Radiotherapy and Oncology,
Год журнала:
2025,
Номер
unknown, С. 110852 - 110852
Опубликована: Март 1, 2025
In
the
HECKTOR
2022
challenge
set
[1],
several
state-of-the-art
(SOTA,
achieving
best
performance)
deep
learning
models
were
introduced
for
predicting
recurrence-free
period
(RFP)
in
head
and
neck
cancer
patients
using
PET
CT
images.
This
study
investigates
whether
a
conventional
DenseNet
architecture,
with
optimized
numbers
of
layers
image-fusion
strategies,
could
achieve
comparable
performance
as
SOTA
models.
The
dataset
comprises
489
oropharyngeal
(OPC)
from
seven
distinct
centers.
It
was
randomly
divided
into
training
(n
=
369)
an
independent
test
120).
Furthermore,
additional
400
OPC
patients,
who
underwent
chemo(radiotherapy)
at
our
center,
employed
external
testing.
Each
patients'
data
included
pre-treatment
CT-
PET-scans,
manually
generated
GTV
(Gross
tumour
volume)
contours
primary
tumors
lymph
nodes,
RFP
information.
present
compared
against
three
developed
on
dataset.
When
inputting
CT,
early
fusion
(considering
them
different
channels
input)
approach,
DenseNet81
(with
81
layers)
obtained
internal
C-index
0.69,
metric
Notably,
removal
input
yielded
same
0.69
while
improving
0.59
to
0.63.
PET-only
models,
when
utilizing
late
(concatenation
extracted
features)
PET,
demonstrated
superior
values
0.68
0.66
both
sets,
better
only
set.
basic
architecture
predictive
par
featuring
more
intricate
architectures
set,
test.
imaging
Artificial Intelligence in Medicine,
Год журнала:
2024,
Номер
154, С. 102900 - 102900
Опубликована: Июнь 5, 2024
With
Artificial
Intelligence
(AI)
increasingly
permeating
various
aspects
of
society,
including
healthcare,
the
adoption
Transformers
neural
network
architecture
is
rapidly
changing
many
applications.
Transformer
a
type
deep
learning
initially
developed
to
solve
general-purpose
Natural
Language
Processing
(NLP)
tasks
and
has
subsequently
been
adapted
in
fields,
healthcare.
In
this
survey
paper,
we
provide
an
overview
how
adopted
analyze
forms
healthcare
data,
clinical
NLP,
medical
imaging,
structured
Electronic
Health
Records
(EHR),
social
media,
bio-physiological
signals,
biomolecular
sequences.
Furthermore,
which
have
also
include
articles
that
used
transformer
for
generating
surgical
instructions
predicting
adverse
outcomes
after
surgeries
under
umbrella
critical
care.
Under
diverse
settings,
these
models
diagnosis,
report
generation,
data
reconstruction,
drug/protein
synthesis.
Finally,
discuss
benefits
limitations
using
transformers
examine
issues
such
as
computational
cost,
model
interpretability,
fairness,
alignment
with
human
values,
ethical
implications,
environmental
impact.
Diagnostics,
Год журнала:
2023,
Номер
13(10), С. 1696 - 1696
Опубликована: Май 11, 2023
Although
handcrafted
radiomics
features
(RF)
are
commonly
extracted
via
software,
employing
deep
(DF)
from
learning
(DL)
algorithms
merits
significant
investigation.
Moreover,
a
"tensor''
paradigm
where
various
flavours
of
given
feature
generated
and
explored
can
provide
added
value.
We
aimed
to
employ
conventional
tensor
DFs,
compare
their
outcome
prediction
performance
RFs.
Polish Journal of Radiology,
Год журнала:
2023,
Номер
88, С. 365 - 370
Опубликована: Авг. 14, 2023
Accurately
segmenting
head
and
neck
cancer
(HNC)
tumors
in
medical
images
is
crucial
for
effective
treatment
planning.
