Middle East Journal of Science,
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 18, 2024
This
study
investigates
the
effectiveness
of
MobileNetV2
transfer
learning
method
and
a
deep
based
Convolutional
Neural
Network
(CNN)
model
in
categorization
malignant
benign
skin
lesions
cancer
diagnosis.
Since
is
disease
that
can
be
cured
with
early
detection
but
fatal
if
delayed,
accurate
diagnosis
great
importance.
The
was
trained
architecture
performed
classification
task
high
accuracy
on
images
lesions.
Metrics
such
as
accuracy,
recall,
precision
F1
score
obtained
during
training
validation
processes
support
performance
model.
92.97%,
Recall
92.71%,
Precision
94.70%
93.47%.
results
show
CNN-based
reliable
effective
tool
for
diagnosis,
small
fluctuations
phase
require
further
data
hyperparameter
optimization
to
improve
generalization
ability
demonstrates
models
enhanced
offer
powerful
solution
medical
image
problems
have
potential
contribute
development
systems
healthcare
field.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Авг. 1, 2024
Abstract
Skin
cancer
is
a
lethal
disease,
and
its
early
detection
plays
pivotal
role
in
preventing
spread
to
other
body
organs
tissues.
Artificial
Intelligence
(AI)-based
automated
methods
can
play
significant
detection.
This
study
presents
an
AI-based
novel
approach,
termed
'DualAutoELM'
for
the
effective
identification
of
various
types
skin
cancers.
The
proposed
method
leverages
network
autoencoders,
comprising
two
distinct
autoencoders:
spatial
autoencoder
FFT
(Fast
Fourier
Transform)-autoencoder.
spatial-autoencoder
specializes
learning
features
within
input
lesion
images
whereas
FFT-autoencoder
learns
capture
textural
distinguishing
frequency
patterns
transformed
through
reconstruction
process.
use
attention
modules
at
levels
encoder
part
these
autoencoders
significantly
improves
their
discriminative
feature
capabilities.
An
Extreme
Learning
Machine
(ELM)
with
single
layer
feedforward
trained
classify
malignancies
using
characteristics
that
were
recovered
from
bottleneck
layers
autoencoders.
'HAM10000'
'ISIC-2017'
are
publicly
available
datasets
used
thoroughly
assess
suggested
approach.
experimental
findings
demonstrate
accuracy
robustness
technique,
AUC,
precision,
values
dataset
being
0.98,
97.68%
97.66%,
0.95,
86.75%
86.68%,
respectively.
highlights
possibility
approach
accurate
cancer.
The
increasing
prevalence
of
skin
diseases
necessitates
accurate
and
efficient
diagnostic
tools.
This
research
introduces
a
novel
disease
classification
model
leveraging
advanced
deep
learning
techniques.
proposed
architecture
combines
the
MobileNet-V2
backbone,
Squeeze-and-Excitation
(SE)
blocks,
Atrous
Spatial
Pyramid
Pooling
(ASPP),
Channel
Attention
Mechanism.
was
trained
on
four
diverse
datasets
such
as
PH2
dataset,
Skin
Cancer
MNIST:
HAM10000
DermNet.
ISIC
dataset.
Data
preprocessing
techniques,
including
image
resizing,
normalization,
played
crucial
role
in
optimizing
performance.
In
this
paper,
backbone
is
implemented
to
extract
hierarchical
features
from
preprocessed
dermoscopic
images.
multi-scale
contextual
information
fused
by
ASPP
for
generating
feature
map.
attention
mechanisms
contributed
significantly,
enhancing
extraction
ability
inter-channel
relationships
discriminative
power
features.
Finally,
output
map
converted
into
probability
distribution
through
softmax
function.
outperformed
several
baseline
models,
traditional
machine
approaches,
emphasizing
its
superiority
with
98.6%
overall
accuracy.
Its
competitive
performance
state-of-the-art
methods
positions
it
valuable
tool
assisting
dermatologists
early
classification.
study
also
identified
limitations
suggested
avenues
future
research,
model's
potential
practical
implementation
field
dermatology.
Big Data and Cognitive Computing,
Год журнала:
2025,
Номер
9(4), С. 97 - 97
Опубликована: Апрель 11, 2025
Skin
cancer,
particularly
melanoma,
is
one
of
the
leading
causes
cancer-related
deaths.
