Journal of medicine and health science.,
Journal Year:
2024,
Volume and Issue:
2(4), P. 90 - 97
Published: Dec. 1, 2024
Diabetes
Mellitus
is
a
paradigmatic
case
of
long-term
care,
being
one
the
most
prevalent
chronic
diseases
worldwide,
with
millions
patients
no
cure.
Health
management
pivotal
in
addressing
diabetes;
however,
lack
personalized
and
tailored
diabetes
intervention
self-care
strategies
has
prevented
from
maximizing
health
outcomes.
The
crowd
profile
technique,
an
effective
tool
for
user
analysis,
combines
artificial
intelligence
big
data
analytics
to
provide
risk
prediction,
management,
digital
consultation
services
diabetes.
This
study
reviews
current
research
on
application
highlighting
potential
benefits
challenges
associated
management.
findings
underscore
critical
need
integrating
into
healthcare
systems
enhance
quality
effectiveness
medRxiv (Cold Spring Harbor Laboratory),
Journal Year:
2024,
Volume and Issue:
unknown
Published: April 8, 2024
Continuous
glucose
monitors
(CGM)
provide
patients
and
clinicians
with
valuable
insights
about
glycemic
control
that
aid
in
diabetes
management.
The
advent
of
large
language
models
(LLMs),
such
as
GPT-4,
has
enabled
real-time
text
generation
summarization
medical
data.
Further,
recent
advancements
have
the
integration
data
analysis
features
chatbots,
raw
can
be
uploaded
analyzed
when
prompted.
Studying
both
accuracy
suitability
LLM-derived
performed
on
time
series
data,
CGM
is
an
important
area
research.
objective
this
study
was
to
assess
strengths
limitations
using
LLM
analyze
produce
summaries
14
days
for
type
1
diabetes.
This
used
simulated
from
10
different
cases.
We
first
evaluated
ability
GPT-4
compute
quantitative
metrics
specific
found
Ambulatory
Glucose
Profile
(AGP).
Then,
two
independent
clinician
graders,
we
accuracy,
completeness,
safety,
qualitative
descriptions
produced
by
across
five
tasks.
demonstrated
performs
well
measures
safety
producing
all
These
results
highlight
capabilities
accurate
safe
narrative
several
work,
including
concerns
related
how
may
misprioritize
highlighting
instances
hypoglycemia
hyperglycemia.
Our
work
serves
a
preliminary
generative
integrated
into
care
through
analysis,
more
broadly,
potential
leverage
LLMs
streamlined
analysis.
medRxiv (Cold Spring Harbor Laboratory),
Journal Year:
2024,
Volume and Issue:
unknown
Published: April 14, 2024
Abstract
Background
The
escalating
global
burden
of
diabetes
necessitates
innovative
management
strategies.
Artificial
intelligence,
particularly
large
language
models
like
GPT-4,
presents
a
promising
avenue
for
improving
guideline
adherence
in
care.
Such
technologies
could
revolutionize
patient
by
offering
personalized,
evidence-based
treatment
recommendations.
Methods
A
comparative,
blinded
design
was
employed,
involving
50
hypothetical
mellitus
case
summaries,
emphasizing
varied
aspects
management.
GPT-4
evaluated
each
summary
adherence,
classifying
them
as
compliant
or
non-compliant,
based
on
the
ADA
guidelines.
medical
expert,
to
GPT-4’s
assessments,
independently
reviewed
summaries.
Concordance
between
and
expert’s
evaluations
statistically
analyzed,
including
calculating
Cohen’s
kappa
agreement.
Results
labelled
30
summaries
20
while
expert
identified
28
22
non-compliant.
Agreement
reached
46
cases,
yielding
0.84,
indicating
near-perfect
demonstrated
92%
accuracy,
with
sensitivity
86.4%
specificity
96.4%.
Discrepancies
four
cases
highlighted
challenges
AI’s
understanding
complex
clinical
judgments
related
medication
adjustments
modifications.
Conclusion
exhibits
potential
support
health-care
professionals
reviewing
plans
adherence.
Despite
high
concordance
instances
non-agreement
underscore
need
AI
refinement
scenarios.
