I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
2
,
A
p
r
il
20
26
,
p
p
.
1
4
8
6
~
1
4
9
6
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
2
.
p
p
1
4
8
6
-
1
4
9
6
1486
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
RBC_
Frame_
Net
:
a
hybrid
deep
le
a
rning
f
ra
mewo
r
k f
o
r
detec
tion o
f
r
ed b
lo
o
d cells in
ma
la
ria
diag
no
stic smea
r
M
uh
a
m
m
a
d Sha
m
ee
m
P
.
,
M
a
t
hia
ra
s
i Ba
la
k
rish
na
n
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
H
i
n
d
u
st
a
n
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
S
c
i
e
n
c
e
,
C
h
e
n
n
a
i
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l
1
,
2
0
2
5
R
ev
is
ed
J
an
13
,
2
0
2
6
Acc
ep
ted
J
an
25
,
2
0
2
6
M
a
laria
c
o
n
ti
n
u
e
s
to
p
o
se
a
m
a
jo
r
g
l
o
b
a
l
h
e
a
lt
h
t
h
re
a
t,
e
sp
e
c
ially
in
a
re
a
s
wh
e
re
ti
m
e
ly
a
n
d
a
c
c
u
ra
te
d
iag
n
o
sis
is
e
ss
e
n
ti
a
l
f
o
r
e
ffe
c
ti
v
e
t
re
a
tme
n
t.
Co
n
v
e
n
ti
o
n
a
l
d
iag
n
o
stic
tec
h
n
i
q
u
e
s,
su
c
h
a
s
m
a
n
u
a
ll
y
e
x
a
m
in
i
n
g
G
iem
sa
-
sta
in
e
d
b
l
o
o
d
sm
e
a
rs,
a
re
o
ften
ti
m
e
-
in
ten
siv
e
,
lab
o
ri
o
u
s,
a
n
d
s
u
sc
e
p
ti
b
le
t
o
h
u
m
a
n
e
rr
o
r.
T
o
o
v
e
rc
o
m
e
t
h
e
se
c
h
a
ll
e
n
g
e
s,
th
is
stu
d
y
p
re
se
n
ts
re
d
b
l
o
o
d
c
e
ll
fra
m
e
n
e
two
rk
(
RBC_
F
ra
m
e
_
Ne
t
)
,
a
n
o
v
e
l
d
e
e
p
-
lea
rn
in
g
f
ra
m
e
wo
rk
th
a
t
c
o
m
b
in
e
s
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
s)
with
tra
n
sfo
rm
e
r
-
b
a
se
d
a
rc
h
it
e
c
tu
re
s,
a
u
g
m
e
n
ted
b
y
a
tt
e
n
t
io
n
m
e
c
h
a
n
ism
s,
f
o
r
t
h
e
a
u
to
m
a
ted
id
e
n
ti
fica
ti
o
n
o
f
RBCs
in
m
a
laria
sm
e
a
r
i
m
a
g
e
s.
Th
e
fra
m
e
wo
rk
lev
e
ra
g
e
s
th
e
c
o
n
v
o
lu
t
io
n
a
l
b
lo
c
k
a
tt
e
n
t
io
n
m
o
d
u
les
(
CBAM
)
-
UN
e
t
m
o
d
e
l
f
o
r
se
g
m
e
n
tatio
n
,
e
n
h
a
n
c
in
g
b
o
th
sp
a
ti
a
l
a
n
d
c
h
a
n
n
e
l
fe
a
tu
re
s
th
r
o
u
g
h
CBAM
a
n
d
in
teg
ra
tes
th
e
d
e
tec
ti
o
n
tra
n
sfo
rm
e
r
(DET
R)
t
o
a
c
c
u
ra
tely
d
e
tec
t
a
n
d
c
las
sify
RBCs
wit
h
in
t
h
e
d
i
a
g
n
o
stic
ima
g
e
s.
T
h
e
m
o
d
e
l
a
c
h
iev
e
d
o
u
tstan
d
in
g
p
e
rfo
rm
a
n
c
e
with
a
se
g
m
e
n
tatio
n
i
n
ters
e
c
ti
o
n
o
v
e
r
u
n
io
n
(Io
U
)
o
f
0
.
9
7
,
a
Dic
e
c
o
e
fficie
n
t
o
f
0
.
9
8
,
a
n
d
n
e
a
r
-
p
e
rfe
c
t
d
e
tec
ti
o
n
re
su
lt
s
(p
re
c
isio
n
:
0
.
9
9
9
,
re
c
a
ll
:
0
.
9
9
8
,
a
n
d
m
e
a
n
a
v
e
ra
g
e
p
re
c
isio
n
(m
AP
):
0
.
9
9
5
)
.
Wh
e
n
c
o
m
p
a
re
d
t
o
lea
d
in
g
m
o
d
e
ls
su
c
h
a
s
YO
LOv
8
,
fa
ste
r
re
g
io
n
-
b
a
se
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
(
F
a
ste
r
R
-
CNN
)
,
a
n
d
Eff
icie
n
tDe
t
-
D3
,
a
n
d
RBC_
F
ra
m
e
_
Ne
t
d
e
m
o
n
stra
te
d
su
p
e
ri
o
r
a
c
c
u
ra
c
y
a
n
d
ro
b
u
st
n
e
ss
.
Th
e
in
c
lu
sio
n
o
f
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
s
a
n
d
a
h
y
b
ri
d
a
rc
h
it
e
c
tu
re
e
n
h
a
n
c
e
it
s
a
d
a
p
tab
il
it
y
,
m
a
k
i
n
g
it
we
ll
-
su
i
ted
fo
r
d
e
p
lo
y
m
e
n
t
in
re
a
l
-
wo
rl
d
,
re
so
u
rc
e
-
li
m
it
e
d
e
n
v
ir
o
n
m
e
n
ts
a
n
d
p
o
siti
o
n
i
n
g
it
a
s
a
v
a
lu
a
b
le
a
ss
e
t
in
a
u
to
m
a
ted
m
a
laria
d
iag
n
o
stics
.
K
ey
w
o
r
d
s
:
Atten
tio
n
m
ec
h
an
is
m
s
Dee
p
lear
n
in
g
Hy
b
r
id
m
o
d
els
Ma
lar
ia
d
iag
n
o
s
is
R
ed
b
lo
o
d
ce
ll d
etec
tio
n
T
r
an
s
f
o
r
m
e
r
s
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mu
h
am
m
ad
Sh
a
m
ee
m
P
.
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
Hin
d
u
s
tan
I
n
s
titu
te
o
f
T
ec
h
n
o
l
o
g
y
a
n
d
Scien
ce
C
h
en
n
ai,
I
n
d
ia
E
m
ail: r
p
.
2
1
7
0
3
0
3
5
@
s
tu
d
en
t.
h
in
d
u
s
tan
u
n
iv
.
ac
.
in
1.
I
NT
RO
D
UCT
I
O
N
Ma
lar
ia
r
em
ain
s
a
m
ajo
r
th
r
e
at
to
h
u
m
an
h
e
alth
wo
r
ld
wid
e
,
p
ar
ticu
lar
ly
in
tr
o
p
ical
an
d
s
u
b
tr
o
p
ical
r
eg
io
n
s
wh
er
e
q
u
ick
an
d
p
r
ec
is
e
d
iag
n
o
s
is
is
e
s
s
en
tial
f
o
r
ef
f
ec
tiv
e
tr
ea
tm
e
n
t
an
d
d
is
ea
s
e
m
itig
atio
n
.
C
o
n
v
en
tio
n
al
d
iag
n
o
s
tic
tech
n
iq
u
es
lik
e
m
an
u
ally
in
s
p
e
ctin
g
Giem
s
a
-
s
tain
ed
b
lo
o
d
s
m
ea
r
s
u
n
d
er
a
m
icr
o
s
co
p
e
ar
e
o
f
te
n
s
lo
w,
r
e
q
u
ir
e
s
ig
n
if
ican
t
ef
f
o
r
t,
an
d
a
r
e
s
u
s
ce
p
tib
le
to
h
u
m
an
m
is
tak
es,
p
ar
ticu
lar
ly
in
u
n
d
er
-
r
eso
u
r
ce
d
en
v
ir
o
n
m
en
t
s
.
I
n
r
esp
o
n
s
e,
s
cien
tis
ts
ar
e
in
cr
ea
s
in
g
ly
e
x
p
lo
r
in
g
t
h
e
u
s
e
o
f
a
u
to
m
ated
,
i
n
tellig
en
t
d
iag
n
o
s
tic
to
o
ls
p
o
wer
ed
b
y
r
ec
en
t
d
e
v
elo
p
m
e
n
ts
in
m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
[
1
]
.
C
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NN
s
)
h
av
e
p
r
o
v
en
h
ig
h
ly
ef
f
ec
tiv
e
in
an
aly
zin
g
m
e
d
ical
im
ag
es,
in
clu
d
in
g
th
o
s
e
u
s
ed
f
o
r
d
iag
n
o
s
in
g
m
a
lar
ia.
Fo
r
e
x
am
p
le,
Sar
k
ar
et
a
l
.
[
2
]
p
r
o
p
o
s
ed
a
lig
h
tweig
h
t
C
NN
m
o
d
el
th
at
m
atch
ed
th
e
ac
cu
r
ac
y
o
f
d
ee
p
er
n
etwo
r
k
s
s
u
ch
as
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
1
6
-
la
y
er
n
et
wo
r
k
(
V
GG
-
16
)
a
n
d
r
esid
u
al
n
etwo
r
k
with
5
0
lay
e
r
s
(
R
esNet
-
50
)
,
b
u
t
with
m
u
c
h
lo
wer
c
o
m
p
u
tatio
n
al
d
em
a
n
d
s
.
I
n
a
n
o
th
er
s
tu
d
y
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
B
C
_
F
r
a
me_
N
et:
a
h
yb
r
id
d
e
ep
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
d
etec
tio
n
o
f red
b
lo
o
d
…
(
Mu
h
a
mma
d
S
h
a
mee
m
P
.
)
1487
B
o
it
an
d
Patil
[
3
]
d
esig
n
ed
a
s
tack
ed
C
NN
f
r
am
ewo
r
k
t
h
at
ac
h
iev
ed
a
r
e
m
ar
k
ab
le
ac
cu
r
a
cy
o
n
th
e
Natio
n
al
I
n
s
titu
tes
o
f
Hea
lth
(
NI
H
)
m
alar
ia
d
ataset,
s
h
o
wca
s
in
g
th
e
ca
p
ab
ilit
y
o
f
d
ee
p
m
o
d
els
to
ex
tr
ac
t
co
m
p
lex
f
ea
tu
r
es f
r
o
m
in
f
ec
ted
b
lo
o
d
c
ells
[
4
]
.
