Multimodal Emotion Recognition on RAVDESS Dataset Using Transfer Learning
Sensors,
Год журнала:
2021,
Номер
21(22), С. 7665 - 7665
Опубликована: Ноя. 18, 2021
Emotion
Recognition
is
attracting
the
attention
of
research
community
due
to
multiple
areas
where
it
can
be
applied,
such
as
in
healthcare
or
road
safety
systems.
In
this
paper,
we
propose
a
multimodal
emotion
recognition
system
that
relies
on
speech
and
facial
information.
For
speech-based
modality,
evaluated
several
transfer-learning
techniques,
more
specifically,
embedding
extraction
Fine-Tuning.
The
best
accuracy
results
were
achieved
when
fine-tuned
CNN-14
PANNs
framework,
confirming
training
was
robust
did
not
start
from
scratch
tasks
similar.
Regarding
recognizers,
framework
consists
pre-trained
Spatial
Transformer
Network
saliency
maps
images
followed
by
bi-LSTM
with
an
mechanism.
error
analysis
reported
frame-based
systems
could
present
some
problems
they
used
directly
solve
video-based
task
despite
domain
adaptation,
which
opens
new
line
discover
ways
correct
mismatch
take
advantage
embedded
knowledge
these
models.
Finally,
combination
two
modalities
late
fusion
strategy,
80.08%
RAVDESS
dataset
subject-wise
5-CV
evaluation,
classifying
eight
emotions.
revealed
carry
relevant
information
detect
users'
emotional
state
their
enables
improvement
performance.
Язык: Английский
A Proposal for Multimodal Emotion Recognition Using Aural Transformers and Action Units on RAVDESS Dataset
Applied Sciences,
Год журнала:
2021,
Номер
12(1), С. 327 - 327
Опубликована: Дек. 30, 2021
Emotion
recognition
is
attracting
the
attention
of
research
community
due
to
its
multiple
applications
in
different
fields,
such
as
medicine
or
autonomous
driving.
In
this
paper,
we
proposed
an
automatic
emotion
recognizer
system
that
consisted
a
speech
(SER)
and
facial
(FER).
For
SER,
evaluated
pre-trained
xlsr-Wav2Vec2.0
transformer
using
two
transfer-learning
techniques:
embedding
extraction
fine-tuning.
The
best
accuracy
results
were
achieved
when
fine-tuned
whole
model
by
appending
multilayer
perceptron
on
top
it,
confirming
training
was
more
robust
it
did
not
start
from
scratch
previous
knowledge
network
similar
task
adapt.
Regarding
recognizer,
extracted
Action
Units
videos
compared
performance
between
employing
static
models
against
sequential
models.
Results
showed
beat
narrow
difference.
Error
analysis
reported
visual
systems
could
improve
with
detector
high-emotional
load
frames,
which
opened
new
line
discover
ways
learn
videos.
Finally,
combining
these
modalities
late
fusion
strategy,
86.70%
RAVDESS
dataset
subject-wise
5-CV
evaluation,
classifying
eight
emotions.
demonstrated
carried
relevant
information
detect
users’
emotional
state
their
combination
allowed
final
performance.
Язык: Английский
AHANet: Adaptive Hybrid Attention Network for Alzheimer’s Disease Classification Using Brain Magnetic Resonance Imaging
T. Illakiya,
R. Karthik,
M. V. Siddharth
и другие.
Bioengineering,
Год журнала:
2023,
Номер
10(6), С. 714 - 714
Опубликована: Июнь 12, 2023
Alzheimer’s
disease
(AD)
is
a
progressive
neurological
problem
that
causes
brain
atrophy
and
affects
the
memory
thinking
skills
of
an
individual.
Accurate
detection
AD
has
been
challenging
research
topic
for
long
time
in
area
medical
image
processing.
Detecting
at
its
earliest
stage
crucial
successful
treatment
disease.
The
proposed
Adaptive
Hybrid
Attention
Network
(AHANet)
two
attention
modules,
namely
Enhanced
Non-Local
(ENLA)
Coordinate
Attention.
