Abstract
Ultra-low-field (0.064T) MRI offers a portable and affordable route to pediatric neuroimaging in low-resource settings, but its reduced signal-to-noise ratio, low tissue contrast, and frequent artifacts make automated analysis difficult. We present our submissions to all three tasks of the MICCAI 2026 LISA challenge. For Task 1a (image quality assessment), we cast the problem as multi-label ordinal classification over seven artifact categories and use a frozen Vision Transformer feature extractor trained slice-wise, followed by scan-level probability pooling and per-artifact ordinal-threshold calibration. For Task 1b (image quality enhancement), we adapt an Unpaired Neural Schrödinger Bridge (UNSB) to translate artifacted images into the artifact-free domain without paired ground truth. For Task 2 (subcortical segmentation), we integrate a SwinUNETR with Euclidean distance-transform supervision, bilateral consistency, and a five-fold ensemble with sagittal test-time augmentation and light morphological post-processing. We report validation-set results and discuss how these established components were adapted to the ultra-low-field challenge setting.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/LISA_014.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=mVjYfyeNYB
BibTex
@InProceedings{HaoYin_Quality_MICCAISAT2026,
author = { Hao, Yingqi AND Sui, Peiwen AND Liu, Mingxuan AND Zhao, Xihai AND Tian, Qiyuan},
title = { { Quality Assessment, Enhancement, and Subcortical Segmentation of Ultra-Low-Field Pediatric Brain MRI: The LISA 2026 Challenge } },
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}
}