NBCDC‐YOLOv8: A new framework to improve blood cell detection and classification based on YOLOv8
IET Computer Vision,
Journal Year:
2025,
Volume and Issue:
19(1)
Published: Jan. 1, 2025
Abstract
In
recent
years,
computer
technology
has
successfully
permeated
all
areas
of
medicine
and
its
management,
it
now
offers
doctors
an
accurate
rapid
means
diagnosis.
Existing
blood
cell
detection
methods
suffer
from
low
accuracy,
which
is
caused
by
the
uneven
distribution,
high
density,
mutual
occlusion
different
types
in
microscope
images,
this
article
introduces
NBCDC‐YOLOv8:
a
new
framework
to
improve
classification
based
on
YOLOv8.
Our
innovates
several
fronts:
uses
Mosaic
data
augmentation
enrich
dataset
add
small
targets,
incorporates
space
depth
convolution
(SPD‐Conv)
tailored
for
cells
that
are
have
resolution,
Multi‐Separated
Enhancement
Attention
Module
(MultiSEAM)
enhance
feature
map
resolution.
Additionally,
integrates
bidirectional
pyramid
network
(BiFPN)
effective
multi‐scale
fusion
includes
four
heads
recognition
accuracy
various
sizes,
especially
target
platelets.
Evaluated
Blood
Cell
Classification
Dataset
(BCCD),
NBCDC‐YOLOv8
obtains
mean
average
precision
(mAP)
94.7%,
thus
surpasses
original
YOLOv8n
2.3%.
Language: Английский
Morphological Analysis and Subtype Detection of Acute Myeloid Leukemia in High-Resolution Blood Smears Using ConvNeXT
AI,
Journal Year:
2025,
Volume and Issue:
6(3), P. 45 - 45
Published: Feb. 24, 2025
(1)
Background:
Acute
Myeloid
Leukemia
(AML)
is
a
complex
hematologic
malignancy
where
accurate
subtype
classification
crucial
for
targeted
treatment
and
improved
patient
outcomes.
Automated
AML
detection
especially
important
underrepresented
subtypes
to
ensure
equitable
diagnostics;
(2)
Methods:
This
study
explores
the
potential
of
ConvNeXt,
an
advanced
convolutional
neural
network
architecture,
classifying
high-resolution
peripheral
blood
smear
images
into
subtypes.
A
deep
learning
pipeline
was
developed,
integrating
Stochastic
Weight
Averaging
(SWA)
model
stability,
Mixup
data
augmentation
enhance
generalization,
Grad-CAM
interpretability,
ensuring
biologically
meaningful
feature
visualization.
Various
models,
including
ResNet50
Vision
Transformers,
were
benchmarked
comparative
performance
analysis;
(3)
Results:
ConvNeXt
outperformed
ResNet50,
achieving
accuracy
95%
compared
91%
81%
transformer-based
models
(Vision
Transformers).
visualizations
provided
interpretable
heatmaps,
enhancing
trust
in
computational
predictions
bridging
gap
between
AI-driven
diagnostics
clinical
decision-making.
Ablation
studies
highlighted
contributions
augmentation,
optimizer
selection,
hyperparameter
tuning,
demonstrating
robustness
adaptability
model;
(4)
Conclusions:
advances
AI’s
role
hematopathology
by
combining
high
performance,
explainability,
scalability.
offers
robust,
interpretable,
scalable
solution
classification,
improving
diagnostic
precision
supporting
These
results
underscore
advancements
efficient
diagnostics.
Language: Английский
The future insights of AI Applications in Hematology diseases diagnosis and prognosis: Review Article
Salud Ciencia y Tecnología,
Journal Year:
2025,
Volume and Issue:
5, P. 1430 - 1430
Published: Feb. 13, 2025
Artificial
intelligence
(AI)
is
rapidly
altering
the
field
of
hematology,
providing
novel
approaches
to
diagnosis,
prognosis,
and
management
hematological
illnesses.
AI
technologies,
including
machine
learning
(ML)
deep
(DL),
allow
for
analysis
massive
volumes
clinical,
genetic,
imaging
data,
resulting
in
more
accurate,
rapid,
individualized
care.
In
diagnostic
transforming
blood
smear
analysis,
bone
marrow
aspirations,
genomic
profiling
by
automating
cell
classification,
detecting
anomalies,
discovering
critical
genetic
changes
associated
with
AI-powered
models
are
also
improving
prognostic
skills
predicting
disease
progression,
treatment
response,
risk
relapse
illnesses
such
as
leukemia,
lymphoma,
anemia,
myeloproliferative
disorders.
Furthermore,
applications
precision
medicine
enable
clinicians
adapt
medicines
based
on
individual
profiles,
thereby
increasing
therapeutic
success
reducing
unwanted
effects.
The
combination
modern
technology
wearable
health
monitors
real-time
tools
promises
improve
patient
proactive
care
via
continuous
monitoring
adaptive
options.
As
develops,
it
has
enormous
potential
enabling
early
identification,
optimizing
regimens,
ultimately
survival
quality
life.
This
study
investigates
future
implications
emphasizing
their
revolutionary
impact
techniques.
Language: Английский
Application of ConvNeXt with Transfer Learning and Data Augmentation for Malaria Parasite Detection in Resource-Limited Settings Using Microscopic Images
medRxiv (Cold Spring Harbor Laboratory),
Journal Year:
2024,
Volume and Issue:
unknown
Published: Nov. 4, 2024
Abstract
Malaria
is
one
of
the
most
widespread
and
deadly
diseases
across
globe,
especially
in
sub-Saharan
Africa
other
parts
developing
world.
This
primarily
because
incorrect
or
late
diagnosis.
Existing
diagnostic
techniques
mainly
depend
on
microscopic
identification
parasites
blood
smear
stained
with
special
dyes,
which
have
drawbacks
such
as
being
time-consuming,
depending
skilled
personnel
vulnerable
to
errors.
work
seeks
overcome
these
challenges
by
proposing
a
deep
learning-based
solution
ConvNeXt
architecture
incorporating
transfer
learning
data
augmentation
automate
malaria
parasite
thin
images.
study’s
dataset
was
set
images
equal
numbers
parasitised
uninfected
samples
drawn
from
public
database
patients
Bangladesh.
To
detect
given
smears,
models
were
fine-tuned.
improve
effectiveness
models,
vast
number
strategies
used
so
that
could
well
various
image
capture
conditions
perform
even
environments
limited
resources.
The
Tiny
model
performed
better,
particularly
re-tuned
version,
than
Swin
Tiny,
ResNet18,
ResNet50,
an
accuracy
95%.
On
hand,
re-modified
version
V2
reached
98%
accuracy.
These
findings
show
potential
implement
ConvNeXt-based
systems
regions
scarce
healthcare
facilities
for
effective
affordable
Language: Английский