Cancers,
Journal Year:
2024,
Volume and Issue:
17(1), P. 96 - 96
Published: Dec. 31, 2024
Background/Objectives:
The
duration
of
the
response
to
radiotherapy-related
treatment
is
a
critical
prognostic
indicator
for
patients
with
nasopharyngeal
carcinoma
(NPC).
Persistent
tumor
status,
including
residual
presence
and
early
recurrence,
associated
poorer
survival
outcomes.
To
address
this,
we
developed
prediction
model
identify
at
high
risk
persistent
status
prior
initiating
treatment.
Methods:
This
retrospective
study
included
104
NPC
receiving
who
had
completed
3-year
follow-up
period;
29
were
classified
into
group
75
disease-free
group.
Radiomic
features
extracted
from
pretreatment
positron
emission
tomography
(PET)
images
used
construct
by
employing
machine
learning
algorithms.
model’s
diagnostic
performance
was
assessed
using
area
under
receiver
operating
characteristic
curve
(AUC),
whereas
SHapley
Additive
exPlanations
(SHAP)
analysis
conducted
determine
contribution
individual
model.
Results:
AdaBoost
algorithm
validated
through
five-fold
cross-validation
achieved
highest
AUC
0.934.
Its
sensitivity,
specificity,
positive
predictive
value,
negative
accuracy
89.66%,
86.67%,
72.22%,
95.59%,
87.5%,
respectively.
SHAP
revealed
that
feature
dependence
low
metabolic
uptake
emphasis50
greatest
impact
on
predictions.
Furthermore,
as
exhibited
markedly
higher
overall
rates
compared
those
status.
Conclusions:
In
conclusion,
proposed
efficiently
identified
radiomic
PET
images.
Seminars in Nuclear Medicine,
Journal Year:
2025,
Volume and Issue:
unknown
Published: Feb. 1, 2025
Nuclear
medicine
has
continuously
evolved
since
its
beginnings,
constantly
improving
the
diagnosis
and
treatment
of
various
diseases.
The
integration
artificial
intelligence
(AI)
is
one
latest
revolutionizing
chapters,
promising
significant
advancements
in
diagnosis,
prognosis,
segmentation,
image
quality
enhancement,
theranostics.
Early
AI
applications
nuclear
focused
on
diagnostic
accuracy,
leveraging
machine
learning
algorithms
for
disease
classification
outcome
prediction.
Advances
deep
learning,
including
convolutional
more
recently
transformer-based
neural
networks,
have
further
enabled
precise
segmentation
as
well
low-dose
imaging,
patient-specific
dosimetry
personalized
treatment.
Generative
AI,
driven
by
large
language
models
diffusion
techniques,
now
allowing
process,
interpretation,
generation
complex
medical
images.
Despite
these
achievements,
challenges
such
data
scarcity,
heterogeneity,
ethical
concerns
remain
barriers
to
clinical
translation.
Addressing
issues
through
interdisciplinary
collaboration
will
pave
way
a
broader
adoption
medicine,
potentially
enhancing
patient
care
optimizing
therapeutic
outcomes.
BMC Cancer,
Journal Year:
2024,
Volume and Issue:
24(1)
Published: Nov. 14, 2024
This
study
aimed
to
develop
a
predictive
model
utilizing
radiomics
and
body
composition
features
derived
from
18F-FDG
PET/CT
scans
forecast
progression-free
survival
(PFS)
overall
(OS)
outcomes
in
patients
with
esophageal
squamous
cell
carcinoma
(ESCC).
We
analyzed
data
91
who
underwent
baseline
imaging.
Radiomic
extracted
PET
CT
images
subsequent
scores
(Rad-scores)
were
calculated.
Body
metrics
also
quantified,
including
muscle
fat
distribution
at
the
L3
level
scans.
Multiparametric
models
constructed
using
Cox
regression
analysis,
their
performance
was
assessed
area
under
time-dependent
receiver
operating
characteristic
(ROC)
curve
(AUC)
concordance
index
(C-index).
