Traditional
treatments
do
not
work
on
33%
of
epileptic
patients.Warning
devices
employing
seizure
prediction
or
forecasting
algorithms
could
bring
patients
a
newfound
quality
life.
These
would
attempt
to
detect
the
preictal
period,
transitional
moment
between
regular
brain
activity
and
seizure,
warn
user.
Several
past
methodologies
have
been
developed,
triggering
an
alarm
when
detecting
but
few
clinically
applicable.
Recent
studies
suggested
paradigm
change
that
takes
probabilistic
approach
instead
crisp
one
prediction.
The
is
substituted
by
constant
risk
assessment
analysis.
To
best
our
knowledge,
no
direct
comparison
using
same
database
has
made.
This
paper
explores
capable
compares
them
with
ones.
Using
data
from
EPILEPSIAE
database,
we
developed
several
patient-specific
different
classifiers
(a
Logistic
Regression,
15
Support
Vector
Machines
ensemble,
Shallow
Neural
Networks
ensemble).
Results
show
increase
sensitivity
in
relative
up
146%
number
displaying
improvement
over
chance
200%.
results
suggest
may
be
more
suitable
for
warning
than
PLoS ONE,
Год журнала:
2025,
Номер
20(3), С. e0320219 - e0320219
Опубликована: Март 25, 2025
Neuroinflammation
is
a
key
feature
of
human
neurodisease
including
neuropathy
and
neurodegenerative
disease
driven
by
the
activation
microglia,
immune
cells
nervous
system.
During
microglia
release
pro-inflammatory
cytokines
as
well
reactive
oxygen
species
(ROS)
that
can
drive
local
neuronal
glial
damage.
Phytocannabinoids
are
an
important
class
naturally
occurring
compounds
found
in
cannabis
plant
(
Cannabis
sativa
)
interact
with
body’s
endocannabinoid
receptor
Cannabidiol
(CBD)
prototype
phytocannabinoid
anti-inflammatory
properties
observed
animal
models.
We
measured
ROS
(HMC3)
using
CellROX,
fluorescent
dynamic
indicator.
tested
effect
CBD
on
level
presence
three
known
activators:
lipopolysaccharide
(LPS),
amyloid
beta
(A
β
42
),
immunodeficiency
virus
(HIV)
glycoprotein
(GP120).
Confocal
microscopy
images
within
were
coupled
to
deep
learning
model
convolutional
neural
network
(CNN)
predict
responses.
Our
study
demonstrates
platform
be
used
assessment
image
measure.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Янв. 3, 2024
Abstract
Almost
one-third
of
epileptic
patients
fail
to
achieve
seizure
control
through
anti-epileptic
drug
administration.
In
the
scarcity
completely
controlling
a
patient’s
epilepsy,
prediction
plays
significant
role
in
clinical
management
and
providing
new
therapeutic
options
such
as
warning
or
intervention
devices.
Seizure
algorithms
aim
identify
preictal
period
that
Electroencephalogram
(EEG)
signals
can
capture.
However,
this
is
associated
with
substantial
heterogeneity,
varying
among
even
between
seizures
from
same
patient.
The
present
work
proposes
patient-specific
algorithm
using
post-processing
techniques
explore
existence
set
chronological
events
brain
activity
precedes
seizures.
study
was
conducted
37
Temporal
Lobe
Epilepsy
(TLE)
EPILEPSIAE
database.
designed
methodology
combines
univariate
linear
features
classifier
based
on
Support
Vector
Machines
(SVM)
two
handle
pre-seizure
temporality
an
easily
explainable
way,
employing
knowledge
network
theory.
Chronological
Firing
Power
approach,
we
considered
sequence
three
separated
time.
Cumulative
assumed
overlapping
events.
These
methodologies
were
compared
approach
typical
machine
learning
pipeline.
We
Prediction
horizon
(SPH)
5
mins
analyzed
several
values
for
Occurrence
Period
(SOP)
duration,
10
55
mins.
Our
results
showed
may
improve
performance.
This
strategy
performed
above
chance
62%
patients,
whereas
only
validated
49%
its
models.