C
NN
-
b
ased
m
o
d
els
n
o
w
in
cl
u
d
e
atten
tio
n
p
r
o
ce
s
s
es
to
im
p
r
o
v
e
d
iag
n
o
s
tic
p
r
ec
is
io
n
.
I
n
o
r
d
e
r
to
im
p
r
o
v
e
s
eg
m
en
tatio
n
r
esu
lt
s
in
m
icr
o
s
c
o
p
ic
im
a
g
e
p
r
o
ce
s
s
in
g
,
Niy
o
g
is
u
b
izo
et
a
l
.
[
5
]
d
e
v
elo
p
e
d
a
n
atten
tio
n
-
g
u
id
e
d
r
esid
u
al
UNe
t
th
at
co
m
b
in
es
s
q
u
ee
ze
-
an
d
-
e
x
citatio
n
(
SE)
b
lo
c
k
s
with
atr
o
u
s
s
p
atial
p
y
r
am
id
p
o
o
lin
g
(
ASPP
)
.
L
ik
ewise,
B
o
it
an
d
Patil
[
3
]
d
ev
elo
p
ed
th
e
ad
ap
tiv
e
d
ee
p
co
n
v
o
lu
tio
n
a
l
n
etwo
r
k
(
ADCN
)
m
o
d
el,
wh
ich
u
tili
ze
d
atten
tio
n
m
o
d
u
les
to
em
p
h
asize
cr
itical
f
ea
tu
r
es
in
m
icr
o
s
co
p
y
im
ag
es,
ac
h
iev
in
g
an
ac
cu
r
ac
y
o
f
9
7
.
4
7
%.
T
r
an
s
f
o
r
m
e
r
-
b
ased
a
r
ch
itectu
r
es
h
av
e
also
g
ain
e
d
atten
ti
o
n
in
th
e
f
ield
o
f
m
alar
ia
d
iag
n
o
s
is
.
L
iu
et
a
l
.
[
6
]
d
ev
elo
p
e
d
th
e
ar
tific
ial
in
tellig
en
ce
(
AI
)
-
b
ased
d
etec
tio
n
s
y
s
tem
f
o
r
m
alar
ia
d
iag
n
o
s
is
f
r
o
m
s
m
ar
tp
h
o
n
e
th
in
-
b
lo
o
d
-
s
m
ea
r
im
ag
es
(
AI
DM
AN
)
s
y
s
tem
,
wh
ich
in
teg
r
ates
YOL
Ov
5
with
tr
an
s
f
o
r
m
er
m
o
d
els
f
o
r
ce
ll
d
etec
tio
n
an
d
class
if
icatio
n
,
ac
h
iev
in
g
a
d
iag
n
o
s
tic
ac
cu
r
ac
y
o
f
9
8
.
6
2
%
f
o
r
in
d
i
v
id
u
al
ce
lls
an
d
9
7
%
f
o
r
en
tire
b
lo
o
d
-
s
m
ea
r
im
ag
es.
T
h
is
h
y
b
r
id
ap
p
r
o
ac
h
illu
s
tr
ates
th
e
ef
f
ec
tiv
en
ess
o
f
co
m
b
in
in
g
C
NNs
w
ith
tr
an
s
f
o
r
m
er
s
to
b
o
o
s
t
d
iag
n
o
s
tic
ac
cu
r
ac
y
.
I
n
p
ar
allel,
th
e
d
ev
elo
p
m
en
t
o
f
lig
h
tweig
h
t
m
o
d
els
is
cr
u
cial
f
o
r
u
s
e
in
lo
w
-
r
eso
u
r
c
e
s
ettin
g
s
.
Nettu
r
et
a
l
.
[
7
]
i
n
tr
o
d
u
ce
d
Ultr
aL
ig
h
tSq
u
ee
ze
N
et
v
ar
ian
ts
,
wh
ich
r
ed
u
ce
d
th
e
n
u
m
b
e
r
o
f
tr
ain
ab
le
p
ar
am
eter
s
b
y
u
p
to
5
4
tim
es
co
m
p
ar
ed
to
Sq
u
ee
ze
Net1
.
1
,
with
o
n
ly
a
s
lig
h
t
d
ec
r
ea
s
e
in
ac
c
u
r
ac
y
.
Similar
ly
,
Ah
m
e
d
et
a
l
.
[
1
]
p
r
esen
ted
m
o
b
ile
m
alar
ia
atten
tio
n
n
etwo
r
k
(
M2
ANE
T
)
,
a
m
o
b
ile
-
o
p
tim
ize
d
m
o
d
el
th
a
t
in
teg
r
ates
Mo
b
ileNetV3
co
m
p
o
n
en
ts
with
an
ad
ap
ted
g
lo
b
al
m
u
lti
-
h
ea
d
s
elf
-
atten
tio
n
m
ec
h
an
is
m
,
d
el
iv
er
in
g
s
u
p
er
i
o
r
p
er
f
o
r
m
a
n
ce
in
b
o
th
ac
cu
r
ac
y
an
d
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
co
m
p
ar
ed
to
o
th
e
r
co
m
p
ac
t a
r
ch
itectu
r
es.
E
n
s
em
b
le
l
ea
r
n
in
g
tech
n
iq
u
es
h
av
e
b
ee
n
ap
p
lied
to
im
p
r
o
v
e
th
e
r
o
b
u
s
tn
ess
an
d
g
en
e
r
a
lizatio
n
o
f
m
alar
ia
d
iag
n
o
s
tic
m
o
d
els.
R
ajar
am
an
et
a
l
.
[
8
]
in
t
r
o
d
u
ce
d
th
e
in
ter
p
r
etab
le
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
I
C
NN
)
-
en
s
em
b
le
m
o
d
el,
wh
ich
ag
g
r
eg
ates
o
u
tp
u
ts
f
r
o
m
s
ev
er
al
C
NN
s
an
aly
zin
g
h
ig
h
-
r
eso
lu
tio
n
im
ag
e
ch
an
n
els,
r
ea
ch
in
g
an
ac
cu
r
ac
y
o
f
9
9
.
6
7
%.
Similar
ly
,
Po
o
s
t
ch
i
et
a
l
.
[
9
]
d
ev
elo
p
e
d
th
e
C
NN
-
d
ee
p
ex
tr
em
e
lear
n
in
g
m
ac
h
in
e
(
DE
L
M
)
m
o
d
el,
wh
ich
f
ea
tu
r
ed
a
p
a
r
asit
e
in
f
lato
r
m
ec
h
an
is
m
d
esig
n
e
d
t
o
en
h
a
n
ce
p
a
r
asit
e
v
is
ib
ilit
y
in
lo
w
-
co
n
tr
ast
im
a
g
es,
ac
h
iev
in
g
an
ac
cu
r
ac
y
o
f
9
9
.
6
6
%.
E
f
f
icien
tNet
-
b
ased
ar
ch
itectu
r
es
h
av
e
also
s
h
o
wn
p
r
o
m
is
e
in
th
is
d
o
m
ain
.
I
n
o
r
d
er
to
s
im
u
ltan
e
o
u
s
ly
ca
p
tu
r
e
s
p
atial
an
d
tem
p
o
r
al
ch
ar
ac
te
r
is
tic
s
f
r
o
m
r
ed
b
lo
o
d
ce
ll
(
R
B
C
)
im
ag
es,
R
ajar
am
an
et
a
l
.
[
4
]
in
v
esti
g
ated
th
e
in
teg
r
atio
n
o
f
C
NNs
with
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs),
esp
ec
ially
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
an
d
g
ated
r
ec
u
r
r
e
n
t
u
n
it
(
GR
U)
lay
er
s
.
T
h
i
s
r
esu
lted
in
an
ac
c
u
r
ac
y
o
f
9
6
.
2
0
%.
T
h
ese
ap
p
r
o
ac
h
es
h
ig
h
lig
h
t
t
h
e
ad
v
a
n
tag
es
o
f
co
m
b
in
in
g
d
if
f
e
r
en
t
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
es
t
o
im
p
r
o
v
e
f
ea
tu
r
e
r
ep
r
esen
tatio
n
an
d
ac
c
o
u
n
t
f
o
r
s
eq
u
e
n
tial
p
atter
n
s
in
im
a
g
in
g
d
ata.
So
m
e
r
esear
ch
e
f
f
o
r
ts
h
a
v
e
co
n
ce
n
t
r
ate
d
o
n
ev
alu
atin
g
th
e
q
u
ality
o
f
b
lo
o
d
s
m
ea
r
im
ag
es
as
p
ar
t
o
f
th
e
d
iag
n
o
s
tic
p
r
o
ce
s
s
.
C
ar
d
en
as
et
a
l
.
[
1
0
]
d
esig
n
e
d
a
co
m
p
r
eh
en
s
iv
e
s
y
s
tem
ca
p
a
b
le
o
f
d
etec
tin
g
an
d
class
if
y
in
g
m
alar
ia
p
ar
asit
es,
ass
es
s
in
g
th
e
q
u
ality
o
f
m
i
cr
o
s
co
p
ic
im
ag
es,
an
d
c
o
u
n
t
in
g
leu
k
o
c
y
tes
in
R
o
m
an
o
wsk
y
-
s
tain
ed
th
ick
b
l
o
o
d
s
m
ea
r
s
.
T
h
is
in
teg
r
ated
a
p
p
r
o
ac
h
o
f
f
er
s
an
all
-
in
-
o
n
e
s
o
lu
tio
n
p
a
r
ticu
lar
ly
s
u
ited
f
o
r
u
s
e
in
l
o
w
-
r
eso
u
r
ce
h
ea
lth
ca
r
e
en
v
i
r
o
n
m
e
n
ts
.
Desp
ite
n
o
tab
le
p
r
o
g
r
ess
,
ac
h
i
ev
in
g
h
ig
h
d
iag
n
o
s
tic
ac
cu
r
ac
y
with
o
u
t
co
m
p
r
o
m
is
in
g
co
m
p
u
tatio
n
al
ef
f
icien
cy
r
em
ain
s
a
k
e
y
c
h
allen
g
e,
esp
ec
ially
in
s
ettin
g
s
with
lim
ited
r
eso
u
r
ce
s
.
I
n
o
r
d
er
t
o
a
d
d
r
ess
th
is
p
r
o
b
lem
,
we
p
r
o
p
o
s
e
r
ed
b
lo
o
d
ce
ll
f
r
am
e
n
etwo
r
k
(
R
B
C
_
Fra
m
e_
Net
)
,
a
h
y
b
r
id
d
ee
p
lear
n
in
g
f
r
am
ewo
r
k
t
h
at
in
co
r
p
o
r
ates
atten
tio
n
m
ec
h
an
is
m
s
an
d
t
h
e
ca
p
ab
ilit
ies
o
f
tr
an
s
f
o
r
m
e
r
ar
ch
itectu
r
es
an
d
C
NNs.