These
modules
extract
global-level
features
local-level
separately
from
Magnetic
Resonance
Imaging
(MRI),
thereby
boosting
feature
extraction
power
network.
ENLA
module
extracts
spatial
contextual
information
on
global
scale
while
also
capturing
important
long-range
dependencies.
captures
local
input
images.
It
embeds
positional
into
channel
mechanism
enhanced
extraction.
Moreover,
Feature
Aggregation
(AFA)
to
fuse
levels
effective
way.
As
result
incorporating
above
architectural
enhancements
DenseNet
architecture,
network
exhibited
better
performance
compared
existing
works.
was
trained
tested
ADNI
dataset,
yielding
classification
accuracy
98.53%.
Язык: Английский
Convolutional Neural Network-Based Parkinson Disease Classification Using SPECT Imaging Data
Mathematics,
Год журнала:
2022,
Номер
10(15), С. 2566 - 2566
Опубликована: Июль 23, 2022
In
this
paper,
we
used
the
single-photon
emission
computerized
tomography
(SPECT)
imaging
technique
to
visualize
deficiency
of
dopamine-generated
patterns
inside
brain.
These
are
establish
a
patient’s
disease
progression,
which
helps
distinguish
patients
into
different
categories.
Furthermore,
convolutional
neural
network
(CNN)
model
classify
based
on
dopamine
level
The
dataset
throughout
paper
is
Parkinson’s
progressive
markers
initiative
(PPMI)
dataset.
collected
was
pre-processed
and
data
amplification
performed
balance
imbalanced
A
CNN-based
defined
input
SPECT
images
four
motivation
behind
proposed
reduce
number
resources
consumed
while
maintaining
performance
classification
model.
This
will
help
healthcare
ecosystem
run
mobile
devices.
contains
14
layers
with
layers,
max-pool
flatten
dense
dimensions.
layer
classifies
categories,
including
PSD,
healthy
control,
scans
without
evidence
dopaminergic
deficit
(SWEDD),
GenReg
PSD
from
entire
dataset,
progression
using
images.
trained
large
58,692
for
training
11,738
validation,
7826
testing.
outperforms
models
surveyed
papers.
model’s
accuracy
0.889,
recall
0.9012,
precision
0.9104,
F1-score
0.9057.
Язык: Английский
Explainable Machine Learning with Pairwise Interactions for Predicting Conversion from Mild Cognitive Impairment to Alzheimer’s Disease Utilizing Multi-Modalities Data
Brain Sciences,
Год журнала:
2023,
Номер
13(11), С. 1535 - 1535
Опубликована: Окт. 31, 2023
Background:
Predicting
cognition
decline
in
patients
with
mild
cognitive
impairment
(MCI)
is
crucial
for
identifying
high-risk
individuals
and
implementing
effective
management.
To
improve
predicting
MCI-to-AD
conversion,
it
necessary
to
consider
various
factors
using
explainable
machine
learning
(XAI)
models
which
provide
interpretability
while
maintaining
predictive
accuracy.
This
study
used
the
Explainable
Boosting
Machine
(EBM)
model
multimodal
features
predict
conversion
of
MCI
AD
during
different
follow-up
periods
providing
interpretability.
Methods:
retrospective
case-control
conducted
data
obtained
from
ADNI
database,
records
1042
2006
2022
included.
The
exposures
included
this
were
MRI
biomarkers,
scores,
demographics,
clinical
features.
main
outcome
was
aMCI
follow-up.
EBM
utilized
converting
based
on
three
feature
combinations,
obtaining
ensuring
Meanwhile,
interaction
effect
considered
model.
combinations
compared
accuracy,
sensitivity,
specificity,
AUC-ROC.
global
local
explanations
are
displayed
by
importance
ranking
plots.
Results:
five-years
prediction
accuracy
reached
85%
(AUC
=
0.92)
both
scores
markers.
Apart
accuracies,
we
features’
periods.
In
early
stage
AD,
markers
play
a
major
role,
middle-term,
more
important.
