Sensors,
Год журнала:
2023,
Номер
23(4), С. 2061 - 2061
Опубликована: Фев. 11, 2023
Adaptive
machine
learning
has
increasing
importance
due
to
its
ability
classify
a
data
stream
and
handle
the
changes
in
distribution.
Various
resources,
such
as
wearable
sensors
medical
devices,
can
generate
with
an
imbalanced
distribution
of
classes.
Many
popular
oversampling
techniques
have
been
designed
for
batch
rather
than
continuous
stream.
This
work
proposes
self-adjusting
window
improve
adaptive
classification
based
on
minimizing
cluster
distortion.
It
includes
two
models;
first
chooses
only
previous
instances
that
preserve
coherence
current
chunk’s
samples.
The
second
model
relaxes
strict
filter
by
excluding
examples
last
chunk.
Both
models
include
generating
synthetic
points
actual
points.
evaluation
proposed
using
Siena
EEG
dataset
showed
their
performance
several
classifiers.
best
results
obtained
Random
Forest
which
Sensitivity
reached
96.83%
Precision
99.96%.
Heliyon,
Год журнала:
2024,
Номер
10(11), С. e31827 - e31827
Опубликована: Май 23, 2024
Epilepsy
is
one
of
the
most
common
brain
disorders,
and
seizures
epilepsy
have
severe
adverse
effects
on
patients.
Real-time
seizure
detection
using
electroencephalography
(EEG)
signals
an
important
research
area
aimed
at
improving
diagnosis
treatment
epilepsy.
This
paper
proposed
a
real-time
approach
based
EEG
signal
for
detecting
STFT
Google-net
convolutional
neural
network
(CNN).
The
CHB-MIT
database
was
used
to
evaluate
performance,
received
results
97.74
%
in
accuracy,
98.90
sensitivity,
1.94
false
positive
rate.
Additionally,
method
implemented
manner
sliding
window
technique.
processing
time
just
0.02
s
every
2-s
episode
achieved
average
9.85-
second
delay
each
onset.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Фев. 20, 2024
Abstract
Brain
disorders
pose
a
substantial
global
health
challenge,
persisting
as
leading
cause
of
mortality
worldwide.
Electroencephalogram
(EEG)
analysis
is
crucial
for
diagnosing
brain
disorders,
but
it
can
be
challenging
medical
practitioners
to
interpret
complex
EEG
signals
and
make
accurate
diagnoses.
To
address
this,
our
study
focuses
on
visualizing
in
format
easily
understandable
by
professionals
deep
learning
algorithms.
We
propose
novel
time–frequency
(TF)
transform
called
the
Forward–Backward
Fourier
(FBFT)
utilize
convolutional
neural
networks
(CNNs)
extract
meaningful
features
from
TF
images
classify
disorders.
introduce
concept
eye-naked
classification,
which
integrates
domain-specific
knowledge
clinical
expertise
into
classification
process.
Our
demonstrates
effectiveness
FBFT
method,
achieving
impressive
accuracies
across
multiple
using
CNN-based
classification.
Specifically,
we
achieve
99.82%
epilepsy,
95.91%
Alzheimer’s
disease
(AD),
85.1%
murmur,
100%
mental
stress
Furthermore,
context
naked-eye
78.6%,
71.9%,
82.7%,
91.0%
AD,
stress,
respectively.
Additionally,
incorporate
mean
correlation
coefficient
(mCC)
based
channel
selection
method
enhance
accuracy
further.
By
combining
these
innovative
approaches,
enhances
visualization
signals,
providing
with
deeper
understanding
images.
This
research
has
potential
bridge
gap
between
image
visual
interpretation,
better
detection
improved
patient
care
field
neuroscience.
International Journal of Neural Systems,
Год журнала:
2024,
Номер
34(10)
Опубликована: Июнь 21, 2024
Seizure
is
a
common
neurological
disorder
that
usually
manifests
itself
in
recurring
seizure,
and
these
seizures
can
have
serious
impact
on
person's
life
health.
Therefore,
early
detection
diagnosis
of
seizure
crucial.
In
order
to
improve
the
efficiency
this
paper
proposes
new
method,
which
based
discrete
wavelet
transform
(DWT)
multi-channel
long-
short-term
memory-like
spiking
neural
P
(LSTM-SNP)
model.
First,
signal
decomposed
into
5
levels
by
using
DWT
obtain
features
components
at
different
frequencies,
series
time-frequency
coefficients
are
extracted.
Then,
used
train
LSTM-SNP
model
perform
detection.
The
proposed
method
achieves
high
accuracy
CHB-MIT
dataset:
98.25%
accuracy,
98.22%
specificity
97.59%
sensitivity.
This
indicates
epilepsy
show
competitive
performance.
Sensors,
Год журнала:
2023,
Номер
24(1), С. 77 - 77
Опубликована: Дек. 22, 2023
Epilepsy
is
a
chronic
neurological
disease
associated
with
abnormal
neuronal
activity
in
the
brain.
Seizure
detection
algorithms
are
essential
reducing
workload
of
medical
staff
reviewing
electroencephalogram
(EEG)
records.
In
this
work,
we
propose
novel
automatic
epileptic
EEG
method
based
on
Stockwell
transform
and
Transformer.
First,
S-transform
applied
to
original
segments,
acquiring
accurate
time-frequency
representations.
Subsequently,
obtained
matrices
grouped
into
different
rhythm
blocks
compressed
as
vectors
these
sub-bands.
After
that,
feature
fed
Transformer
network
for
selection
classification.
Moreover,
series
post-processing
methods
were
introduced
enhance
efficiency
system.
When
evaluating
public
CHB-MIT
database,
proposed
algorithm
achieved
an
accuracy
96.15%,
sensitivity
96.11%,
specificity
96.38%,
precision
96.33%,
area
under
curve
(AUC)
0.98
segment-based
experiments,
along
96.57%,
false
rate
0.38/h,
delay
20.62
s
event-based
experiments.
These
outstanding
results
demonstrate
feasibility
implementing
seizure
future
clinical
applications.