T
h
is
m
o
d
el
is
d
esig
n
ed
to
ef
f
ec
ti
v
ely
d
etec
t
RBC
s
in
m
alar
i
a
d
iag
n
o
s
tic
s
m
ea
r
s
wh
ile
m
ain
tain
in
g
a
b
alan
ce
b
etwe
en
p
er
f
o
r
m
an
ce
an
d
ef
f
icien
cy
,
m
a
k
in
g
it
well
-
s
u
ited
f
o
r
p
r
ac
tical
d
ep
lo
y
m
en
t
in
r
ea
l
-
wo
r
ld
,
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
R
ec
en
t
ad
v
an
ce
m
e
n
ts
in
d
ee
p
lear
n
i
n
g
h
av
e
g
r
ea
tly
im
p
r
o
v
ed
au
to
m
ated
d
etec
tio
n
o
f
m
alar
ia
-
in
f
ec
ted
R
B
C
s
.
Yan
g
et
a
l
.
[
1
1
]
p
r
o
p
o
s
ed
a
C
NN
-
b
ased
f
r
am
ewo
r
k
f
o
r
m
ala
r
ia
p
a
r
asit
e
d
etec
tio
n
in
th
in
b
lo
o
d
s
m
ea
r
im
a
g
es,
lev
er
a
g
in
g
tr
an
s
f
er
lear
n
i
n
g
with
p
r
e
-
tr
ain
ed
n
etwo
r
k
s
an
d
e
x
ten
s
iv
e
d
ata
au
g
m
en
tatio
n
.
T
h
eir
m
o
d
el
ac
h
iev
ed
an
ac
cu
r
ac
y
a
b
o
v
e
9
5
%,
in
d
icatin
g
r
o
b
u
s
t
p
er
f
o
r
m
a
n
ce
in
d
if
f
er
en
ti
atin
g
in
f
ec
te
d
ce
lls
u
n
d
er
v
ar
ie
d
im
a
g
in
g
co
n
d
itio
n
s
.
T
h
e
UNet
ar
ch
itectu
r
e
wa
s
f
ir
s
t
in
tr
o
d
u
ce
d
b
y
R
o
n
n
eb
e
r
g
er
et
a
l
.
[
1
2
]
an
d
h
as
s
u
b
s
eq
u
en
tly
b
ee
n
ex
ten
s
i
v
ely
u
s
ed
an
d
en
h
an
ce
d
f
o
r
b
i
o
m
ed
ical
im
ag
e
s
eg
m
e
n
tatio
n
.
Ad
d
in
g
t
o
th
is
,
a
n
u
m
b
er
o
f
s
tu
d
ies
in
c
o
r
p
o
r
at
ed
atten
tio
n
m
ec
h
a
n
is
m
s
to
en
h
an
ce
f
ea
tu
r
e
r
e
p
r
esen
tatio
n
.
W
o
o
et
a
l
.
[
1
3
]
p
r
o
p
o
s
ed
th
e
co
n
v
o
lu
tio
n
al
b
lo
ck
a
tten
tio
n
m
o
d
u
le
(
C
B
AM
)
,
wh
ich
,
wh
en
co
m
b
in
e
d
with
UNet
an
d
d
etec
tio
n
tr
an
s
f
o
r
m
er
s
(
DE
T
R
)
,
r
esu
lted
in
s
u
p
er
i
o
r
s
eg
m
en
tatio
n
in
ter
s
ec
tio
n
o
v
er
u
n
i
o
n
(
I
o
U
)
o
f
0
.
9
7
a
n
d
d
etec
tio
n
p
r
ec
is
io
n
clo
s
e
to
1
.
0
,
o
u
tp
e
r
f
o
r
m
in
g
b
aselin
e
m
o
d
els
in
b
o
th
s
p
atial
a
n
d
ch
an
n
e
l
f
ea
tu
r
e
r
ef
in
em
en
t.
Hy
b
r
id
m
o
d
els
c
o
m
b
in
in
g
C
NN
b
ac
k
b
o
n
es
with
tr
an
s
f
o
r
m
er
-
b
ased
d
etec
tio
n
f
r
a
m
ewo
r
k
s
h
av
e
b
ee
n
in
cr
e
asin
g
ly
p
o
p
u
la
r
.
C
ar
io
n
et
a
l
.
[
1
4
]
p
r
esen
ted
DE
T
R
,
wh
ich
em
p
lo
y
s
tr
an
s
f
o
r
m
er
atten
tio
n
f
o
r
en
d
-
to
-
e
n
d
o
b
ject
d
etec
tio
n
a
n
d
s
eg
m
en
tatio
n
.
Ap
p
lied
to
m
alar
ia
ce
ll
d
etec
tio
n
,
DE
T
R
d
em
o
n
s
tr
ated
r
o
b
u
s
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
8
6
-
1
4
9
6
1488
lo
ca
lizatio
n
ca
p
ab
ilit
ies,
ac
h
iev
in
g
m
ea
n
av
er
a
g
e
p
r
ec
is
io
n
(
m
AP)
s
co
r
es
ar
o
u
n
d
0
.
9
1
,
with
p
o
ten
tial
to
f
u
r
th
er
im
p
r
o
v
e
wh
en
in
teg
r
ated
with
atten
tio
n
-
b
ased
s
eg
m
en
ts
.
Ho
war
d
et
a
l.
[
1
5
]
ex
p
lo
r
ed
ef
f
icien
t
ar
ch
itectu
r
es
f
o
r
d
ep
l
o
y
m
en
t
in
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
i
r
o
n
m
en
ts
b
y
in
tr
o
d
u
cin
g
Mo
b
il
eNe
t,
a
lig
h
tweig
h
t
C
NN
o
p
tim
ized
f
o
r
s
p
ee
d
an
d
ac
cu
r
ac
y
.
C
o
m
b
i
n
ed
with
s
in
g
le
s
h
o
t
m
u
ltib
o
x
d
etec
to
r
(
S
SD)
,
th
is
ap
p
r
o
ac
h
ac
h
iev
ed
d
etec
tio
n
ac
cu
r
ac
y
o
v
er
9
2
%
with
in
f
er
e
n
ce
tim
es
u
n
d
e
r
3
0
m
s
p
er
im
ag
e,
m
ak
in
g
it
p
r
ac
tical
f
o
r
r
ea
l
-
tim
e
d
iag
n
o
s
tic
s
u
p
p
o
r
t
in
lo
w
-
r
eso
u
r
ce
s
ettin
g
s
.
Alzu
b
aid
i
et
a
l
.
[
1
6
]
tar
g
ete
d
s
ick
le
ce
ll
an
em
ia
d
etec
tio
n
b
y
i
n
t
eg
r
atin
g
s
h
a
p
e
d
escr
ip
to
r
s
with
C
NN
class
if
ier
s
to
id
en
tify
m
o
r
p
h
o
lo
g
ic
al
ab
n
o
r
m
alities
in
R
B
C
s
.
T
h
eir
h
y
b
r
id
ap
p
r
o
ac
h
y
ield
ed
class
if
icatio
n
ac
cu
r
ac
y
ex
ce
ed
in
g
9
5
%,
u
n
d
er
s
c
o
r
in
g
th
e
v
alu
e
o
f
co
m
b
in
in
g
tr
ad
itio
n
al
h
an
d
cr
a
f
ted
f
ea
tu
r
es with
d
ee
p
lear
n
in
g
.
O
r
tet
et
a
l
.
[
1
7
]
p
r
o
p
o
s
ed
a
m
u
lti
-
task
lear
n
in
g
n
etwo
r
k
th
at
s
im
u
ltan
eo
u
s
ly
s
eg
m
en
ts
an
d
class
if
ies
R
B
C
s
b
y
s
h
ar
in
g
en
co
d
er
lay
er
s
a
n
d
u
s
in
g
task
-
s
p
ec
if
ic
d
ec
o
d
er
s
.
T
h
is
m
eth
o
d
im
p
r
o
v
ed
b
o
th
s
eg
m
en
tatio
n
Dice
s
co
r
es
an
d
class
if
icatio
n
r
ec
all,
illu
s
tr
atin
g
th
e
ad
v
an
tag
e
o
f
jo
in
t
lea
r
n
in
g
f
o
r
co
m
p
r
e
h
en
s
iv
e
R
B
C
an
aly
s
i
s
.
Sh
o
r
ten
an
d
Kh
o
s
h
g
o
f
taar
[
1
8
]
s
y
s
tem
atica
lly
s
tu
d
ied
d
ata
au
g
m
en
tatio
n
tec
h
n
iq
u
es
s
u
ch
as
r
o
tatio
n
,
f
lip
p
in
g
,
an
d
co
lo
r
jitt
er
in
g
to
en
h
a
n
ce
R
B
C
d
etec
tio
n
g
en
er
aliza
tio
n
.
T
h
eir
r
esu
lts
s
h
o
wed
p
r
ec
is
io
n
im
p
r
o
v
em
en
t
s
o
f
u
p
to
4
%
o
n
u
n
s
ee
n
test
s
am
p
les,
co
n
f
i
r
m
i
n
g
au
g
m
en
tatio
n
'
s
r
o
le
in
r
ed
u
c
in
g
o
v
er
f
itti
n
g
.
Han
et
a
l
.
[
1
9
]
f
o
cu
s
ed
o
n
m
o
d
el
c
o
m
p
r
ess
io
n
s
tr
ateg
ies,
in
clu
d
in
g
p
r
u
n
in
g
a
n
d
q
u
an
ti
za
tio
n
,
to
d
e
v
elo
p
e
f
f
icien
t
C
NNs
f
o
r
R
B
C
d
etec
tio
n
wi
th
m
in
im
al
lo
s
s
in
ac
cu
r
ac
y
(
ab
o
v
e
9
0
%).
T
h
ei
r
wo
r
k
is
p
ar
ticu
lar
ly
r
elev
an
t
f
o
r
d
ep
lo
y
in
g
m
o
d
els
o
n
m
o
b
ile
d
e
v
ices
an
d
em
b
ed
d
e
d
s
y
s
tem
s
.
E
f
f
icien
tDet,
p
r
o
p
o
s
ed
b
y
T
an
et
a
l
.
[
2
0
]
,
b
alan
ce
s
d
ete
ctio
n
ac
cu
r
ac
y
an
d
co
m
p
u
tat
io
n
al
co
s
t.
