Abstract
Automatic segmentation of eleven subcortical structures in ultra-low-field (ULF; 0.064T) pediatric brain MRI is challenging owing to low signal-to-noise ratio, weak tissue contrast, and a limited annotated cohort (n=79). We present a dual-annotation, boundary-aware multi-task framework that exploits both high-field segmentation masks (HF-Seg) and paired low-field masks (LF-Seg) as complementary supervision signals. A shared 3-D encoder feeds three parallel decoders—Seg, LF-Seg, and Edge—whose intermediate features and outputs are integrated by a FusionModule augmented with a left-right symmetry (LR-Sym) constraint. ANTs-based preprocessing (N4 bias correction, skull stripping, intensity normalisation) and patch-based training are applied throughout. Five-fold cross-validation on all 79 subjects yields mean DSC 0.7895±0.0046, HD95 2.49 ± 0.56mm, and ASD 0.838±0.036mm, surpassing all single-modal baselines. Ablation results indicate a 27% reduction in volume error when LF-Seg auxiliary supervision is incorporated (|RVE|% 11.64 → 8.51), and LR-Sym cuts HD95 by 37% (4.55 → 2.84mm).
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/LISA_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=Vpd0TyEoqR
BibTex
@InProceedings{PiaZhe_DualAnnotation_MICCAISAT2026,
author = { Piao, Zhegao AND Xu, Rongxu AND Gu, Yeong Hyeon},
title = { { Dual-Annotation Assisted Boundary-Aware Segmentation for Ultra-Low-Field Pediatric Brain MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17257},
month = {pending},
page = {pending}
}
