Abstract Book

S1159

ESTRO 37

Results

records and database. The phenotype of the primary tumor was quantified on pre-treatment CT scans by 435 radiomics features, which was extracted from segmented volumes of each tumor. 62 of these, which were considered stable and independent features, were included in this analysis. Univariate and multivariate analysis was performed to evaluate radiomics performance using the concordance index (CI) Results High level of NSE （ OR=1.945 ， P=0.014 ） , differentiation （ OR=1.654 ， p<0.001 ） and smoking status (OR=1.452 ， p=0.032 ） were independent clinical and pathological factors for brain metastasis prediction. Statistically significant differences were found in 21 radiomics features between brain metastasis and non- brain metastasis groups in univariate analysis (CI=0.642, p=0.031). A multiple logistic regression model illustrated that adding radiomics features to a clinical model resulted in a significant improvement of predicting power, the AUC increased from 0.682 to 0.711 (P=0.023). Conclusion This study shown that radiomics features could capture useful information about tumor phenotype, and the model we discovered can be used to predict predictive brain metastasis through pre-treatment CT in SCLC patients. In our future work we focus on inner and external validations for this model. D. Vuong 1 , M. Bogowicz 1 , M. Huellner 2 , P. Veit-Haibach 2 , N. Andratschke 1 , J. Unkelbach 1 , M. Guckenberger 1 , S. Tanadini-Lang 1 1 University Hospital Zurich, Radiation Oncology, Zürich, Switzerland 2 University Hospital Zurich, Nuclear Medicine, Zürich, Switzerland Purpose or Objective Radiomics is a promising tool for identification of new prognostic biomarkers. However, image reconstruction settings and test-retest variability may influence the absolute values of radiomic features. Unstable radiomic features cannot be used as reliable biomarkers. PET/MR is becoming more available and often replaces PET/CT. The aim of this study was to quantify to what extend [18F]-FDG PET/CT radiomics models can be transferred to PET/MR. Material and Methods Nine patients with non-small cell lung cancer underwent first an [18F]-FDG PET/MR scan (SIGNA PET/MR, GE Healthcare, Waukesha) followed by an [18F]-FDG PET/CT scan (Discovery 690, GE Healthcare) with a delay time of 38 min ± 5 min. All patients had one single FDG injection for both scans. The primary tumors were segmented independently on the PET dataset from PET/CT and PET/MR with two semi-automated methods (gradient- based and threshold-based). The resolution of the scans was 2.73x2.73x3.27 mm for the PET/CT and 2.34x2.34x2.78 mm for the PET/MR. Resampling was performed to 3.27 mm. In total, 1355 radiomic features were calculated, i.e. shape (n=18), intensity (n=17), texture (n=136) and wavelets (n=1184). The intra-class correlation coefficient (ICC(3,2)) was calculated to compare the radiomic features in both imaging modalities. An ICC larger than 0.9 was considered stable among both types of PET scans. An average linkage hierarchical clustering was performed to identify classes of stable and uncorrelated features with a cut-off value of 0.7. EP-2108 Robustness study on radiomic features in [18F]-FDG PET/CT and [18F]-FDG PET/MR

The median of the relative volume difference of the primary tumor segmented on PET/CT and PET/MR was 4.8% (range 0.4% to 39.9%) for the gradient-based method and 18.0% (range 0.7% to 71.2%) for the threshold-based method. The fraction of features stable between the scans is shown in Table 1. A larger number of radiomic features was stable when segmentation was performed using the gradient-based compared to the threshold- based method, which is in agreement with the improved reproducibility of tumor volume using gradient-based method. More than 75% of shape and intensity features yielded an ICC>0.9 between the scans for both segmentation methods. However, only 51.5% of the texture and 34.2% of the wavelet features reached this criterion (for gradient-based method and even less in threshold-based method). In the wavelet features analysis, more features were robust in the smoothed images (low-pass filtering) in comparison to images with emphasized heterogeneity (high-pass filtering) (Fig.1). Hierarchical clustering revealed 11 uncorrelated groups of stable features. Conclusion Shape and intensity radiomic features were robust when comparing the two types of [18F]-FDG PET scans (PET/CT and PET/MR). In contrast, substantially worse stability was observed for texture and wavelet features, which needs to be considered for their use in prognostic modelling. This instability can be caused by different factors, such as the different attenuation correction methods or test-retest variability. EP-2109 Functional diffusion maps to assess treatment response in head and neck tumors B. Peltenburg 1 , T. Schakel 1 , C.H.J. Terhaard 1 , R. Bree 2 , M. Philippens 1 1 UMC Utrecht, Radiation Oncology, Utrecht, The Netherlands 2 UMC Utrecht, Head and Neck Surgical Oncology, Utrecht, The Netherlands Purpose or Objective The apparent diffusion coefficient (ADC) determined by diffusion weighted magnetic resonance imaging (DW-MRI) is a surrogate measurement of cellularity and stromal component of a tumor. ADC changes during treatment