Abstract
The BraTS-PEDs Challenge requires accurate multi-class segmentation of pediatric brain tumors under high biological heterogeneity and a small training cohort. We present a postprocessing-centric pipeline that combines (i) a data-driven per-class champion source composition chain that assigns each tumor sub-region to the model that performs best on that class and combines them under Hierarchical Post-Processing (HPP_v5) constraints, and (ii) a boundary anchor that lifts the non-enhancing tumor core (NET) normalized surface Dice ceiling by adding peripheral voxels along the predicted surface. We characterize a sensitivity of the original K-th cutoff form of the anchor to tied distance values and propose a tied-aware stable variant whose output depends only on the threshold distance, not the selection order. On the BraTS-PEDs 2026 validation set (91 cases), the pipeline achieves competitive across-class metrics; an ablation against the K-th variant shows the stable form preserves them while restoring deterministic behavior.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_PEDs_007.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=HP0YClod1B
BibTex
@InProceedings{YooSeo_Stable_MICCAISAT2026,
author = { Yoon, Seoyoung},
title = { { Stable Boundary Anchoring and Per-Class Champion Source Composition for Pediatric Brain Tumor Segmentation in BraTS-PEDs } },
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}
}
