ESTRO 38 Abstract book

S1058 ESTRO 38

treatment. Xgboost was applied to build a model to predict the correlation between parameters and treatment response. Results There were total of 42 radiomics features and 18 dosimetric parameters extracted for these 94 cases, together with sex, age and radiotherapy modalities were included in the modeling. According the results of data visualization using parallel coordinates, prescription dose, GTV, heart and cord related dosimetri parameters, as well GlobalMean X.33.1, Correlation, Coarseness, Skewness have a strong correlation with treatment response. The prediction accuracy and precise of training and validation data were 0.9, 0.86 and 0.54, 0.6, respectively. Remodeling after principal components analysis (PCA), the prediction accuracy and precise of training and validation data were 0.79, 0.76 and 0.75, 0.76, respectively. According to the Gain calculated for each parameter in the prediction model, image features of GlobalMean X.333.1, Coarseness, Skewness, GlobalStd were contributed most to the model. Dosimetric parameters of PTV HI, Cord Dmax, Prescription dose, Heart-Dmean and Heart-V50 also have a strong contribution to the model. The Area under curve (AUC) of receiver operating characteristics (ROCs) for dosimetric features alone and combined dosimetric features and dosimetric parameters were 0.6 and 0.75, respectively. Conclusion Models combining radiomics features and dosimetric parameters is capable of and outperform those with radiomics features alone in predicting the treatment response for EC patients underwent CRT. EP-1943 Intra-fractional respiration monitoring for patients undergoing lung SBRT E.P.S. Sande 1 , T.P. Hellebust 1 1 Oslo University Hospital, Dep. of Medical Physics, Oslo, Norway Purpose or Objective The Varian TrueBeam (TB) Respiratory Gating System (RGS) tracks respiratory motion via a reflector block placed on the patient’s chest. The system may interrupt the beam when detected motion exceeds pre-defined thresholds, due to sudden or gradual involuntary patient motion – such as coughing or baseline drift. The purpose of this study was to investigate the feasibility of using the TB RGS for intra-fractional monitoring during free breathing (FB) stereotactic lung radiotherapy (SBRT) to possibly increase the accuracy in dose delivery. Material and Methods Firstly, intra-fractional respiratory motion curves were recorded for 31 lung SBRT treatment sessions (9 patients). Thresholds were not applied, ensuring no RGS-related beam interruptions. Post-treatment analysis assessed the fraction of treatment sessions in which the beam would have been interrupted due to sudden motion or baseline drift, if thresholds had been set prior to treatment. Findings were used to determine suitable thresholds, given possible clinical implementation of RGS intra- fractional monitoring. Secondly, possible dosimetric disadvantages of interrupting the delivery of SBRT plans were investigated. A 6 MV FFF VMAT plan and a 10 MV FFF static plan (PTV D min 15 Gy/fraction) were delivered to a stationary ArcCheck phantom (Sun Nuclear Corp.). Respiratory motion of the reflector block was simulated during delivery using a CIRS (CIRS Inc.) respiratory motion phantom. Plans were delivered using three duty cycles (DC) – defined as the ratio of beam-on time to the total Electronic Poster: Physics track: Intra-fraction motion management

Image Informatics, Tokushima, Japan ; 3 University of Tokyo, Department of Neurosurgery, Tokyo, Japan Purpose or Objective Gliomas with 1p/19q codeletion are diagnosed as oligodendrogliomas and associated with better prognosis than their 1p/19q nondeleted counterparts. The purpose of this study was to investigate feasibility for predicting the 1p/19q codeletion status of the gliomas based on the radiogenomic analysis using T2-weighted magnetic resonance (MR) images (T2WIs). Material and Methods We analyzed the institutional database, for adult patients with a diagnosis of World Health Organization grade II and III gliomas from 1995 to 2017. Based on multiplex ligation- dependent probe amplification (MLPA) or microsatellite analysis, 38 patients underwent testing for 1p/19q codeletion (23 with codeleted and 15 with nondeleted). Pretreatment T2WIs of all patients were retrospectively evaluated. Gross tumor volumes (GTVs) were manually contoured and radiogenomic features (shape, size, histogram, and texture features) were extracted in the GTVs. Data were analyzed using L1-norm regularized logistic regression. A leave-one-out cross validation was employed in order to evaluate performance of a prediction model. Accuracy, sensitivity, specificity, and area under the receiver operating characteristics (ROC) curve (AUC) value were calculated as evaluation indices. This retrospective data evaluation was approved by the local ethics committee. Results Total number of the radiogenomic features was 476. Accuracy, sensitivity, specificity, and AUC values of the prediction models were 0.711, 0.697, 0.733, and 0.736, respectively. The 1p/19q codeletion could be moderately predicted using the proposed framework. Conclusion We developed a radiogenomics-based framework for non- invasively predicting the 1p/19q codeletion in grade II-III gliomas using the T2WIs. The proposed framework would be feasible to preoperatively predict genomic status of the gliomas using the MR images. Further validation studies at multi-centric level are mandatory and scheduled to confirm our findings. EP-1942 Combining radiomic and dosimetric parameters to predict chemoradation response of EC patients X. Jin 1 , C. Xie 1 1 The 1st Affiliated Hospital of Wenzhou Medical University, Radiation and Medical Oncology, Wenzhou, China Purpose or Objective The treatment response prediction capability of radiomics features extracted from PET and CT images had been extensive investigated. However, radiotherapy dosimetric parameters, which are the key parameters that affect the treatment response, were not included in the response studies. The purpose of this study is to investigate the prediction feasibility and accuracy of an integrated model with combined radiomic features and dosimetric parameters for esophageal cancer (EC) patients underwent concurrent chemoradiation (CRT) using machine learning techniques. Material and Methods Ninety-four patients underwent CRT for primary EC between Oct 2012 and Oct 2015 were enrolled in this study. The primary target volumes (GTVs) of these patients were delineated by two expert oncologists. DICOMS files were exported from TPS to an open infrastructure software platform: IBEX for preprocessing and feature extraction. Dosimetric parameters were extracted from DVH. The response of these patients to treatment was evaluated three months after the