Abstract
Ultra-low-field (ULF) MRI offers a portable and cost-effective alternative to conventional high-field neuroimaging, but its limited signal-to-noise ratio and spatial resolution pose challenges for automated analysis. This work presents three independent task-specific approaches submitted to the LISA Challenge, addressing image quality improvement, quality assurance, and multi-structure neuroanatomical segmentation in pediatric brain MRI acquired at 0.064T. For image quality improvement, a physics-guided enhancement model (ULF-Synth) is applied to synthesize high-field-like appearance from ULF acquisitions in a zero-shot setting. Quality assurance is performed using an ensemble of radiomics-based XGBoost classifiers and a DINOv3 vision foundation model adapted with low-rank adaptation for artifact classification. For neuroanatomical segmentation, an ensemble of convolutional and transformer-based architectures is trained on enhanced ULF images, including a modified dual-route MedNeXt variant incorporating high-frequency edge processing. Experimental results demonstrate reliable performance across all three tasks under strong domain shift conditions. Quality assurance achieves a micro-averaged F1 score of 0.798, while segmentation attains an average Dice score of 0.82 across major subcortical structures. Image enhancement metrics indicate measurable distributional and perceptual differences between enhanced and reference domains. Overall, the proposed methods highlight the complementary role of physics-informed enhancement, radiomics-based modeling, and large-scale vision transformers for effective analysis of ultra-low-field neuroimaging.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/LISA_016.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=AZ96OOmuzj
BibTex
@InProceedings{AkiOlu_Enhancing_MICCAISAT2026,
author = { Akinmuleya, Oluwatobi Iyanuoluwa AND Musah, Toufiq AND Li, Xiaomeng AND Sharma, Ajay AND Geethanath, Sairam},
title = { { Enhancing Neuroanatomical Segmentation and Quality Assurance in Pediatric Ultra-Low-Field 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}
}