Ap
p
lied
to
R
B
C
d
etec
tio
n
,
E
f
f
icien
tDet
-
D3
v
ar
ian
t
a
ch
iev
e
d
p
r
ec
is
io
n
o
f
0
.
9
7
5
a
n
d
in
f
er
en
ce
tim
e
o
f
6
5
m
s
,
d
em
o
n
s
tr
atin
g
a
p
r
ac
tical
s
p
e
ed
-
ac
cu
r
ac
y
tr
ad
e
-
o
f
f
s
u
itab
le
f
o
r
clin
ical
ap
p
licatio
n
s
.
Niy
o
g
is
u
b
izo
et
a
l
.
[
5
]
in
tr
o
d
u
ce
d
an
in
n
o
v
ativ
e
ap
p
r
o
ac
h
b
y
co
m
b
in
in
g
a
n
atten
tio
n
-
g
u
id
e
d
r
esid
u
al
UNet
with
SE
co
n
n
ec
tio
n
s
an
d
ASPP
f
o
r
en
h
an
ce
d
ce
ll
s
eg
m
en
tatio
n
in
m
icr
o
s
co
p
y
im
a
g
es.
T
h
eir
m
eth
o
d
in
teg
r
ates
d
ee
p
lear
n
in
g
with
wate
r
s
h
ed
-
b
ased
tech
n
iq
u
es,
a
ch
iev
in
g
s
u
p
er
io
r
s
eg
m
en
tatio
n
ac
cu
r
ac
y
co
m
p
ar
ed
to
tr
ad
iti
o
n
a
l
m
eth
o
d
s
.
T
h
e
in
co
r
p
o
r
atio
n
o
f
atten
tio
n
m
e
ch
an
is
m
s
an
d
m
u
lti
-
s
ca
le
co
n
tex
t
ag
g
r
e
g
atio
n
allo
ws
f
o
r
p
r
ec
is
e
d
elin
ea
tio
n
o
f
ce
ll b
o
u
n
d
ar
ies,
d
em
o
n
s
tr
atin
g
th
e
ef
f
ec
tiv
e
n
ess
o
f
h
y
b
r
id
m
o
d
els in
b
io
m
e
d
ical
im
ag
e
a
n
aly
s
is
.
R
ec
en
t
ad
ap
tatio
n
s
o
f
th
is
ap
p
r
o
ac
h
i
n
te
g
r
ate
d
ee
p
lear
n
in
g
f
o
r
m
o
r
e
r
o
b
u
s
t
b
o
u
n
d
ar
y
d
eli
n
ea
tio
n
i
n
cr
o
wd
ed
ce
ll
r
e
g
io
n
s
.
B
r
eim
an
in
tr
o
d
u
ce
d
r
an
d
o
m
f
o
r
est
class
if
ier
s
u
s
in
g
h
an
d
cr
af
te
d
R
B
C
f
ea
tu
r
es
f
o
r
co
u
n
tin
g
an
d
class
if
icatio
n
,
ac
h
iev
in
g
s
tr
o
n
g
c
o
r
r
elatio
n
s
with
m
an
u
al
a
n
n
o
tatio
n
s
.
M
o
r
e
r
e
ce
n
t
wo
r
k
s
h
av
e
en
h
an
ce
d
th
ese
m
eth
o
d
s
with
d
ee
p
f
ea
tu
r
es
f
o
r
b
etter
s
ca
lab
ilit
y
an
d
ac
cu
r
ac
y
.
Dietter
ich
h
ig
h
lig
h
te
d
th
e
ef
f
ec
tiv
en
ess
o
f
en
s
em
b
le
lear
n
in
g
to
im
p
r
o
v
e
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
b
y
co
m
b
in
in
g
o
u
tp
u
ts
f
r
o
m
m
u
ltip
le
m
o
d
els.
E
n
s
em
b
les
o
f
C
NNs
an
d
t
r
an
s
f
o
r
m
er
s
h
av
e
led
to
m
AP
im
p
r
o
v
em
en
ts
u
p
t
o
0
.
9
7
in
R
B
C
d
etec
t
io
n
task
s
.
C
ar
io
n
et
a
l
.
[
1
4
]
also
em
p
h
asized
th
e
b
en
e
f
its
o
f
en
d
-
to
-
e
n
d
tr
a
n
s
f
o
r
m
e
r
m
o
d
e
ls
f
o
r
s
im
p
lify
in
g
p
ip
elin
es,
th
o
u
g
h
n
o
ted
th
e
s
lo
wer
co
n
v
er
g
e
n
ce
co
m
p
a
r
e
d
to
C
NN
-
b
ased
m
eth
o
d
s
,
in
d
icatin
g
r
o
o
m
f
o
r
f
u
r
th
er
o
p
tim
izatio
n
.
Ma
q
s
o
o
d
et
a
l
.
[
2
1
]
d
em
o
n
s
tr
ated
t
h
e
u
tili
ty
o
f
p
r
etr
ain
ed
Den
s
eNe
t
ar
ch
itectu
r
es
f
in
e
-
tu
n
e
d
f
o
r
m
alar
ia
d
etec
tio
n
,
ac
h
iev
i
n
g
h
i
g
h
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
b
y
e
x
p
lo
iti
n
g
d
ee
p
er
f
ea
tu
r
e
h
ier
ar
ch
ies.
Ok
tay
et
a
l.
[
2
2
]
d
ev
el
o
p
ed
a
f
r
am
ewo
r
k
in
teg
r
atin
g
atten
tio
n
-
g
u
id
ed
C
NNs
f
o
r
im
p
r
o
v
e
d
s
eg
m
en
tatio
n
o
f
R
B
C
b
o
u
n
d
ar
ies,
r
ep
o
r
tin
g
Dice
co
ef
f
icien
t
s
ex
ce
ed
in
g
0
.
9
6
,
u
n
d
er
s
co
r
in
g
th
e
im
p
o
r
ta
n
ce
o
f
s
p
atial
atten
tio
n
in
m
ed
ical
im
ag
e
s
eg
m
en
tatio
n
.
3.
M
E
T
H
O
D
I
n
th
is
s
ec
tio
n
,
co
n
v
en
tio
n
al
UNe
t
ar
ch
itectu
r
e,
C
B
AM
,
D
E
T
R
,
an
d
p
r
o
p
o
s
ed
C
B
AM
-
U
Net+
DE
T
R
f
r
am
ewo
r
k
f
o
r
d
etec
tio
n
o
f
RBC
s
in
m
alar
ia
s
m
ea
r
im
ag
es is
p
r
esen
ted
.
3
.
1
.
UNet
a
rc
hite
ct
ure
UNet
as
d
ep
icted
in
Fig
u
r
e
1
,
en
h
an
ce
s
tr
ad
itio
n
al
C
NNs
an
d
f
u
lly
c
o
n
v
o
lu
tio
n
al
n
etwo
r
k
s
(
FC
Ns)
b
y
in
co
r
p
o
r
atin
g
a
s
y
m
m
etr
ic
en
co
d
er
-
d
ec
o
d
e
r
ar
ch
itectu
r
e
co
n
n
ec
ted
th
r
o
u
g
h
s
k
ip
p
ath
way
s
.
T
h
is
d
esig
n
m
ak
es
it
esp
ec
ially
ef
f
ec
tiv
e
f
o
r
s
eg
m
en
tin
g
m
e
d
ical
im
ag
es.
UNet
'
s
s
y
m
m
etr
ic
en
co
d
er
-
d
ec
o
d
er
d
esig
n
allo
ws
it
to
ef
f
ec
tiv
ely
lear
n
b
o
th
g
l
o
b
al
c
o
n
tex
t
an
d
f
in
e
-
g
r
ain
ed
d
etails
with
in
an
im
ag
e,
wh
ich
is
cr
u
cial
f
o
r
pr
ec
is
e
s
eg
m
en
tatio
n
.
I
n
o
r
d
e
r
to
p
r
o
v
id
e
c
o
m
p
r
eh
en
s
iv
e
s
eg
m
en
tatio
n
o
u
tp
u
ts
,
th
e
en
co
d
er
co
m
p
r
ess
es
th
e
in
p
u
t
b
y
e
x
tr
ac
tin
g
f
ea
tu
r
es
an
d
r
ed
u
cin
g
s
p
atial
d
im
en
s
io
n
s
,
wh
ile
th
e
d
ec
o
d
er
g
r
a
d
u
ally
r
esto
r
es
th
e
s
p
atial
r
eso
lu
tio
n
.
B
y
s
en
d
i
n
g
f
ea
t
u
r
e
m
ap
s
s
tr
aig
h
t
f
r
o
m
th
e
e
n
co
d
er
to
th
e
a
p
p
r
o
p
r
iate
d
e
co
d
er
lay
e
r
s
,
s
k
ip
co
n
n
ec
tio
n
s
r
etain
cr
u
cial
s
p
a
tial
in
f
o
r
m
atio
n
th
at
is
lo
s
t
d
u
r
in
g
d
o
wn
s
am
p
lin
g
.
T
h
is
im
p
r
o
v
es
s
eg
m
en
tatio
n
ac
cu
r
ac
y
an
d
p
r
eser
v
es
h
ig
h
-
r
eso
lu
tio
n
f
ea
tu
r
es.
UNet
,
d
er
iv
ed
f
r
o
m
FC
Ns,
ca
n
p
r
o
ce
s
s
im
ag
es
o
f
v
ar
io
u
s
s
izes
an
d
g
en
er
ate
s
eg
m
en
tatio
n
m
ap
s
t
h
at
m
atch
t
h
e
in
p
u
t
d
im
en
s
io
n
s
,
o
f
f
er
in
g
f
lex
ib
ilit
y
in
m
e
d
ical
im
ag
e
an
aly
s
is
.
Ad
d
itio
n
ally
,
g
iv
en
t
h
at
m
ed
ical
d
atasets
o
f
ten
h
a
v
e
lim
ited
s
am
p
les,
UNet
is
d
esig
n
ed
to
p
er
f
o
r
m
well
with
s
m
all
d
ataset
s
b
y
lev
er
ag
in
g
d
ata
au
g
m
en
tatio
n
a
n
d
its
ar
ch
itectu
r
e
to
i
m
p
r
o
v
e
g
en
er
aliza
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
B
C
_
F
r
a
me_
N
et:
a
h
yb
r
id
d
e
ep
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
d
etec
tio
n
o
f red
b
lo
o
d
…
(
Mu
h
a
mma
d
S
h
a
mee
m
P
.
)
1489
Fig
u
r
e
1
.
F
lo
wch
ar
t
o
f
th
e
AI
-
b
ased
m
o
d
els an
d
ex
p
e
r
im
en
t
al
m
eth
o
d
s
ap
p
lied
3
.
2
.
