% 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:1033582,
author = {Schiffer, Christian and Amunts, Katrin and Dickscheid,
Timo},
title = {{C}yto{N}et: {A} {D}eep {N}eural {N}etwork for
{W}hole-brain {C}haracterization of {H}uman
{C}ytoarchitecture},
reportid = {FZJ-2024-06462},
year = {2024},
abstract = {The characterization of cytoarchitecture in the human brain
provides an essential building block for the creation of a
high-resolution multi-modal brain atlas. Cytoarchitecture is
defined by the spatial organization of neuronal cells,
including their shape, density, size, cell type, as well as
their columnar and laminar arrangement, which differ between
brain regions. High-throughput light-microscopic scanning of
large, cell-body stained histological sections obtained by
sectioning postmortem human brains enables detailed
examination of cytoarchitectonic organizational principles
across multiple brain samples, which is mandatory to capture
the highly variable cytoarchitectonic organization. The
limited scalability of existing methods to image and analyze
datasets in the terabyte to petabyte range motivates current
developments of AI methods for data-driven characterization
and classification of human cytoarchitecture at large
scale.In this work, we present CytoNet, a deep neural
network model that enables data-driven characterization of
cytoarchitecture in the human brain. CytoNet is a
convolutional neural network that is trained on one million
image patches (2048px@2μm/px) extracted from 4115
histological sections of 9 postmortem brains. The model is
trained using a novel contrastive learning objective that
derives the similarity relationship between image samples
from their spatial distance in a common reference brain
space. Using this loss, CytoNet is trained to map spatially
close image samples, which likely show similar
cytoarchitectonic structures, to similar feature
representations.We demonstrate that feature representations
extracted by CytoNet allow classifying cytoarchitectonic
areas and cortical layers, predicting spatial and
morphological features, studying inter-individual
variations, and enabling data-driven quantification and
query-based exploration of microstructural principles at
whole-brain level. Moreover, we show that the latent space
learned by CytoNet exhibits an anatomically highly plausible
structure that facilitates intuitive exploration of brain
organization. CytoNet significantly extends existing methods
for cytoarchitecture analysis and thus provides the
foundation for novel analysis workflows that have the
potential to facilitate studies relating the brain’s
microstructure to connectivity and function.},
month = {Nov},
date = {2024-11-19},
organization = {INM Retreat 2024, Jülich (Germany),
19 Nov 2024 - 20 Nov 2024},
subtyp = {After Call},
cin = {INM-1},
cid = {I:(DE-Juel1)INM-1-20090406},
pnm = {5251 - Multilevel Brain Organization and Variability
(POF4-525) / 5254 - Neuroscientific Data Analytics and AI
(POF4-525) / HIBALL - Helmholtz International BigBrain
Analytics and Learning Laboratory (HIBALL) (InterLabs-0015)
/ EBRAINS 2.0 - EBRAINS 2.0: A Research Infrastructure to
Advance Neuroscience and Brain Health (101147319) /
Helmholtz AI - Helmholtz Artificial Intelligence
Coordination Unit – Local Unit FZJ (E.40401.62) / X-BRAIN
(ZT-I-PF-4-061)},
pid = {G:(DE-HGF)POF4-5251 / G:(DE-HGF)POF4-5254 /
G:(DE-HGF)InterLabs-0015 / G:(EU-Grant)101147319 /
G:(DE-Juel-1)E.40401.62 / G:(DE-HGF)ZT-I-PF-4-061},
typ = {PUB:(DE-HGF)24},
url = {https://juser.fz-juelich.de/record/1033582},
}