S2430
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
morbidity while promoting standardized, patient- centered radiotherapy planning practices. Keywords: diffusion MRI, tractography, brain metastases
Brachytherapy Department, CHR Metz-Thionville, Metz, France
Purpose/Objective: Damage to critical white matter pathways remains a major cause of cognitive and functional morbidity after stereotactic radiotherapy (SRT) for brain metastases. Diffusion MRI (dMRI) and tractography offer a non-invasive approach to identify and potentially spare functionally relevant white matter bundles, but their clinical use is currently limited by variability in acquisition protocols, reconstruction methods, and treatment planning system interfaces. The PROTECT-DTI project aims to establish a harmonized, clinically scalable workflow for white matter tract preservation during SRT across multiple radiotherapy centers in different European regions. Material/Methods: A three-year transnational consortium was formed involving six radiotherapy departments across four countries, supported by academic research laboratories and European regional development funding. The project is organized into four work packages: (WP1) harmonization and optimization of dMRI acquisition protocols across scanners with different field strengths and vendors; (WP2) development of standardized tract extraction pipelines and interoperability formats for treatment planning systems; (WP3) clinical implementation and prospective data collection in patients undergoing SRT for brain metastases; and (WP4) training, dissemination, and evaluation of inter-site reproducibility.Initial technical validation builds upon previously conducted pilot work demonstrating feasibility of shortened dMRI protocols and multicenter reproducibility of key tract bundles in healthy volunteers and phantoms. The clinical phase will evaluate white matter sparing strategies in treatment planning and document neurocognitive outcomes, workflow practicality, and clinical acceptance. Results: Preliminary data indicate that harmonized dMRI protocols can yield reproducible tract reconstructions across sites, with >90% tract overlap when applying standard uncertainty margins. Shortened acquisition sequences maintain clinically relevant bundle visualization while remaining compatible with routine SRT workflows. Early pipeline integration tests confirm successful tract export and visualization in multiple commercial planning systems. Conclusion: PROTECT-DTI establishes the infrastructure for the first coordinated, multicenter clinical evaluation of tractography-guided white matter sparing in brain SRT across multiple European healthcare systems. The project aims to reduce treatment-related neurological
Mini-Oral 1738 Radiomics and Dosiomics-Based Machine Learning Models for Predicting Radiation Esophagitis: Evidence from the Prospective RTOG 0617 and REQUITE Trials Lukas Manuel Reuter 1,2 , Kim Melanie Kraus 1,3 , Stefan Michael Fischer 1,2 , Danai Pletzer 1,2 , Annika Domres 1 , Mai Nguyen 1 , Denise Bernhardt 1,3 , Stephanie Elisabeth Combs 1,3 , Julia Anne Schnabel 2,4 , Jan Caspar Peeken 1,3 1 Department of Radiation Oncology, School of Medicine and TUM University Hospital rechts der Isar, Technical University of Munich (TUM), Munich, Germany. 2 Institute of Machine Learning in Biomedical Imaging (IML), Helmholtz Zentrum München (HMGU) GmbH, Neuherberg, Germany. 3 Institute of Radiation Medicine (IRM), Helmholtz Zentrum München (HMGU) GmbH, Neuherberg, Germany. 4 School of Biomedical Engineering and Imaging Sciences, King’s College London, London, United Kingdom Purpose/Objective: Radiation-induced esophagitis is a common side effect associated with thoracic radiation therapy (RT). Severe manifestations (CTCAE ≥ 3) frequently necessitate hospitalization and may result in treatment interruptions. This study aims to predict the occurrence of esophagitis grade 3 or higher using radiomic and dosiomic models with subsequent Prospective data from 748 multicentric lung cancer patients were evaluated, with 58 developing severe esophagitis. The training cohort is based on the NRG Oncology/RTOG 0617 trial (n=465) and the external test set on the REQUITE study (n=283). Adverse effects were evaluated based on CTCAE v3 (RTOG 0617) and v4 (REQUITE). The clinical characteristics assessed included age, gender, smoking status, tumor histology, and radiation technique employed. Radiomics and dosiomics features of the esophagus were extracted from pre-treatment CT scans, 3D dose, and EQD2 volumes in accordance with IBSI guidelines. Various combinations of radiomics, dosiomics, DVH, and clinical features were analyzed using a 10x5-fold nested cross-validation (nCV), utilizing a Random Forest classifier and the minimum-redundancy- maximum-relevance algorithm. Feature selection was conducted within the inner loop, while class imbalances were addressed through synthetic oversampling and undersampling utilizing SMOTE- external validation. Material/Methods:
Made with FlippingBook - Share PDF online