Leukemia
is
a
critical
disease
that
requires
early
and
accurate
diagnosis.
type
of
blood
cancer
mainly
occurring
when
bone
marrow
builds
extra
white
cells
in
the
human
body.
This
affects
adults
common
among
children.
paper
presents
deep-learning
approach
using
EfficientNetB5
to
classify
The
Cancer
Imaging
Archive
(TCIA)
with
more
than
10,000
images
from
118
patients.
achieved
confusion
matrix
will
contribute
improving
research
diagnosing
cancer.
PeerJ Computer Science,
Год журнала:
2024,
Номер
10, С. e1813 - e1813
Опубликована: Фев. 2, 2024
Blood
diseases
such
as
leukemia,
anemia,
lymphoma,
and
thalassemia
are
hematological
disorders
that
relate
to
abnormalities
in
the
morphology
concentration
of
blood
elements,
specifically
white
cells
(WBC)
red
(RBC).
Accurate
efficient
diagnosis
these
conditions
significantly
depends
on
expertise
hematologists
pathologists.
To
assist
pathologist
diagnostic
process,
there
has
been
growing
interest
utilizing
computer-aided
(CAD)
techniques,
particularly
those
using
medical
image
processing
machine
learning
algorithms.
Previous
surveys
this
domain
have
narrowly
focused,
often
only
addressing
specific
areas
like
segmentation
or
classification
but
lacking
a
holistic
view
segmentation,
classification,
feature
extraction,
dataset
utilization,
evaluation
matrices,
etc.
This
survey
aims
provide
comprehensive
systematic
review
existing
literature
research
work
field
analysis
deep
techniques.
It
focuses
techniques
algorithms
excel
morphological
characterization
WBCs
RBCs.
The
is
structured
cover
four
main
areas:
methodologies,
descriptive
selection,
parameters,
selection
for
Our
reveals
several
interesting
trends
preferences
among
researchers.
Regarding
approximately
50%
related
WBC
60%
RBC
opted
manually
obtaining
images
rather
than
predefined
dataset.
When
it
comes
45%
previous
chose
ALL-IDB
dataset,
while
significant
73%
researchers
focused
decided
obtain
from
institutions
instead
datasets.
In
terms
features
were
most
popular,
being
chosen
55%
80%
studies
respectively.
accuracy
blood-related
can
be
enhanced
through
effective
use
CAD
which
evolved
considerably
recent
years.
provides
broad
in-depth
employed,
utilization
selection.
inconsistency
suggests
need
standardized,
high-quality
datasets
strengthen
capabilities
further.
Additionally,
popularity
indicates
future
could
further
explore
innovate
direction.
International Journal of Imaging Systems and Technology,
Год журнала:
2023,
Номер
34(1)
Опубликована: Авг. 25, 2023
Abstract
One
of
the
most
fatal
and
prevalent
diseases
central
nervous
system
is
a
brain
tumour.
Different
subgrades
exist
for
each
type
tumour
because
broad
variety
tumours
pathologies.
Manual
diagnosis
may
be
error‐prone
time‐consuming,
both
which
are
becoming
more
challenging
as
medical
community's
workload
grows.
There
need
automatic
diagnosis.
In
this
study,
we
have
proposed
deep
learning
model
(MultiFeNet)
based
on
convolutional
neural
network
classification
tumours.
MultiFeNet
uses
multi‐scale
feature
scaling
extraction
in
magnetic
resonance
imaging
(MRI)
images.
Multi‐scaling
helps
to
learn
better
representation
MRI
image
enhanced
performance.
To
evaluate
model,
3064
scans
three
distinct
categories
(meningiomas,
gliomas
pituitary
tumours)
were
used.
The
obtained
96.4%
sensitivity,
F1‐score,
precision
accuracy
benchmark
Figshare
dataset.
addition,
an
ablation
study
conducted
with
objective
evaluating
role
multi‐scaling
Acute
leukemia
(AL),
classified
as
acute
myeloid
(AML)
and
lymphocytic
(ALL),
is
a
hematologic
malignancy
caused
by
the
uncontrolled
proliferation
of
leucocytes
in
bone
marrow.
Early
detection
AL
crucial
for
clinical
treatment.