It
essential
to
detect
and
start
treatment
in
early
stages
for
it
be
effective
improve
survival
rates.
This
study
developed
evaluated
a
deep
learning-based
classification
model
classify
skin
lesion
images
as
benign
(non-cancerous)
malignant
(cancerous).
In
this
study,
we
used
ISIC
2016
dataset
train
segmentation
Kaggle
10,000
model.
We
applied
different
data
pre-processing
techniques
enhance
robustness
our
generate
binary
mask
with
corresponding
pre-processed
image
by
overlaying
its
edges
highlight
region,
before
feeding
transfer
learning,
using
ResNet-50
backbone
feedforward
network.
achieved
an
accuracy
92.80%,
precision
98.64%,
recall
86.80%.
From
have
found
that
integrating
learning
proper
improves
model’s
performance.
Future
work
will
focus
on
expanding
datasets
testing
more
architectures
performance
metrics
IJIIS International Journal of Informatics and Information Systems,
Год журнала:
2024,
Номер
7(2), С. 46 - 54
Опубликована: Март 31, 2024
Skin
cancer
is
a
type
of
that
can
cause
death,
where
skin
included
in
the
15
common
cancers
occur
Indonesia.
The
number
sufferers
was
around
6,170
cases
non-melanoma
and
1,392
melanoma
2018
Therefore,
research
related
to
classification
increasing.
This
done
as
an
initial
step
detecting
whether
lesion
be
said
cancerous
or
not.
deep
learning
approach
has
certainly
shown
promising
results
carrying
out
classification,
so
this
proposes
learning-based
method
used
for
classification.
proposed
involves
Convolutional
Neural
Networks
with
ISIC
2017
dataset.
models
are
InceptionV3,
EfficientNetB0,
ResNet50,
MobileNetV2,
NASNetMobile.
highest
accuracy
single
model
produced
reached
69.3%
using
MobileNetV2
model.
An
ensemble
combining
five
also
tested
compared
other
result
80.6%.
Computation,
Год журнала:
2023,
Номер
11(12), С. 246 - 246
Опубликована: Дек. 5, 2023
This
research
paper
presents
a
deep-learning
approach
to
early
detection
of
skin
cancer
using
image
augmentation
techniques.
We
introduce
two-stage
process
utilizing
geometric
and
generative
adversarial
network
(GAN)
differentiate
categories.
The
public
HAM10000
dataset
was
used
test
how
well
the
proposed
model
worked.
Various
pre-trained
convolutional
neural
(CNN)
models,
including
Xception,
Inceptionv3,
Resnet152v2,
EfficientnetB7,
InceptionresnetV2,
VGG19,
were
employed.
Our
demonstrates
an
accuracy
96.90%,
precision
97.07%,
recall
96.87%,
F1-score
96.97%,
surpassing
performance
other
state-of-the-art
methods.
also
discusses
use
Shapley
Additive
Explanations
(SHAP),
interpretable
technique
for
diagnosis,
which
can
help
clinicians
understand
reasoning
behind
diagnosis
improve
trust
in
system.
Overall,
method
promising
automated
that
could
patient
outcomes
reduce
healthcare
costs.
2020 International Seminar on Application for Technology of Information and Communication (iSemantic),
Год журнала:
2023,
Номер
unknown, С. 485 - 489
Опубликована: Сен. 16, 2023
The
field
of
dermatology
faces
considerable
challenges
when
it
comes
to
early
detection
skin
cancer.
Our
study
focused
on
using
different
datasets,
including
original
data,
augmented
and
SMOTE
oversampled
identify
dataset
consisted
images
lesions
from
the
MNIST
Skin
Cancer
(HAM
10000),
samples
both
cancerous
benign
cases
in
dataset.
We
employed
data
augmentation
expand
dataset's
size
increase
diversity
lesion
features.
Furthermore,
tackle
class
imbalance
dataset,
we
applied
oversampling
technique
generate
synthetic
for
under-represented
group.
With
original,
augmented,
trained
a
Convolutional
Neural
Network
(CNN)
model.
performance
model
was
evaluated
accuracy,
recall,
precision,
F1-score.
comparison
between
results
obtained
clearly
revealed
distinctions
performance.
findings
demonstrate
that
employing
can
significantly
enhance
efficacy
cancer
detection.