ESTRO 2025 - Abstract Book

S2526

Physics - Autosegmentation

ESTRO 2025

3490

Digital Poster Development of a fully automated CTV segmentation model for resection cavities of brain metastases in a multi-center patient cohort Mai Q. Nguyen 1 , Elena Belli 1 , Tobi Martins 1 , Ivan Ezhov 2,3 , Claus Zimmer 4 , Bernhard Meyer 5 , Rami A. El Shafie 6,7,8 , Jürgen Debus 6,7 , Robert Wolff 9,10 , Oliver Blanck 9,11 , Kerstin A. Eitz 1,12,13 , Benedikt Wiestler 3,4 , Denise Bernhardt 1,12 , Stephanie E. Combs 1,12,13 , Jan C. Peeken 1,12,13 1 Department of Radiation Oncology, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany. 2 Department of Informatics, Technical University of Munich, Munich, Germany. 3 TranslaTUM - Central Institute for Translational Cancer Research, Technical University of Munich, Munich, Germany. 4 Department of Diagnostic and Interventional Neuroradiology, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany. 5 Department of Neurosurgery, Klinikum rechts der Isar, Technical University of Munich, Munich, Germany. 6 Department of Radiation Oncology, Heidelberg University Hospital, Heidelberg, Germany. 7 Heidelberg Institute for Radiation Oncology (HIRO), National Center for Radiation Oncology (NCRO), Heidelberg, Germany. 8 Department of Radiation Oncology, University Medical Center Göttingen, Göttingen, Germany. 9 Saphir Radiosurgery, Center Frankfurt and Northern Germany, Kiel, Germany. 10 Department of Neurosurgery, University Hospital Frankfurt, Frankfurt, Germany. 11 Department of Radiation Oncology, University Medical Center Schleswig Holstein, Kiel, Germany. 12 Deutsches Konsortium für Translationale Krebsforschung (DKTK), Partner Site Munich, Munich, Germany. 13 Institute of Radiation Medicine (IRM), Helmholtz Center Munich, Munich, Germany Purpose/Objective: Clinical target volume (CTV) segmentation is a time-consuming task in radiation treatment planning and is susceptible to interrater variability. We propose an automated workflow for delineating brain metastases resection cavities (RC) and subsequently defining the CTV by adding intra-parenchymal and dural margins to the segmented RC according to an international guideline [1]. Material/Methods: Data was collected within the multicenter study, including post-operative T1c-MR images from 124 patients across three centers. As the first step, automated segmentation of the RC based on deep learning was established. Manual RC delineations were performed by a resident following established guidelines, with expert review provided by a board-certified radiation oncologist. For automated segmentation, we trained an nnUNet v2 model using 58 cases from center 1 and 42 from center 2, for 250 epochs. A 5-fold cross-validation approach was employed with the 3D full-resolution setting. The model was tested on an independent external dataset comprising 24 cases from center 3. To generate the CTV, we trained an additional nnUNet v2 model for automated dura segmentation. For this, 24 cases from center 3 were manually annotated, including the falx cerebri and tentorium cerebelli. The same nnUNet v2 settings mentioned above were applied for training. Python-based scripting was then used to add intra-parenchymal margins to the automatically segmented RCs and adaptable dural margins to the automatically segmented dura.