Cervix-Cancer-CBCT-Segmentator v1.0.0

Le résumé fourni par la source

## Introduction Cone-Beam CT (CBCT) remains the main image-of-the-day modality in cervical cancer photon radiotherapy. Segmentation of the Clinical Target Volume (CTV) and Organs At Risk (OARs) on these images enables daily treatment plan selection and adaptation. Fast and accurate automatic segmentation is highly needed, since the time required for manual delineation makes it clinically impractical. In this codebase, we share with the community a segmentation model based on nnU-Net v2 (Isensee et al., https://github.com/DIAGNijmegen/nnUNet_v2) for segmentation of: - CTVt (the whole uterus) - OARs including the bladder and rectum The model was trained on a single-center manually annotated dataset, including: - 23 patients - 73 CT images - 265 CBCT images Training was performed using nnU-Net v2 version 2.4.2. The default 5-fold cross-validation configuration was used, and the five checkpoints are jointly used as an ensemble model during inference. In our study, we found that fine-tuning this model with an annotated planning CT from the same patient can improve segmentation accuracy on that patient's CBCT images. ## Citation If you use this model, please cite our manuscript: > Songyue Han*, Delphine Lebret*, Caroline Lafond*, Julie Leseur, Cedric Hemon, Anaïs Baratea, Jean-Claude Nunes, Karine Peignaux, Nathalie Mesgouez-Nebout, Magali Le Blanc-Onfro, Chantal Hanzen, Nedjla Allouach, Sophie Renard-Oldrini, Florence Le Tinier, Renaud de Crevoisier, Antoine Simon > Evaluation of Training Strategies for U-Net-Based Cervical Cancer CBCT Segmentation in a Multi-Institutional Setting > Physics and Imaging in Radiation Oncology, Under Review

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.