PeerJ Computer Science,
Год журнала:
2024,
Номер
10, С. e2587 - e2587
Опубликована: Дек. 19, 2024
Gastrointestinal
(GI)
disorders
are
common
and
often
debilitating
health
issues
that
affect
a
significant
portion
of
the
population.
Recent
advancements
in
artificial
intelligence,
particularly
computer
vision
algorithms,
have
shown
great
potential
detecting
classifying
medical
images.
These
algorithms
utilize
deep
convolutional
neural
network
architectures
to
learn
complex
spatial
features
images
make
predictions
for
similar
unseen
The
proposed
study
aims
assist
gastroenterologists
making
more
efficient
accurate
diagnoses
GI
patients
by
utilizing
its
two-phase
transfer
learning
framework
identify
diseases
from
endoscopic
Three
pre-trained
image
classification
models,
namely
Xception,
InceptionResNetV2,
VGG16,
fine-tuned
on
publicly
available
datasets
annotated
tract.
Additionally,
two
custom
networks
constructed
fully
trained
comparative
analysis
their
performance.
Four
different
tasks
examined
based
categories.
architecture
employing
InceptionResNetV2
achieves
most
consistent
generalized
performance
across
tasks,
yielding
accuracy
scores
85.7%
general
tract
(eight-category
classification),
97.6%
three-diseases
classification,
99.5%
polyp
identification
(binary
74.2%
binary
esophagitis
severity
results
indicate
effectiveness
clinical
use
enhance
diseases,
aiding
early
diagnosis
treatment.
Gastrointestinal
tract-related
cancers
pose
a
significant
health
burden,
with
high
mortality
rates.
In
order
to
detect
the
anomalies
of
gastrointestinal
tract
that
may
progress
cancer,
video
capsule
endoscopy
procedure
is
employed.
The
number
endoscopic
(
$$\mathcal
{VCE}$$
)
images
produced
per
examination
enormous,
which
necessitates
hours
analysis
by
clinicians.
Therefore,
there
pressing
need
for
automated
computer-aided
lesion
classification
techniques.
Computer-aided
systems
utilize
deep
learning
(DL)
techniques,
as
they
can
potentially
enhance
anomaly
detection
However,
most
DL
techniques
available
in
literature
utilizes
static
frames
purpose,
uses
only
spatial
information
image.
addition,
perform
binary
classification.
Thus,
presented
work
proposes
framework
multi-class
using
dynamic
images.
proposed
algorithm
combination
fractional
variational
model
and
model.
captures
estimating
optical
flow
color
maps.
Optical
maps
are
fed
training.
performs
task
localizes
region
interest
maximum
class
score.
inspired
Faster
RCNN
approach,
its
backbone
architecture
EfficientNet
B0.
achieves
average
AUC
value
0.98,
mAP
0.93,
0.878
balanced
accuracy
value.
Hence,
efficient
image
interest.
BMC Medical Informatics and Decision Making,
Год журнала:
2025,
Номер
25(1)
Опубликована: Март 31, 2025
Deep
learning
has
significantly
contributed
to
medical
imaging
and
computer-aided
diagnosis
(CAD),
providing
accurate
disease
classification
diagnosis.
However,
challenges
such
as
inter-
intra-class
similarities,
class
imbalance,
computational
inefficiencies
due
numerous
hyperparameters
persist.
This
study
aims
address
these
by
presenting
a
novel
deep-learning
framework
for
classifying
localizing
gastrointestinal
(GI)
diseases
from
wireless
capsule
endoscopy
(WCE)
images.
The
proposed
begins
with
dataset
augmentation
enhance
training
robustness.
Two
architectures,
Sparse
Convolutional
DenseNet201
Self-Attention
(SC-DSAN)
CNN-GRU,
are
fused
at
the
network
level
using
depth
concatenation
layer,
avoiding
costs
of
feature-level
fusion.
Bayesian
Optimization
(BO)
is
employed
dynamic
hyperparameter
tuning,
an
Entropy-controlled
Marine
Predators
Algorithm
(EMPA)
selects
optimal
features.
These
features
classified
Shallow
Wide
Neural
Network
(SWNN)
traditional
classifiers.
Experimental
evaluations
on
Kvasir-V1
Kvasir-V2
datasets
demonstrate
superior
performance,
achieving
accuracies
99.60%
95.10%,
respectively.
offers
improved
accuracy,
precision,
efficiency
compared
state-of-the-art
models.
addresses
key
in
GI
diagnosis,
demonstrating
its
potential
efficient
clinical
applications.
Future
work
will
explore
adaptability
additional
optimize
complexity
broader
deployment.
BMC Medical Imaging,
Год журнала:
2024,
Номер
24(1)
Опубликована: Ноя. 12, 2024
The
early
detection
and
diagnosis
of
gastrointestinal
tract
diseases,
such
as
ulcerative
colitis,
polyps,
esophagitis,
are
crucial
for
timely
treatment.
Traditional
imaging
techniques
often
rely
on
manual
interpretation,
which
is
subject
to
variability
may
lack
precision.
Current
methodologies
leverage
conventional
deep
learning
models
that,
while
effective
an
extent,
suffer
from
overfitting
generalization
issues
medical
image
datasets
due
the
intricate
subtle
variations
in
disease
manifestations.
These
typically
do
not
fully
utilize
potential
transfer
or
advanced
data
augmentation,
leading
less-than-optimal
performance,
especially
diverse
real-world
scenarios
where
high.
This
study
introduces
a
robust
model
using
EfficientNetB5
architecture
combined
with
sophisticated
augmentation
strategy.
tailored
high
details
present
images.
By
integrating
maximal
pooling
extensive
regularization,
aims
enhance
diagnostic
accuracy
reduce
overfitting.
proposed
achieved
test
98.89%,
surpassing
traditional
methods
by
incorporating
regularization
techniques.
application
horizontal
flipping
dynamic
scaling
during
training
significantly
improved
model's
ability
generalize,
evidenced
low-test
loss
0.230
precision
metrics
across
all
classes.
framework
demonstrates
superior
performance
automated
classification
diseases
data.
addressing
key
limitations
existing
through
innovative
techniques,
this
contributes
enhancement
processes
imaging,
potentially
more
accurate
interventions.
