Abstract
Connectional brain templates (CBTs) summarize population-level brain connectivity into a unified reference for studying neurodevelopmental disorders. All existing methods reduce the population to a \emph{single} static connectivity matrix, discarding temporal dynamics, intra-block heterogeneity, and brain-state transition structure that carry disorder-specific information beyond any static estimate. We introduce \textbf{MetaCBT}, a framework that redefines the population template as a structured six-layer representation jointly encoding static inter-network coupling, dynamic variability, connectivity dispersion, dynamic lift (network-space projection of state-transition structure), co-activation bias, and transition grammar. Rather than averaging uniformly across subjects, MetaCBT anchors each population template to its most representative members through medoid-based weighting, requiring no supervision beyond the diagnostic labels already available. Evaluated on ABIDE-I, MetaCBT matches the discriminative accuracy of established single-layer baselines while producing markedly more reproducible biomarkers: the disorder-linked connectivity patterns it identifies remain far more stable across independent cross-validation folds than those recovered by any competing method. Crucially, layer ablation reveals a dissociation between classification-oriented and population-template-level discriminability, a structural property inaccessible to any single-layer representation. These results position multi-layer connectivity modeling as a more faithful and more reproducible foundation for biomarker discovery in heterogeneous neurodevelopmental conditions. Our source code is available at \url{https://github.com/basiralab/MetaCBT}.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/GRAIL_019.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=4L0kCF5k1j
BibTex
@InProceedings{SouMay_MetaCBT_MICCAISAT2026,
author = { Soussia, Mayssa AND Mahjoub, Mohamed Ali AND Rekik, Islem},
title = { { MetaCBT: A Multi-Layer Connectional Brain Template for Reproducible Population-Level Biomarker Discovery } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}