Abstract
Pediatric brain tumor subregion segmentation is challeng ing because of substantial morphological variability, severe class imbal ance, and the instability of small-volume regions. We propose Hierarchi cal Region-Expert Fusion (HREF), an nnU-Net-based framework that retains a stable whole-label backbone while introducing task-specific region experts and conservative inference-time fusion. The whole-label backbone produces a complete mutually exclusive base segmentation. A TC-subregion expert provides probabilities for enhancing tumor (ET), non-enhancing tumor core (NETC), and cystic component (CC), while a six-region metric-aligned expert supplies complementary probabilities for the principal body regions, including whole tumor (WT), tumor core (TC), and NETC. This paper adopts a two-stage hierarchical safe fusion process. Stage 1 applies high-confidence, spatially constrained NETC-to ET repair. Stage 2 repairs WT/TC/NETC body boundaries through re stricted background-to-NETC and NETC-to-background updates while locking ET, CC, and edema (ED). On the official BraTS-PEDs 2026 validation set, the core HREF configuration achieves an average perfor mance of 0.5118, computed as the unweighted mean of the six DSC and six NSD entries. These results indicate that region-expert probabilities can support localized refinement while preserving sensitive subregions and the global label hierarchy.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_PEDs_009.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=szd3a8B0gI
BibTex
@InProceedings{AnXin_HREF_MICCAISAT2026,
author = { An, Xinjun AND Qu, Qingchao AND Li, Chengwei AND Zhang, Yixin AND Zhu, Liying AND Wang, Hongjuan},
title = { { HREF: Hierarchical Region-Expert Fusion Framework for Accurate Pediatric Brain Tumor Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17253},
month = {pending},
page = {pending}
}
