Abstract
Low-field pediatric brain magnetic resonance imaging (MRI) is often affected by reduced signal-to-noise ratio, limited contrast, and acquisition-related artifacts, which can hinder downstream image analysis. In this work, we present a unified Masked Autoencoder (MAE)-based framework for the LISA 2026 Challenge, covering MRI quality assessment, image restoration, and multi-structure segmentation. For Task 1a, a pretrained MAE encoder with a lightweight classification head was used to predict the severity of seven MRI quality-related artifact categories. The final model achieved an official validation accuracy and micro F1 score of 0.811. For Task 1b, the MAE encoder was adapted into a residual restoration network using synthetic degradations, identity-preserving samples, and edge-aware optimization. The final restoration model achieved a mean FID of 125.031, LPIPS of 0.462, PSNR of 11.955, and FRD of 7.076. For Task 2, an MAE encoder and lightweight convolutional decoder were used for segmentation of 11 anatomical structures. Class-wise largest connected component filtering was applied after three-dimensional reconstruction to remove isolated false-positive regions. The final segmentation pipeline achieved a mean Dice score of 0.78, with mean HD95 and ASSD values of 2.23 and 0.63, respectively. These results demonstrate that MAE-based representations can be effectively adapted across quality assessment, image restoration, and anatomical segmentation through task-specific heads, decoders, and post-processing strategies.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/LISA_003.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=rxPiI6NUjM
BibTex
@InProceedings{ZhuLey_AUnified_MICCAISAT2026,
author = { Zhu, Leyan AND Tian, Song},
title = { { A Unified Masked Autoencoder Framework for Quality Assessment, Image Restoration, and Multi-Structure Segmentation in Low-Field 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}
}