Feature
risk
scoring
plots
demonstrated
insightful
nonlinear
interactive
associations
between
selected
outcome.
one-year
prediction,
lower
right
inferior
temporal
volume
(<9000)
significantly
associated
conversion.
For
two-year
low
left
thickness
(<2)
most
critical.
three-year
higher
FAQ
(>4)
During
four-year
APOE4
five-year
entorhinal
(<1000)
critical
feature.
Conclusions:
established
glass-box
EBMs
superior
ability
detailed
MCI.
Multi
significant
identified.
Further
may
be
significance
determine
whether
tool
would
management
patients.
Язык: Английский
Parkinson’s Disease Detection by Processing Different ANN Architecture Using Vocal Dataset
Mohammed Yusra,
Snwr,
J. Mohammed
и другие.
Eurasian Journal of Science and Engineering,
Год журнала:
2023,
Номер
9(2)
Опубликована: Янв. 1, 2023
Parkinson's
Disease
(PD)
is
a
long-standing
neurodegenerative
condition
of
the
central
nervous
system
that
mainly
affects
motor
and
origins
full
or
partial
damage
in
behavior,
speech,
reflexes,
mental
processing,
other
energetic
functions.Doctors
use
different
types
datasets
such
as
movement
images
from
people
to
diagnose
disease.In
this
paper,
speech
dataset
collected
with
without
PD
detect
disease.The
voice
recording
samples
are
analyzed
feature
vectors
extracted
samples.A
supervised
ANN
Multi-Layer
Perceptron
backpropagation
algorithm
presented
accurately
distinguish
between
healthy
individuals.Different
Architecture
diverse
neuron
numbers
hidden
layers
tested
utilize
model
result
each
architecture
compared
select
best
for
recognition.So
far,
our
score
highest
which
93%
testing
dataset.
Язык: Английский
Human Activity Recognition on Smartphones using Innovative Logistic Regression and Comparing Accuracy of Naive Bayes Algorithm
L. Anand Kumar Reddy,
Prasanna Sadagopan
E3S Web of Conferences,
Год журнала:
2024,
Номер
491, С. 03023 - 03023
Опубликована: Янв. 1, 2024
The
objective
of
this
study
is
to
compare
the
Naive
Bayes
algorithm
with
Innovative
Logistic
Regression
in
order
enhance
human
activity
identification
for
sitting
and
walking.
To
predict
activity,
are
used
different
training
testing
splits.
From
each
group,
ten
sets
samples
selected,
yielding
a
total
twenty
samples.
About
80%
data
from
an
independent
sample
T
test
were
utilized
Gpower
(g
power
setup
parameters:
α
=
0.05
0.80,
β
0.2).
Compared
(90.7210%),
(95.5680%)
has
higher
accuracy,
statistical
significance
value
P
0.003
(p
<
0.05).
When
compared
Bayes,
accuracy.
Язык: Английский
AI for Automated Thoracic Disease Assessment from X-Ray Imaging: a Review
Опубликована: Окт. 21, 2023
With
the
increasing
availability
of
digital
X-ray
imaging,
artificial
intelligence
(AI)
has
emerged
as
a
promising
tool
for
automating
assessment
thoracic
diseases.
The
objective
this
study
is
to
systematically
review
and
deep
learning
methods
proposed
automated
diseases
from
chest
images.
A
thorough
search
relevant
literature
was
conducted,
studies
that
met
inclusion
criteria
were
critically
reviewed.
Information
on
datasets,
model
architectures,
evaluation
metrics,
results
extracted.
Convolutional
neural
networks
are
prevalent,
achieving
state-of-the-art
classification
performance.
Recent
have
explored
more
complex
tasks
such
disease
localization,
segmentation,
report
generation.
multitask
multimodal
approaches
promising.
Challenges
related
data,
evaluations,
clinical
adoption
identified.
This
prevails
there
significant
progress
in
using
analysis.
Further
research
needed
validate
these
models
real-world
settings
facilitate
their
integration
into
workflows.
Язык: Английский