ESTRO 2020 Abstract book

S887 ESTRO 2020

assessed every 3 months after the completion of radiation course, and scored in a prospective database. A total of 21 shape, 228 radiomic and 228 dosimetric textural features were calculated in the PTV and breast volume (954 in total). The clinical variables fractionation, follow up and age were also considered. Imbalance of the dataset was minimized by data augmentation using SMOTE technique. Sequential feature selection was used to identify a subset of features that best predict the data and remove redundant or not significant predictors. A support vector machine classifier was trained on the patient dataset for prediction of occurrence of LSF during follow-up. The predictive power of the model was assessed using sensitivity (probability that test is positive on patients with LSB) and specificity in five-fold cross validation on the augmented dataset. Results 41 patients experienced LSB after average 5 years of follow up. Nine features (3 radiomic from breast and 3 from PTV, 1 shape of PTV, 1 dose from the breast and 1 from PTV). The model with nine radiomic and dosiomic features and number of fractions scored sensitivity 97.6% (95%CI 95.1-99.4), specificity 81.0% (95%CI 76.3-84.8%) and accuracy 93.2% (95%CI 91.0-95.1 %) in the cross-validation. Conclusion These findings show that radiomic and dose textural variables extracted from the breast and PTV volumes in the CT can predict LSF and may help to better select patients candidate to exclusive partial breast irradiation. PO-1548 Can the tumorlet model explain why 2 x 13.5Gy is not the equivalent of 1 x 19.5Gy? A. Urdaneta 1 , D. Todor 1 1 VCU Health, Radiation Oncology, Richmond, USA Purpose or Objective Fractionated high dose rate prostate brachytherapy (HDRPB) achieves an excellent biochemical control (BC) comparable if not better to external beam radiation therapy when adjusted for risk stratification with lower rates of chronic toxicity. However single fraction (fx) HDRPB has shown unsatisfactory 5 year BC with low nominal doses being considered the main culprit, despite an apparent BED equivalence with multiple fraction regimens. Using a tumorlet (TL) model, we are showing that what seems inadequate is the biological effect to disease foci, rather than the peripheral dose. Material and Methods We developed and validated a comprehensive software platform enabling in-silico growth of realistic prostate tumor-like volumes of random shape and morphology and controlled volume and position. Published probabilities of disease occurrence in various prostate zonal anatomies are taken into account for TL generation. 500 TL for each of the 9 discrete volumes, ranging from 0.1 to 10 cm 3 are generated. Each time a TL is grown, it is overlapped with the dose distribution of a specific patient and radiobiological integral parameters (EUBED, gBEUD) are computed. Data from 25 international radiotherapy outcome datasets were used to generate dose-response curves, which are then used in our model to convert integral metric parameters into a probability of cure. An US image based case already treated with 2 fx of 13.5Gy was re-planed for peripheral dose of 19.5, 21 and 24Gy, which under the LQ/BED orthodoxy for an α/β=3Gy or lower, are at least equivalent if not higher to 27Gy in 2Fx. The geometry of the implant is the same and the plan was created using inverse optimization. Each batch of TL of a given volume but diverse spatial position was averaged and converted in a probability of cure. Results The initial treatment was delivered in two implants, each receiving a peripheral prescription of 13.5Gy, with the same number of needles but neither their position nor

Results Two-hundred-forty-eight radiomic features are extracted from the MRI image for each ROIs. For train dataset, AUC value for liver metastases is 0.64 [0.62, 0.66] ； AUC value for 0.5cm-wide ring is 0.98 [0.98 0.99]; 1cm-wide ring is 0.83 [0.81 0.842]. For validation dataset, AUC value for liver metastases is 0.58 [0.53 0.63] ； AUC value for 0.5cm- wide ring is 0.56 [0.51 0.61]; 1cm-wide ring is 0.69 [0.64 0.73](figure 2).

Conclusion The use of liver metastasis MRI-derived radiomic features to predict KRAS mutation is feasible in colorectal cancer. The surrounding area of liver metastases have something to do with KRAS mutation state. PO-1547 Prediction of late subcutaneous fibrosis after partial breast irradiation by radiomics and dosiomics M. Avanzo 1 , G. Pirrone 1 , L. Vinante 2 , A. Caroli 2 , S. Massarut 3 , M. Mileto 3 , J. Stancanello 4 , M. Urbani 5 , A. Drigo 1 , M. Trovo 6 , I. El Naqa 7 , A. De Paoli 2 , G. Sartor 1 1 Centro di Riferimento Oncologico di Aviano CRO IRCCS, Medical Physics Department, Aviano, Italy ; 2 Centro di Riferimento Oncologico di Aviano CRO IRCCS, Radiation Oncology Department, Aviano, Italy ; 3 Centro di Riferimento Oncologico di Aviano CRO IRCCS, Breast Surgery Department, Aviano, Italy ; 4 Guerbet SA, Guerbet SA, Villepinte, France ; 5 Centro di Riferimento Oncologico di Aviano CRO IRCCS, Radiology Department, Aviano, Italy ; 6 Udine General Hospital, Department of Radiation Oncology, Udine, Italy ; 7 University of Michigan, Department of Radiation Oncology, Ann Harbor, USA Purpose or Objective to predict the occurrence of late subcutaneous fibrosis (LSF) after partial breast irradiation for early-stage invasive breast carcinoma eligible for partial breast radiotherapy (PBI) by use of computed tomography (CT)- based radiomic and dose (“dosiomic”) features from breast and planning target volume (PTV). Material and Methods 167 patients treated with breast conservative surgery for an early stage ductal carcinoma underwent external PBI following an hypo-fractionation protocol consisting in 40Gy/10 (73) , 35/7 (60) and 28/8 (34) fractions. Physicians evaluated toxicity at regular intervals by the RTOG scale and by the CTAE (version 3.0). LSB was