npj Digital Medicine,
Journal Year:
2024,
Volume and Issue:
7(1)
Published: Aug. 24, 2024
Radiological
imaging
is
a
globally
prevalent
diagnostic
method,
yet
the
free
text
contained
in
radiology
reports
not
frequently
used
for
secondary
purposes.
Natural
Language
Processing
can
provide
structured
data
retrieved
from
these
reports.
This
paper
provides
summary
of
current
state
research
on
Large
Model
(LLM)
based
approaches
information
extraction
(IE)
We
conduct
scoping
review
that
follows
PRISMA-ScR
guideline.
Queries
five
databases
were
conducted
August
1st
2023.
Among
34
studies
met
inclusion
criteria,
only
pre-transformer
and
encoder-based
models
are
described.
External
validation
shows
general
performance
decrease,
although
LLMs
might
improve
generalizability
IE
approaches.
Reports
related
to
CT
MRI
examinations,
as
well
thoracic
reports,
prevail.
Most
common
challenges
reported
missing
external
augmentation
described
methods.
Different
reporting
granularities
affect
comparability
transparency
Research Square (Research Square),
Journal Year:
2024,
Volume and Issue:
unknown
Published: Oct. 30, 2024
Abstract
Electronic
medical
records
(EMRs)
are
essential
in
clinical
practice.
Although
current
large
language
models
(LLMs)
excel
tasks
like
US
Medical
Licensing
Examination,
they
struggle
with
real-world
applications
due
to
insufficient
large-scale
EMR
data
their
training,
hindering
expertise.
To
address
this
limitation,
we
proposed
EMR-LLM,
an
LLM
for
practice
using
EMRs.
Firstly,
continually
pre-trained
a
general
on
corpora
enhance
its
domain
knowledge.
Then,
designed
three
categories
of
instruction
EMRs:
structure
understanding,
numerical
and
downstream
tasks.
Finally,
introduced
ability-boosting
instruction-tuning
method,
which
mimics
human
learning,
progressing
from
simple
complex
while
introducing
replay
strategy
retain
learned
Experimental
results
demonstrated
that
EMR-LLM
outperformed
strong
competitors
six
tasks,
nine
benchmarks,
open-domain
benchmarks.
Moreover,
discharge
summary
generation,
achieved
performance
levels
close
those
expert
clinicians.
