Abstract
Ultra-low-field MRI presents an accessible solution for bedside clinical use in developing countries; however, the resulting 64mT imaging data suffer from various imaging artifacts and reduced spatial resolution, which creates a substantial hurdle for segmenting fine brain structures. As part of our participation in the 2026 Low-Field Pediatric Brain Magnetic Resonance Image Segmentation and Quality Assurance Challenge, we developed two deep learning architectures to overcome these low-field imaging limitations. First, for the quality assessment of 64mT T2 brain MRI data, we adopted artifact-specific hierarchical DenseNet pipelines jointly optimized with an auxiliary axis-classification loss to promote orientation-aware feature learning across axial, coronal, and sagittal acquisitions. Second, we propose a robust 3D SegResNet pipeline for the multi-class anatomical segmentation of 11 brain sub-structures, featuring a foreground-focused patch sampling strategy and a novel VolumeAsymmetryLoss to enforce left-right volumetric constraints during training. On the official challenge validation set, our quality assessment pipeline achieved a Micro F1-score of 0.803, while our segmentation pipeline achieved a mean Dice similarity coefficient of 0.80 across 11 target structures. With these models, we expect to alleviate the accessibility barrier to ultra-low-field brain MRI.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/LISA_018.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=g1d1Z7D7wp
BibTex
@InProceedings{OnSun_Axisaware_MICCAISAT2026,
author = { On, Sungchul AND Lee, Hyunki AND Kim, Ian AND Heo, Juhyun AND Kim, Hyun Gi AND Kim, Namkug},
title = { { Axis-aware Image Artifact Evaluation and Asymmetry-aware Multi-Label Segmentation of Subcortical Structures in Low-Resolution 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}
}