Epilepsia Open,
Год журнала:
2023,
Номер
8(2), С. 285 - 297
Опубликована: Апрель 19, 2023
Many
state-of-the-art
methods
for
seizure
prediction,
using
the
electroencephalogram,
are
based
on
machine
learning
models
that
black
boxes,
weakening
trust
of
clinicians
in
them
high-risk
decisions.
Seizure
prediction
concerns
a
multidimensional
time-series
problem
performs
continuous
sliding
window
analysis
and
classification.
In
this
work,
we
make
critical
review
which
explanations
increase
models'
decisions
predicting
seizures.
We
developed
three
methodologies
to
explore
their
explainability
potential.
These
contain
different
levels
model
transparency:
logistic
regression,
an
ensemble
15
support
vector
machines,
convolutional
neural
networks.
For
each
methodology,
evaluated
quasi-prospectively
performance
40
patients
(testing
data
comprised
2055
hours
104
seizures).
selected
with
good
poor
explain
Then,
grounded
theory,
how
these
helped
specialists
(data
scientists
working
epilepsy)
understand
obtained
dynamics.
four
lessons
better
communication
between
clinicians.
found
goal
is
not
system's
but
improve
system
itself.
Model
transparency
most
significant
factor
explaining
decision
prediction.
Even
when
intuitive
features,
it
hard
brain
dynamics
relationship
models.
achieve
understanding
by
developing,
parallel,
several
systems
explicitly
deal
signal
changes
help
develop
complete
formulation.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Март 7, 2024
Epilepsy
affects
around
1%
of
the
population
worldwide.
Anti-epileptic
drugs
are
an
excellent
option
for
controlling
seizure
occurrence
but
do
not
work
one-third
patients.
Warning
devices
employing
prediction
or
forecasting
algorithms
could
bring
patients
new-found
comfort
and
quality
life.
These
would
attempt
to
detect
a
seizure's
preictal
period,
transitional
moment
between
regular
brain
activity
seizure,
relay
this
information
user.
Over
years,
many
studies
using
Electroencephalogram-based
methodologies
have
been
developed,
triggering
alarm
when
detecting
period.
Recent
suggested
shift
in
view
from
forecasting.
Seizure
takes
probabilistic
approach
problem
question
instead
crisp
prediction.
In
field
study,
triggered
symbolize
detection
period
is
substituted
by
constant
risk
assessment
analysis.
The
present
aims
explore
capable
establish
comparison
with
results.
Using
40
EPILEPSIAE
database,
we
developed
several
patient-specific
different
classifiers
(a
Logistic
Regression,
15
Support
Vector
Machines
ensemble,
Shallow
Neural
Networks
ensemble).
Results
show
increase
sensitivity
relative
up
146%
number
that
displayed
improvement
over
chance
300%.
results
suggest
methodology
may
be
more
suitable
warning
than
one.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Апрель 8, 2024
Abstract
Seizure
prediction
remains
a
challenge,
with
approximately
30%
of
patients
unresponsive
to
conventional
treatments.
Addressing
this
issue
is
crucial
for
improving
patients’
quality
life,
as
timely
intervention
can
mitigate
the
impact
seizures.
In
research
field,
it
critical
identify
preictal
interval,
transition
from
regular
brain
activity
seizure.
While
previous
studies
have
explored
various
Electroencephalogram
(EEG)
based
methodologies
prediction,
few
been
clinically
applicable.
Recent
underlined
dynamic
nature
EEG
data,
characterised
by
data
changes
time,
known
concept
drifts,
highlighting
need
automated
methods
detect
and
adapt
these
changes.
study,
we
investigate
effectiveness
automatic
drift
adaptation
in
seizure
prediction.
Three
patient-specific
approaches
10-minute
horizon
are
compared:
algorithm
incorporating
window
adjustment
method
optimising
performance
Support
Vector
Machines
(Backwards-Landmark
Window),
data-batch
(seizures)
selection
using
logistic
regression
(Seizure-batch
Regression),
integration
classifiers
(Dynamic
Weighted
Ensemble).