Co
nv
o
lutio
na
l blo
ck
a
t
t
ent
io
n m
o
du
le
T
h
e
atten
tio
n
m
ec
h
a
n
is
m
w
o
r
k
s
b
y
im
itatin
g
h
o
w
h
u
m
an
s
v
is
u
ally
f
o
cu
s
o
n
k
e
y
ar
ea
s
o
f
in
f
o
r
m
atio
n
,
allo
win
g
m
o
d
els
to
p
r
io
r
itize
im
p
o
r
tan
t
p
ar
ts
o
f
th
e
in
p
u
t
an
d
im
p
r
o
v
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
.
I
n
n
eu
r
al
n
etwo
r
k
s
,
th
is
h
elp
s
h
ig
h
lig
h
t r
elev
an
t d
ata
f
o
r
th
e
s
p
e
cif
ic
task
wh
ile
r
ed
u
cin
g
th
e
i
m
p
ac
t o
f
ir
r
elev
a
n
t
n
o
is
e.
C
B
AM
,
is
a
lig
h
tweig
h
t
y
et
ef
f
ec
tiv
e
atten
tio
n
m
o
d
u
le
m
ad
e
to
en
h
a
n
ce
C
NN
f
ea
tu
r
e
r
ep
r
esen
tatio
n
.
I
n
o
r
d
e
r
to
ad
ap
tiv
el
y
r
ef
in
e
t
h
e
f
ea
tu
r
es,
it
cr
ea
tes
atten
tio
n
m
ap
s
in
two
d
im
en
s
io
n
s
—
ch
an
n
el
an
d
s
p
atial
—
an
d
ap
p
lies
th
ese
m
ap
s
to
th
e
i
n
p
u
t f
ea
tu
r
e
m
ap
s
.
Fig
u
r
e
2
d
e
p
icts
th
e
ar
ch
itectu
r
e
o
f
th
e
C
B
AM
.
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
C
B
A
M
T
h
e
ch
an
n
el
atten
tio
n
m
o
d
u
le
(
C
AM
)
p
r
o
d
u
ce
s
th
e
ch
a
n
n
el
atten
tio
n
m
ap
b
y
ca
p
t
u
r
in
g
th
e
r
elatio
n
s
h
ip
s
b
etwe
en
v
ar
io
u
s
f
ea
tu
r
e
ch
an
n
els.
First,
it
u
s
e
s
b
o
th
av
er
ag
e
p
o
o
lin
g
an
d
m
ax
im
u
m
p
o
o
lin
g
to
ex
tr
ac
t
g
lo
b
al
in
f
o
r
m
atio
n
f
r
o
m
th
e
f
ea
tu
r
e
m
ap
.
T
h
e
ch
a
n
n
el
atten
tio
n
m
ap
is
s
u
b
s
eq
u
en
tly
p
r
o
d
u
ce
d
b
y
p
ass
ed
th
ese
co
m
b
in
ed
f
in
d
in
g
s
v
ia
co
m
m
o
n
m
u
lti
-
lay
e
r
p
e
r
ce
p
tio
n
(
ML
P)
.
T
h
e
C
AM
is
s
h
o
wn
in
Fig
u
r
e
3
.
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
o
f
C
AM
T
h
e
s
p
atial
r
elatio
n
s
h
ip
s
with
in
th
e
f
ea
t
u
r
es
ar
e
th
e
m
ain
f
o
cu
s
o
f
th
e
s
p
atial
atten
ti
o
n
m
o
d
u
le
(
SAM)
.
T
wo
d
is
tin
ct
2
D
f
ea
t
u
r
e
m
a
p
s
ar
e
cr
ea
ted
b
y
ap
p
l
y
in
g
a
v
er
ag
e
p
o
o
lin
g
an
d
m
a
x
p
o
o
lin
g
ac
r
o
s
s
th
e
in
p
u
t
f
ea
tu
r
e
m
a
p
'
s
ch
an
n
el
d
im
en
s
io
n
.
T
h
e
s
p
atial
atten
tio
n
m
ap
is
th
en
cr
ea
ted
b
y
co
m
b
i
n
in
g
an
d
p
r
o
ce
s
s
in
g
th
ese
m
ap
s
u
s
in
g
a
co
n
v
o
l
u
tio
n
al
lay
er
.
T
h
e
SA
M
's ar
ch
itectu
r
e
is
s
h
o
wn
in
F
ig
u
r
e
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
8
6
-
1
4
9
6
1490
Fig
u
r
e
4
.
Ar
c
h
itectu
r
e
o
f
SAM
I
n
p
r
ac
tical
ap
p
licatio
n
s
,
th
e
C
AM
an
d
SAM
ar
e
ty
p
ically
ap
p
lied
o
n
e
a
f
ter
th
e
o
th
e
r
.
First,
C
A
M
p
r
o
ce
s
s
es
th
e
in
p
u
t
f
ea
tu
r
e
m
ap
,
an
d
th
e
r
ef
in
ed
o
u
tp
u
t
is
th
en
p
ass
ed
to
SAM.
T
h
e
r
es
u
ltin
g
f
ea
tu
r
e
m
ap
,
en
h
an
ce
d
b
y
b
o
th
atten
tio
n
m
o
d
u
les,
is
s
u
b
s
eq
u
en
tly
u
s
ed
i
n
f
u
r
th
er
co
n
v
o
lu
tio
n
al
lay
er
s
.
E
x
ten
s
iv
e
test
in
g
o
n
b
en
c
h
m
ar
k
d
atasets
s
u
ch
a
s
I
m
ag
eNe
t
-
1
K,
MS
C
OC
O
,
an
d
VOC
2
0
0
7
h
as
s
h
o
wn
th
at
in
co
r
p
o
r
atin
g
C
B
AM
n
o
tab
ly
b
o
o
s
ts
p
er
f
o
r
m
an
ce
in
task
s
lik
e
im
ag
e
cla
s
s
if
icatio
n
an
d
o
b
ject
d
etec
tio
n
.
T
h
is
im
p
r
o
v
em
en
t
co
m
es
with
o
n
ly
a
m
in
o
r
in
cr
ea
s
e
in
co
m
p
u
tatio
n
al
co
s
t
an
d
p
ar
am
eter
s
,
in
d
icatin
g
t
h
at
C
B
AM
ca
n
b
e
ef
f
ec
tiv
ely
in
teg
r
ated
in
to
a
v
a
r
iety
o
f
C
NN
ar
ch
itectu
r
es to
i
m
p
r
o
v
e
ac
cu
r
ac
y
with
o
u
t sacr
if
icin
g
ef
f
icien
c
y
.
3
.
3
.
Det
ec
t
io
n
t
ra
ns
f
o
r
m
er
DE
T
R
is
an
ad
v
an
ce
d
o
b
ject
d
etec
tio
n
m
o
d
el
d
ev
elo
p
ed
b
y
Face
b
o
o
k
AI
th
at
r
eim
ag
in
es
th
e
ty
p
ical
d
ete
ctio
n
wo
r
k
f
lo
w.
DE
T
R
h
a
n
d
les
o
b
ject
id
e
n
tific
atio
n
as
a
d
ir
ec
t
s
et
p
r
e
d
ictio
n
p
r
o
b
le
m
in
s
tead
o
f
r
ely
in
g
o
n
m
a
n
u
ally
co
n
s
tr
u
cte
d
ele
m
en
ts
lik
e
an
c
h
o
r
b
o
x
es,
r
e
g
io
n
s
u
g
g
esti
o
n
s
,
o
r
n
o
n
-
m
a
x
im
u
m
s
u
p
p
r
ess
io
n
(
NM
S).
I
t
r
ec
o
g
n
izes
a
p
r
e
d
et
er
m
in
ed
n
u
m
b
er
o
f
o
b
jec
ts
in
an
im
ag
e
in
a
s
in
g
le
p
ass
u
s
in
g
en
c
o
d
er
-
d
ec
o
d
e
r
m
eth
o
d
b
ased
o
n
tr
an
s
f
o
r
m
er
s
.
W
ith
th
is
m
eth
o
d
,
th
e
m
o
d
el
m
ay
p
r
o
v
id
e
class
lab
els
an
d
b
o
u
n
d
in
g
b
o
x
es
with
o
u
t r
eq
u
i
r
in
g
f
u
r
th
er
p
o
s
t
-
p
r
o
ce
s
s
in
g
s
tep
s
.
T
h
e
s
tr
u
ctu
r
e
o
f
DE
T
R
in
Fig
u
r
e
5
.
is
b
u
ilt
ar
o
u
n
d
th
r
ee
p
r
i
m
ar
y
p
ar
ts
:
a
C
NN
b
ac
k
b
o
n
e
,
co
m
m
o
n
l
y
R
esNet
f
o
r
ex
tr
ac
tin
g
im
ag
e
f
ea
tu
r
es
a
tr
an
s
f
o
r
m
er
en
co
d
er
-
d
ec
o
d
e
r
f
o
r
m
o
d
elin
g
th
e
g
lo
b
al
co
n
te
x
t
with
in
th
e
im
ag
e,
an
d
a
f
ee
d
-
f
o
r
war
d
n
etwo
r
k
th
at
h
a
n
d
les
o
b
jec
t
class
if
icat
io
n
an
d
lo
ca
lizatio
n
.
Af
ter
th
e
C
NN
p
r
o
ce
s
s
es
th
e
im
ag
e,
th
e
r
esu
ltin
g
f
ea
tu
r
e
m
ap
s
ar
e
f
latten
ed
an
d
p
ass
ed
th
r
o
u
g
h
th
e
tr
an
s
f
o
r
m
er
en
c
o
d
e
r
alo
n
g
with
p
o
s
itio
n
al
in
f
o
r
m
atio
n
.
T
h
e
d
ec
o
d
er
t
h
en
o
p
er
ates
u
s
in
g
a
f
ix
e
d
n
u
m
b
e
r
o
f
o
b
ject
q
u
er
ies
th
at
f
o
cu
s
o
n
th
e
en
c
o
d
ed
f
ea
tu
r
e
s
to
p
r
ed
ict
o
b
je
ct
in
s
tan
ce
s
.
E
ac
h
q
u
er
y
is
r
esp
o
n
s
ib
le
f
o
r
d
etec
tin
g
a
s
in
g
le
o
b
ject,
m
ea
n
in
g
th
e
n
u
m
b
e
r
o
f
q
u
er
ies lim
its
th
e
m
ax
im
u
m
d
etec
tab
le
o
b
jects.
Du
r
in
g
tr
ai
n
in
g
,
DE
T
R
alig
n
s
th
e
p
r
ed
icted
o
u
tp
u
ts
with
g
r
o
u
n
d
-
tr
u
th
an
n
o
tatio
n
s
th
r
o
u
g
h
b
ip
a
r
tite
m
atch
in
g
u
s
in
g
t
h
e
Hu
n
g
a
r
i
an
alg
o
r
ith
m
.
