ESTRO 2025 - Abstract Book

S1466

Clinical – Mixed sites & palliation

ESTRO 2025

ulceration four and twelve months after irradiations of bulky LNM (GTVs: 136 and 167 ccm; doses: 30Gy and 32Gy in five fractions each, respectively). The 2-years local control-rate was 85.8%. In univariate analysis, smaller target volumes, prostate histology, pelvic localization, and higher BED were associated with improved overall survival, distant control rates and freedom from progression (p-values <0.05), but not with local control rates. In multivariate analysis, no association remained. Conclusion: MRI-guided SBRT for abdominal and pelvic LNM achieved durable local tumor control rates of almost 90% after two years and was associated with a favorable toxicity profile. The optimal dose and fractionation regimens remain to be determined.

Keywords: MR-Linac, SBRT, Lymph node metastasis

636

Digital Poster A.I. generated prediction model of response to SBRT treatment of liver lesions through clinical and radiomic features Cecilia Pace 1 , Savino Cilla 2 , Gabriella Macchia 1 , Donato Pezzulla 1 , Carmela Romano 2 , Mariangela Boccardi 1 , Marica Ferro 1 , Paolo Bonome 1 , Vincenzo Picardi 1 , Milly Buwenge 3 , Gian Carlo Mattiucci 4,5 , Alessio Giuseppe Morganti 3,6 , Francesco Deodato 1,5 1 Radiation Oncology Department, Responsible Research Hospital, Campobasso, Italy. 2 Medical Physics Unit, Responsible Research Hospital, Campobasso, Italy. 3 Department of Medical and Surgical Sciences-DIMEC, Alma Mater Studiorum University of Bologna, Bologna, Italy. 4 Radiation Oncology Department, Mater Olbia Hospital, Olbia, Italy. 5 Istituto di Radiologia, Università Cattolica del Sacro Cuore Roma, Roma, Italy. 6 Radiation Oncology, IRCCS Azienda Ospedaliero-Universitaria di Bologna, Bologna, Italy Purpose/Objective: Stereotactic body radiation therapy (SBRT) is a key therapeutic modality for patients with liver oligometastases. This study aims to develop a predictive model based on machine learning to predict complete response (CR) in patients with liver oligometastases undergoing SBRT. Material/Methods: Computed Tomography (CT) images of 77 hepatic oligometastases from 54 patients treated with SBRT were analyzed. Gross Tumor Volume (GTV) was contoured on each image. Patients who achieved CR at 4 months were labeled as “Responder.” From each GTV, 107 radiomic features were extracted using PyRadiomics software. To ensure the robustness of the features, concordance correlation coefficients (CCC) of features extracted from ROIs obtained from two segmentations were calculated. The association of clinical variables and radiomic features with complete response (CR) was assessed by univariate logistic regression. Finally, a supervised machine learning model, the Classification and Regression Tree (CART) algorithm, was trained to predict the CR. The model was validated by 5-fold Cross-Validation, and model performance was evaluated using the area under the ROC curve (AUC), as well as accuracy, precision, and recall metrics. Results: 31 lesions (40.3%) reported CR. CR was associated with four radiomic characteristics, such as surface area/volume ratio, Uniformity, MCC, and IDM. Significant correlations emerged with two clinical variables, the number of lesions and a simplified oligometastases classification. Using these variables, the CART model was trained and evaluated, reporting accuracy, precision, and recall of 0.792, 0.784, and 0.803 in the training set and 0.750, 0.651, and 0.754 in the validation set, respectively. The area under the ROC curve (AUC) for CR prediction was 0.841 in the training set and 0.804 in the validation set. CART analysis classified “repeated” lesions to have 0% probability of CR. Lesions