ESTRO 2020 Abstract Book

S294 ESTRO 2020

Conclusion Geometric distortion was assessed on 11 scanners used for RT across the UK and for all scanners distortion increased with distance to isocentre. Variance of geometric distortions between scanners was found to be low (<0.8) and increased with distance to isocentre. The vast majority of scanners showed mean distortions of > 2 mm at a distance ≥ 250 mm from the isocentre, but these values have limited clinical relevance. This assessment demonstrates feasibility of the technique to be repeated in a UK wide audit of all MRI scanners used clinically for RT. PH-0532 Standardization of brain MRI across machines and protocols: bridging the gap for MRI-based radiomics A. Carré 1,2,3 , G. Klausner 1,3 , M. Edjlali 4,5,6 , M. Lerousseau 1,2,7 , J. Briend-Diop 1,2 , R. Sun 1,2,3,7 , S. Ammari 8 , S. Reuzé 1,2,3 , E. Alvarez-Andres 1,2,9 , T. Estienne 1,2,7 , S. Niyoteka 1,2 , E. Battistella 1,2,7 , M. Vakalopoulou 2,7 , F. Dhermain 3 , N. Paragios 9 , E. Deutsch 1,3 , C. Oppenheim 4,5,6 , J. Pallud 5,6,10 , C. Robert 1,2,3 1 Molecular Radiotherapy - Gustave Roussy - Inserm, Paris-Sud University - Paris-Saclay University, 94805 - Villejuif, France ; 2 Department of Medical Physics - Gustave Roussy, Paris-Saclay University, 94805 - Villejuif, France ; 3 Department of Radiotherapy - Gustave Roussy, Paris-Saclay University, 94805 - Villejuif, France ; 4 Department of Neuroradiology, Sainte-Anne Hospital, 75014 - Paris, France ; 5 Paris Descartes University, Sorbonne Paris Cité, Paris, France ; 6 UMR 1266 INSERM - Ima-Brain, Institute of Psychiatry and Neurosciences of Paris, Paris, France ; 7 Mathématiques et Informatique pour la Complexité et les Systèmes, CentraleSupélec - Paris-Saclay University, 91190 - Gif-sur-Yvette, France ; 8 Department of Radiology - Gustave Roussy, Paris-Saclay University, 94805 - Villejuif, France ; 9 TheraPanacea, Radiotherapy, Paris, France ; 10 Department of Neurosurgery, Sainte-Anne Hospital, 75014 - Paris, France Purpose or Objective Magnetic Resonance Images (MRI) properties are highly dependent on acquisition and reconstruction parameters. Today, there is no consensus about the optimal pre- processing of MR images which is critical to the generalizability of published radiomic models. Radiomics consists in the extraction of a wide variety of quantitative image-based features to provide decision support. This study aims at assessing the impact of three standardization methods namely Nyul (Nyul et al. 1999), WhiteStripe (Shinohara et al., 2014) and Z-Score (Zero mean and unit variance) used in combination with two methods for intensity discretization (Fixed Bin Size and Fixed Bin Number) on first and second-order radiomic features from brain MRI and at proposing a unified methodology for Two independent MRI datasets were used. The first one included 20 institutional patients with WHO grades II and III gliomas who underwent T1w-gd and T2w-flair sequences on two different MR devices (1.5T and 3.0T) with a one- month delay. The Jensen Shannon Divergence was used to compare pairs of intensity histograms before and after normalization. Stability of first-order and second-order features across the two acquisitions was analyzed using the concordance correlation coefficient and the intra-class correlation coefficient. The second dataset was extracted from the public TCIA database and included 108 patients with low-grade glioma and 135 patients with future radiomic studies. Material and Methods

glioblastomas. Impact of normalization and discretization methods was evaluated based on a tumor grade classification task (accuracy measurement) using five machine learning algorithms (random forest, naïve bayes, logistic regression, support vector machine and neural networks Multi-layer Perception) (Figure 1).

Results Normalization highly improves the robustness of first-order features and the performance of subsequent models on classification. For the T1w-gd sequence, the accuracy performance of the tumor grade classification model based on first-order features has increased from 0.68 to 0.83, 0.80, 0.82 respectively for the different standardization methods Nyul, WhiteStripe, Z-Score. Relative discretization makes unnecessary the use of normalization for textural features. The bin number has no major impact on classification performances when ranging from 16 to 128. There is a maximum variation of 10% in the percentages of robust features and a maximum variation of 9% in accuracy for bincount discretization in T2w-flair sequence for tumor grade classification based on texture (Figure 2).

Conclusion A standardized pre-processing pipeline is proposed. For models based on first and second order features, we recommend to normalize images with the Z-Score method and to prefer an absolute discretization. For second-order features only, relative discretization can be used without prior normalization. In both cases, a number of bins between 16 and 128 is recommended. This study may paves the way for the multicentric development and validation of MR-based radiomic biomarkers.