Abstract
Ultra-low-field (ULF) MRI at 0.064T extends neuroimaging to resource-limited and pediatric settings, but low SNR, weak contrast and coarse resolution challenge automated analysis. We report a single-team submission to three LISA Challenge 2026 tracks, organized around one question: at this field strength, how far a well-configured baseline goes under leakage-free evaluation, faithful validation and honest curation. For Task 1A (multi-label artifact grading over seven classes at three ordinal severity levels) we model the target in its native ordinal form with two cumulative heads, an ordinal focal loss and a monotonicity penalty, and evaluate a ten-model cross-validated ensemble on out-of-fold predictions, reaching micro-accuracy 0.839; a per-plane analysis and a slice-sampling ablation quantify how unevenly the fixed three-slice input captures spatially localized artifacts across planes. For Task 1B (low-field image-quality enhancement) we show through a controlled study that a carefully normalized native-resolution pass-through is a strong, overfitting-free baseline that our four learned enhancers fail to beat; we diagnose three causes: self-inflicted preprocessing damage, a cross-field registration ceiling near 0.5-0.6 correlation that leaves the target only partially aligned, and metric-overfitting (Goodhart) in checkpoint selection. For Task 2 (3D segmentation of eleven annotated subcortical labels) a full-cohort residual-encoder nnU-Net with stock components attains mean Dice 0.784 over eleven labels (0.785 over the nine scored under our submission-time reading), and structure-specific connected-component post-processing proves locally optimal, while uniform filtering deletes real anatomy. Across all three tasks a well-configured, honestly validated pipeline is competitive, and such baselines deserve reporting before novelty is claimed.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/LISA_011.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=Tk9Y0u5jQt
BibTex
@InProceedings{MarRon_Rigor_MICCAISAT2026,
author = { Marca, Ronald AND Guerra, Gabriel AND Ortiz-Puerta, David AND Chabert, Steren AND Salas, Rodrigo},
title = { { Rigor over Novelty in Ultra-Low-Field Pediatric Brain MRI: Quality Control and Subcortical Segmentation for the LISA Challenge 2026 } },
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}
}