T
h
is
p
r
o
ce
s
s
m
in
im
izes
a
co
m
b
in
ed
l
o
s
s
f
u
n
ctio
n
t
h
at
tak
es
b
o
u
n
d
in
g
b
o
x
p
r
ec
is
io
n
an
d
cl
ass
if
icatio
n
ac
cu
r
ac
y
in
to
ac
c
o
u
n
t.
T
h
e
u
s
e
o
f
s
et
-
b
ased
p
r
e
d
ictio
n
en
s
u
r
es
th
at
ea
ch
p
r
ed
icted
o
b
ject
is
u
n
iq
u
ely
m
atch
ed
to
a
g
r
o
u
n
d
-
tr
u
th
co
u
n
ter
p
a
r
t,
elim
in
atin
g
r
ed
u
n
d
an
t d
etec
tio
n
s
an
d
s
im
p
lify
in
g
th
e
p
r
ed
ictio
n
p
r
o
ce
s
s
.
Alth
o
u
g
h
DE
T
R
r
eq
u
ir
es
ex
ten
s
iv
e
tr
ain
in
g
tim
e
an
d
d
ata
d
u
e
to
its
g
lo
b
al
atten
tio
n
m
ec
h
an
is
m
,
it
d
eli
v
er
s
r
o
b
u
s
t
p
e
r
f
o
r
m
an
ce
,
p
a
r
ticu
lar
ly
in
co
m
p
lex
s
ce
n
es
wh
er
e
tr
ad
itio
n
a
l
ass
u
m
p
tio
n
s
ab
o
u
t o
b
ject
lay
o
u
t m
ay
n
o
t a
p
p
l
y
.
Fig
u
r
e
5
.
Ar
c
h
itectu
r
e
o
f
th
e
DE
T
R
3
.
4
.
P
r
o
po
s
ed
CB
AM
-
U
Net
wit
h DE
T
R
f
o
r
RB
C
det
ec
t
i
o
n
T
h
e
h
y
b
r
id
C
B
AM
-
UNet
in
teg
r
ated
with
DE
T
R
ar
ch
itectu
r
e
lev
er
ag
es
th
e
s
eg
m
en
tatio
n
ca
p
ab
ilit
ies
o
f
UNe
t,
th
e
atten
tio
n
en
h
an
ce
m
en
t
f
r
o
m
C
B
AM
,
an
d
th
e
o
b
ject
d
etec
tio
n
p
o
wer
o
f
D
E
T
R
to
ac
cu
r
ately
id
en
tify
a
n
d
lo
ca
te
R
B
C
s
in
m
alar
ia
s
m
ea
r
im
ag
es.
T
h
e
d
etec
tio
n
o
f
R
B
C
u
s
in
g
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
co
m
p
r
is
es
o
f
m
u
ltip
le
s
tag
es
as
d
ep
icted
i
n
Fig
u
r
e
6
.
T
h
e
p
h
ases
in
clu
d
e
th
e
in
p
u
t
m
alar
ia
s
m
ea
r
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
,
en
co
d
in
g
u
s
in
g
UNe
t
m
o
d
el,
wh
ich
s
er
v
e
as
a
f
ea
tu
r
e
ex
tr
ac
to
r
,
f
ea
tu
r
e
r
ef
in
em
e
n
t
u
s
in
g
C
B
AM
,
d
ec
o
d
in
g
an
d
d
etec
tio
n
o
f
th
e
R
B
C
s
f
r
o
m
th
e
r
ec
o
n
s
tr
u
cted
im
ag
es g
e
n
er
ated
b
y
th
e
d
ec
o
d
er
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
B
C
_
F
r
a
me_
N
et:
a
h
yb
r
id
d
e
ep
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
d
etec
tio
n
o
f red
b
lo
o
d
…
(
Mu
h
a
mma
d
S
h
a
mee
m
P
.
)
1491
Fig
u
r
e
6
.
Ar
c
h
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
C
B
AM
-
UNe
t w
ith
DE
T
R
f
o
r
R
B
C
d
etec
tio
n
3
.
4
.
1
.
Da
t
a
s
et
I
n
th
is
r
esear
ch
,
th
e
NI
H
m
al
ar
ia
d
ataset
is
u
s
ed
to
tr
ain
an
d
ev
alu
ate
th
e
p
r
o
p
o
s
ed
f
r
am
e
wo
r
k
.
T
h
e
d
ataset
is
an
o
p
en
ac
ce
s
s
r
eso
u
r
ce
p
u
b
licly
av
ailab
le
to
r
esear
ch
er
s
b
y
th
e
NI
H.
I
t
h
as
b
ee
n
a
cr
itical
r
eso
u
r
ce
em
p
lo
y
ed
b
y
r
esear
ch
e
r
s
f
o
r
ad
v
an
cin
g
AI
ap
p
licatio
n
s
in
m
alar
ia
tr
ea
tm
en
ts
.
T
h
e
NI
H
d
ataset
co
n
s
is
ts
o
f
ap
p
r
o
x
im
ately
2
7
,
0
0
0
an
n
o
ta
ted
im
ag
es
o
f
in
d
iv
id
u
al
R
B
C
’
s
ex
tr
ac
ted
f
r
o
m
th
in
b
lo
o
d
s
m
ea
r
s
lid
es
o
f
v
ar
io
u
s
p
atien
ts
.
T
h
e
im
ag
es
ar
e
ca
teg
o
r
ized
in
to
p
ar
asit
ized
an
d
u
n
in
f
ec
te
d
.
T
h
e
p
ar
asit
ized
im
ag
es
co
n
tai
n
P
la
s
mo
d
iu
m
fa
lcip
a
r
u
m
,
wh
ic
h
ca
u
s
es
m
alar
ia
in
h
u
m
an
s
,
wh
ile
th
e
u
n
in
f
ec
ted
im
a
g
es
in
d
icate
th
e
ab
s
en
ce
o
f
P
la
s
mo
d
iu
m
.
T
h
e
d
ataset
ca
p
tu
r
es
a
wid
e
r
an
g
e
o
f
n
at
u
r
al
v
ar
iab
ilit
y
in
ter
m
s
o
f
s
tain
in
g
tech
n
iq
u
es,
illu
m
in
atio
n
co
n
d
itio
n
s
,
an
d
ce
llu
lar
m
o
r
p
h
o
lo
g
y
,
o
f
f
e
r
in
g
a
r
ea
lis
tic
an
d
d
iv
er
s
e
s
et
o
f
s
am
p
les
f
o
r
d
ev
elo
p
in
g
an
d
ev
alu
atin
g
a
u
to
m
ated
d
iag
n
o
s
tic
m
o
d
els.
E
ac
h
im
ag
e
is
p
r
o
v
id
ed
in
R
GB
f
o
r
m
at
with
a
r
eso
lu
tio
n
o
f
1
2
8
×
1
2
8
p
ix
els
,
wh
ich
en
ab
les
s
tr
aig
h
t
f
o
r
w
ar
d
in
te
g
r
atio
n
in
to
d
ee
p
lea
r
n
in
g
f
r
am
ewo
r
k
s
with
o
u
t
th
e
n
ee
d
f
o
r
s
u
b
s
tan
ti
al
p
r
e
-
p
r
o
ce
s
s
in
g
.
T
h
e
d
ataset
was
cu
r
ated
to
s
u
p
p
o
r
t r
esear
ch
in
m
ed
ical
im
a
g
e
class
if
icatio
n
,
p
ar
ticu
lar
ly
f
o
c
u
s
in
g
o
n
t
h
e
ap
p
licatio
n
o
f
C
NNs
f
o
r
d
etec
tin
g
p
ar
asit
ic
in
f
ec
tio
n
s
.
I
ts
b
alan
ce
d
co
m
p
o
s
itio
n
o
f
in
f
ec
ted
an
d
u
n
in
f
ec
ted
s
am
p
les
m
a
k
es
it
well
-
s
u
ited
f
o
r
b
in
ar
y
class
if
icatio
n
task
s
,
wh
ile
th
e
h
ig
h
q
u
ality
o
f
its
an
n
o
tatio
n
s
en
s
u
r
es
r
eliab
ilit
y
in
b
o
th
tr
ain
in
g
an
d
ev
al
u
atio
n
p
r
o
ce
s
s
es.
As
a
r
esu
lt,
th
e
NI
H
m
alar
ia
d
ataset
h
as
b
ec
o
m
e
a
wid
ely
ad
o
p
ted
b
en
c
h
m
ar
k
in
co
m
p
u
tatio
n
al
p
ath
o
l
o
g
y
an
d
co
n
tin
u
es
to
co
n
tr
ib
u
te
s
ig
n
if
ica
n
tly
to
r
ese
ar
ch
ef
f
o
r
t
s
in
g
lo
b
al
h
ea
lth
in
f
o
r
m
atics.
3
.
4
.
2
.
P
re
pro
ce
s
s
ing
I
n
th
is
p
h
ase,
th
e
m
alar
ia
s
m
ea
r
im
ag
es
ar
e
p
r
e
p
ar
ed
f
o
r
an
aly
s
is
as
th
e
im
ag
es
wer
e
ty
p
ically
ca
p
tu
r
ed
u
s
in
g
a
lig
h
t
m
ic
r
o
s
co
p
e,
t
h
u
s
h
av
in
g
d
if
f
er
en
t
q
u
ality
,
r
eso
lu
tio
n
a
n
d
lev
el
o
f
n
o
is
e.
T
o
ad
d
r
es
s
th
ese
v
ar
iatio
n
s
,
th
e
im
ag
e
d
atasets
ar
e
n
o
r
m
alize
d
f
o
r
p
ix
el
in
te
n
s
ity
an
d
au
g
m
en
ted
[
2
3
]
u
s
in
g
au
g
m
en
tatio
n
tech
n
iq
u
es
lik
e
f
lip
p
in
g
[
1
8
]
,
r
o
tatio
n
[
2
4
]
,
an
d
co
n
tr
ast
ad
ju
s
tm
en
t
[
2
5
]
t
o
e
n
ab
le
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
b
ec
o
m
e
r
o
b
u
s
t
to
v
ar
iatio
n
s
in
ce
ll
o
r
ien
tatio
n
an
d
lig
h
t
en
i
n
g
,
th
u
s
en
s
u
r
in
g
a
co
n
s
is
ten
t
p
er
f
o
r
m
an
ce
o
n
r
ea
l lif
e
im
ag
es.