Multivariate
analysis
identified
Rad-scorePFS
(P
=
0.003),
sarcopenia
<
0.001),
visceral
adipose
tissue
(VATI)
0.001)
as
independent
predictors
of
PFS.
For
OS,
Rad-scoreOS
0.002),
VATI
0.037),
stage
0.042),
mass
(BMI)
0.008)
confirmed
prognostic
factors.
Integration
Rad-score
clinical
variables
parameters
enhanced
accuracy,
yielding
C-indices
0.810
(95%
CI:
0.737–0.884)
for
PFS
0.806
0.720–0.891)
OS.
underscored
potential
combining
refine
assessment
ESCC
patients.
Scientific Reports,
Journal Year:
2025,
Volume and Issue:
15(1)
Published: Jan. 13, 2025
The
present
study
analyzed
the
impact
of
age
on
causes
death
(CODs)
in
patients
with
nasopharyngeal
carcinoma
(NPC)
undergoing
chemoradiotherapy
(CRT)
using
machine
learning
approaches.
A
total
2841
(1037
classified
as
older,
≥
60
years
and
1804
younger,
<
years)
were
enrolled.
Variations
CODs
between
two
groups
before
after
applying
inverse
probability
treatment
weighting
(IPTW).
Additionally,
seven
different
models
employed
predictive
tools
to
identify
key
variables
assess
therapeutic
outcomes
NPC
receiving
CRT.
younger
group
exhibited
a
significantly
longer
overall
survival
(OS)
than
older
group,
both
IPTW
adjustment
(140
vs.
50
months,
P
0.001)
(137
53
0.001).
After
IPTW,
was
associated
worse
5-,
10-,
15-year
cumulative
incidences
terms
NPC-related
deaths
(30,
34,
38%
21,
27,
30%;
0.001),
cardiovascular
disease
(CVD;
4.1,
7.2,
8.8%
0.5,
1.8,
3.0%;
other
(8.3,
17,
24%
8.7,
12%;
However,
secondary
malignant
neoplasms
comparable
(P
=
0.100).
random
forest
(RF)
model
demonstrated
highest
concordance
index
0.701
among
all
models.
Time-dependent
variable
importance
plots
indicated
that
most
influential
factor
affecting
3-,
10-year
survival,
followed
by
metastasis
tumor
stage.
Younger
had
OS
their
counterparts.
Older
higher
likelihood
dying
from
non-NPC-related
causes,
particularly
CVDs.
RF
showed
best
accuracy,
identifying
critical
influencing
Physics and Imaging in Radiation Oncology,
Journal Year:
2025,
Volume and Issue:
33, P. 100733 - 100733
Published: Jan. 1, 2025
Deep
learning
(DL)
models
can
extract
prognostic
image
features
from
pre-treatment
PET/CT
scans.
The
study
objective
was
to
explore
the
potential
benefits
of
incorporating
pathologic
lymph
node
(PL)
spatial
information
in
addition
that
primary
tumor
(PT)
DL-based
for
predicting
local
control
(LC),
regional
(RC),
distant-metastasis-free
survival
(DMFS),
and
overall
(OS)
oropharyngeal
cancer
(OPC)
patients.
included
409
OPC
patients
treated
with
definitive
(chemo)radiotherapy
between
2010
2022.
Patient
data,
including
scans,
manually
contoured
PT
(GTVp)
PL
(GTVln)
structures,
clinical
variables,
endpoints,
were
collected.
Firstly,
a
method
employed
segment
tumours
PET/CT,
resulting
predicted
probability
maps
(TPMp)
(TPMln).
Secondly,
different
combinations
CT,
PET,
manual
contours
300
used
train
outcome
prediction
each
endpoint
through
5-fold
cross
validation.
Model
performance,
assessed
by
concordance
index
(C-index),
evaluated
using
test
set
100
Including
improved
C-index
results
all
endpoints
except
LC.