Detection
methods
are
currently
blood
tests,
marrow
imaging,
spinal
fluid
tests.
However,
these
tests
have
drawbacks,
such
high
cost
time
consumption.
Liquid
biopsy
using
biological
fluids
or
serum
an
emerging
technique
noninvasive
cancer
monitoring.
Surface-enhanced
Raman
spectroscopy
(SERS),
which
enhanced
signals
interaction
plasmonic
nanostructures
with
analyte,
highly
sensitive
specific
method
simple
sample
preparation
that
has
been
used
combination
machine
learning
techniques
to
analyze
liquid
biopsy.
In
this
study,
we
developed
SERS-based
approach
enables
accurate
classification
AML
ALL
subtypes
prediction
disease
recurrence.
SERS
spectra
samples
from
24
healthy
individuals,
43
patients,
18
patients
were
obtained
Ag-based
substrate
clustered
hierarchical
cluster
analysis
(HCA).
The
then
three
commonly
classifiers,
namely,
support
vector
(SVM),
random
forest
(RF),
k-nearest
neighbor
(kNN).
Our
findings
demonstrate
RF
classifier
highest
accuracy
values,
96.1,
95.5,
98.5%
classifying
groups
predicting
recurrence
ALL,
respectively.
algorithms
represents
remarkable
advancement
realm
hematological
diagnostics,
particularly
ALL.
This
not
only
facilitates
precise
differentiation
but
also
introduces
novel
capability
prognosticating
Frontiers in Oncology,
Год журнала:
2023,
Номер
13
Опубликована: Дек. 6, 2023
Acute
lymphoblastic
leukemia
(ALL)
poses
a
significant
health
challenge,
particularly
in
pediatric
cases,
requiring
precise
and
rapid
diagnostic
approaches.
This
comprehensive
review
explores
the
transformative
capacity
of
deep
learning
(DL)
enhancing
ALL
diagnosis
classification,
focusing
on
bone
marrow
image
analysis.
Examining
ten
studies
conducted
between
2013
2023
across
various
countries,
including
India,
China,
KSA,
Mexico,
synthesis
underscores
adaptability
proficiency
DL
methodologies
detecting
leukemia.
Innovative
models,
notably
Convolutional
Neural
Networks
(CNNs)
with
Cat-Boosting,
XG-Boosting,
Transfer
Learning
techniques,
demonstrate
notable
Some
models
achieve
outstanding
accuracy,
one
CNN
reaching
100%
cancer
cell
classification.
The
incorporation
novel
algorithms
like
Cat-Swarm
Optimization
specialized
architectures
contributes
to
superior
classification
accuracy.
Performance
metrics
highlight
these
achievements,
consistently
outperforming
traditional
methods.
For
instance,
Cat-Boosting
attains
while
others
hover
around
99%,
showcasing
models'
robustness
diagnosis.
Despite
acknowledged
challenges,
such
as
need
for
larger
more
diverse
datasets,
findings
underscore
DL's
potential
reshaping
diagnostics.
high
numerical
accuracies
accentuate
promising
trajectory
toward
efficient
accurate
clinical
settings,
prompting
ongoing
research
address
challenges
refine
optimal
integration.
2021 International Conference on System, Computation, Automation and Networking (ICSCAN),
Год журнала:
2023,
Номер
unknown
Опубликована: Ноя. 17, 2023
Cancer
continues
to
be
a
global
health
challenge,
demanding
innovative
solutions
improve
early
detection
and
treatment
outcomes.
This
research
project
harnesses
the
power
of
deep
learning
in
field
medical
imaging
investigate
applicability
YOLOv8
(You
Only
Look
Once
version
8)
algorithm
for
diagnosing
various
cancer
types,
such
as
Acute
Lymphoblastic
Leukemia,
Cervical,
Lung,
Colon,
Oral,
Skin
cancers.
The
algorithm,
renowned
its
real-time
object
prowess,
represents
promising
candidate
automating
identification
classification
cancerous
regions
within
images.
study
encompasses
comprehensive
methodology,
starting
with
collection
preprocessing
diverse
well-annotated
image
datasets.
is
then
fine-tuned
trained
on
these
datasets,
capitalizing
capabilities
discern
lesions.
model's
performance
undergoes
evaluation
using
established
metrics,
guaranteeing
dependability
precision
clinical
setting.
findings
this
have
potential
offer
insightful
information
YOLOv8.
