Representation learning of human cortical folding to reveal long lasting neurodevelopmental signatures
Abstract
The human brain folds in utero, primarily during late gestation. Shortly after birth, cortical folding patterns are established and remain stable thereafter, making them promising early neurodevelopmental markers. Yet it is unclear whether the representations given by current neuroimaging foundation models capture cortical folding variability. Here, we introduce Champollion, a self-supervised learning framework that learns interpretable local representations of cortical folding from structural MRI. Optimized on representative folding-related tasks, Champollion accurately captures known folding patterns across cortical regions and external datasets. In a comprehensive benchmark, it consistently outperforms neuroimaging and general-purpose foundation models. Furthermore, Champollion reveals richer genetic associations than conventional morphometric descriptors and identifies localized folding signatures associated with incomplete hippocampal inversion, prematurity, and maternal smoking. These results establish cortical folding as a rich and largely untapped source of neurodevelopmental information, and Champollion provides a unified framework for discovering, localizing and interpreting long lasting cortical folding signatures.
Related
- FlatClip: Reusing Image Foundation Models for fMRI Representation Learning via Cortical Flatmaps
- Self-supervised representations reveal the genetic architecture of human cortical folding
- In Utero MRI of Radial Glial Fibers Highlights Their Role in Modulating Fetal Cortical Folding
- Vision In SyndrOmic Neurodevelopmental disorders-15q (VISiON-15q) study: eye tracking to supplement clinical examination for detection of cerebral/cortical visual impairment in children with Angelman and Duplication of 15q Syndromes
- Three-hinge gyri link individualized folding topology to cortical and superficial white-matter diffusion organization
BibTeX
@misc{laval2026representation,
title = {{Representation learning of human cortical folding to reveal long lasting neurodevelopmental signatures}},
author = {Julien Laval and Robin Guiavarch and Antoine Dufournet and Racim Menasria and Barthélémy Drabczuk and Cristobal Mendoza and Saeb Tounsi and Chikh Abdelghani Baroud and Merieme Bourenane and Vanessa Troiani and William Snyder and Marisa A Patti and Mylène Moyal and Marion Plaze and Arnaud Cachia and Federica Santacroce and Giorgia Committeri and Claire Cury and Kevin De Matos and Olivier Colliot and Zhong Yi Sun and Clara Fischer and Vincent Frouin and Pietro Gori and Denis Rivière and Joël Chavas and Jean-François Mangin},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.05438},
eprint = {2609.05438},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.05438},
}
RIS
TY - UNPB
TI - Representation learning of human cortical folding to reveal long lasting neurodevelopmental signatures
AU - Julien Laval
AU - Robin Guiavarch
AU - Antoine Dufournet
AU - Racim Menasria
AU - Barthélémy Drabczuk
AU - Cristobal Mendoza
AU - Saeb Tounsi
AU - Chikh Abdelghani Baroud
AU - Merieme Bourenane
AU - Vanessa Troiani
AU - William Snyder
AU - Marisa A Patti
AU - Mylène Moyal
AU - Marion Plaze
AU - Arnaud Cachia
AU - Federica Santacroce
AU - Giorgia Committeri
AU - Claire Cury
AU - Kevin De Matos
AU - Olivier Colliot
AU - Zhong Yi Sun
AU - Clara Fischer
AU - Vincent Frouin
AU - Pietro Gori
AU - Denis Rivière
AU - Joël Chavas
AU - Jean-François Mangin
PY - 2026
DA - 2026/09/10
PB - arXiv
DO - 10.48550/arxiv.2609.05438
UR - https://arxiv.org/abs/2609.05438
AB - The human brain folds in utero, primarily during late gestation. Shortly after birth, cortical folding patterns are established and remain stable thereafter, making them promising early neurodevelopmental markers. Yet it is unclear whether the representations given by current neuroimaging foundation models capture cortical folding variability. Here, we introduce Champollion, a self-supervised learning framework that learns interpretable local representations of cortical folding from structural MRI. Optimized on representative folding-related tasks, Champollion accurately captures known folding patterns across cortical regions and external datasets. In a comprehensive benchmark, it consistently outperforms neuroimaging and general-purpose foundation models. Furthermore, Champollion reveals richer genetic associations than conventional morphometric descriptors and identifies localized folding signatures associated with incomplete hippocampal inversion, prematurity, and maternal smoking. These results establish cortical folding as a rich and largely untapped source of neurodevelopmental information, and Champollion provides a unified framework for discovering, localizing and interpreting long lasting cortical folding signatures.
ER -