ESTRO 36 Abstract Book

S485 ESTRO 36 2017 _______________________________________________________________________________________________

Material and Methods The mechanistic prediction is based on a radiobiological TCP model describing the interplay between tumor cell proliferation and hypoxia (Jeong et al., PMB 2013). The study presented here (see Sup. Figure 1) focuses on a cohort of 35 head and neck cancer patients treated with chemoradiotherapy which received baseline FDG PET and FMISO dynamic PET, and intra-treatment FMISO dynamic PET scans, and which excluded subjects having a significant increase in hypoxia during treatment. The model is used to predict the radiobiological evolution of each tumor voxel of the baseline image up until the intra- treatment scan (9.2±3.4 days). The main inputs to the model are the initial fractions of proliferative and hypoxic tumor cells in each voxel, obtained from an approximate solution to a system of linear equations relating cell fractions to voxel-level FDG uptake, perfusion (FMISO K 1 ) and hypoxia (FMISO k 3 ). For each lesion, the predicted levels of intra-treatment hypoxia are compared to the measured k 3 from the intra-treatment scan. A single global parameter (the average fraction of extremely hypoxic cells that take up FMISO) is determined from a training subset of 29 lesions by minimizing the average discrepancy between each lesion’s measured and predicted intra- treatment k 3 histograms (Cramér-von Mises criterion). A validation subset of 10 lesions is held out to test the resulting model.

Conclusion This work presents a methodology to e stimate the parameters of a mechanistic, radiobiological l TCP model based on pre-treatment FMISO and FDG PET scans. The method is able to predict mean and median values of intra-treatment hypoxia for each of the lesions in a validation dataset held out from the analysis. This could potentially be used in the future to, for example, select patients for a de-escalation protocol based on their expected response. More patients will be added to the analysis in order to refine the prediction, find the defining characteristics of the outliers, and consolidate the results. PO-0891 Quality assessment of target volume delineation and dose planning in the Skagen Trial 1 G. Francolini 1 , M. Thomsen 2 , E. Yates 2 , C. Kirkove 3 , I. Jensen 4 , E. Blix 5 , C. Kamby 6 , M. Nielsen 7 , M. Krause 8 , M. Berg 9 , I. Mjaaland 10 , A. Schreiber 11 , U. Kasti 12 , K. Boye 13 , B. Offersen 14 1 Azienda Ospedaliera Universitaria Careggi, Department of Radiation oncology, Firenze, Italy 2 Aarhus University hospital, Department of Medical physics, Aarhus, Denmark 3 Catholic University of Louvain, Department of Radiation Oncology, Brussels, Belgium 4 Aalborg University Hospital, Department of Medical Physics, Aalborg, Denmark 5 University Hospital of North Norway, Department of Oncology, Tromso, Norway 6 Rigshospitalet, Department of Oncology, Copenhagen, Denmark 7 Odense University Hospital, Department of Oncology, Odense, Denmark 8 University Hospital Carl Gustav Carus, Department of Radiation Oncology, Dresden, Germany 9 Hospital of Vejle, Department of Physics, Vejle, Denmark 10 Stavanger University Hospital, Department of Oncology, Stavanger, Norway 11 Praxis für Strahlentherapie, Department of Radiation oncology, Dresden, Germany 12 Sørlandet Sykehus HF, Department of Oncology, Kristiansand, Norway 13 Zealand University Hospital, Department of Oncology, Naestved, Denmark 14 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark Purpose or Objective Skagen Trial 1 is a multicenter, non-inferiority trial randomising early breast cancer patients to loco-regional irradiation with 50 Gy/25 fractions vs 40 Gy/15fractions. Primary endpoint is arm lymphedema. The protocol has pre-specified criteria for target volume delineation and dose planning, and quality assessment of this is reported. Inter-observer variability in delineation and its impact on dose parameters were assessed. Automated atlas-based segmentation was used in order to streamline assessment procedure. Material and Methods

Results The average fraction of extremely hypoxic cells that take up FMISO is 0.15 (95% CI 0.05 – 0.30 on bootstrap). In the training subset, the model predicts the mean, median and standard deviation of each lesion’s intra-treatment k 3 histograms (Pearson’s linear correlation coefficients between predicted and measured values of ρ=0.62, 0.60 and 0.69 respectively, all with positive 95% CI on bootstrap – see Sup. Table 1). In the validation subset, only the predictions of the intra-treatment mean and median k 3 of each lesion are significant (ρ=0.59 and 0.60 respectively).