European Radiology,
Год журнала:
2025,
Номер
unknown
Опубликована: Янв. 15, 2025
This
study
aimed
to
develop
an
open-source
multimodal
large
language
model
(CXR-LLaVA)
for
interpreting
chest
X-ray
images
(CXRs),
leveraging
recent
advances
in
models
(LLMs)
potentially
replicate
the
image
interpretation
skills
of
human
radiologists.
For
training,
we
collected
592,580
publicly
available
CXRs,
which
374,881
had
labels
certain
radiographic
abnormalities
(Dataset
1)
and
217,699
provided
free-text
radiology
reports
2).
After
pre-training
a
vision
transformer
with
Dataset
1,
integrated
it
LLM
influenced
by
LLaVA
network.
Then,
was
fine-tuned,
primarily
using
2.
The
model's
diagnostic
performance
major
pathological
findings
evaluated,
along
acceptability
radiologic
radiologists,
gauge
its
potential
autonomous
reporting.
demonstrated
impressive
test
sets,
achieving
average
F1
score
0.81
six
MIMIC
internal
set
0.56
external
set.
scores
surpassed
those
GPT-4-vision
Gemini-Pro-Vision
both
sets.
In
radiologist
evaluations
set,
achieved
72.7%
success
rate
reporting,
slightly
below
84.0%
ground
truth
reports.
highlights
significant
LLMs
CXR
interpretation,
while
also
acknowledging
limitations.
Despite
these
challenges,
believe
that
making
our
will
catalyze
further
research,
expanding
effectiveness
applicability
various
clinical
contexts.
Question
How
can
be
adapted
interpret
X-rays
generate
reports?
Findings
developed
CXR-LLaVA
effectively
detects
generates
higher
accuracy
compared
general-purpose
models.
Clinical
relevance
demonstrates
support
radiologists
autonomously
generating
reports,
reducing
workloads
improving
efficiency.
BioMed Research International,
Год журнала:
2022,
Номер
2022, С. 1 - 22
Опубликована: Апрель 25, 2022
Dense
convolutional
network
(DenseNet)
is
a
hot
topic
in
deep
learning
research
recent
years,
which
has
good
applications
medical
image
analysis.
In
this
paper,
DenseNet
summarized
from
the
following
aspects.
First,
basic
principle
of
introduced;
second,
development
and
analyzed
five
aspects:
broaden
structure,
lightweight
dense
unit,
connection
mode,
attention
mechanism;
finally,
application
field
analysis
three
pattern
recognition,
segmentation,
object
detection.
The
structures
are
systematically
certain
positive
significance
for
DenseNet.
ACM Transactions on Interactive Intelligent Systems,
Год журнала:
2023,
Номер
13(4), С. 1 - 35
Опубликована: Март 14, 2023
eXplainable
AI
(XAI)
involves
two
intertwined
but
separate
challenges:
the
development
of
techniques
to
extract
explanations
from
black-box
models
and
way
such
are
presented
users,
i.e.,
explanation
user
interface.
Despite
its
importance,
second
aspect
has
received
limited
attention
so
far
in
literature.
Effective
interfaces
fundamental
for
allowing
human
decision-makers
take
advantage
oversee
high-risk
systems
effectively.
Following
an
iterative
design
approach,
we
present
first
cycle
prototyping-testing-redesigning
explainable
technique
interface
clinical
Decision
Support
Systems
(DSS).
We
XAI
that
meets
technical
requirements
healthcare
domain:
sequential,
ontology-linked
patient
data,
multi-label
classification
tasks.
demonstrate
applicability
explain
a
DSS,
prototype
Next,
test
with
providers
collect
their
feedback
two-fold
outcome:
First,
obtain
evidence
increase
users’
trust
system,
second,
useful
insights
on
perceived
deficiencies
interaction
can
re-design
better,
more
human-centered
Radiology Artificial Intelligence,
Год журнала:
2021,
Номер
3(6)
Опубликована: Окт. 7, 2021
Data-driven
approaches
have
great
potential
to
shape
future
practices
in
radiology.
The
most
straightforward
strategy
obtain
clinically
accurate
models
is
use
large,
well-curated
and
annotated
datasets.
However,
patient
privacy
constraints,
tedious
annotation
processes,
the
limited
availability
of
radiologists
pose
challenges
building
such
This
review
details
model
training
strategies
scenarios
with
data,
insufficiently
labeled
and/or
expert
resources.
discusses
enlarge
data
sample,
decrease
time
burden
manual
supervised
labeling,
adjust
neural
network
architecture
improve
performance,
apply
semisupervised
approaches,
leverage
efficiencies
from
pretrained
models.
Keywords:
Computer-aided
Detection/Diagnosis,
Transfer
Learning,
Limited
Annotated
Data,
Augmentation,
Synthetic
Semisupervised
Federated
Few-Shot
Class
Imbalance
Diagnostics,
Год журнала:
2021,
Номер
11(7), С. 1155 - 1155
Опубликована: Июнь 24, 2021
Since
December
2019,
the
global
health
population
has
faced
rapid
spreading
of
coronavirus
disease
(COVID-19).
With
incremental
acceleration
number
infected
cases,
World
Health
Organization
(WHO)
reported
COVID-19
as
an
epidemic
that
puts
a
heavy
burden
on
healthcare
sectors
in
almost
every
country.
The
potential
artificial
intelligence
(AI)
this
context
is
difficult
to
ignore.
AI
companies
have
been
racing
develop
innovative
tools
contribute
arm
world
against
pandemic
and
minimize
disruption
it
may
cause.
main
objective
study
survey
decisive
role
technology
used
fight
pandemic.
Five
significant
applications
for
were
found,
including
(1)
diagnosis
using
various
data
types
(e.g.,
images,
sound,
text);
(2)
estimation
possible
future
spread
based
current
confirmed
cases;
(3)
association
between
infection
patient
characteristics;
(4)
vaccine
development
drug
interaction;
(5)
supporting
applications.
This
also
introduces
comparison
datasets.
Based
limitations
literature,
review
highlights
open
research
challenges
could
inspire
application
COVID-19.
