Abstract
Pediatric ultra-low-field brain magnetic resonance imaging (MRI) requires methods that remain reliable despite low signal-to-noise ratio, motion, limited contrast, and age-dependent anatomy. We study pretrained and self-configuring ensembles for three coupled problems: image-quality assessment, quality improvement, and multi-structure segmentation. Quality assessment used an ImageNet-pretrained 2.5D ResNet50 model and a six-model validation-weighted ensemble. Quality improvement was driven principally by route-tuned block-matching four-dimensional filtering (BM4D); a sparse, low-weight Noise-Level Adaptive Diffusion (Nila) branch altered only five validation images and produced nearly identical proxy metrics. Both Task 1 systems were subsequently notified as top-three test-phase entries. Segmentation combined self-configuring three-dimensional U-Net models with residual-encoder scaling, pretrained Scalable and Transferable U-Net (STU-Net) fine-tuning, mixed-label supervision, and soft-probability ensembling. On external validation, the four-model segmentation ensemble achieved average Dice similarity coefficient 0.82, Hausdorff distance 3.46, 95th-percentile Hausdorff distance 1.76, average symmetric surface distance 0.48, and relative volume error 0.14. These results support careful data organization, heterogeneous pretraining, conservative enhancement, and calibrated ensembling for small pediatric ultra-low-field MRI cohorts.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/LISA_004.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=Rll6iOtCAS
BibTex
@InProceedings{ZhaBo_Pretrained_MICCAISAT2026,
author = { Zhang, Bo AND Lin, Junjie AND He, Youquan AND Zhang, Zhiyuan AND Cai, Jiacai AND Lyu, Mengye},
title = { { Pretrained and Self-Configuring Ensembles for Pediatric Ultra-Low-Field Brain MRI Analysis } },
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}
}