Future
research
should
aim
at
enhancing
reasoning
capabilities
exploring
its
integration
other
improved
healthcare
delivery.
The Visual Computer,
Journal Year:
2024,
Volume and Issue:
unknown
Published: May 10, 2024
Abstract
Image
quality
assessment
(IQA)
of
fundus
images
constitutes
a
foundational
step
in
automated
disease
analysis.
This
process
is
pivotal
supporting
the
automation
screening,
diagnosis,
follow-up,
and
related
academic
research
for
diabetic
retinopathy
(DR).
study
introduced
deep
learning-based
approach
IQA
ultra-widefield
optical
coherence
tomography
angiography
(UW-OCTA)
patients
with
DR.
Given
novelty
technology,
its
limited
prevalence,
high
costs
associated
equipment
operational
training,
concerns
regarding
ethics
patient
privacy,
UW-OCTA
datasets
are
notably
scarce.
To
address
this,
we
initially
pre-train
vision
transformer
(ViT)
model
on
dataset
comprising
6
mm
×
OCTA
images,
enabling
to
acquire
fundamental
understanding
image
characteristics
indicators.
Subsequent
fine-tuning
12
aims
enhance
accuracy
assessment.
transfer
learning
strategy
leverages
generic
features
learned
during
pre-training
adjusts
evaluate
effectively.
Experimental
results
demonstrate
that
our
proposed
method
achieves
superior
performance
compared
ResNet18,
ResNet34,
ResNet50,
an
AUC
0.9026
Kappa
value
0.7310.
Additionally,
ablation
studies,
including
omission
substitution
backbone
network
ViT
base
version,
resulted
varying
degrees
decline
values,
confirming
efficacy
each
module
within
methodology.
Research Square (Research Square),
Journal Year:
2024,
Volume and Issue:
unknown
Published: May 30, 2024
Abstract
We
present
a
prompt
learning
framework
designed
to
enhance
the
performance
in
computer
vision
task
considering
particular
use
case
where
training
image
dataset
is
confronted
with
highly
imbalanced
categorical
distributions.
By
formulating
as
variational
problem,
our
model
capable
of
generating
multiple
prompts
describe
semantic
(i.e,
class).
The
motivation
behind
originates
from
heuristic
that
voting
ensemble
establishes
more
robust
aggregated
algorithm
which
potentially
benefits
tail
classes
number
sample
scarce.
Unlike
previous
techniques,
are
often
restricted
by
using
fixed
set
during
and
test
phase,
we
propose
learn
distribution
an
arbitrary
can
be
sampled
whenever
required,
named
method
“Prompt
Distribution
Learning
(PDL)”.
will
discuss
contrast
various
ways
formulate
variation
thoroughly
compare
their
performances
against
state-of-the-art
solutions
for
long-tailed
visual
recognition.
Our
empirical
study
suggests
proposed
prompt-learning
beneficial
transferring
pre-trained
vision-language
downstream
recognition
tasks
while
being
sufficiently
flexible
accommodating
different
designs
prompt-generating
functions.
code
publicly
available
at
https:
//github.com/Walter-pixel/Prompt-Distribution-of-CLIP-Long-Tailed-Data.
Research Square (Research Square),
Journal Year:
2024,
Volume and Issue:
unknown
Published: May 31, 2024
Abstract
Efficient
medical
image
segmentation
plays
an
important
role
in
computer-aided
diagnosis
(CAD).
Deep
mining
of
pixel
semantics
is
crucial
for
segmentation.
However,
previous
works
on
semantic
usually
overlook
the
importance
embedding
subspace,
and
lacked
latent
space
direction
information.
In
this
work,
we
constructed
global
orthogonal
basis
channel
space,
which
can
significantly
enhance
feature
representation.
We
propose
a
novel
distance-based
method
that
decouples
into
sub-embedding
spaces
different
classes,
then
implements
level
classification
based
distance
between
its
features
origin
subspace.
Experiments
various
public
benchmarks
show
effectiveness
our
model
as
compared
to
state-of-the-art
methods.
The
code
will
be
published
at
https://github.com/lxt0525/LSDENet.
