ESTRO 2022 - Abstract Book

S1594

Abstract book

ESTRO 2022

(MDADI) scores were collected at baseline and at time intervals up to two years post treatment. The degree of dysphagia was found not to be strongly affected by the treatment arm. This study seeks to determine anatomical locations associated with dysphagia as measured by the change in overall MDADI score relative to baseline. Approaches that do not rely on delineated structures were employed because various structures such as the oral mucosa, pharyngeal mucosa and constrictor muscles have all been identified as possible contributors to dysphagia. In this exploratory analysis we consider three possible approaches to this problem and explore the merits of each. Materials and Methods A total of 160 patients had all the required data available for analysis. Each patient CT scan was registered against a reference scan using deformable registration. Delineated contours and dose grids were subsequently propagated through the same deformation. Variation in structure position was used to quantify the registration uncertainty. We compared three approaches to image based outcomes analysis; a cluster detection approach, voxel-wise regression and a regression method based on Principal Component Analysis, Partial Least Squares (PLS) regression, controlled for tumour volume using the GTV volume. The correlation between each dose voxel and the change in toxicity was calculated alongside the t-statistic. Permutation testing was used to identify statistically significant clusters. In parallel; two regression models were tried. A voxel-wise Huber regression was used to fit the change in outcome to dose, tumour volume and the volume of the 60 Gy isodose at each point in space. The second method, a PLS regression, related to principal component regression, identifies the latent scores of the dependent and independent variables having maximum correlation, five components were used. GTV volume was used a surrogate measure of tumour burden. Results Regions of correlation were identified close to the salivary glands and at the level of the larynx. These regions were significant with a t-test. After applying permutation testing due to the very large number of comparisons, these regions, although anatomically plausible, were found to not be statistically significant (p > 0.05). Regression identified similar regions as important. Conclusion Correlations between delivered dose and dysphagia in anatomical regions that may be expected to correlate were identified. However, these were not statistically significant either using a correlation cluster based approach or a partial least squares regression approach. The partial least squares method has the advantage of being able to decompose the image into only a few components rather than having to fit thousands of voxels individually. D. Vuong 1 , K. Daetwyler 2 , M. Bogowicz 1 , P. Radojewski 3 , R. Meier 4 , M. Reyes 5,6 , N. Saltybaeva 1 , A. Depeursinge 7,8 , M. Bach 9 , M. Piccirelli 10 , M. Guckenberger 1 , S. Tanadini-Lang 1 , R. Wies 11 1 University Hospital Zurich and University of Zurich, Department of Radiation Oncology, Zurich, Switzerland; 2 University of Bern, Inselspital Bern University Hospital, Support Center for Advanced Neuroimaging (SCAN), University Institute of Diagnostic and Interventional Neuroradiology, Bern, Switzerland; 3 University of Bern, Inselspital Bern University Hospital, Support Center for Advanced Neuroimaging (SCAN), University Institute of Diagnostic and Interventional Neuroradiology Bern , Bern, Switzerland; 4 University of Bern, Inselspital Bern University Hospital, Support Center for Advanced Neuroimaging (SCAN), University Institute of Diagnostic and Interventional Neuroradiology, Bern, Switzerland; 5 Inselspital Bern University Hospital, Insel Data Science Center, Bern, Switzerland; 6 University of Bern, ARTORG Center for Biomedical Research, Bern, Switzerland; 7 University of Applied Sciences Western Switzerland (HES-SO), Institute of Information Systems, Sierre, Switzerland; 8 Lausanne University Hospital, Department of Nuclear Medicine and Molecular Imaging, Lausanne, Switzerland; 9 University Hospital Basel and University of Basel, Department of Research & Analytic Services, Basel, Switzerland; 10 University Hospital Zurich and University of Zurich, Department of Neuroradiology, Zurich, Switzerland; 11 University of Bern, Inselspital Bern University Hospital, Support Center for Advanced Neuroimaging (SCAN), University Institute of Diagnostic and Interventional Neuroradiology, , Bern, Switzerland Purpose or Objective Radiomics is a method that extracts a large number of quantitative features from medical images. This virtual profile builds a tumor phenotype and might help in patient stratification. Delineation of the lesion’s volume of interests (VOI) is an essential step that is often performed manually. Deep learning (DL) segmentation methods have shown high accuracy in performing this time-consuming task prone to inter-observer variability (IOV). We studied the robustness of MR radiomics against six DL segmentation methods in glioblastoma patients. Materials and Methods In total, 30 glioblastoma patients from three centers were included. Four MR sequences (T1, T1ce, T2, T2 FLAIR) were collected retrospectively for each patient. Six DL segmentation methods delineated four VOIs: necrosis and central non- enhancing tumor (CNEH), peripheral non-enhancing components (often referred to as edema, PNEH), contrast enhancing tumor (CET) and the combination thereof (Combined). Images were resized to 2 mm 3 voxels using trilinear interpolation and normalized using histogram matching to a reference patient. IOV among DL segmentation methods was assessed with Dice and Hausdorff distance. Intensity (n=17), texture (n=137), and wavelets (n=1232) based radiomic features were calculated using an in-house developed software implementation (Z-Rad, Python v3.7). Features were considered robust when the intra-class correlation coefficient > 0.9. PO-1787 Impact of deep learning segmentation methods on the robustness of MR glioblastoma radiomics

Results