By
bridging
gap
between
cutting-edge
technology
practice,
seeks
advance
provide
foundation
more
precise,
efficient,
accessible
methods.
Ultimately,
goal
enhance
diagnosis
cancer,
offering
new
possibilities
timely
intervention
improved
patient
PLoS ONE,
Год журнала:
2023,
Номер
18(10), С. e0292172 - e0292172
Опубликована: Окт. 9, 2023
Cancer
is
a
serious
public
health
concern
worldwide
and
the
leading
cause
of
death.
Blood
cancer
one
most
dangerous
types
cancer.
Leukemia
type
that
affects
blood
cell
bone
marrow.
Acute
leukemia
chronic
condition
fatal
if
left
untreated.
A
timely,
reliable,
accurate
diagnosis
at
an
early
stage
critical
to
treating
preserving
patients'
lives.
There
are
four
leukemia,
namely
acute
lymphocytic
myelogenous
in
extracting,
leukemia.
Recognizing
these
cancerous
development
cells
often
done
via
manual
analysis
microscopic
images.
This
requires
extraordinarily
skilled
pathologist.
symptoms
might
include
lethargy,
lack
energy,
pale
complexion,
recurrent
infections,
easy
bleeding
or
bruising.
One
challenges
this
area
identifying
subtypes
for
specialized
treatment.
Study
carried
out
increase
precision
assist
personalized
plans
treatment,
improve
general
leukemia-related
healthcare
practises.
In
research,
we
used
gene
expression
data
from
Curated
Microarray
Database
(CuMiDa).
Microarrays
ideal
studying
cancer,
however,
categorizing
pattern
microarray
information
can
be
challenging.
proposed
study
uses
feature
selection
methods
machine
learning
techniques
predict
classify
CuMiDa
(GSE9476).
research
work
utilized
linear
programming
(LP)
as
machine-learning
technique
classification.
Linear
model
classifies
predicts
Bone_Marrow_CD34,
Bone
Marrow,
AML,
PB,
PBSC
CD34.
Before
using
LP
model,
selected
25
features
given
dataset
22283
features.
These
significant
were
distinguishing
The
classification
accuracy
98.44%.
PLoS ONE,
Год журнала:
2025,
Номер
20(1), С. e0313277 - e0313277
Опубликована: Янв. 30, 2025
The
incidence
of
acute
myeloid
leukemia
(AML)
is
increasing
annually,
and
timely
diagnostic
treatments
can
substantially
improve
patient
survival
rates.
AML
typing
traditionally
relies
on
manual
microscopy
for
classifying
counting
cells,
which
time-consuming,
laborious,
subjective.
Therefore,
developing
a
reliable
automated
model
cell
classification
imperative.
This
study
evaluated
the
performance
five
widely-used
models
largest
publicly
available
bone
marrow
dataset
(BM).
However,
accuracy
significantly
affected
by
imbalance
in
distribution
types.
To
address
this
issue,
analyzed
different
Loss
functions
seven
attention
mechanisms.
When
chosen,
Swin
Transformer
V2
was
found
to
perform
best.
lightweight
RegNetX-3.2gf
had
fewer
parameters
faster
inference
speed
than
V2,
its
F1
Score
only
0.032
lower
that
V2.
Accordingly,
strongly
recommended
practical
applications.
During
evaluation
function
mechanism,
Cost-Sensitive
Function
(CS)
channel
mechanism
Squeeze-and-Excitation
Networks
(SE)
demonstrated
superior
performance.
optimal
(RegNetX-3.2gf
+
CS
SE)
achieved
an
average
precision
68.183%,
recall
63.722%,
65.155%.
exhibited
improved
compared
original
results,
achieving
enhancement
17.183%
10.655%
Score.
Finally,
class
activation
maps
demonstrate
our
focused
cells
themselves,
especially
nucleus
when
making
classifications.
It
proved
reliable.
provided
important
reference
application
model,
promoting
development
intelligent
AML.