ESTRO 2020 Abstract book

S253 ESTRO 2020

PD-0429 CT radiomics differentiates levels of radiocurability in tumor subvolumes in head and neck cancer M. Bogowicz 1 , M. Pavic 1 , O. Riesterer 1 , T. Finazzi 1 , H. Garcia Schüler 1 , E. Holz-Sapra 1 , L. Rudofsky 1 , S. Glatz 1 , L. Basler 1 , M. Spaniol 1 , M. Hüllner 2 , M. Guckenberger 1 , S. Tanadini-Lang 1 1 University Hospital Zurich and University of Zurich, Radiation Oncology, Zurich, Switzerland ; 2 University Hospital Zurich and University of Zurich, Nuclear Medicine, Zurich, Switzerland Purpose or Objective Radiomics was proposed as prognostic biomarker in head and neck cancer (HNC); however it is not clear what treatment could be offered to the patients with worse prognosis based on existing radiomic signatures. Here we investigated if radiomics can distinguish between tumor subvolumes with different levels of radiocurability, to potentially increase the dose to these parts. Material and Methods We have retrospectively collected data from 28 HNC patients treated with definitive radiochemotherapy fulfilling the following inclusion criteria: contrast- enhanced CT imaging prior to treatment, local in-field tumor recurrence confirmed by biopsy and FDG-PET/CT imaging at the time of recurrence. The recurrent tumor was contoured on the PET/CT image (g_rec) and the contours were rigidly transferred to the planning CT. In a first step, two volumes were analyzed with radiomics: recurrence region (overlap between primary tumor (PT) and g_rec) and control region (PT minus recurrence). 162 intensity and texture features were extracted from the planning CT. Principal component analysis and multivariate logistic regression with backward selection were used to distinguish between control and recurrence regions. The final model was tested in a cohort of 12 patients from a phase II study selected based on the same inclusion criteria. In a second step, we tested if radiomics allows, not only for differentiation of the levels of radiocurability, but also potentially for their detection. To that end the radiomics analysis was performed on 8 subvolumes (Figure 1) of the primary tumor by splitting the tumor bounding box in 8 equal parts. If the subvolume consisted of more than 50% of voxels identified as g_rec it was classified as radioresistant subvolume. The model training and validation was performed analogous to the first step.

For the 8 subvolumes analysis, 3 features were selected in the training cohort. This model showed slightly inferior performance AUC=0.70 (95%CI: 0.53–0.86) in the validation cohort than the first model (Figure 2b). However, for all 4 patients, who presented at least 1 out of 8 radioresistant subvolumes (more than 50% voxels identified as g_rec), at least one subvolume was identified correctly by the model.

Conclusion We have shown for the first time that pretreatment radiomics can differentiate levels of radiosensitivity in HNC. This is a potential first step towards radiomics-based dose painting. PD-0430 Radiation induced dyspnea in lung cancer patients treated with stereotactic body radiation therapy G. Palma 1 , S. Monti 1 , M. Thor 2 , A. Rimner 3 , J. Deasy 2 , L. Cella 1 1 Institute of Biostructure and Bioimaging-CNR, National Research Council of Italy, Napoli, Italy ; 2 Memorial Sloan Kettering Cancer Center, Department of Medical Physics, New York, USA ; 3 Memorial Sloan Kettering Cancer Center, Department of of Radiation Oncology, New York, USA Purpose or Objective Thoracic radiation therapy (RT) is often associated to the risk of developing acute or late radiation induced lung damage (RILD) which may lead to dyspnea, lung fibrosis, and impaired quality of life [1]. A range of inconsistent clinical and dosimetric parameters have previously been shown to be predictive of RILD, in particular for hypofractionated RT. In this work, prognostic factors of radiation induced dyspnea were investigated in a cohort of patients treated for Non-Small-Cell Lung Cancer (NSCLC) with Stereotactic Body RT (SBRT). Material and Methods NSCLC patients with required clinical and dosimetric data elements treated with SBRT at a single institution from 2012 and 2015 were reviewed (N=106). The median prescription dose was 50 Gy (range: 40-54 Gy), delivered in a median of 4 fractions (range: 3-12). Median patient age was 75 years (range: 32–93 years). Dyspnea within 6 months after SBRT was scored according to CTCAE v.4.0 [2]. A voxelwise conversion of physical doses to EQD2Gy using α/β = 3 Gy was performed, and dose volume histograms (DVHs) for lungs and heart were extracted. Multivariable (MV) logistic regression analysis using bootstrapping was performed. Models were evaluated by Spearman Rs coefficient and the area under the receiver operator characteristic curve (ROC-AUC). Results Fifty-six patients (52.8%) out of 106 developed dyspnea of any grade within 6 months after SBRT (25/56 cases were of grade ≥2). Chronic obstructive pulmonary disease (COPD) and heart volume were the only clinical variables

Results The final model, to distinguish the recurrent vs controlled tumor subvolumes in the pretreatment imaging, comprised two features, indicating that recurrence regions are more heterogeneous than control regions prior to any treatment. The model showed good performance in the validation cohort AUC=0.88 (95%CI: 0.72–1.00), sensitivity=0.75, specificity=0.83, Figure 2a.