These
incorporate
retraining
process
after
each
use
combination
univariate
linear
features
SVM
classifiers.
The
Firing
Power
was
used
post-processing
technique
generate
alarms
before
were
compared
control
approach
on
typical
machine
learning
pipeline,
considering
group
37
Temporal
Lobe
Epilepsy
EPILEPSIAE
database.
best-performing
Window)
achieved
results
0.75
±
0.33
sensitivity
1.03
1.00
false
positive
rate
per
hour.
This
new
strategy
performed
above
chance
89%
surrogate
predictor,
whereas
only
validated
46%.
Brain Sciences,
Год журнала:
2024,
Номер
14(4), С. 306 - 306
Опубликована: Март 25, 2024
Epilepsy
is
a
neurological
disease
with
one
of
the
highest
rates
incidence
worldwide.
Although
EEG
crucial
tool
for
its
diagnosis,
manual
detection
epileptic
seizures
time
consuming.
Automated
methods
are
needed
to
streamline
this
process;
although
there
already
several
works
that
have
achieved
this,
process
by
which
it
executed
remains
black
box
prevents
understanding
ways
in
machine
learning
algorithms
make
their
decisions.
A
state-of-the-art
deep
model
seizure
and
three
databases
were
chosen
study.
The
developed
models
trained
evaluated
under
different
conditions
(i.e.,
distinct
levels
overlap
among
data
windows).
classifiers
best
performance
selected,
then
Shapley
Additive
Explanations
(SHAPs)
Local
Interpretable
Model-Agnostic
(LIMEs)
employed
estimate
importance
value
each
channel
Spearman’s
rank
correlation
coefficient
was
computed
between
features
signals
values.
results
show
database
training
may
affect
classifier’s
performance.
most
significant
accuracy
0.84,
0.73,
0.64
CHB-MIT,
Siena,
TUSZ
datasets,
respectively.
In
addition,
displayed
negligible
or
low
Finally,
concluded
values
(generated
SHAP
LIME)
been
absent
even
high-performance
models.
IEEE Transactions on Biomedical Engineering,
Год журнала:
2024,
Номер
71(8), С. 2341 - 2351
Опубликована: Фев. 21, 2024
Seizure
prediction
is
a
promising
solution
to
improve
the
quality
of
life
for
drug-resistant
patients,
which
concerns
nearly
30%
patients
with
epilepsy.
The
present
study
aimed
ascertain
impact
incorporating
sleep-wake
information
in
seizure
prediction.
Frontiers in Neuroscience,
Год журнала:
2024,
Номер
18
Опубликована: Июль 15, 2024
Epilepsy
affects
1%
of
the
global
population,
with
approximately
one-third
patients
resistant
to
anti-seizure
medications
(ASMs),
posing
risks
physical
injuries
and
psychological
issues.
Seizure
prediction
algorithms
aim
enhance
quality
life
for
these
individuals
by
providing
timely
alerts.
This
study
presents
a
patient-specific
seizure
algorithm
applied
diverse
databases
(EPILEPSIAE,
CHB-MIT,
AES,
Ecosystem).
The
proposed
undergoes
standardized
framework,
including
data
preprocessing,
feature
extraction,
training,
testing,
postprocessing.
Various
necessitate
adaptations
in
algorithm,
considering
differences
availability
characteristics.
exhibited
variable
performance
across
databases,
taking
into
account
sensitivity,
FPR/h,
specificity,
AUC
score.
distinguishes
between
sample-based
approaches,
which
often
yield
better
results
disregarding
temporal
aspect
seizures,
alarm-based
simulate
real-life
conditions
but
produce
less
favorable
outcomes.
Statistical
assessment
reveals
challenges
surpassing
chance
levels,
emphasizing
rarity
events.
Comparative
analyses
existing
studies
highlight
complexity
assessments,
given
methodologies
dataset
variations.
Rigorous
aiming
outcomes,
importance
realistic
assumptions
comprehensive,
long-term,
systematically
structured
datasets
future
research.