technique using radiomics features and a visual evaluation method for predicting the LC of brain metastases (BMs) treated by stereotactic radi- osurgery (SRS). Materials/Methods: Data of 115 tumors of 45 patients with BMs, who underwent GKRS at the University of Minnesota from 2006 to 2012 were analyzed. The treatment outcomes of these tumors were known prior to the study. The data were split into two sets: 80 tumors for training of models and 35 tumors for model testing. The LC was classified into two groups by using the RECIST definition. The first group (group I) was a Complete Response (CR) + Partial Response (PR). The second group (group II) was Stable Disease (SD) + Progressive Disease (PD). A total number of 705 radiomics features were extracted from radiotherapy planning MRI scans (Gd-contrast T1-weighted) by using an open infrastructure software pro- gram. An experienced radiation oncologist classified the tumors into three types of patterns (homogeneous, heterogeneous, or ring) by visually inspecting the MR images. Based on the well-accepted knowledge, we assigned the predicted LCs of the homogeneous tumors and tumors with heterogeneous or ring enhancement to the success group (group I) and the failure group (group II), respectively. For the machine learning method, a prediction model was constructed by using a convolution neural network (NN) classifier coupled with a least absolute shrinkage and selection operator (LASSO). We trained the model (NN-LASSO) by using the training data set. The test data set was used to evaluate the prediction capability of the NN-LASSO model and the visual evaluation technique. We calculated the precision, accuracy, sensitivity by generating confusion matrices. Furthermore, the area under the receiver operating characteristic curve (AUC) was obtained for the NN-LASSO method. Results: By the LASSO analysis of the training data, we found seven radiomics features (45-7ClusterShade, 225-7ClusterShade, 45-7Informa- tionMeasureCorr1, 225-7InformationMeasureCorr1, 90-4Informa- tionMeasureCorr2, 225-7Energy and 315-5Energy) useful for the classification. The accuracy, specificity and sensitivity of the visual eval- uation method were 44.0%, 54.3% and 23.7%. On the other hand, the accuracy, specificity and sensitivity of the NN-LASSO model were 82.8%, 80.0%, 90.2%, and the AUC was 0.78. Conclusion: The NN-LASSO model using the radiomics features of tumor image was more accurate than the visual evaluation method using the image heterogeneity information in predicting the LC of brain metastases after GKRS. Because of the good prediction ability of the method, the method can be used to assist physicians to have more accurate expectation of the treatment outcome than the traditional method. Author Disclosure: D. Kawahara: None. X. Tang: None. C.K. Lee: None. Y. Nagata: None. Y. Watanabe: Honoraria; ASTRO. Travel Expenses; ASTRO. 1115 NTCP Models for Permanent Radiation Induced Alopecia in Brain Tumor Patients Treated with Scanned Proton Beams G. Palma, 1 A. Taffelli, 2 F. Fellin, 3 V. D’Avino, 1 D. Scartoni, 4 F. Tommasino, 5 E. Scifoni, 5 M. Durante, 6 M. Amichetti, 3 M. Schwarz, 4 D. Amelio, 3 and L. Cella 7 ; 1 National Research Council, Institute of Biostructures and Bioimaging, Napoli, Italy, 2 Universita`degli Studi di Trento, Trento, Italy, 3 Proton Therapy Center, Azienda Provinciale per i Servizi Sanitari, Trento, Italy, 4 Proton Therapy Center, APSS, Trento, Italy, 5 Istituto Nazionale di Fisica Nucleare (INFN-TIFPA), Trento, Italy, 6 GSI Helmholtz Centre for Heavy Ion Research, Department of Biophysics, Darmstadt, Germany, 7 National Research Council, Institute of Biostructures and Bioimaging, Napoli, Campania, Italy Purpose/Objective(s): Radiation therapy (RT) of brain tumors may induce radiation-induced alopecia (RIA) impacting on treatment cosmetic outcome and patient quality of life. Aim of this study was to develop normal tissue complication probability (NTCP) models for permanent RIA in brain tumor patients treated with proton therapy (PT). Materials/Methods: We analyzed 103 adult brain tumor patients (median age 56 yr; range 27-84) treated at a single institution and undergoing scanning beam PT (median dose 54 GyRBE; range 36-72) for permanent (>12 months after RT) RIA of grade >Z 2 according to CTCAE v.4 grading system. In order to accurately evaluate the dose to the scalp hair follicles (on average 4 mm deep from the surface of skin), the scalp structure was automatically defined by an ad hoc segmentation algorithm. For each patient, the dose-surface histogram (DSH) of the scalp was then extracted and used for Lyman-Kutcher-Burman (LKB) modelling. DSH metrics and non-dosimetric variables (gender, surgery, chemotherapy and previous irradiation) were moreover included in a multivariable logistic regression NTCP model. Model performances were evaluated by the leave-one-out cross-validated (CV) area under the receiver operator char- acteristic curve (ROC-AUC) and calibration plot parameters (slope and intercept of the linear regression). Results: Permanent RIA was observed in 20 of 103 (19%) patients. A large spectrum of DSH parameters showed a strong correlation with RIA status. Both LKB and logistic models achieved high predictive performances, with CV ROC-AUCs of 0.86 and 0.87, respectively (Table). The LKB model showed a weak dose-surface effect (nZ0.09), with a tolerance dose value of 44 GyRBE. Consistently, the multivariable modeling selected the scalp near maximum dose (D 2% - cut-off value of 48 GyRBE) as RIA predictor, while none of the significantly correlated clinical variables was included in the final logistic model. Conclusion: Following different modelling approaches, we derived two NTCP models for permanent RIA after PT. They achieved very good prediction performances only based on dose metrics of the scalp. The obtained findings provide a coherent picture of the analyzed morbidity and can be easily exploited for treatment planning optimization in order to improve patient quality of life. Author Disclosure: G. Palma: None. A. Taffelli: None. F. Fellin: None. V. D’Avino: None. D. Scartoni: None. F. Tommasino: None. E. Scifoni: None. M. Durante: None. M. Amichetti: None. M. Schwarz: None. D. Amelio: None. L. Cella: None. 1116 Assessment of Dosimetric Associations with Patient- Reported Toxicities Using Machine Learning Methods X. Pan, 1,2 J. Huang, 1 and X. Qi 3 ; 1 School of Computer Science and Technology, Xi’an University of Posts & Telecommunications, Xi’an, China, 2 Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing, Xi’an University of Posts and Telecommunications, Xi’an, Abstract 1115; Table 1 Permanent RIA Logistic Model variables Coefficient p LKB parameters Estimate p D 2% (GyRBE) 0.105  0.025 <10 -4 n 0.09 (0.01-0.33) m 0.29 (0.19-0.47) constant -5.8  1.2 <10 -5 TD50 (Gy) 44 (27-56) Performance Estimate Performance Estimate AUC 0.89 (0.81-0.95) AUC 0.88 (0.78-0.94) CV-AUC 0.87 (0.77-0.94) CV-AUC 0.86 (0.76-0.93) Calibration slope 1.00  0.11 <10 -4 CV-Calibration slope 0.95  0.06 <10 -5 Calibration intercept 0.004  0.032 0.91 CV-Calibration intercept 0.008  0.017 0.63 Abbreviations: LKB: Lyman-Kutcher-Burman, D 2% : dose released to highest 2% surface, AUC: Area under the Roc curve, CV: Cross-validation. International Journal of Radiation Oncology  Biology  Physics S168