Graph
neural
networks
(GNNs)
have
recently
been
applied
to
develop
useful
diagnostic
tools
for
psychiatric
disorders.
However,
due
the
lack
of
interpretability,
clinicians
are
hard
identify
quantifiable
and
personalizable
biomarkers
which
provide
biologically
clinically
relevance.
We
introduce
three
proposed
GNN-based
disorders
models,
namely
BrainIB,
Graph-PRI
CI-GNN,
from
an
information-theoretic
perspective.
These
models
able
discriminate
patients
healthy
controls
predictive
subgraph,
a.k.a.
biomarkers,
solely
observations.
demonstrate
their
improved
classification
accuracy
interpretability
on
ABIDE
database.
also
put
forward
proposals
future
research.
In
this
paper,
we
introduce
graph
neural
networks
(GNNs)
and
automatic
stock
clustering
to
improve
market
prediction
accuracy
in
the
of
Taiwan
Stock
Exchange.
GNNs
capture
intricate
inter-stock
relationships,
enhancing
accuracy.
Automatic
based
on
behavior
ensures
adaptability.
Results
demonstrate
significant
improvements,
with
potential
applicability
beyond
Taiwan's
market,
advancing
financial
methodologies.
medRxiv (Cold Spring Harbor Laboratory),
Год журнала:
2024,
Номер
unknown
Опубликована: Окт. 29, 2024
Abstract
The
wide
variation
in
symptoms
of
neurological
disorders
among
patients
necessitates
uncovering
individual
pathologies
for
accurate
clinical
diagnosis
and
treatment.
Current
methods
attempt
to
generalize
specific
biomarkers
explain
pathology,
but
they
often
lack
analysis
the
underlying
pathogenic
mechanisms,
leading
biased
unreliable
diagnoses.
To
address
this
issue,
we
propose
a
motif-induced
subgraph
generative
learning
model
(MSGL),
which
provides
multi-tiered
facilitates
explainable
diagnoses
disorders.
MSGL
uncovers
mechanisms
by
exploring
representative
connectivity
patterns
within
brain
net-works,
offering
motif-level
tackle
challenge
heterogeneity.
Furthermore,
it
utilizes
information
generate
enhanced
network
subgraphs
as
personalized
identifying
pathology.
Experimental
results
demonstrate
that
outperforms
baseline
models.
identified
align
with
recent
neuroscientific
findings,
enhancing
their
applicability.
IECE transactions on intelligent systematics.,
Год журнала:
2024,
Номер
1(2), С. 58 - 68
Опубликована: Сен. 23, 2024
The
integration
of
graph
neural
networks
(GNNs)
with
brain
functional
network
analysis
is
an
emerging
field
that
combines
neuroscience
and
machine
learning
to
enhance
our
understanding
complex
dynamics.
We
first
briefly
introduce
the
fundamentals
networks,
followed
by
overview
Graph
Neural
Network
principles
architectures.
review
then
focuses
on
applications
these
address
current
challenges
in
field,
such
as
need
for
interpretable
models
effective
multi-modal
neuroimaging
data.
also
highlight
potential
GNNs
clinical
perimenopausal
areas
depression
research,
demonstrating
broad
applicability
this
approach.
concludes
outlining
future
research
directions,
including
development
more
sophisticated
architectures
large-scale,
heterogeneous
graphs,
exploration
causal
inference
networks.
By
synthesizing
recent
advances
identifying
key
aims
summarize
focal
points
GNNs,
explore
their
integration,
provide
a
reference
advancing
interdisciplinary
field.
NeuroImage,
Год журнала:
2024,
Номер
305, С. 120951 - 120951
Опубликована: Дек. 4, 2024
The
apolipoprotein
E
(APOE)
ɛ4
allele
is
a
recognized
genetic
risk
factor
for
Alzheimer's
Disease
(AD).
Studies
have
shown
that
APOE
mediates
the
modulation
of
intrinsic
functional
brain
networks
in
cognitively
normal
individuals
and
significantly
disrupts
whole-brain
topological
structure
AD
patients.
However,
how
regulates
connectivity
(FC)
consequently
affects
levels
cognitive
impairment
patients
remains
unknown.
In
this
study,
we
systematically
analyzed
magnetic
resonance
imaging
(fMRI)
data
from
two
distinct
cohorts:
an
In-house
dataset
includes
59
(73.37
±
6.42
years),
ADNI
117
(74.91
7.91
years).
Experimental
comparisons
were
conducted
by
grouping
based
on
both
status
AD.
Network-Based
Statistic
(NBS)
method
Graph
Neural
Network
Explainer
(GNN-Explainer)
combined
to
identify
significant
FC
changes
across
different
comparisons.
Importantly,
GNN-Explainer
was
introduced
as
enhancement
over
NBS
better
model
complex
high-order
nonlinear
characteristics
discovering
features
contribute
classification
tasks.
results
showed
primarily
influenced
temporal
lobe
FCs,
while
it
adjusting
prefrontal-parietal
FCs.
These
findings
validated
p-values
<
0.05
method,
5-fold
cross-validation
along
with
ablation
studies
method.
conclusion,
our
provide
new
insights
into
role
altering
dynamics
during
progression
AD,
highlighting
potential
targets
early
intervention.
