ESTRO 2020 Abstract book

S879 ESTRO 2020

compared to the reference predictive models. Their use is recommended in future studies in presence of scarcely represented data, such as rare side effect and high grade toxicities. PO-1536 RadiomiK: a phantom to test repeatability and reproducibility of CT-derived Radiomic Features S. Pallotta 1 , D. Cusumano 2 , A. Taddeucci 3 , M. Benelli 4 , R. Sulejmeni 1 , J. Lenkowicz 2 , S. Calusi 1 , L. Marrazzo 5 , C. Talamonti 1 , G. Belli 3 , M. De Spirito 2 , A. Barucci 6 , N. Zoppetti 6 1 Azienda Ospedaliero Universitaria Careggi - University of Florence, Department of Biomedical Experimental and Clinical Sciences "Mario Serio", Florence, Italy ; 2 Fondazione Policlinico Universitario A. Gemelli IRCCS, Medical Physics, Rome, Italy ; 3 AOU-Careggi, Health Physics, Florence, Italy ; 4 Hospital of Prato, Bioinformatics Unit, Prato, Italy ; 5 Azienda Ospedaliero Universitaria Careggi, Medical Physics Unit, Florence, Italy ; 6 CNR Florence Research Area-, Institute of Applied Physics "Nello Carrara", Florence, Italy Purpose or Objective Radiomics has been demonstrated to have a role in several clinical processes. Although radiomic approach is interesting it suffers from several noise sources, associated with image acquisition and post-processing (1). Understanding noise sources due to image acquisition permits to guide the creation of local prospective imaging protocols. Aim of this work is to present a phantom developed to test reproducibility and repeatability of radiomic features extracted from CT images. Material and Methods The phantom - RadiomiK (fig. 1a) comprises 23 elements imbedded in a layer of epoxy resin. The shape, materials and filling textures were chosen to produce a wide range of radiomics feature values capable to mimic those found in CT images of human being. 15 elements with cubic, cylindrical or conical shape, (1cm side, diameter and high) were fabricated with a 3D printer using PLA, FLEX and PETG. Honeycomb and gyroid textures, with air-filled holes and different filling percentage were employed. 3 cubes and 2 cylinders were made using slabs of Solid Water, Cortical Bone and Lung (Gammex-RMI, Middleton, WI, USA). Finally, 3 elements made with 25 mini-cubes (2mm side) were assembled creating three different patterns.

different minority class oversampling techniques and to investigate whether they can provide improved predictive power of urinary side-effects, following prostate cancer radiotherapy. Material and Methods A dataset of 254 prostate cancer patients treated with IMRT/IGRT was used and different urinary symptoms were analyzed. Depending on the symptom, datasets presented different levels of class imbalance. A large number of parameters, including clinical-/treatment-related and dosimetric extracted from the bladder's dose-volume histograms (DVH), were analyzed in a nested 5-fold cross- validation pipeline, using different popular classification methods. The performance of these classifiers (using the area under the ROC curve (AUC) and the F-measure) was first evaluated on the original dataset and compared with the logistic regression model (considered as reference). Then, we computed and compared the classifiers’ performance on datasets that had been balanced with different oversampling techniques. Results Table 1 shows the improvement brought by different oversampling techniques averaged over all the classifiers (cross-comparison). The alternative hypothesis was that the oversampling techniques 1 performed better than the oversampling techniques 2. Oversampling with Synthetic Minority Oversampling Technique (SMOTE) followed by Edited Nearest Neighbour algorithm (ENN) outperformed other oversampling techniques. Using SMOTE+ENN, increased the prediction performance (AUC) by 22% to 89%, compared to the original dataset without oversampling. The performance of the classifiers was symptom- dependent. For late retention, for example, the performance of the logistic regression model on the original dataset was AUC=0.57 whereas the best performance was obtained with PLS-DA classifier (AUC=0.65). On the oversampled datasets, the Regularized Discriminant Analysis (RDA) and the partial least square discriminant analysis (PLS-DA) scored higher, with average AUCs ranging from 0.67 to 0.71 and from 0.61 to 0.70, respectively. Table 2 shows the average performance of each classifier and for each dataset on the hold-out test sets, for predicting late retention.

Conclusion Synthetic oversampling approaches significantly increased the prediction performance of machine learning methods,