Abstract
Pediatric brain tumor subregion segmentation is strongly imbalanced: whole-tumor anatomy is comparatively stable, whereas enhancing tumor (ET), cystic component (CC), and edema (ED) may be absent or small. We present an anatomy-preserving system for BraTS-PEDs 2026 Task 2. Two independent five-fold residual-encoder nnU-Net ensembles supply tumor anatomy, one defining the submitted segmentation and one whose tumor core localizes an enhancing-tumor specialist, and the specialist reaches the output only through a constrained transplant: it may relabel non-enhancing tumor core (NETC) as ET and nothing else, the whole case is rolled back once a proposal exceeds 800 voxels, and a terminal tumor-core guard closes the assembly, so tumor core, whole tumor and the cystic mask are invariant by construction. On the 91-case challenge validation set the frozen submission achieved Dice 0.588, 0.917, 0.204, 0.000, 0.944 and 0.944 for ET, NETC, CC, ED, tumor core and whole tumor. We also report a negative result: four learned rare-subtype specialists, each with a large ground-truth oracle ceiling, transferred almost nothing under strict held-out evaluation, and importing a challenge winner degraded CC. Pulled back from the registry by digest, the submitted image reprocessed all 91 cases and differs from the frozen reference in ten of them by twelve voxels in total, leaving presence, connected components and instance topology unchanged. Code, container recipe and the verification record are released publicly.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_PEDs_001.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/BraTS_PEDs_001_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=qkrnudegtE
BibTex
@InProceedings{RaiHar_AnatomyPreserving_MICCAISAT2026,
author = { Rai, Hari Mohan AND Omarbekov, Dauren},
title = { { Anatomy-Preserving Rare-Subtype Adaptation for 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}
}