Journal of Clinical Medicine,
Год журнала:
2024,
Номер
13(8), С. 2323 - 2323
Опубликована: Апрель 17, 2024
Background:
This
study
evaluates
the
performance
of
a
vision
transformer
(ViT)
model,
ViT-b16,
in
classifying
ischemic
stroke
cases
from
Moroccan
MRI
scans
and
compares
it
to
Visual
Geometry
Group
16
(VGG-16)
model
used
prior
study.
Methods:
A
dataset
342
scans,
categorized
into
‘Normal’
’Stroke’
classes,
underwent
preprocessing
using
TensorFlow’s
tf.data
API.
Results:
The
ViT-b16
was
trained
evaluated,
yielding
an
impressive
accuracy
97.59%,
surpassing
VGG-16
model’s
90%
accuracy.
Conclusions:
research
highlights
superior
classification
capabilities
for
diagnosis,
contributing
field
medical
image
analysis.
By
showcasing
efficacy
advanced
deep
learning
architectures,
particularly
context
this
underscores
potential
real-world
clinical
applications.
Ultimately,
our
findings
emphasize
importance
further
exploration
AI-based
diagnostic
tools
improving
healthcare
outcomes.
EAI Endorsed Transactions on Pervasive Health and Technology,
Год журнала:
2023,
Номер
9
Опубликована: Сен. 21, 2023
INTRODUCTION:
Nowadays
one
of
the
primary
causes
permanent
blindness
is
glaucoma.
Due
to
trade-offs,
it
makes
in
terms
portability,
size,
and
cost,
fundus
imaging
most
widely
used
glaucoma
screening
technique.
OBJECTIVES:To
boost
accuracy,focusing
on
less
execution
time,
resources
consumption,
we
have
proposed
a
vision
transformer-based
model
with
data
pre-processing
techniques
which
fix
classification
problems.
METHODS:
Convolution
“local”
technique
by
CNNs
that
restricted
limited
area
around
an
image.
Self-attention,
Vision
Transformers,
“global”
action
since
gathers
from
whole
This
possible
for
ViT
successfully
collect
far-off
semantic
relevance
Several
optimizers,
including
Adamax,
SGD,
RMSprop,
Adadelta,
Adafactor,
Nadam,
Adagrad,
were
studied
this
paper.
We
trained
tested
Transformer
IEEE
Fundus
image
dataset
having
1750
Healthy
Glaucoma
images.
Additionally,
was
preprocessed
using
resizing,
auto-rotation,
auto-adjust
contrast
adaptive
equalization.
RESULTS:
Results
also
show
Nadam
Optimizer
increased
accuracy
up
97%
equalized
preprocessing
followed
auto
rotate
resizing
operations.
CONCLUSION:
The
experimental
findings
shows
transformer
based
spurred
revolution
computer
reduced
time
training
classification.
Journal of Clinical Medicine,
Год журнала:
2024,
Номер
13(8), С. 2323 - 2323
Опубликована: Апрель 17, 2024
Background:
This
study
evaluates
the
performance
of
a
vision
transformer
(ViT)
model,
ViT-b16,
in
classifying
ischemic
stroke
cases
from
Moroccan
MRI
scans
and
compares
it
to
Visual
Geometry
Group
16
(VGG-16)
model
used
prior
study.
Methods:
A
dataset
342
scans,
categorized
into
‘Normal’
’Stroke’
classes,
underwent
preprocessing
using
TensorFlow’s
tf.data
API.
Results:
The
ViT-b16
was
trained
evaluated,
yielding
an
impressive
accuracy
97.59%,
surpassing
VGG-16
model’s
90%
accuracy.
Conclusions:
research
highlights
superior
classification
capabilities
for
diagnosis,
contributing
field
medical
image
analysis.
By
showcasing
efficacy
advanced
deep
learning
architectures,
particularly
context
this
underscores
potential
real-world
clinical
applications.
Ultimately,
our
findings
emphasize
importance
further
exploration
AI-based
diagnostic
tools
improving
healthcare
outcomes.
