ESTRO 2020 Abstract Book

S844 ESTRO 2020

1 Institut Claudius Regaud/Institut Universitaire du Cancer de Toulouse – Oncopôle, Radiation oncology, Toulouse, France ; 2 ToNIC- Toulouse NeuroImaging Center- Université de Toulouse- Inserm- UPS- Inserm 1214, Research, Toulouse, France ; 3 Institut Claudius Regaud/Institut Universitaire du Cancer de Toulouse – Oncopôle, Biostatistics, Toulouse, France ; 4 Institut Claudius Regaud/Institut Universitaire du Cancer de Toulouse – Oncopôle, Engineering and Medical Physics, Toulouse, France ; 5 CHU Toulouse, Nuclear Medicine, Toulouse, France ; 6 CHU Toulouse, Neurosurgery, Toulouse, France ; 7 CHU Toulouse, Radiology, Toulouse, France ; 8 Centre Paul Strauss, Radiation oncology, Strasbourg, France ; 9 Centre Georges-François Leclerc, Radiation Oncology, Dijon, France ; 10 Centre Léon- Bérard, Radiation oncology, Lyon, France ; 11 Institut du Cancer de Montpellier, Radiation Oncology, Montpellier, France ; 12 Institut de Cancérologie de la Loire Lucien Neuwirth, Radiation oncology, Saint-Priest-en-Jarez, France ; 13 Inserm U1037- Centre de Recherches contre le Cancer de Toulouse, Research, Toulouse, France Gliomics is a research project with the goal of extracting radiomics from multimodal imaging data of multicenter SPECTRO GLIO prospective phase III trial for newly diagnosed glioblastoma (NCT01507506) [1]. The aim of this study is to identify combinations of features derived from pretreatment imaging modalities associated with 2-year survival. Material and Methods Methods Rigid registration was used to align T1-Gd, FLAIR WI, ADC and rCBV map (extracted from diffusion and perfusion MRI respectively) with their corresponding CT images. Imaging and RT structure were imported into IBEX (Imaging Biomarker Explorer software) [4] and automatically processed using in-house developed data analytics. Shape, first and second order statistics features [4] were extracted from GTV (Low dose) volumes. The primary endpoint was a binary outcome (with 2 years survival cut-off) and we excluded patients alive and followed less than 2 years. For each imaging modality (T1- Gd, FLAIR WI, ADC, rCBV and CT), a radiomics risk score was created based on the linear predictor given by a multivariable penalized-likelihood logistic regression model (Elastic Net (Zhou et al, 2005)). Prognostic performances of the radiomics scores were assessed by the area under the curve (AUC) and internal validations were performed using 10 folds cross validation. Multivariable regression analyses with clinical factors were carried out to test the independent prognostic ability of radiomics Eighty-two patients were analyzed in this study, 53.7 % were in Arm A, with Stupp protocol and 46.3% were in Arm B with an additional SIB of 72Gy/2.4Gy [1, 2, 3]. 46 deaths were observed within the first 24 months. Fifteen prognostic features from ADC, 11 from rCBV, 12 from FLAIR, 15 from T1-Gd and 13 from CT images were selected with Area under the Curve (AUC) values of 0.82, 0.84, 0.84, 0.87 and 0.79 respectively and medians of AUC from internal validation were 0.60, 0.67, 0.71, 0.61 and 0.56. For each modality a model combining the radiomics score and clinical data showed that all scores were significantly associated with 2 years survival (p<0.001, <0.001, <0.001, <0.001 and 0.001, respectively) Conclusion The analysis of the data from this prospective trial, showed that combination of anatomical imaging and functional MRI radiomics features are associated with 2-year survival after adjustment for clinical factors. The work is in progress to incorporate spectroscopic MRI data. [1] Laprie et al, BMC Cancer, 2019 Purpose or Objective Background scores. Results

with respect to time and accounts for cell diffusion (D) and cell proliferation (ρ). The diffusion coefficient assumes different values for white and grey matters (D W , D G ) and is zero everywhere else. An important parameter is the ratio D/ρ which determines the gradient of the tumour cell density function. There is no consensus in the literature on the values of D/ρ and D W /D G . The model for the tumour spread was implemented on a contrast enhanced T1 MRI brain scan. The image was segmented with respect to the diffusion coefficients. The model was solved in MATLAB using a finite-domain time-difference method. The values of D/ρ used ranged between 0.25 mm 2 and 25 mm 2 and the values of D W /D G were 5, 10, and 100, as found in the literature. The simulations were run until the volume encompassed by the tumour cell density visibility threshold isoline was reached (8000 cells/mm 3 ). The volumes encompassed by the isoline with 1000 cells/mm 3 (V 1000 ) were investigated as a function of D/ρ and D W /D G . Results Figure 1a shows isolines/contours of simulated tumours for various values of D/ρ and D W /D G = 10. Despite having roughly the same detectable volumes (encompassed by the red isoline), the actual tumour extension (the green and blue isolines) increases with increasing D/ρ. Figure 1b shows V 1000 (normalized) plotted as a function of D/ρ and D W /D G . There is a clear trend of an increase in tumour volume with an increase in D/ρ. The tumour invasiveness does not appear to be dependent on D W /D G .

Conclusion Mathematical modelling of the invasiveness and tumour extension of high-grade gliomas could be used for guiding the delineation of stereotactic radiosurgery targets with invasive character. The actual invasiveness and extension of the tumour depend on the cell diffusion and cell proliferation ratio, D/ρ, but do not appear to be influenced by D W /D G . PO-1559 Survival prediction in GBM using radiomics of multimodal imaging F. Tensaouti 1,2 , J. Bailleul 1,2 , E. Martin 3 , F. Desmoulin 2 , S. Ken 4 , J. Desrousseaux 1 , L. Vieillevigne 4 , J. Lotterie 2,5 , V. Lubrano 6 , I. Catalaa 7 , G. Noël 8 , G. Truc 9 , M. Sunyach 10 , M. Charissoux 11 , N. Magné 12 , T. Filleron 3 , P. Péran 2 , E. Cohen-Jonathan Moyal 1,13 , A. Laprie 1,2