However,
current
methods
HNC
segmentation
are
limited
their
accuracy
efficiency.
The
present
study
aimed
to
design
a
model
three-dimensional
(3D)
positron
emission
tomography
(PET)
using
Non-Local
Means
(NLM)
morphological
operations.The
proposed
was
tested
data
from
the
HECKTOR
challenge
public
dataset,
which
included
408
patient
with
tumors.
NLM
utilized
image
noise
reduction
preservation
of
critical
information.
Following
pre-processing,
operations
were
used
assess
similarity
intensity
edge
information
within
images.
Dice
score,
Intersection
Over
Union
(IoU),
evaluate
manual
predicted
results.The
achieved
an
average
score
81.47
±
3.15,
IoU
80
4.5,
94.03
4.44,
demonstrating
its
effectiveness
PET
images.The
algorithm
provides
capability
produce
patient-specific
tumor
without
interaction,
addressing
limitations
segmentation.
has
potential
improve
planning
aid
development
personalized
medicine.
Additionally,
this
can
be
extended
effectively
segment
other
organs
annotated
Cancers,
Год журнала:
2023,
Номер
15(7), С. 1932 - 1932
Опубликована: Март 23, 2023
Automatic
delineation
and
detection
of
the
primary
tumour
(GTVp)
lymph
nodes
(GTVn)
using
PET
CT
in
head
neck
cancer
recurrence-free
survival
prediction
can
be
useful
for
diagnosis
patient
risk
stratification.
We
used
data
from
nine
different
centres,
with
524
359
cases
training
testing,
respectively.
utilised
posterior
sampling
weight
space
proposed
segmentation
model
to
estimate
uncertainty
false
positive
reduction.
explored
prognostic
potential
radiomics
features
extracted
predicted
GTVp
GTVn
SHAP
analysis
explainability.
evaluated
bias
models
respect
age,
gender,
chemotherapy,
HPV
status,
lesion
size.
achieved
an
aggregate
Dice
score
0.774
0.760
on
test
set
GTVn,
observed
a
per
image
reduction
19.5%
7.14%
threshold
Radiomics
both
are
most
prognostic,
our
achieves
C-index
0.672
set.
Our
framework
incorporates
estimation,
fairness,
explainability,
demonstrating
accurate
European Journal of Nuclear Medicine and Molecular Imaging,
Год журнала:
2023,
Номер
50(13), С. 3996 - 4009
Опубликована: Авг. 19, 2023
Abstract
Purpose
Prognostic
prediction
is
crucial
to
guide
individual
treatment
for
locoregionally
advanced
nasopharyngeal
carcinoma
(LA-NPC)
patients.
Recently,
multi-task
deep
learning
was
explored
joint
prognostic
and
tumor
segmentation
in
various
cancers,
resulting
promising
performance.
This
study
aims
evaluate
the
clinical
value
of
LA-NPC
Methods
A
total
886
patients
acquired
from
two
medical
centers
were
enrolled
including
data,
[
18
F]FDG
PET/CT
images,
follow-up
progression-free
survival
(PFS).
We
adopted
a
model
(DeepMTS)
jointly
perform
(DeepMTS-Score)
FDG-PET/CT
images.
The
DeepMTS-derived
masks
leveraged
extract
handcrafted
radiomics
features,
which
also
used
(AutoRadio-Score).
Finally,
we
developed
learning-based
radiomic
(MTDLR)
nomogram
by
integrating
DeepMTS-Score,
AutoRadio-Score,
data.
Harrell's
concordance
indices
(C-index)
time-independent
receiver
operating
characteristic
(ROC)
analysis
discriminative
ability
proposed
MTDLR
nomogram.
For
patient
stratification,
PFS
rates
high-
low-risk
calculated
using
Kaplan–Meier
method
compared
with
observed
probability.
