CAAI Transactions on Intelligence Technology,
Год журнала:
2024,
Номер
9(4), С. 837 - 849
Опубликована: Апрель 8, 2024
Abstract
Convolutional
neural
networks
depend
on
deep
network
architectures
to
extract
accurate
information
for
image
super‐resolution.
However,
obtained
of
these
convolutional
cannot
completely
express
predicted
high‐quality
images
complex
scenes.
A
dynamic
super‐resolution
(DSRNet)
is
presented,
which
contains
a
residual
enhancement
block,
wide
feature
refinement
block
and
construction
block.
The
composed
enhanced
architecture
facilitate
hierarchical
features
To
enhance
robustness
model
scenes,
achieves
learn
more
robust
applicability
an
varying
prevent
interference
components
in
utilises
stacked
accurately
features.
Also,
learning
operation
embedded
the
long‐term
dependency
problem.
Finally,
responsible
reconstructing
images.
Designed
heterogeneous
can
not
only
richer
structural
information,
but
also
be
lightweight,
suitable
mobile
digital
devices.
Experimental
results
show
that
our
method
competitive
terms
performance,
recovering
time
complexity.
code
DSRNet
at
https://github.com/hellloxiaotian/DSRNet
.
International Journal of Machine Learning and Cybernetics,
Год журнала:
2024,
Номер
15(9), С. 3579 - 3597
Опубликована: Март 5, 2024
Abstract
Serious
consequences
due
to
brain
tumors
necessitate
a
timely
and
accurate
diagnosis.
However,
obstacles
such
as
suboptimal
imaging
quality,
issues
with
data
integrity,
varying
tumor
types
stages,
potential
errors
in
interpretation
hinder
the
achievement
of
precise
prompt
diagnoses.
The
rapid
identification
plays
pivotal
role
ensuring
patient
safety.
Deep
learning-based
systems
hold
promise
aiding
radiologists
make
diagnoses
swiftly
accurately.
In
this
study,
we
present
an
advanced
deep
learning
approach
based
on
Swin
Transformer.
proposed
method
introduces
novel
Hybrid
Shifted
Windows
Multi-Head
Self-Attention
module
(HSW-MSA)
along
rescaled
model.
This
enhancement
aims
improve
classification
accuracy,
reduce
memory
usage,
simplify
training
complexity.
Residual-based
MLP
(ResMLP)
replaces
traditional
Transformer,
thereby
improving
speed,
parameter
efficiency.
We
evaluate
Proposed-Swin
model
publicly
available
MRI
dataset
four
classes,
using
only
test
data.
Model
performance
is
enhanced
through
application
transfer
augmentation
techniques
for
efficient
robust
training.
achieves
remarkable
accuracy
99.92%,
surpassing
previous
research
models.
underscores
effectiveness
Transformer
HSW-MSA
ResMLP
improvements
innovative
diagnostic
offering
support
diagnosis,
ultimately
outcomes
reducing
risks.
Scientific Reports,
Год журнала:
2025,
Номер
15(1)
Опубликована: Янв. 10, 2025
Brain
tumors
present
a
significant
global
health
challenge,
and
their
early
detection
accurate
classification
are
crucial
for
effective
treatment
strategies.
This
study
presents
novel
approach
combining
lightweight
parallel
depthwise
separable
convolutional
neural
network
(PDSCNN)
hybrid
ridge
regression
extreme
learning
machine
(RRELM)
accurately
classifying
four
types
of
brain
(glioma,
meningioma,
no
tumor,
pituitary)
based
on
MRI
images.
The
proposed
enhances
the
visibility
clarity
tumor
features
in
images
by
employing
contrast-limited
adaptive
histogram
equalization
(CLAHE).
A
PDSCNN
is
then
employed
to
extract
relevant
tumor-specific
patterns
while
minimizing
computational
complexity.
RRELM
model
proposed,
enhancing
traditional
ELM
improved
performance.
framework
compared
with
various
state-of-the-art
models
terms
accuracy,
parameters,
layer
sizes.
achieved
remarkable
average
precision,
recall,
accuracy
values
99.35%,
99.30%,
99.22%,
respectively,
through
five-fold
cross-validation.
PDSCNN-RRELM
outperformed
pseudoinverse
(PELM)
exhibited
superior
introduction
led
enhancements
performance
parameters
sizes
those
models.
Additionally,
interpretability
was
demonstrated
using
Shapley
Additive
Explanations
(SHAP),
providing
insights
into
decision-making
process
increasing
confidence
real-world
diagnosis.