Af
ter
th
e
p
r
ep
r
o
ce
s
s
in
g
,
th
e
p
r
ep
r
o
ce
s
s
ed
im
ag
es a
r
e
tr
an
s
f
o
r
m
ed
in
to
te
n
s
o
r
v
ec
to
r
s
u
itab
le
f
o
r
th
e
p
r
o
p
o
s
ed
d
ee
p
lear
n
in
g
m
o
d
el.
T
o
a
v
o
id
l
o
s
s
o
f
v
ital
in
f
o
r
m
atio
n
d
u
e
to
s
m
all
s
ize
n
atu
r
e
o
f
th
e
R
B
C
’
s
an
d
p
ar
a
s
ites
,
we
m
ain
tain
a
h
ig
h
r
eso
lu
tio
n
o
f
th
e
im
ag
es.
T
h
e
ten
s
o
r
g
e
n
er
ated
in
th
is
p
h
ase
is
f
ed
in
to
t
h
e
en
c
o
d
er
p
o
r
tio
n
o
f
th
e
UNe
t
m
o
d
el
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
q
u
ality
an
d
co
n
s
is
ten
cy
o
f
p
r
ep
r
o
ce
s
s
in
g
d
ir
ec
tly
in
f
l
u
e
n
ce
th
e
ef
f
e
ctiv
e
n
ess
o
f
f
ea
tu
r
e
lear
n
in
g
,
esp
ec
ially
in
d
etec
tin
g
s
m
all
o
r
m
o
r
p
h
o
lo
g
ically
s
im
ilar
R
B
C
s
an
d
p
a
r
asit
es.
3
.
4
.
3
.
Da
t
a
enco
din
g
I
n
th
is
p
h
ase,
th
e
p
r
ep
r
o
ce
s
s
ed
im
ag
es
ar
e
en
co
d
ed
u
s
in
g
th
e
d
o
wn
s
am
p
lin
g
p
ath
o
f
th
e
UNe
t
m
o
d
el.
T
h
e
UNe
t
ar
ch
itectu
r
e
s
er
v
e
s
as
f
ea
tu
r
e
e
x
tr
ac
to
r
t
h
at
ex
tr
ac
ts
r
elev
an
t
in
f
o
r
m
ati
o
n
f
r
o
m
th
e
s
m
ea
r
im
ag
es.
T
h
e
en
co
d
er
co
m
p
r
is
es
o
f
a
s
er
ies
o
f
co
n
v
o
l
u
tio
n
al
b
lo
ck
s
in
teg
r
ated
with
f
ea
t
u
r
e
d
o
wn
s
am
p
lin
g
lay
er
s
.
I
n
th
is
r
esear
ch
,
ea
ch
c
o
n
v
o
lu
ti
o
n
b
lo
c
k
is
d
esig
n
ed
t
o
co
n
tain
two
3
b
y
3
c
o
n
v
o
lu
t
io
n
s
,
a
s
in
g
le
b
atch
n
o
r
m
aliza
tio
n
a
n
d
a
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
ac
tiv
atio
n
.
T
h
e
c
o
n
v
o
lu
tio
n
b
lo
ck
s
e
x
tr
ac
t
lo
ca
l
f
ea
tu
r
es
f
r
o
m
th
e
p
r
e
p
r
o
ce
s
s
ed
im
ag
e
s
in
th
e
p
r
ev
io
u
s
s
tag
e,
wh
ich
in
clu
d
e
th
e
im
ag
e
ce
ll
b
o
u
n
d
a
r
ies,
tex
tu
r
es,
p
atter
n
ch
ar
ac
ter
is
tics
o
f
th
e
R
B
C
’
s
an
d
p
o
s
s
ib
le
p
ar
asit
ic
in
f
ec
tio
n
s
.
T
h
e
d
o
wn
s
am
p
l
in
g
in
th
e
en
c
o
d
er
g
r
ad
u
ally
r
ed
u
ce
s
th
e
s
p
atial
r
eso
lu
tio
n
wh
ile
in
cr
ea
s
in
g
th
e
r
ec
ep
tiv
e
f
ield
an
d
f
ea
tu
r
e
d
ep
t
h
.
As
th
e
p
r
ep
r
o
ce
s
s
ed
im
ag
es
m
o
v
es
d
ee
p
er
in
to
th
e
UNe
t
m
o
d
el
en
co
d
er
,
th
e
n
etwo
r
k
ca
p
tu
r
es
i
n
cr
ea
s
in
g
ly
ab
s
tr
ac
t
f
ea
tu
r
es
b
eg
in
n
i
n
g
with
th
e
i
m
ag
e
ed
g
es
an
d
te
x
tu
r
es,
p
r
o
g
r
ess
in
g
to
ca
p
tu
r
e
m
o
r
e
s
o
p
h
is
ticated
s
h
ap
es
lik
e
cir
cu
lar
R
B
C
’
s
an
d
th
e
co
r
r
esp
o
n
d
in
g
p
ar
asit
e
s
ig
n
atu
r
es
w
ith
in
th
e
ce
lls
.
T
h
e
f
ea
tu
r
e
m
a
p
s
at
th
e
in
d
iv
i
d
u
al
lev
el
ar
e
s
to
r
ed
an
d
u
tili
ze
d
b
y
th
e
d
ec
o
d
e
r
v
ia
s
k
ip
co
n
n
ec
tio
n
s
,
th
u
s
p
r
eser
v
in
g
f
in
e
-
g
r
a
in
ed
s
p
atial
d
etails.
T
h
e
en
c
o
d
er
lay
s
th
e
g
r
o
u
n
d
wo
r
k
f
o
r
s
em
an
tic
s
eg
m
en
tati
o
n
an
d
f
ea
tu
r
e
d
etec
tio
n
b
y
c
o
n
v
er
tin
g
th
e
im
ag
e
in
to
a
h
ig
h
-
d
im
en
s
io
n
al
f
ea
tu
r
e
s
p
ac
e.
3
.
4
.
4
.
F
ea
t
ure
re
f
ine
m
ent
I
n
th
is
p
h
ase,
th
e
en
co
d
ed
f
ea
t
u
r
es
f
r
o
m
th
e
en
co
d
er
ar
e
r
ef
i
n
ed
b
ef
o
r
e
r
ec
o
n
s
tr
u
ctio
n
in
th
e
d
ec
o
d
er
p
ar
t
o
f
th
e
UNe
t
m
o
d
el.
I
n
o
r
d
er
to
d
o
t
h
is
,
a
C
B
AM
is
in
tr
o
d
u
ce
d
b
etwe
en
th
e
en
c
o
d
er
an
d
th
e
d
ec
o
d
er
at
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
8
6
-
1
4
9
6
1492
th
e
UNe
t's
b
o
ttl
en
ec
k
.
T
h
is
co
m
p
o
n
e
n
t a
p
p
lies
atten
tio
n
m
ec
h
an
is
m
to
th
e
f
ea
tu
r
es o
b
tain
e
d
f
r
o
m
t
h
e
en
co
d
er
m
o
d
u
le
an
d
r
ef
i
n
es
th
em
b
ef
o
r
e
p
ass
in
g
th
em
to
t
h
e
d
e
co
d
e
r
f
o
r
r
ec
o
n
s
tr
u
ctio
n
.
C
B
AM
atten
tio
n
m
ec
h
an
is
m
is
d
esig
n
ed
to
co
m
p
r
is
e
two
s
eq
u
en
tial
s
u
b
m
o
d
u
les,
n
am
e
c
h
an
n
el
atten
tio
n
a
n
d
SAM
s
.
T
h
e
C
AM
s
lear
n
th
e
f
ea
tu
r
e
ch
an
n
els
th
at
ar
e
m
o
r
e
r
elev
an
t
an
d
d
ef
in
e
th
e
f
ea
tu
r
es
to
f
o
cu
s
o
n
.
T
h
e
f
ea
tu
r
es
to
co
n
ce
n
tr
ate
o
n
ar
e
d
eter
m
in
ed
b
y
th
e
C
AM
s
,
wh
ich
also
lear
n
w
h
ich
f
ea
tu
r
e
ch
an
n
els
ar
e
m
o
r
e
p
er
tin
en
t.
Ap
p
ly
in
g
g
lo
b
al
a
v
er
ag
e
p
o
o
lin
g
an
d
m
ax
p
o
o
lin
g
ac
r
o
s
s
s
p
atial
d
im
en
s
io
n
s
,
f
ee
d
in
g
th
em
v
ia
m
u
lti
-
lay
er
p
er
ce
p
t
r
o
n
,
a
n
d
p
r
o
d
u
cin
g
ch
a
n
n
el
-
wis
e
atten
tio
n
m
ap
s
ar
e
h
o
w
th
e
C
AM
s
d
o
th
is
.
T
h
e
m
ap
o
b
tain
ed
h
elp
s
th
e
m
o
d
el
t
o
s
u
p
p
r
ess
th
e
ir
r
elev
an
t r
eg
io
n
s
o
f
th
e
im
ag
e
f
ea
tu
r
es
an
d
em
p
h
asize
o
n
th
e
cr
itical
ar
ea
s
.
C
B
AM
en
h
an
ce
s
th
e
m
o
d
el’
s
f
o
cu
s
,
in
c
r
ea
s
in
g
it
s
ca
p
ab
ilit
y
to
d
etec
t
s
u
b
tle
ab
n
o
r
m
alities
an
d
im
p
r
o
v
in
g
th
e
ac
cu
r
ac
y
o
f
d
o
wn
s
tr
ea
m
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
.
3
.
4
.
5
.
F
ea
t
ure
re
c
o
ns
t
ruct
io
n
I
n
th
is
p
h
ase,
th
e
s
p
atial
r
es
o
lu
tio
n
o
f
th
e
im
ag
e
is
r
ec
o
n
s
tr
u
cted
f
r
o
m
th
e
co
m
p
r
ess
ed
f
ea
tu
r
e
r
ep
r
esen
tatio
n
.
T
h
is
em
p
lo
y
s
a
d
ec
o
d
er
th
at
p
er
f
o
r
m
s
u
p
s
am
p
lin
g
f
o
llo
wed
b
y
co
n
v
o
lu
tio
n
lay
er
s
.
E
ac
h
u
p
s
am
p
lin
g
p
h
ase
is
p
air
ed
with
a
s
k
ip
co
n
n
ec
tio
n
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
en
c
o
d
er
lev
el
,
th
u
s
en
s
u
r
in
g
th
at
th
e
s
p
atial
in
f
o
r
m
atio
n
lo
s
t
at
th
e
co
r
r
esp
o
n
d
in
g
d
o
wn
s
am
p
le
p
h
ase
ar
e
r
ec
o
v
er
e
d
.
