pipette
ENEnglish

RC-aware nnU-Netv2 for Pre-treatment and Post-treatment Glioma Segmentation Using Multimodal MRI

Lin Qu, Ziqi Chen, Anqi Wu, Jinyao Shen, Hai Shu

PreprintUso en el mundo real

En palabras de los autores

BraTS 2025 Lighthouse Challenge Task 1 (BraTS-GLI 2025) evaluates glioma segmentation in pre-treatment and post-treatment multimodal MRI. The resection cavity (RC) is applicable only to post-treatment cases, creating different target definitions across the two cohorts. We developed RC-aware nnU-Netv2, a framework that combines lesion- and boundary-aware single-cohort training with an RC-aware joint objective and treatment-status-guided routing. During pooled training, the joint objective applies both standard and boundary-weighted binary cross-entropy to all four region channels, while using standard Dice supervision for channels with a non-empty target in the current mini-batch and a controlled false-positive penalty for channels that are empty across the mini-batch. This empty-target handling is particularly relevant to the all-zero RC target in pre-treatment cases. Beyond the standard nnU-Netv2 augmentation pipeline, the submission does not use synthetic tumor generation, on-the-fly GliGAN augmentation, model-level probability averaging, voting, multi-fold fusion, or multi-architecture ensembling. Each case is routed to exactly one specialized model. Our submission ranked second in BraTS-GLI 2025. On the official blind test set, our method achieved mean lesion-wise Dice scores of 0.7878, 0.8709, 0.7923, and 0.8715 and mean NSD@1.0 scores of 0.8309, 0.8712, 0.7980, and 0.8336 for enhancing tumor, RC, tumor core, and whole tumor, respectively. In paired post-treatment analysis, the RC-aware joint model improved mean lesion-wise Dice by 0.042 (95% CI: [0.028, 0.057]) and mean NSD@1.0 by 0.043 (95% CI: [0.028, 0.059]) compared with joint baseline training.

Resultado principalEl resumen no menciona limitaciones.

Apareció: viernes, 2 de octubre. arXiv. Preprint, todavía sin revisión por pares.

Comentario de los autores: The proposed method won second place in the MICCAI 2025 BraTS Lighthouse Challenge on Glioma Segmentation on Pre- and Post-Treatment MRI (BraTS-GLI 2025)