Journal of Statistical Computation and Simulation,
Год журнала:
2024,
Номер
unknown, С. 1 - 24
Опубликована: Окт. 18, 2024
Classification
in
high
dimensions
has
been
highlighted
for
the
past
two
decades
since
Fisher's
linear
discriminant
analysis
(LDA)
is
not
optimal
a
smaller
sample
size
n
comparing
number
of
covariates
p,
i.e.
p>n,
which
mostly
due
to
singularity
covariance
matrix.
Rather
than
modifying
how
estimate
and
mean
vector
constructing
classifier,
we
build
types
high-dimensional
classifiers
using
data
splitting,
single
splitting
(SDS)
multiple
(MDS).
Moreover,
introduce
weighted
version
MDS
classifier
that
improves
classification
performance
as
illustrated
numerical
studies.
Each
split
sets
compared
so
LDA
applicable,
results
can
be
combined
with
respect
minimizing
misclassification
rate.
We
present
theoretical
justification
backing
up
our
proposed
methods
by
rates
dimension.
also
conduct
wide
range
simulations
analyse
four
microarray
sets,
demonstrates
outperform
some
existing
or
at
least
yield
comparable
performances.
Journal of Innovative Image Processing,
Год журнала:
2025,
Номер
7(2), С. 315 - 332
Опубликована: Июнь 1, 2025
In
recent
decades,
Diabetic
Macular
Edema
(DME)
has
emerged
as
a
significant
cause
of
vision
loss
among
diabetic
patients
due
to
retinal
fluid
leakage.
To
address
this
challenge,
reliable
and
efficient
diagnostic
methods
are
essential.
The
proposed
methodology
aims
facilitate
early
detection
through
multi-stage
process,
including
feature
extraction,
selection,
classification.For
we
introduce
the
H2A2Net
model,
which
incorporates
Dense
Spectral-Spatial
Module
(DSSM)
that
employs
3D
convolutional
DenseNet-inspired
layers
extract
spectral-spatial
features.
This
is
complemented
by
Hybrid
Resolution
(HRM)
designed
achieve
fine
spatial
detail
multi-scale
process.
Additionally,
Double
Attention
(DAM)
implemented
capture
global
cross-channel
interactions,
utilizing
both
pixel-wise
channel-wise
attenuation.
Feature
selection
conducted
using
Cuckoo
Search
Spider
Monkey
Optimization
(CSSMO),
effectively
processes
local
searches
enable
high-value
classification
phase,
hybrid
AdaBoost-Backpropagation
Neural
Network
(BPNN)
model
employed,
where
BPNNs
function
weak
classifiers
whose
outputs
iteratively
boosted
create
strong
ensemble.
Experimental
results
on
CUHK
dataset
demonstrate
method
achieves
an
accuracy
97.4%,
recall
97.6%,
specificity
97%,
F1-score
98%.
These
outcomes
surpass
those
existing
state-of-the-art
methods,
indicating
approach
offers
enhanced
robustness
efficiency
for
DME
classification.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Июль 17, 2024
Abstract
The
microarray
gene
expression
data
poses
a
tremendous
challenge
due
to
their
curse
of
dimensionality
problem.
sheer
volume
features
far
surpasses
available
samples,
leading
overfitting
and
reduced
classification
accuracy.
Thus
the
must
be
with
efficient
feature
extraction
methods
reduce
extract
meaningful
information
enhance
accuracy
interpretability.
In
this
research,
we
discover
uniqueness
applying
STFT
(Short
Term
Fourier
Transform),
LASSO
(Least
Absolute
Shrinkage
Selection
Operator),
EHO
(Elephant
Herding
Optimisation)
for
extracting
significant
from
lung
cancer
reducing
database.
is
performed
using
following
classifiers:
Gaussian
Mixture
Model
(GMM),
Particle
Swarm
Optimization
(PSO)
GMM,
Detrended
Fluctuation
Analysis
(DFA),
Naive
Bayes
classifier
(NBC),
Firefly
Support
Vector
Machine
Radial
Basis
Kernel
(SVM-RBF)
Flower
Pollination
(FPO)
GMM.
FPO-GMM
attained
highest
in
range
96.77,
an
F1
score
97.5,
MCC
0.92
Kappa
0.92.
reported
results
underline
significance
utilizing
STFT,
LASSO,
data.
These
methodologies
also
help
improved
early
diagnosis
enhanced
Sensors,
Год журнала:
2024,
Номер
24(8), С. 2383 - 2383
Опубликована: Апрель 9, 2024
Classification-based
myoelectric
control
has
attracted
significant
interest
in
recent
years,
leading
to
prosthetic
hands
with
advanced
functionality,
such
as
multi-grip
hands.
