ESTRO 2020 Abstract Book

S408 ESTRO 2020

response, only tumor grading was seen to predict response to nCRT significantly (p=0.034). Thereby, the rate of patients with moderately differentiated primary tumors was significantly higher in the subgroup of responders (67%) than in the subgroup of non-responders, where predominantly poorly differentiated tumors occurred (72.7%). In the cox-regression model the textural features Busyness (p<0.01; Exp. B=1.56), ContrastM3 (p<0.01; Exp. B=9.34) and SUVmax (p=0.019; Exp. B=0.93) were significantly associated with OS. We observed a significant prognosis deterioration for a value >1.76 for Busyness and a value >0.407 for Contrast respectively. Conclusion In our analysis, only tumor differentiation was significantly associated with response to nCRT. However, the 18F-FDG- PET/CT derived textural features Busyness, Contrast and SUVmax provide statistical value for predicting overall survival in ESCC patients undergoing nCRT and surgery and should therefore be further investigated in future studies. PH-0720 delta-radiomics based on MRI predicts response to concurrent chemoradiotherapy in esophageal cancer D. An 1 , B. Li 1 , Q. Cao 2 , W. Yin 3 1 Shandong Cancer Hospital and Institute- Shandong First Medical University and Shandong Academy of Medical Sciences, department of radiation oncology, Jinan, China ; 2 Southeast University, Laboratory of Image Science and In this prospective, observational study, ESCC patients treated with cCRT were enrolled and assigned to the training set and the validation set according to the beginning of radiotherapy in a 2:1 ratio. Each patient underwent sequential diffusion-weighted magnetic resonance imaging (DW-MRI) at pretreatment, 5th radiation completed and 10th radiation completed. Whole- tumor ADC values and 851 RFs were extracted within a volume of interest. Bland-Altman method and the minimum redundancy/maximum relevance algorithm were performed in pretreatment images on the training set for selecting RFs. The changes in those selected values were calculated as relative differences between each time point (delta-ADC values and delta-RFs). Afterwards, the identified RFs were used to train Support Vector Machine (SVM) classifiers for building a radiomics signature, to associate the treatment response and validated in the validation set. Radiomics signatures were developed from both RFs extracted from pretreatment images and delta- RFs calculated during 1st week, 2nd week and 2 weeks, respectively. Receiver operating characteristic (ROC) curve analysis was performed in the training and validation sets. Univariate analysis was performed to explore association between clinical characteristics, ADC values, radiomics signature and their changes with treatment response. The meaningful results were used to build treatment response-related multivariate logistic regress models. Finally, the performances of each model was assessed by the ROC curves. Technology, Nanjing, China ; 3 Inspur Electronic Information Industry Co.-Ltd, Department of Architecture Research, Jinan, China Purpose or Objective This study aimed to investigate the association between the radiomics features (RFs) extracted from the whole- tumor apparent diffusion coefficient (ADC) map during the early treatment course and response to concurrent chemoradiotherapy (cCRT) in patients with esophageal squamous cell carcinoma (ESCC). Material and Methods

Results 85 consecutive patients were enrolled. 9 patients were excluded because of poor-quality DW-MRI or interruption of radiotherapy. 54 patients experienced clinical partial response (sensitive group) and 22 with stable disease (resistant group). 51 patients were assigned to the training set and the subsequent 25 were allocated to the validation set. Primary tumor site was associated with treatment response in our study with a p -value of 0.030. None of ADC values and delta-ADC values were significantly correlated with treatment response. Furthermore, radiomics signature built from the training set which included 30 delta-RFs during the time range of 2 weeks showed a strong relationship to treatment response in the validation set ( p =0.032). The AUC value of the signature was 0.823 and 0.812 respectively ( p -value=0.942 in Delong test) in two sets within this time range. The combined model of tumor primary site and radiomics signature performed better and could successfully predict patients' response to cCRT. Conclusion The ADC map-based delta-radiomics during the early course of treatment can successfully predicts response to cCRT in patients with ESCC. PH-0721 Predicting overall survival in central NSCLC treated with SBRT: nomogram development and validation M. Duijm 1 , E. Oomen - de Hoop 1 , N. Van der Voort van Zyp 2 , H. Tekatli 3 , P. Van de Vaart 2 , M. Hoogeman 1 , S. Senan 3 , J. Nuyttens 1 1 Erasmus MC Cancer Institute, Radiation Oncology, Rotterdam, The Netherlands ; 2 Haaglanden MC, Department of Radiation Oncology, The Hague, The Netherlands ; 3 Amsterdam University Medical Center, Department of Radiation Oncology, Amsterdam, The Netherlands Purpose or Objective The treatment of centrally located NSCLC with SBRT is associated with higher probabilities of toxicity compared to peripheral lesions. A specific nomogram for this patient group could prevent that patients with low survival expectations are exposed to high risks of high-grade toxicities. Therefore, we developed and externally validated a nomogram to predict overall survival (OS)