Institute, School of Medicine, University of Electronic Science and Technology of China, Radiation Oncology, Chengdu, China; 2 Fondazione Policlinico Universitario Agostino Gemelli IRCCS, ART Radioterapia Oncologica, Rome, Italy; 3 Institute of Nuclear Science and Technology Sichuan University, Key Laboratory of Radiation Physics and Technology, Chengdu, China Purpose or Objective The identification of clinically meaninfull features is of crucial importance for the development of the objective radiomics model. The number of features that can be extracted from an image is large, but not all of them have sufficient robustness to sources of variability. This work aims to reinforce the feature selection process of an external validated radiomic-based model, to enhance its prediction of pathological complete response (pCR) in locally advanced rectum cancer (LARC) patients. Materials and Methods A generalized linear model (GLM) to predict pCR in LARC patients based on two radiomic features and two clinical parameters previously trained in Europe and validated with external inter-continental cohort (59 patients) was object of this study. MR imaging datasets of further 88 LARC patients were acquired with a 3T scanner and images from T2-weighted fast spin echo acquisition protocol were analysed. Patients received chemo-radiotherapy (CRT), with a long or short course RT, followed by total mesorectal excision; response to CRT was determined using tumor regression grade (TRG) according to Mandard classification, as pCR (TRG = 1), or non-responder (TRG > 1). Patient and tumor characteristics are shown in Table 1. The patients were divided homogenously into training group (75%) and validation group (25%). For features extraction, images were preprocessed with Laplacian of Gaussian convolution with kernel filter; sigma from 0.1 to 1.0 (step size 0.05) were applied, then statistical, morphological, textural features, gray Level Co0occurrence Matrix, Gray Level Size Zone Matrix, and Gray Level Run Length Matrix features were extracted by Moddicom software. Mann-Whitney test was used to find features significant for pCR. To avoid the multicollinearity among radiomics features, the optimal subset of radiomic features was selected by the least absolute shrinkage and selection operator (LASSO) binary logistic regression model through tuning the lambda parameter by 5-fold cross-validation, while evaluating area under curve (AUC) of receiver operating curve (ROC). The performances of the new and original model were compared through AUC of the ROC.