JAMA Network Open,
Год журнала:
2023,
Номер
6(3), С. e231671 - e231671
Опубликована: Март 6, 2023
Importance
Neuroimaging-based
artificial
intelligence
(AI)
diagnostic
models
have
proliferated
in
psychiatry.
However,
their
clinical
applicability
and
reporting
quality
(ie,
feasibility)
for
practice
not
been
systematically
evaluated.
Objective
To
assess
the
risk
of
bias
(ROB)
neuroimaging-based
AI
psychiatric
diagnosis.
Evidence
Review
PubMed
was
searched
peer-reviewed,
full-length
articles
published
between
January
1,
1990,
March
16,
2022.
Studies
aimed
at
developing
or
validating
diagnosis
disorders
were
included.
Reference
lists
further
suitable
original
studies.
Data
extraction
followed
CHARMS
(Checklist
Critical
Appraisal
Extraction
Systematic
Reviews
Prediction
Modeling
Studies)
PRISMA
(Preferred
Reporting
Items
Meta-analyses)
guidelines.
A
closed-loop
cross-sequential
design
used
control.
The
PROBAST
(Prediction
Model
Risk
Bias
Assessment
Tool)
modified
CLEAR
Evaluation
Image-Based
Artificial
Intelligence
Reports)
benchmarks
to
evaluate
ROB
quality.
Findings
total
517
studies
presenting
555
included
Of
these
models,
461
(83.1%;
95%
CI,
80.0%-86.2%)
rated
as
having
a
high
overall
based
on
PROBAST.
particular
analysis
domain,
including
inadequate
sample
size
(398
[71.7%;
68.0%-75.6%]),
poor
model
performance
examination
(with
100%
lacking
calibration
examination),
lack
handling
data
complexity
(550
[99.1%;
98.3%-99.9%]).
None
perceived
be
applicable
practices.
Overall
completeness
number
reported
items/number
items)
61.2%
(95%
60.6%-61.8%),
poorest
technical
assessment
domain
with
39.9%
38.8%-41.1%).
Conclusions
Relevance
This
systematic
review
found
that
feasibility
challenged
by
Particularly
should
addressed
before
application.
IEEE Transactions on Neural Systems and Rehabilitation Engineering,
Год журнала:
2022,
Номер
31, С. 710 - 719
Опубликована: Дек. 16, 2022
Due
to
the
limited
perceptual
field,
convolutional
neural
networks
(CNN)
only
extract
local
temporal
features
and
may
fail
capture
long-term
dependencies
for
EEG
decoding.
In
this
paper,
we
propose
a
compact
Convolutional
Transformer,
named
Conformer,
encapsulate
global
in
unified
classification
framework.
Specifically,
convolution
module
learns
low-level
throughout
one-dimensional
spatial
layers.
The
self-attention
is
straightforwardly
connected
correlation
within
features.
Subsequently,
simple
classifier
based
on
fully-connected
layers
followed
predict
categories
signals.
To
enhance
interpretability,
also
devise
visualization
strategy
project
class
activation
mapping
onto
brain
topography.
Finally,
have
conducted
extensive
experiments
evaluate
our
method
three
public
datasets
EEG-based
motor
imagery
emotion
recognition
paradigms.
experimental
results
show
that
achieves
state-of-the-art
performance
has
great
potential
be
new
baseline
general
code
been
released
https://github.com/eeyhsong/EEG-Conformer.
Frontiers in Molecular Neuroscience,
Год журнала:
2022,
Номер
15
Опубликована: Окт. 4, 2022
Autism
spectrum
disorder
(ASD)
is
a
brain
condition
characterized
by
diverse
signs
and
symptoms
that
appear
in
early
childhood.
ASD
also
associated
with
communication
deficits
repetitive
behavior
affected
individuals.
Various
detection
methods
have
been
developed,
including
neuroimaging
modalities
psychological
tests.
Among
these
methods,
magnetic
resonance
imaging
(MRI)
are
of
paramount
importance
to
physicians.
Clinicians
rely
on
MRI
diagnose
accurately.
The
non-invasive
include
functional
(fMRI)
structural
(sMRI)
methods.
However,
diagnosing
fMRI
sMRI
for
specialists
often
laborious
time-consuming;
therefore,
several
computer-aided
design
systems
(CADS)
based
artificial
intelligence
(AI)
developed
assist
specialist
Conventional
machine
learning
(ML)
deep
(DL)
the
most
popular
schemes
AI
used
ASD.
This
study
aims
review
automated
using
AI.
We
CADS
ML
techniques
diagnosis
modalities.
There
has
very
limited
work
use
DL
develop
diagnostic
models
A
summary
studies
provided
Supplementary
Appendix.
Then,
challenges
encountered
during
described
detail.
Additionally,
graphical
comparison
automatically
discussed.
suggest
future
approaches
detecting
ASDs
neuroimaging.
Sensors,
Год журнала:
2021,
Номер
21(22), С. 7710 - 7710
Опубликована: Ноя. 19, 2021
Epilepsy
is
a
brain
disorder
disease
that
affects
people's
quality
of
life.
Electroencephalography
(EEG)
signals
are
used
to
diagnose
epileptic
seizures.
This
paper
provides
computer-aided
diagnosis
system
(CADS)
for
the
automatic
seizures
in
EEG
signals.
The
proposed
method
consists
three
steps,
including
preprocessing,
feature
extraction,
and
classification.
In
order
perform
simulations,
Bonn
Freiburg
datasets
used.
Firstly,
we
band-pass
filter
with
0.5-40
Hz
cut-off
frequency
removal
artifacts
datasets.
Tunable-Q
Wavelet
Transform
(TQWT)
signal
decomposition.
second
step,
various
linear
nonlinear
features
extracted
from
TQWT
sub-bands.
this
statistical,
frequency,
based
on
fractal
dimensions
(FDs)
entropy
theories.
classification
different
approaches
conventional
machine
learning
(ML)
deep
(DL)
discussed.
CNN-RNN-based
DL
number
layers
applied.
have
been
fed
input
CNN-RNN
model,
satisfactory
results
reported.
K-fold
cross-validation
k
=
10
employed
demonstrate
effectiveness
procedure.
revealed
achieved
an
accuracy
99.71%
99.13%,
respectively.