For
LC,
comparable
C-indices
(around
0.66)
observed
trained
only
those
as
additional
structure.
Models
combined
into
single
structure
achieved
highest
0.65
0.80
RC
DMFS
prediction,
respectively.
these
target
structures
separate
entities
0.70
OS.
Incorporating
performance
RC,
DMFS,
Frontiers in Oncology,
Journal Year:
2025,
Volume and Issue:
15
Published: May 2, 2025
Objective
Early
detection
and
timely
surgical
intervention
are
crucial
in
reducing
mortality
rates
associated
with
clinically
significant
prostate
cancer
(csPCa).
Currently,
clinical
diagnostics
primarily
depend
on
magnetic
resonance
imaging
(MRI)
nuclear
medicine,
the
potential
diagnostic
value
of
abdominal
computed
tomography
(CT)
remaining
underexplored.
This
study
aims
to
evaluate
effectiveness
multi-task
deep
learning
neural
networks
identifying
early-stage
using
CT
scans.
Methods
In
this
study,
we
enrolled
539
patients
from
Department
Radiology
(N=461)
Nuclear
Medicine
(N=78).
We
utilized
a
network
model
(MTDL),
based
3DUnet
architecture,
segment
analyze
collected
plain
images.
The
predictive
performance
was
compared
radiomics
single-task
ResNet18.
A
nomogram
then
developed
approach,
incorporating
prediction
results
PSAD,
age.
different
models
evaluated
receiver
operating
characteristic
(ROC)
curve
area
under
(AUC).
Results
461
were
divided
into
training
test
sets
at
ratio
6:4,
while
formed
validation
set.
Our
MTDL
demonstrated
AUCs
0.941
(95%
confidence
interval
[CI]:
0.905valceedi
0.912
CI:
0.904valceedi
0.932
0.883valceed
training,
test,
cohorts,
respectively.
indicates
that
combining
effectively
diagnoses
csPCa,
offering
superior
models.
Additionally,
outperformed
both
accuracy.
Conclusion
can
accurately
predict
presence
scans,
for
early
diagnosis
cancer.
Radiology Imaging Cancer,
Journal Year:
2025,
Volume and Issue:
7(3)
Published: May 1, 2025
MultiRecNet,
a
fully
automatic
multitask
deep
learning
network,
accurately
predicted
disease-free
survival
in
patients
with
locally
advanced
rectal
cancer
after
neoadjuvant
chemoradiotherapy,
using
multimodal
MRI.
npj Precision Oncology,
Journal Year:
2024,
Volume and Issue:
8(1)
Published: Oct. 14, 2024
Early
survival
prediction
is
vital
for
the
clinical
management
of
cancer
patients,
as
tumors
can
be
better
controlled
with
personalized
treatment
planning.
Traditional
methods
are
based
on
radiomics
feature
engineering
and/or
indicators
(e.g.,
staging).
Recently,
models
advances
in
deep
learning
techniques
have
achieved
state-of-the-art
performance
end-to-end
by
exploiting
features
derived
from
medical
images.
However,
existing
heavily
reliant
prognostic
information
within
primary
and
cannot
effectively
leverage
out-of-tumor
characterizing
local
tumor
metastasis
adjacent
tissue
invasion.
Also,
sub-optimal
leveraging
multi-modality
images
they
rely
empirically
designed
fusion
strategies
to
integrate
information,
where
pre-defined
domain-specific
human
prior
knowledge
inherently
limited
adaptability.
Here,
we
present
an
Adaptive
Multi-modality
Segmentation-to-Survival
model
(AdaMSS)
The
AdaMSS
self-adapt
its
strategy
training
data
also
adapt
focus
regions
capture
outside
tumors.
Extensive
experiments
two
large
datasets
(1380
patients
nine
centers)
show
that
our
surmounts
(C-index:
0.804
0.757),
demonstrating
potential
facilitate