Abstract
Accurately predicting brain age from infant cortical surfaces provides essential neurodevelopmental biomarkers, but is constrained by the scarcity of labeled infant MRI. We propose CoSMa (Cortical Surface Mamba), which couples self-supervised contrastive pre-training with a Bidirectional Mamba backbone and achieves lower prediction error than the evaluated GNN and Transformer-based baselines. Building on this, we systematically evaluate two design choices widely assumed to matter: spatial scanning order and mesh resolution. Across five scanning strategies, no paired scanning-order comparison reached the Bonferroni-corrected significance threshold for any evaluated architecture, although the magnitude of the observed effects varied across model configurations. A permutation-invariant baseline performed significantly worse than CoSMa, supporting a benefit from inter-patch relational modeling even when no corrected effect of the exact serialization strategy was detected. We further find that Ico-2 provides the best performance among the tested resolutions in this data-constrained regime. Together, these results support moderate-resolution bidirectional sequence modeling without evidence that a specialized scanning strategy is required under the evaluated conditions.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ShapeMI_040.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=UQBFC9g5kb
BibTex
@InProceedings{YooHye_Rethinking_MICCAISAT2026,
author = { Yoon, Hye Jin AND Gahm, Jin Kyu},
title = { { Rethinking Sequence Modeling on Cortical Surfaces: Robustness, Spatial Ordering, and Resolution in Infant Brain Age Prediction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17259},
month = {pending},
page = {pending}
}