Clinical Gastroenterology and Hepatology,
Journal Year:
2021,
Volume and Issue:
20(7), P. 1499 - 1507.e4
Published: Sept. 14, 2021
Background
&
AimsArtificial
intelligence-based
computer-aided
polyp
detection
(CADe)
systems
are
intended
to
address
the
issue
of
missed
polyps
during
colonoscopy.
The
effect
CADe
screening
and
surveillance
colonoscopy
has
not
previously
been
studied
in
a
United
States
(U.S.)
population.MethodsWe
conducted
prospective,
multi-center,
single-blind
randomized
tandem
study
evaluate
deep-learning
based
system
(EndoScreener,
Shanghai
Wision
AI,
China).
Patients
were
enrolled
across
4
U.S.
academic
medical
centers
from
2019
through
2020.
presenting
for
colorectal
cancer
or
first
high-definition
white
light
(HDWL)
first,
followed
immediately
by
other
procedure
fashion
same
endoscopist.
primary
outcome
was
adenoma
miss
rate
(AMR),
secondary
outcomes
included
sessile
serrated
lesion
(SSL)
adenomas
per
(APC).ResultsA
total
232
patients
entered
study,
with
116
undergo
HDWL
first.
After
exclusion
9
patients,
cohort
223
patients.
AMR
lower
CADe-first
group
compared
HDWL-first
(20.12%
[34/169]
vs
31.25%
[45/144];
odds
ratio
[OR],
1.8048;
95%
confidence
interval
[CI],
1.0780-3.0217;
P
=
.0247).
SSL
(7.14%
[1/14])
(42.11%
[8/19];
.0482).
First-pass
APC
higher
(1.19
[standard
deviation
(SD),
2.03]
0.90
[SD,
1.55];
.0323).
ADR
50.44%
43.64
%
(P
.3091).ConclusionIn
this
multicenter
controlled
trial,
we
demonstrate
decrease
an
increase
first-pass
use
CADe-system
when
alone.
Artificial
population.
We
(APC).
A
.3091).
In
npj Digital Medicine,
Journal Year:
2021,
Volume and Issue:
4(1)
Published: Jan. 8, 2021
Abstract
A
decade
of
unprecedented
progress
in
artificial
intelligence
(AI)
has
demonstrated
the
potential
for
many
fields—including
medicine—to
benefit
from
insights
that
AI
techniques
can
extract
data.
Here
we
survey
recent
development
modern
computer
vision
techniques—powered
by
deep
learning—for
medical
applications,
focusing
on
imaging,
video,
and
clinical
deployment.
We
start
briefly
summarizing
a
convolutional
neural
networks,
including
tasks
they
enable,
context
healthcare.
Next,
discuss
several
example
imaging
applications
stand
to
benefit—including
cardiology,
pathology,
dermatology,
ophthalmology–and
propose
new
avenues
continued
work.
then
expand
into
general
highlighting
ways
which
workflows
integrate
enhance
care.
Finally,
challenges
hurdles
required
real-world
deployment
these
technologies.
npj Digital Medicine,
Journal Year:
2021,
Volume and Issue:
4(1)
Published: April 7, 2021
Deep
learning
(DL)
has
the
potential
to
transform
medical
diagnostics.
However,
diagnostic
accuracy
of
DL
is
uncertain.
Our
aim
was
evaluate
algorithms
identify
pathology
in
imaging.
Searches
were
conducted
Medline
and
EMBASE
up
January
2020.
We
identified
11,921
studies,
which
503
included
systematic
review.
Eighty-two
studies
ophthalmology,
82
breast
disease
115
respiratory
for
meta-analysis.
Two
hundred
twenty-four
other
specialities
qualitative
Peer-reviewed
that
reported
on
using
imaging
included.
Primary
outcomes
measures
accuracy,
study
design
reporting
standards
literature.
Estimates
pooled
random-effects
In
AUC's
ranged
between
0.933
1
diagnosing
diabetic
retinopathy,
age-related
macular
degeneration
glaucoma
retinal
fundus
photographs
optical
coherence
tomography.
imaging,
0.864
0.937
lung
nodules
or
cancer
chest
X-ray
CT
scan.
For
0.868
0.909
mammogram,
ultrasound,
MRI
digital
tomosynthesis.
Heterogeneity
high
extensive
variation
methodology,
terminology
outcome
noted.
This
can
lead
an
overestimation
There
immediate
need
development
artificial
intelligence-specific
EQUATOR
guidelines,
particularly
STARD,
order
provide
guidance
around
key
issues
this
field.
arXiv (Cornell University),
Journal Year:
2020,
Volume and Issue:
unknown
Published: Jan. 1, 2020
ML
models
often
exhibit
unexpectedly
poor
behavior
when
they
are
deployed
in
real-world
domains.
We
identify
underspecification
as
a
key
reason
for
these
failures.
An
pipeline
is
underspecified
it
can
return
many
predictors
with
equivalently
strong
held-out
performance
the
training
domain.
Underspecification
common
modern
pipelines,
such
those
based
on
deep
learning.
Predictors
returned
by
pipelines
treated
equivalent
their
domain
performance,
but
we
show
here
that
behave
very
differently
deployment
This
ambiguity
lead
to
instability
and
model
practice,
distinct
failure
mode
from
previously
identified
issues
arising
structural
mismatch
between
this
problem
appears
wide
variety
of
practical
using
examples
computer
vision,
medical
imaging,
natural
language
processing,
clinical
risk
prediction
electronic
health
records,
genomics.
Our
results
need
explicitly
account
modeling
intended
any
Cancer Cell,
Journal Year:
2022,
Volume and Issue:
40(10), P. 1095 - 1110
Published: Oct. 1, 2022
In
oncology,
the
patient
state
is
characterized
by
a
whole
spectrum
of
modalities,
ranging
from
radiology,
histology,
and
genomics
to
electronic
health
records.
Current
artificial
intelligence
(AI)
models
operate
mainly
in
realm
single
modality,
neglecting
broader
clinical
context,
which
inevitably
diminishes
their
potential.
Integration
different
data
modalities
provides
opportunities
increase
robustness
accuracy
diagnostic
prognostic
models,
bringing
AI
closer
practice.
are
also
capable
discovering
novel
patterns
within
across
suitable
for
explaining
differences
outcomes
or
treatment
resistance.
The
insights
gleaned
such
can
guide
exploration
studies
contribute
discovery
biomarkers
therapeutic
targets.
To
support
these
advances,
here
we
present
synopsis
methods
strategies
multimodal
fusion
association
discovery.
We
outline
approaches
interpretability
directions
AI-driven
through
interconnections.
examine
challenges
adoption
discuss
emerging
solutions.
