ESTRO 37 Abstract book

S295

ESTRO 37

incontinence, the AUC improved from 0.69 to 0.70 (+DVH) and 0.73 (+TA). Conclusion 3D dosimetric texture analysis features in predictive modelling of GI and GU toxicity rates in prostate cancer radiotherapy improved prediction performance.

model’s predictive accuracy. 1-year survival was used as the outcome. Results Twenty-four out of 78 (~31%) patients had a survival time over one year. The AUC of the validated model was 0.67 (95% CI: 0.54-0.80) (figure 1). The accuracy of the model was 0.69 (95% CI: 0.58-0.80).

Figure 1: A: View on the 3D rectum dose distribution of one of the included patients, B-F: TA histograms and matrices derived from this dose distribution. B: grey level frequency histogram, C: grey level co-occurence matrix (GLCM), D: grey level run length matrix (GLRLM), E: grey level size zone matrix (GLSZM), F: neighbourhood grey tone difference matrix (NGTDM). In C and D, intensities are displayed in logarithmic scale PV-0566 Survival prediction with radiomics of patients with brain metastases of non-small cell lung cancer S. Keek 1 , A. Jochems 1 , J. Zindler 2 , P. Lambin 1 1 University of Maastricht GROW Research Institute, Faculty of Health- Medicine and Life Sciences, Maastricht, The Netherlands 2 Maastro clinic, Oncology MAASTRO, Maastricht, The Netherlands Purpose or Objective Clinical nomograms allow for evaluation of individualized probabilities of clinical endpoints. Nomograms enable individualized prediction of long-term survival of patients treated with stereo-tactic radiosurgery (SRS) for brain metastases (BM) of non-small cell lung cancer (NSCLC). These nomograms consist of clinical factors only and do not utilize the information captured in planning CT- images made routinely when performing SRS. A prognostic radiomics signature is available for lung and head- and neck-cancer which allows prediction of survival in non-metastatic cancer patients. In this work, we retroactively investigate if the signature developed on primary lung tumors is able to predict 1-year survival in patients with BM of NSCLCs. Material and Methods 78 patients treated with a maximum of 3 BMs of NSCLC treated with SRS at clinic x were selected, 50% male and with a mean age of 62. Contrast-enhanced CT images used to plan SRS were collected. The data set was used to validate an existing radiomics signature. The largest BM was delineated, and the resulting volumes of interest were used to extract the signature required for the prediction model, consisting of the following 4 radiomics features: first order statistics energy, shape compactness, grey level non-uniformity and wavelet decomposed grey level non-uniformity. The prediction model used was a cox-proportional hazard model predictive of long-term survival, trained on NSCLC patients(N=422) and validated externally on lung and head and neck data sets. The performance of the model was determined using the area under the curve (AUC) of the receiver operator characteristic and through the

Conclusion The previously developed radiomics signature model built on primary non-metastatic lung cancer, can surprisingly predict 1-year survival of BM of NSCLC above the chance level. While the existing model shows decent predictive performance, a model specifically tailored to BMs from primary lung tumors could further improve predictive performance. A combination of radiomics image features, clinical and semantic features may be optimal for prediction of long-term survival, and warrants future investigation. PV-0567 Minibeam radiation therapy in a commercial irradiator spares normal rat brain Y. Prezado 1 , M. Dos Santos 1 , W. Gonzalez 1 , G. Jouvion 2 , C. Guardiola 1 , S. Heinrich 3 , D. Labiod 3 , M. Juchaux 1 , L. Jourdain 4 , C. Sebrie 4 , F. Pouzoulet 3 1 Centre National de la Recherche Scientifique, Imagerie et Modélisation en Neurobiologie et Cancérologie, Orsay, France 2 Institut Pasteur, Anatomopathology and animal models, paris, France 3 Institut Curie, Experimental Radiotherapy Platform, Orsay, France 4 University Paris Sud, IR4M, Orsay, France Purpose or Objective Minibeam radiation therapy (MBRT) is an innovative synchrotron radiotherapy technique able to shift the normal tissue complication probability curves to very high doses [1]. MBRT seem to involve different biological mechanisms (not well understood yet) different from those in standard RT. However, its exploration was hindered due to the limited and expensive beamtime at synchrotrons. The aim of this work was to evaluate the feasibility of the implementation of MBRT into cost- effective equipment. This would permit the realization of systematic radiobiological studies to evaluate the tumour control effectiveness for various tumour sites as well as to unravel the distinct biological mechanisms involved. Material and Methods A series of modifications of a small animal irradiator (Small Animal Radiation Research Platform-XSTRAHL Ltd.) were performed to make it suitable for MBRT experiments. In particular, an adapted collimator was designed by means of Monte Carlo simulations (Geant4). Peak to valley dose ratio (PVDR) values and full width half at maximum (FWHM) similar to those obtained at the European synchrotron radiation facility (ESRF) [2] were used as figure of merit. As a proof of concept, two groups of animals were irradiated: a first group (series 1) received conventional (broad beam) irradiations, the