Sensors,
Год журнала:
2024,
Номер
24(23), С. 7710 - 7710
Опубликована: Дек. 2, 2024
This
work
aims
to
develop
a
novel
convolutional
neural
network
(CNN)
named
ResNet50*
detect
various
gastrointestinal
diseases
using
new
ResNet50*-based
deep
feature
engineering
model
with
endoscopy
images.
The
novelty
of
this
is
the
development
ResNet50*,
variant
ResNet
model,
featuring
convolution-based
residual
blocks
and
pooling-based
attention
mechanism
similar
PoolFormer.
Using
image
dataset
was
trained,
an
explainable
(DFE)
developed.
DFE
comprises
four
primary
stages:
(i)
extraction,
(ii)
iterative
selection,
(iii)
classification
shallow
classifiers,
(iv)
information
fusion.
self-organizing,
producing
14
different
outcomes
(8
classifier-specific
6
voted)
selecting
most
effective
result
as
final
decision.
During
heatmaps
are
identified
gradient-weighted
class
activation
mapping
(Grad-CAM)
features
derived
from
these
regions
via
global
average
pooling
layer
pretrained
ResNet50*.
Four
selectors
employed
in
selection
stage
obtain
distinct
vectors.
classifiers
k-nearest
neighbors
(kNN)
support
vector
machine
(SVM)
used
produce
specific
outcomes.
Iterative
majority
voting
voted
top
determined
by
greedy
algorithm
based
on
accuracy.
presented
trained
augmented
version
Kvasir
dataset,
its
performance
tested
Kvasir,
2,
wireless
capsule
(WCE)
curated
colon
disease
datasets.
Our
proposed
demonstrated
accuracy
more
than
92%
for
all
three
datasets
remarkable
99.13%
WCE
dataset.
These
findings
affirm
superior
ability
confirm
generalizability
developed
architecture,
showing
consistent
across
IEEE Access,
Год журнала:
2024,
Номер
12, С. 113972 - 113987
Опубликована: Янв. 1, 2024
Early
detection
of
Gastrointestinal
(GI)
tract
diseases
is
essential
for
effective
healthcare
management,
treatment,
and
prevention,
ultimately
lowering
morbidity
mortality
rates
worldwide.
Current
classification
models
lack
spatial
feature
arrangement
consideration,
diminishing
discriminative
power
leading
to
misdiagnosis
esophagitis
ulcerative
colitis
due
overlapping
visual
characteristics
with
other
GI
diseases.
Hence,
a
novel
Hierarchical
Spatio
Pyramid
TranfoNet
featuring
Spatial
Transformer
Network
(STN)
pyramid
pooling
introduced,
which
enhances
in
distinguishing
between
disease
characteristics.
Enhancing
Dyed
Lifted
Polyps
(DLP)
Resection
Margins
(DRM)
endoscopy
images
critical
precise
gastrointestinal
diagnosis,
tackling
challenges
posed
by
complexity
inter-class
confounders.
PitTree
Fusion
Algorithm,
combining
Minimum
Spanning
Tree
(MST)
analysis
Kudo's
pit
pattern
introduced
accurately
locate
differentiate
normal
tissue
from
dyed
regions
like
DLPs
DRMs
images.
Then,
Efficient-CondConv
SwishNet
enhance
extracting
informative
features
endoscopic
images,
utilizing
EfficientNet-CondConv
Swish
activation.
After
classification,
heatmaps
highlighting
influential
are
produced
via
gradient-weighted
class
activation
mapping,
or
Grad-CAM,
provides
information
about
decisions.
The
results
show
that
the
suggested
model
outperforms
current
showing
increased
accuracy,
precision,
recall,
sensitivity,
specificity,
F1
score,
reduced
loss
rate.
PLoS ONE,
Год журнала:
2024,
Номер
19(11), С. e0310721 - e0310721
Опубликована: Ноя. 6, 2024
Non-linear
and
non-stationary
signals
are
analyzed
processed
in
the
time-frequency
(TF)
domain
due
to
interpretation
simplicity.
Wigner-Ville
distribution
(WVD)
delivers
a
very
sharp
resolution
of
TF
domain.
However,
cross-terms
occur
between
true
frequency
modes
their
bilinear
nature.
Masked
WVD
reduces
by
multiplying
representation
(TFR)
obtained
from
with
TFR
same
signal
another
method,
while
S-transform
(ST)
is
linear
analysis
method
that
combines
advantages
short-time
Fourier
transform
(STFT)
wavelet
(WT).
This
paper
investigated
masking
both
original
modified
STs
compare
cross-term
reduction
results.
Moreover,
additional
parameters
integrated
into
ST
deliver
better
and,
consequently,
more
satisfactory
reduction.
these
must
be
carefully
optimized
expert
users
respective
application
fields.
Sensors,
Год журнала:
2024,
Номер
24(23), С. 7710 - 7710
Опубликована: Дек. 2, 2024
This
work
aims
to
develop
a
novel
convolutional
neural
network
(CNN)
named
ResNet50*
detect
various
gastrointestinal
diseases
using
new
ResNet50*-based
deep
feature
engineering
model
with
endoscopy
images.
The
novelty
of
this
is
the
development
ResNet50*,
variant
ResNet
model,
featuring
convolution-based
residual
blocks
and
pooling-based
attention
mechanism
similar
PoolFormer.
Using
image
dataset
was
trained,
an
explainable
(DFE)
developed.
DFE
comprises
four
primary
stages:
(i)
extraction,
(ii)
iterative
selection,
(iii)
classification
shallow
classifiers,
(iv)
information
fusion.
self-organizing,
producing
14
different
outcomes
(8
classifier-specific
6
voted)
selecting
most
effective
result
as
final
decision.
During
heatmaps
are
identified
gradient-weighted
class
activation
mapping
(Grad-CAM)
features
derived
from
these
regions
via
global
average
pooling
layer
pretrained
ResNet50*.
Four
selectors
employed
in
selection
stage
obtain
distinct
vectors.
classifiers
k-nearest
neighbors
(kNN)
support
vector
machine
(SVM)
used
produce
specific
outcomes.
Iterative
majority
voting
voted
top
determined
by
greedy
algorithm
based
on
accuracy.
presented
trained
augmented
version
Kvasir
dataset,
its
performance
tested
Kvasir,
2,
wireless
capsule
(WCE)
curated
colon
disease
datasets.
Our
proposed
demonstrated
accuracy
more
than
92%
for
all
three
datasets
remarkable
99.13%
WCE
dataset.
These
findings
affirm
superior
ability
confirm
generalizability
developed
architecture,
showing
consistent
across