ESTRO 2025 - Abstract Book

S1855

Clinical – Upper GI

ESTRO 2025

Conclusion: Although this acute “immediate” post-SBRT adverse effect manifests rarely in patients with liver lesions (1.6%) and passes relatively quickly regardless of whether symptomatic treatment is applied, we must take it into consideration.

Keywords: Liver, SBRT, toxicity

References: 1.Majeed H, Gupta V. Adverse Effects of Radiation Therapy. 2023 Aug 14. In: StatPearls [Internet]. Treasure Island (FL): StatPearls Publishing; 2024 Jan–. PMID: 33085406. 2.Pollom EL, Chin AL, Diehn M, Loo BW, Chang DT. Normal Tissue Constraints for Abdominal and Thoracic Stereotactic Body Radiotherapy. Semin Radiat Oncol. 2017 Jul;27(3):197-208. doi: 10.1016/j.semradonc.2017.02.001. Epub 2017 Feb 20. PMID: 28577827.

4070

Digital Poster A Decision Tree Model for Predicting Local Control in Liver Metastases Treated with SBRT Lisa Seyfried, Michael Eble, Ahmed Allam Mohamed Radiation Oncology, Faculty of Medicine, RWTH Aachen University, Aachen, Germany Purpose/Objective: Liver metastases are a common and challenging clinical issue in oncology, with surgical resection being the standard therapy for resectable cases. However, surgery is not suitable for all patients, especially those with multifocal disease or poor liver function. Stereotactic body radiotherapy (SBRT) has emerged as an effective, non invasive treatment option for patients with limited liver metastases, offering durable local control (LC) while preserving liver function. Predicting LC following SBRT is crucial for treatment decisions. In this study, we developed a Decision Tree classifier to evaluate the impact of key clinical and dosimetric factors on LC outcomes. The model aims to provide a reliable tool for predicting patient outcomes following SBRT. Material/Methods: This retrospective study included patients who received SBRT for liver metastases between 2015 and 2024 at the Department of Radiation Oncology, RWTH Aachen University. Clinical, dosimetric, and treatment-related data were collected, and the most relevant predictors for local control were identified based on a correlation matrix. A Decision Tree Classifier was developed to predict local control. Model performance was evaluated and validated using bootstrap resampling to assess robustness. Results: A total of 71 patients with 91 liver metastases treated with SBRT were included in the analysis. Local control rates at 6, 12, and 24 months were 85.3%, 81.3%, and 81.3%, respectively. A Decision Tree model was developed using PTV D50% (EQD2), IGRT (CBCT DIBH vs. CBCT FB), and tumor volume as predictors for local control. Feature selection was based on a correlation matrix, with highly correlated predictors removed using a Variance Inflation Factor threshold of 10. The model was trained with data-driven thresholds that minimized Gini impurity, allowing for the identification of the most significant splits to predict local control outcomes (figure 1). The Decision Tree model demonstrated strong predictive performance, achieving ROC-AUC of 0.92. Bootstrap resampling (1000 iterations) validated the model’s robustness, with a mean ROC-AUC of 0.95 and a low standard deviation of 0.03, indicating consistent performance across different data subsets. The model showed high accuracy in predicting local control, with a recall for local control failure of 84.6% and a precision of 52% for classifying local control success (figure 2).