ESTRO 2021 Abstract Book

S1527

ESTRO 2021

Belgium)) for non-clinical use. 165 features were extracted based on the prostatic index lesion. The features extracted were morphological features, local intensity features, first-order statistics, intensity histogram features, fractal features and the textural feature families of GLCM, GLRLM, GLSZM, GLZDM, NGTDM and NGLDM. Hypoxia was modelled as PIMO <3 (not hypoxic) and PIMO score >/= 3 (hypoxic). Highly correlated features (ρ > 0.9), features with near zero variance and linear combinations between features were first eliminated from further analysis. Multivariable logistic regression with Elastic Net regularization was utilised using 10 times repeated 10-fold cross-validation to select the optimal model hyperparameters, optimizing for area under the receiver operating characteristic curve (AUC). All features were standardized before modelling. The simplest candidate model (i.e., the model with the fewest non-zero coefficients) within one standard error of the best performing model was selected. Results The average sample performance based on the repeated cross validation yielded an area under the receiver operating characteristic curve (AUC) of 0.61 ±0.2 for radiomics features only using the BEST model. The most important features were Shape-based. Results for the ONESE model were similar with Shape-based features the most important and an AUC of 0.60±0.2. Conclusion This preliminary MRI-radiomics study of the index lesion in intermediate and high risk prostate cancer provides a basis for the hypothesis that radiomics may have a future role in the identification of hypoxia in prostate radiation therapy. Further study with enhanced texture analysis is warranted.