Abstract
Constructing a four-dimensional (4D) infant cerebellum atlas is critical for modeling spatiotemporal brain development and identifying early cerebellar abnormalities. However, this task is particularly challenging due to extremely complex folia folding patterns, dynamic growth, rapid myelination, and pronounced variations in MR image appearance across developmental stages. These factors compromise population‐level anatomical correspondences and introduce longitudinal inconsistencies, causing existing atlas construction methods to produce blurred tissue boundaries and fail to preserve fine‐grained cerebellar structures. To address these challenges, we propose a structure-preserving and cycle-consistent atlas learning (SCALE) framework for 4D infant cerebellum atlas construction. SCALE jointly optimizes atlas generation and deformable registration through a dual-network architecture. Specifically, it incorporates a Hessian-based enhanced structural representation that effectively captures the accordion‐like laminar sheets of cerebellar folia, enabling more accurate alignment. To promote longitudinal consistency, SCALE introduces a cross-temporal cycle-consistency constraint that forms closed deformation loops. This constraint enforces agreement between direct within-subject longitudinal transformation paths and atlas-mediated transformation paths, thereby regularizing atlas evolution to be consistent with individual growth trajectories. Extensive experiments on the UNC/UMN Baby Connectome Project dataset demonstrate that SCALE generates a high‐fidelity 4D infant cerebellum atlas with sharper structural details, more stable population‐level alignment, and greater longitudinal consistency across development. These improvements may facilitate early detection of atypical developmental deviations of the cerebellum.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_042.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=naRPX3JY0x
BibTex
@InProceedings{ShaQin_SCALE_MICCAISAT2026,
author = { Sha, Qingrui AND Liu, Jiameng AND Hu, Dan AND Cheng, Jiale AND Sun, Yue AND Wu, Zhengwang AND Lin, Weili AND Wang, Li AND Li, Gang},
title = { { SCALE: Structure-preserving Cycle-consistent Atlas Learning for 4D Infant Cerebellum Atlas Construction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}