Thus
far,
high
classification
accuracies
have
been
achieved
by
increasing
the
number
of
surface
electromyography
(sEMG)
electrodes
or
adding
other
sensing
mechanisms.
While
many
prescribed
still
adopt
two-electrode
sEMG
systems,
detailed
studies
on
signal
processing
and
performance
are
lacking.
In
this
study,
nine
able-bodied
participants
were
recruited
perform
six
typical
hand
actions,
from
which
signals
two
acquired
using
a
Delsys
Trigno
Research+
acquisition
system.
Signal
machine
learning
algorithms,
specifically,
linear
discriminant
analysis
(LDA),
k-nearest
neighbors
(KNN),
support
vector
machines
(SVM),
used
study
accuracies.
Overall
accuracy
93
±
2%,
action-specific
97
F1-score
87
7%
achieved,
comparable
those
reported
multi-electrode
systems.
The
highest
SVM
algorithm
compared
LDA
KNN
algorithms.
A
logarithmic
relationship
between
features
was
revealed,
plateaued
at
five
features.
These
comprehensive
findings
may
potentially
contribute
strategies
for
commonly
systems
further
improve
functionality.
International Journal of Imaging Systems and Technology,
Год журнала:
2024,
Номер
34(5)
Опубликована: Сен. 1, 2024
ABSTRACT
Colorectal
adenocarcinoma,
the
most
prevalent
form
of
colon
cancer,
originates
in
glandular
structures
intestines,
presenting
histopathological
abnormalities
affected
tissues.
Accurate
gland
segmentation
is
crucial
for
identifying
these
potentially
fatal
abnormalities.
While
recent
methodologies
have
shown
success
segmenting
glands
benign
tissues,
their
efficacy
diminishes
when
applied
to
malignant
tissue
segmentation.
This
study
aims
develop
a
robust
learning
algorithm
using
convolutional
neural
network
(CNN)
segment
histology
images.
The
methodology
employs
CNN
based
on
U‐Net
architecture,
augmented
by
weighted
ensemble
that
integrates
DenseNet
169,
Inception
V3,
and
Efficientnet
B3
as
backbone
models.
Additionally,
segmented
boundaries
are
refined
watershed
algorithm.
Evaluation
Warwick‐QU
dataset
demonstrates
promising
results
model,
achieving
an
F1
score
0.928
0.913,
object
dice
coefficient
0.923
0.911,
Hausdorff
distances
38.97
33.76
test
sets
A
B,
respectively.
These
compared
with
outcomes
from
GlaS
challenge
(MICCAI
2015)
existing
research
findings.
Furthermore,
our
model
validated
publicly
available
named
LC25000,
visual
inspection
reveals
results,
further
validating
approach.
proposed
underscores
advantages
amalgamating
diverse
models,
highlighting
potential
techniques
enhance
tasks
beyond
individual
capabilities.
PLoS ONE,
Год журнала:
2024,
Номер
19(9), С. e0310748 - e0310748
Опубликована: Сен. 27, 2024
Brain
tumors
are
one
of
the
leading
diseases
imposing
a
huge
morbidity
rate
across
world
every
year.
Classifying
brain
accurately
plays
crucial
role
in
clinical
diagnosis
and
improves
overall
healthcare
process.
ML
techniques
have
shown
promise
classifying
based
on
medical
imaging
data
such
as
MRI
scans.
These
aid
detecting
planning
treatment
early,
improving
patient
outcomes.
However,
image
datasets
frequently
affected
by
significant
class
imbalance,
especially
when
benign
outnumber
malignant
number.
This
study
presents
an
explainable
ensemble-based
pipeline
for
tumor
classification
that
integrates
Dual-GAN
mechanism
with
feature
extraction
techniques,
specifically
designed
highly
imbalanced
data.
facilitates
generation
synthetic
minority
samples,
addressing
imbalance
issue
without
compromising
original
quality
Additionally,
integration
different
methods
capturing
precise
informative
features.
proposes
novel
deep
ensemble
(DeepEFE)
framework
surpasses
other
benchmark
learning
models
accuracy
98.15%.
focuses
achieving
high
while
prioritizing
stable
performance.
By
incorporating
Grad-CAM,
it
enhances
transparency
interpretability
research
identifies
most
relevant
contributing
parts
input
images
toward
accurate
outcomes
enhancing
reliability
proposed
pipeline.
The
significantly
improved
Precision,
Sensitivity
F1-Score
demonstrate
effectiveness
handling
accuracy.
Furthermore,
explainability
process
to
establish
reliable
model
classification,
encouraging
their
adoption
practice
promoting
trust
decision-making
processes.