% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@INPROCEEDINGS{Schiffer:1043530,
author = {Schiffer, Christian and Boztoprak, Zeynep and Kropp,
Jan-Oliver and Thönnißen, Julia and Behr, Katja and
Spitzer, Hannah and Amunts, Katrin and Dickscheid, Timo},
title = {{C}yto{N}et: {A} {F}oundation {M}odel for {M}icroscopic
{A}nalysis of {C}ytoarchitecture in the {H}uman {B}rain},
reportid = {FZJ-2025-02906},
year = {2025},
abstract = {Studying the structure of biological networks in the human
brain is key to decoding the mechanisms underlying brain
function, dysfunction, and behavior. Imaging and mapping
distributions of cells and nerve fibers at the micrometer
scale across entire human brains can bridge the gap between
nanoscale imaging of small fields of view (e.g., by EM) and
in vivo imaging of the whole brain (e.g., MRI, fMRI, DWI),
the latter capturing structure and function at the
millimeter scale across large numbers of subjects. A
fundamental organizational principle in the cerebral cortex
is cytoarchitecture; defined by the columnar and laminar
arrangement of cells as well as their shape, density, size,
and type. High-throughput microscopic imaging of whole human
brain sections allows to map cytoarchitecture at whole-brain
level, but implies to process petabyte-scale image datasets
to capture inter-individual microstructures variability of
different brains. To leverage the rich information of such
large datasets for brain research, we propose CytoNet, a
foundation model for cytoarchitecture in the human cerebral
cortex. CytoNet is trained with a specifically designed
selfsupervised learning task that exploits the relationship
between spatial proximity and architectural similarity in
the brain to promote the extraction of cytoarchitectonic
features from microscopic image patches extracted along the
cortical ribbon. The model learns to extract anatomically
plausible latent features in a fully data-driven fashion,
capturing fundamental properties of cytoarchitecture with
their regional variance and inter-subject variability. The
features are comparable across brain regions and subjects,
can be computed at arbitrarily dense sampling locations in
the cerebral cortex of different brains, and facilitate a
broad range of neuroscientific analysis tasks. In
particular, we demonstrate state-of-the-art performance on
brain area classification, cortical layer segmentation,
estimation of morphological parameters, and unsupervised
parcellation. As a foundation model, CytoNet offers new
perspectives for characterizing microscopic architecture
across subjects, establishing the foundation for holistic
analyses of cytoarchitecture and its relationship to other
organizational and functional principles at the whole-brain
level.},
month = {Jun},
date = {2025-06-25},
organization = {Helmholtz Imaging Conference 2025,
Potsdam (Germany), 25 Jun 2025 - 27 Jun
2025},
subtyp = {After Call},
cin = {INM-1},
cid = {I:(DE-Juel1)INM-1-20090406},
pnm = {5254 - Neuroscientific Data Analytics and AI (POF4-525) /
X-BRAIN (ZT-I-PF-4-061) / Helmholtz AI - Helmholtz
Artificial Intelligence Coordination Unit – Local Unit FZJ
(E.40401.62) / JL SMHB - Joint Lab Supercomputing and
Modeling for the Human Brain (JL SMHB-2021-2027) / EBRAINS
2.0 - EBRAINS 2.0: A Research Infrastructure to Advance
Neuroscience and Brain Health (101147319) / HIBALL -
Helmholtz International BigBrain Analytics and Learning
Laboratory (HIBALL) (InterLabs-0015)},
pid = {G:(DE-HGF)POF4-5254 / G:(DE-HGF)ZT-I-PF-4-061 /
G:(DE-Juel-1)E.40401.62 / G:(DE-Juel1)JL SMHB-2021-2027 /
G:(EU-Grant)101147319 / G:(DE-HGF)InterLabs-0015},
typ = {PUB:(DE-HGF)6},
url = {https://juser.fz-juelich.de/record/1043530},
}