Abstract Book

S36

ESTRO 37

OC-0075 A MRI radiomic signature for predicting brachytherapy outcomes in locally advanced cervical cancer S. Reuzé 1,2,3 , A. Alexis 1,2 , C. Chargari 2,3,4,5,6 , S. Bockel 4 , K. Berthelot 4 , A. Escande 4 , I. Dumas 1 , F. Orlhac 7 , C. Haie- Meder 4 , E. Deutsch 2,3,4 , C. Robert 1,2,3 1 Gustave Roussy, Radiotherapy Department- Medical Physics Unit, Villejuif, France 2 Paris-Saclay University, Faculty of Medicine, Le Kremlin- Bicêtre, France 3 INSERM U1030, Molecular Radiotherapy, Villejuif, France 4 Gustave Roussy, Radiotherapy Department, Villejuif, France 5 Institut de Recherche Biomédicale des Armées, Paris, France 6 French Military Health Services Academy, Ecole du Val- de-Grâce, Paris, France 7 CEA-SHFJ, IMIV, Orsay, France cervical cancer (LACC) consists of concomitant chemoradiation followed by brachytherapy (BT). The recent implementation of image-guided adaptive brachytherapy (IGABT) has shown a significant improvement in local control rates while limiting toxicity. Conventional prognostic factors for local control have been identified such as the volume of the high-risk clinical target volume (HR-CTV) at time of BT or the overall treatment time. However, with the advent of IGABT, tumor texture analysis could be a powerful tool providing additional quantitative information to refine the prediction of outcome. The aim of this study was to identify a radiomic signature of LACC relapse based on per-BT fast-spin echo T2 (FSE T2) MRI realized with vaginal mold, after a 45 Gy radiotherapy. A methodological study was first carried out to identify the most robust and informative textural features. Material and Methods Two groups of patients with LACC treated with pulsed- dose-rate IGABT after initial concomitant chemoradiation were retrospectively included (TS: N=60, training; VS: N=40, validation). FSE T2 MRI were used for BT planning and were acquired on the same device with similar acquisition parameters. Using LIFEx freeware, we extracted 30 textural features from the HR-CTV delineated on per-BT MRI. To this end, voxel intensities were first resampled (absolute method: fixed bin size in [0-6000]). The relationships between features, their correlation to tumor volume and their robustness with respect to the intensity discretization step were studied. The ability of features to predict relapse was afterwards assessed through univariate and multivariate statistical analysis. Results Discretization affects the feature values and resampling using at least 30 gray levels (bins) should be used for texture feature calculation for FSE T2 MRI. LGZE and LZLGE were highly sensitive to variation of discretization step. Using 40 bins, 11 groups of highly correlated features were identified (Spearman correlation coefficient |ρ|>0.75) and 4 features were shown to be highly correlated to the volume of HR-CTV (Fig.1). Therefore, only 9 groups including 24 robust features were considered for the clinical analysis. In univariate analysis, five features were statistically significant discriminators of relapsing patients from non-relapsing patients (TS: Wilcoxon test, p<0.05). A 5-features signature predicting relapse was identified and validated (TS: AUC=0.88, p<0.00001, VS: AUC=0.78, p<0.01, Fig.2) and performed better than the five features determined from univariate analysis and the volume of HR-CTV (TS, Delong’s test, p<0.05). Purpose or Objective The standard treatment of locally advanced

Conclusion This study showed promising results concerning the use of per-BT MRI textural features for predicting relapse in LACC. A methodology was proposed and a powerful signature of recurrence was established on routine acquisitions. This method has to be validated in an external cohort before considering a clinical use for treatment personalization. OC-0076 MR-guided vs CT-guided brachytherapy more effective and less costly in locally advanced cervical cancer J. Skliarenko 1 , D. D'Souza 2 , J. Perdrizet 3 , M. Ang 4 , L. Barbera 5 , E. Gutierrez 4 , A. Ravi 6 , K. Tanderup 7 , P. Warde 4 , K. Chan 4 , W. Isaranuwatchai 8 , M. Milosevic 1 1 Princess Margaret Cancer Centre, Radiation Medicine Program, Toronto, Canada 2 London Health Sciences Centre, Radiation Oncology, London, Canada 3 St. Michael’s Hospital Centre for Excellence in Economic Analysis Research, Health Economics, Toronto, Canada