medRxiv (Cold Spring Harbor Laboratory),
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 10, 2024
Abstract
Deep
learning
models
have
shown
substantial
promise
in
assisting
medical
diagnosis,
offering
the
potential
to
improve
patient
outcomes
and
reduce
clinician
workloads.
However,
widespread
adoption
of
these
clinical
practice
has
been
hindered
by
concerns
surrounding
their
trustworthiness,
transparency,
interpretability.
Addressing
challenges
requires
not
only
development
explainable
AI
(xAI)
techniques
but
also
quantitative
metrics
evaluate
effectiveness.
This
study
presents
a
comprehensive
framework
for
training,
explaining,
quantitatively
assessing
deep
skin
cancer
diagnosis.
Leveraging
HAM10000
dataset
seven
diagnostic
lesion
categories,
multiple
convolutional
neural
network
architectures—including
custom
CNNs,
DenseNet,
MobileNet,
ResNet—were
trained
optimized
using
augmentation,
oversampling,
hyperparameter
tuning.
Following
model
explainability
such
as
SHAP,
LIME,
Integrated
Gradients
were
deployed
generate
post
hoc
explanations.
Critically,
primary
contribution
this
work
is
evaluation
explanation
methods
related
faithfulness,
robustness,
complexity.
All
code,
models,
results
are
publicly
available,
providing
reproducible
pathway
toward
more
trustworthy,
tools.
Indonesian Journal of Computer Science,
Год журнала:
2024,
Номер
13(3)
Опубликована: Июнь 15, 2024
The
sharp
increase
in
cases
of
melanoma
and
other
skin
cancers
worldwide
highlights
the
urgent
need
for
improved
diagnostic
methods.
Because
lesions
vary
widely
access
to
dermatological
knowledge
is
limited
resource-poor
areas,
traditional
methods
-
which
rely
on
visual
inspection
clinical
experience
have
difficulty
identifying
diseases
accurately.
This
situation
requires
innovative
approaches
improve
accessibility
accuracy.
To
address
these
issues,
this
work
uses
deep
learning
(DL)
convolutional
neural
networks
(CNNs).
paper
trying
transform
cancer
diagnosis
through
use
large
databases
dermoscopic
images
advanced
artificial
intelligence
algorithms.
In
order
evaluate
effectiveness
CNNs
DL
diseases,
we
conducted
a
comprehensive
analysis
literature,
focusing
accuracy
type
classification.
Our
approach
focused
model
architectures,
data
preparation
methods,
performance
indicators
while
examining
existing
research
using
AI
algorithms
diagnose
cancer.
With
ultimate
goal
improving
patient
outcomes
early
detection
accurate
classification
conditions,
not
only
underscores
great
potential
CNN
transcending
limitations,
but
also
continued
development
AI-based
tools
pathology.
Dermatology.
Diagnosis.
medRxiv (Cold Spring Harbor Laboratory),
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 10, 2024
Abstract
Deep
learning
models
have
shown
substantial
promise
in
assisting
medical
diagnosis,
offering
the
potential
to
improve
patient
outcomes
and
reduce
clinician
workloads.
However,
widespread
adoption
of
these
clinical
practice
has
been
hindered
by
concerns
surrounding
their
trustworthiness,
transparency,
interpretability.
Addressing
challenges
requires
not
only
development
explainable
AI
(xAI)
techniques
but
also
quantitative
metrics
evaluate
effectiveness.
This
study
presents
a
comprehensive
framework
for
training,
explaining,
quantitatively
assessing
deep
skin
cancer
diagnosis.
Leveraging
HAM10000
dataset
seven
diagnostic
lesion
categories,
multiple
convolutional
neural
network
architectures—including
custom
CNNs,
DenseNet,
MobileNet,
ResNet—were
trained
optimized
using
augmentation,
oversampling,
hyperparameter
tuning.
Following
model
explainability
such
as
SHAP,
LIME,
Integrated
Gradients
were
deployed
generate
post
hoc
explanations.
Critically,
primary
contribution
this
work
is
evaluation
explanation
methods
related
faithfulness,
robustness,
complexity.
All
code,
models,
results
are
publicly
available,
providing
reproducible
pathway
toward
more
trustworthy,
tools.