Results
Our
achieved
C-index
0.818
(95%
confidence
interval
(CI):
0.785–0.851),
0.752
CI:
0.638–0.865),
0.717
0.641–0.793)
area
under
curve
(AUC)
0.859
0.822–0.895),
0.769
0.642–0.896),
0.730
0.634–0.826)
training,
internal
validation,
external
validation
cohorts,
showed
statistically
significant
improvement
over
conventional
nomograms.
divided
into
significantly
different
groups.
Conclusion
demonstrated
that
can
reliable
accurate
patients,
enabled
better
could
facilitate
personalized
planning.
Medical Physics,
Год журнала:
2023,
Номер
51(3), С. 2096 - 2107
Опубликована: Сен. 30, 2023
Abstract
Background
Radiotherapy
(RT)
combined
with
cetuximab
is
the
standard
treatment
for
patients
inoperable
head
and
neck
cancers.
Segmentation
of
(H&N)
tumors
a
prerequisite
radiotherapy
planning
but
time‐consuming
process.
In
recent
years,
deep
convolutional
neural
networks
(DCNN)
have
become
de
facto
automated
image
segmentation.
However,
due
to
expensive
computational
cost
associated
enlarging
field
view
in
DCNNs,
their
ability
model
long‐range
dependency
still
limited,
this
can
result
sub‐optimal
segmentation
performance
objects
background
context
spanning
over
long
distances.
On
other
hand,
Transformer
models
demonstrated
excellent
capabilities
capturing
such
information
several
semantic
tasks
performed
on
medical
images.
Purpose
Despite
impressive
representation
capacity
vision
transformer
models,
current
transformer‐based
suffer
from
inconsistent
incorrect
dense
predictions
when
fed
multi‐modal
input
data.
We
suspect
that
power
self‐attention
mechanism
may
be
limited
extracting
complementary
exists
To
end,
we
propose
novel
model,
debuted,
Cross‐modal
Swin
(SwinCross),
cross‐modal
attention
(CMA)
module
incorporate
feature
extraction
at
multiple
resolutions.
Methods
architecture
3D
two
main
components:
(1)
integrating
modalities
(PET
CT),
(2)
shifted
window
block
learning
modalities.
evaluate
efficacy
our
approach,
conducted
experiments
ablation
studies
HECKTOR
2021
challenge
dataset.
compared
method
against
nnU‐Net
(the
backbone
top‐5
methods
2021)
state‐of‐the‐art
including
UNETR
UNETR.
The
employed
five‐fold
cross‐validation
setup
using
PET
CT
Results
Empirical
evidence
demonstrates
proposed
consistently
outperforms
comparative
techniques.
This
success
attributed
CMA
module's
enhance
inter‐modality
representations
between
during
head‐and‐neck
tumor
Notably,
SwinCross
surpasses
across
all
five
folds,
showcasing
its
proficiency
varying
resolutions
through
modules.
Conclusions
introduced
automating
delineation
Our
incorporates
cross‐modality
module,
enabling
exchange
features
experimental
results
establish
superiority
improved
correlations
Furthermore,
methodology
holds
applicability
involving
different
imaging
like
SPECT/CT
or
PET/MRI.
Code:
https://github.com/yli192/SwinCross_CrossModalSwinTransformer_for_Medical_Image_Segmentation
Advances in Radiation Oncology,
Год журнала:
2024,
Номер
9(7), С. 101521 - 101521
Опубликована: Апрель 21, 2024
Historically,
clinician-derived
contouring
of
tumors
and
healthy
tissues
has
been
crucial
for
radiotherapy
(RT)
planning.
In
recent
years,
advances
in
artificial
intelligence
(AI),
predominantly
deep
learning
(DL),
have
rapidly
improved
automated
RT
applications,
particularly
routine
organs-at-risk
1–3.
Despite
research
efforts
actively
promoting
its
broader
acceptance,
clinical
adoption
auto-contouring
is
not
yet
standard
practice.