Abstract

Low-field (LF) magnetic resonance imaging (MRI) expands access to neuroimaging through portable, cost-effective scanners but remains susceptible to image quality degradation, particularly in pediatric imaging where patient motion and limited scan compliance are common. We address LISA 2026 Task 1a, which requires automated severity assessment of seven image artifact types on a three-level ordinal scale for each acquisition plane of low-field pediatric brain MRI. We propose a modular quality assessment framework that combines a plane-aware three- dimensional convolutional neural network (CNN) with artifact-specific prediction heads and a BrainFM foundation-model specialist for positioning assessment. The proposed artifact-level expert-merge strategy selectively incorporates BrainFM predictions only for the positioning artifact while preserving the remaining CNN outputs. On the official LISA hold-out evaluation, our plane-aware CNN achieved a weighted F1 of 0.810, a weighted F2 of 0.813, a weighted precision of 0.815, a weighted recall of 0.818, and a weighted accuracy of 81.8%. The proposed expert-merge model, which alters only the positioning predictions, left the weighted F1 unchanged at 0.810 and produced small, targeted gains in weighted F2 (0.815) and weighted accuracy (82.1%), with a marginal decrease in weighted precision (0.815 to 0.814).

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/LISA_006.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=M8wGxiaECR

BibTex

@InProceedings{ÖzeCan_Automated_MICCAISAT2026,
        author = { Özer, Caner AND Aytimur, Ayda AND Xu, Ruru AND Oksuz, Ilkay},
        title = { { Automated Artifact Quality Control for Pediatric Low-Field MRI via Plane-Aware CNNs and BrainFM Feature-Pyramid Expert Merging } },
        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}
}


back to top