Technology and Health Care,
Год журнала:
2024,
Номер
32(3), С. 1947 - 1965
Опубликована: Фев. 23, 2024
BACKGROUND:
Osteoporosis
is
a
medical
disorder
that
causes
bone
tissue
to
deteriorate
and
lose
density,
increasing
the
risk
of
fractures.
Applying
Neural
Networks
(NN)
analyze
imaging
data
detect
presence
or
severity
osteoporosis
in
patients
known
as
classification
using
Deep
Learning
(DL)
algorithms.
DL
algorithms
can
extract
relevant
information
from
images
discover
intricate
patterns
could
indicate
osteoporosis.
OBJECTIVE:
DCNN
biases
must
be
initialized
carefully,
much
like
their
weights.
Biases
are
incorrectly
might
affect
network’s
learning
dynamics
hinder
model’s
ability
converge
an
ideal
solution.
In
this
research,
Convolutional
(DCNNs)
used,
which
have
several
benefits
over
conventional
ML
techniques
for
image
processing.
METHOD:
One
key
DCNNs
automatically
Feature
Extraction
(FE)
raw
data.
time-consuming
procedure
During
training
phase
DCNNs,
network
learns
recognize
characteristics
straight
The
Squirrel
Search
Algorithm
(SSA)
makes
use
combination
Local
(LS)
Random
(RS)
inspired
by
foraging
habits
squirrels.
RESULTS:
method
made
it
possible
efficiently
explore
search
space
find
prospective
values
while
promising
areas
refine
improve
solutions.
Effectively
recognizing
optimum
nearly
optimal
solutions
depends
on
balancing
exploration
exploitation.
weight
optimized
with
help
SSA,
enhances
performance
classification.
CONCLUSION:
comparative
analysis
state-of-the-art
shows
proposed
SSA-based
highly
accurate,
96.57%
accuracy.
Deleted Journal,
Год журнала:
2024,
Номер
37(1), С. 308 - 338
Опубликована: Янв. 10, 2024
In
the
realm
of
medical
diagnostics,
utilization
deep
learning
techniques,
notably
in
context
radiology
images,
has
emerged
as
a
transformative
force.
The
significance
artificial
intelligence
(AI),
specifically
machine
(ML)
and
(DL),
lies
their
capacity
to
rapidly
accurately
diagnose
diseases
from
images.
This
capability
been
particularly
vital
during
COVID-19
pandemic,
where
rapid
precise
diagnosis
played
pivotal
role
managing
spread
virus.
DL
models,
trained
on
vast
datasets
have
showcased
remarkable
proficiency
distinguishing
between
normal
COVID-19-affected
cases,
offering
ray
hope
amidst
crisis.
However,
with
any
technological
advancement,
vulnerabilities
emerge.
Deep
learning-based
diagnostic
although
proficient,
are
not
immune
adversarial
attacks.
These
attacks,
characterized
by
carefully
crafted
perturbations
input
data,
can
potentially
disrupt
models'
decision-making
processes.
context,
such
could
dire
consequences,
leading
misdiagnoses
compromised
patient
care.
To
address
this,
we
propose
two-phase
defense
framework
that
combines
advanced
image
filtering
techniques.
We
use
modified
algorithm
enhance
model's
resilience
against
examples
training
phase.
During
inference
phase,
apply
JPEG
compression
mitigate
cause
misclassification.
evaluate
our
approach
three
models
based
ResNet-50,
VGG-16,
Inception-V3.
perform
exceptionally
classifying
images
(X-ray
CT)
lung
regions
into
normal,
pneumonia,
pneumonia
categories.
then
assess
vulnerability
these
targeted
attacks:
fast
gradient
sign
method
(FGSM),
projected
descent
(PGD),
basic
iterative
(BIM).
results
show
significant
drop
model
performance
after
greatly
improves
resistance
maintaining
high
accuracy
examples.
Importantly,
ensures
reliability
diagnosing
clean
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Июнь 9, 2024
Image
segmentation
techniques
play
a
vital
role
in
aiding
COVID-19
diagnosis.
Multi-threshold
image
methods
are
favored
for
their
computational
simplicity
and
operational
efficiency.
Existing
threshold
selection
multi-threshold
segmentation,
such
as
Kapur
based
on
exhaustive
enumeration,
often
hamper
efficiency
accuracy.
The
whale
optimization
algorithm
(WOA)
has
shown
promise
addressing
this
challenge,
but
issues
persist,
including
poor
stability,
low
efficiency,
accuracy
segmentation.
To
tackle
these
issues,
we
introduce
Latin
hypercube
sampling
initialization-based
multi-strategy
enhanced
WOA
(CAGWOA).
It
incorporates
COS
initialization
strategy
(COSI),
an
adaptive
global
search
approach
(GS),
all-dimensional
neighborhood
mechanism
(ADN).
COSI
leverages
probability
density
functions
created
from
sampling,
ensuring
even
solution
space
coverage
to
improve
the
stability
of
model.
GS
widens
exploration
scope
combat
stagnation
during
iterations
ADN
refines
convergence
around
optimal
individuals
CAGWOA's
performance
is
validated
through
experiments
various
benchmark
function
test
sets.
Furthermore,
apply
CAGWOA
alongside
similar
model
comparative
lung
X-ray
images
infected
patients.
results
demonstrate
superiority,
better
detail
preservation,
clear
boundaries,
adaptability
across
different
levels.
Intelligence-Based Medicine,
Год журнала:
2024,
Номер
10, С. 100156 - 100156
Опубликована: Янв. 1, 2024
Coronavirus
disease
2019
(COVID-19)
has
become
a
pandemic
all
over
the
world
and
spread
rapidly.
To
distinguish
between
infected
non-infected
areas,
there
is
critical
need
for
segmentation
methods
that
can
identify
areas
from
Chest
Computed
Tomography
(CT)
scans.
In
recent
years,
deep
learning
most
widely
used
approach
medical
image
segmentation,
including
identification
of
in
CT
We
propose
an
encoder-decoder
based
on
U-NET
architecture
segmenting
MedSeg
dataset,
which
contains
lung
study
effect
input
dimensions
model's
output
results,
we
gave
images
with
224
×
224,
256
256,
512
as
inputs
to
model.
The
results
showed
achieved
higher
compared
512,
dicecoef
81.36,
accuracy
87.65,
sensitivity
84.71,
specificity
88.35.
Additionally,
proposed
model
highest
number
correct
answers
previous
U-net
methods.
be
applied
effective
screening
tool
help
primary
service
staff
better
refer
suspected
patients
specialists.