T
h
e
s
e
s
k
ip
co
n
n
ec
tio
n
s
in
teg
r
ate
h
ig
h
-
lev
el
s
em
an
ti
c
d
etails
with
lo
w
-
lev
el
s
p
atial
d
etails,
wh
ich
is
p
ar
am
o
u
n
t
f
o
r
ac
c
u
r
ate
r
ec
o
g
n
itio
n
o
f
t
h
e
b
o
u
n
d
a
r
ies o
f
RBC
s
an
d
af
f
ec
ted
r
eg
io
n
s
.
As
th
e
d
ec
o
d
er
ad
v
a
n
ce
s
,
it
r
ec
o
n
s
tr
u
cts
f
ea
tu
r
e
m
a
p
s
with
h
ig
h
r
eso
lu
tio
n
t
h
at
m
atch
es
th
e
in
itial
im
ag
e
d
im
en
s
io
n
s
.
T
h
e
last
c
o
n
v
o
lu
ti
o
n
lay
er
o
f
th
e
d
ec
o
d
er
g
en
e
r
ates
th
e
s
eg
m
en
tatio
n
m
ask
with
ea
ch
p
ix
el
ca
teg
o
r
ized
to
a
s
p
ec
if
ic
class
.
T
h
i
s
s
eg
m
en
tatio
n
m
ap
p
r
o
v
id
es
p
ix
el
-
lev
el
lo
ca
li
za
tio
n
o
f
RBC
s
an
d
allo
ws
f
o
r
p
r
ec
is
e
m
o
r
p
h
o
lo
g
ical
an
aly
s
is
.
T
h
e
d
ec
o
d
er
,
en
h
an
ce
d
b
y
C
B
AM
-
r
ef
in
ed
f
ea
tu
r
es,
h
elp
s
in
d
etec
tin
g
an
d
class
if
y
in
g
RBC
s
with
b
o
th
g
l
o
b
al
co
n
tex
t a
n
d
f
in
e
d
etail.
3
.
4
.
6
.
Red
blo
o
d c
ells
det
ec
t
io
n
I
n
th
is
p
h
a
s
e
,
th
e
RBC
s
ar
e
d
ete
ct
ed
.
T
h
i
s
i
s
ac
h
i
ev
ed
u
s
in
g
th
e
DE
T
R
m
o
d
e
l,
an
ad
v
an
ce
d
o
b
je
ct
d
ete
ct
io
n
m
o
d
e
l
th
a
t
r
e
im
ag
i
n
es
th
e
ty
p
i
ca
l
d
et
ec
tio
n
w
o
r
k
f
lo
w.
T
o
a
ch
i
ev
e
a
cc
u
r
at
e
d
ete
ct
io
n
,
th
e
f
ea
tu
r
e
m
ap
s
f
r
o
m
th
e
C
B
AM
-
U
Ne
t
ar
e
f
ed
in
to
th
e
DE
T
R
m
o
d
e
l
f
o
r
o
b
j
ec
t
d
et
ec
tio
n
.
H
eu
r
is
t
ic
s
lik
e
an
ch
o
r
b
o
x
e
s
an
d
NM
S
ar
e
n
o
t
n
ec
es
s
ar
y
s
i
n
ce
t
h
e
m
o
d
e
l
co
n
s
id
er
s
o
b
jec
t
id
en
t
if
ica
t
io
n
a
s
a
d
ir
e
ct
s
et
o
f
p
r
ed
ic
t
io
n
p
r
o
b
le
m
.
T
h
e
m
o
d
el
e
m
p
lo
y
s
a
tr
an
s
f
o
r
m
er
en
co
d
er
-
d
ec
o
d
er
s
tr
u
ctu
r
e
c
ap
ab
le
o
f
at
ten
d
i
n
g
to
a
ll
r
eg
io
n
s
o
f
th
e
im
ag
e
s
a
t
t
h
e
s
a
m
e
tim
e,
ca
p
tu
r
in
g
lo
n
g
r
an
g
e
d
ep
en
d
en
c
ie
s
.
I
n
o
r
d
er
to
d
et
er
m
in
e
th
e
r
e
la
tio
n
s
h
ip
s
b
et
wee
n
v
ar
io
u
s
p
ar
t
s
o
f
th
e
i
n
p
u
t
im
ag
e
,
th
e
in
p
u
t
f
e
atu
r
e
m
ap
i
s
f
la
tt
en
ed
in
to
a
s
er
i
e
s
o
f
to
k
en
s
,
ea
ch
o
f
wh
i
ch
co
n
ta
in
s
p
o
s
it
io
n
al
en
co
d
in
g
s
.
T
h
e
to
k
en
s
a
r
e
th
en
p
a
s
s
ed
th
r
o
u
g
h
t
r
an
s
f
o
r
m
er
l
ay
er
s
.
A
p
r
ed
e
ter
m
in
ed
s
et
o
f
le
ar
n
ed
o
b
je
ct
q
u
er
ie
s
i
s
s
en
t
in
to
th
e
tr
an
s
f
o
r
m
er
d
e
co
d
e
r
in
DE
T
R
,
wh
i
ch
d
ec
o
d
e
s
th
em
i
n
to
p
r
o
b
ab
l
e
o
b
je
ct
r
e
p
r
e
s
en
ta
tio
n
s
w
it
h
in
th
e
im
a
g
e.
I
t
g
en
er
a
te
s
a
co
ll
ec
tio
n
o
f
b
o
u
n
d
in
g
b
o
x
e
s
;
ea
ch
l
in
k
ed
to
a
s
p
ec
if
i
c
cl
as
s
la
b
e
l.
T
h
e
m
o
d
e
l
i
s
tr
ain
ed
u
s
i
n
g
b
ip
ar
t
it
e
m
at
ch
in
g
u
s
in
g
th
e
Hu
n
g
ar
ian
lo
s
s
m
eth
o
d
t
o
en
s
u
r
e
a
o
n
e
-
to
-
o
n
e
co
r
r
e
s
p
o
n
d
en
c
e
b
et
we
en
th
e
p
r
ed
ic
ted
an
d
ac
tu
al
o
b
j
ec
t
s
.
T
h
e
u
s
e
o
f
a
tr
an
s
f
o
r
m
er
a
ll
o
w
s
DE
T
R
to
ef
f
ec
tiv
ely
c
ap
t
u
r
e
co
m
p
le
x
s
p
at
ia
l
d
ep
en
d
en
ci
es
a
n
d
i
n
ter
ac
tio
n
s
am
o
n
g
RBC
s
an
d
p
ar
a
s
it
e
s
,
wh
ich
i
s
e
s
p
ec
ia
l
ly
b
en
ef
ic
ia
l
in
h
a
n
d
lin
g
o
v
er
l
ap
p
i
n
g
c
el
l
s
an
d
d
en
s
e
ly
p
o
p
u
l
at
ed
s
m
e
ar
r
eg
io
n
s
.
4.
E
XP
E
R
I
M
E
N
T
AND
R
E
SU
L
T
S
4
.
1
.
E
x
perim
ent
a
l set
up
T
h
e
p
r
o
p
o
s
ed
ex
p
e
r
im
en
tal
f
r
am
ewo
r
k
f
o
r
im
p
lem
en
tin
g
t
h
e
h
y
b
r
i
d
C
B
AM
-
UNe
t
in
teg
r
ated
with
th
e
DE
T
R
ar
ch
itectu
r
e
u
tili
z
es
a
d
u
al
-
s
tr
ea
m
ap
p
r
o
ac
h
t
ailo
r
ed
f
o
r
co
n
c
u
r
r
en
t
R
B
C
s
eg
m
en
tatio
n
an
d
d
etec
tio
n
in
m
alar
ia
s
m
ea
r
im
ag
es.
I
n
itially
,
all
im
ag
es f
r
o
m
th
e
m
alar
ia
d
atase
t a
r
e
r
esize
d
to
1
2
8
×
1
2
8
p
i
x
els
an
d
n
o
r
m
alize
d
to
s
tan
d
ar
d
ize
p
ix
el
in
ten
s
ity
v
alu
es.
Data
a
u
g
m
en
tatio
n
m
eth
o
d
s
in
cl
u
d
in
g
r
o
tatio
n
,
r
an
d
o
m
f
lip
p
in
g
,
an
d
co
lo
r
p
e
r
tu
r
b
ati
o
n
s
ar
e
u
s
ed
to
im
p
r
o
v
e
th
e
m
o
d
el'
s
ca
p
ab
ilit
ies
f
o
r
g
en
e
r
aliza
tio
n
.
Af
ter
th
at,
th
e
d
ataset
is
d
iv
id
ed
f
o
r
test
in
g
,
v
alid
atio
n
,
an
d
tr
ain
in
g
.
I
n
th
is
s
etu
p
,
th
e
C
B
AM
-
UN
et
m
o
d
el
h
an
d
les
p
ix
el
-
lev
el
s
eg
m
en
tatio
n
b
y
in
co
r
p
o
r
atin
g
atten
tio
n
m
o
d
u
les
th
at
r
ef
in
e
b
o
th
s
p
ati
al
an
d
ch
an
n
el
-
wis
e
f
ea
tu
r
es.
Simu
ltan
eo
u
s
ly
,
th
e
DE
T
R
m
o
d
el
p
e
r
f
o
r
m
s
o
b
ject
d
etec
tio
n
,
lev
er
a
g
i
n
g
tr
a
n
s
f
o
r
m
er
-
b
ased
atten
tio
n
m
ec
h
an
is
m
s
an
d
b
ip
ar
tite
m
atch
i
n
g
to
i
d
en
tify
an
d
lo
ca
lize
in
f
ec
ted
ce
lls
.
T
r
ain
in
g
is
ca
r
r
ied
o
u
t
in
a
GPU
-
s
u
p
p
o
r
ted
en
v
ir
o
n
m
e
n
t
u
s
in
g
Py
T
o
r
ch
o
n
Go
o
g
le
C
o
lab
,
with
C
UDA
p
r
o
v
id
in
g
co
m
p
u
tatio
n
al
ac
ce
ler
atio
n
.
T
h
e
s
y
s
tem
c
o
n
f
i
g
u
r
atio
n
in
clu
d
es
an
NVI
DI
A
T
esla
T
4
GPU
with
1
6
GB
VR
AM
,
1
2
th
-
g
e
n
er
atio
n
I
n
tel
Xeo
n
C
PU
b
ac
k
en
d
,
an
d
2
5
GB
o
f
av
aila
b
le
R
AM
,
en
s
u
r
in
g
s
u
f
f
icien
t
r
eso
u
r
ce
s
f
o
r
ef
f
ic
ien
t
tr
ain
in
g
an
d
ev
alu
atio
n
.
T
h
e
o
p
tim
izatio
n
p
r
o
ce
s
s
u
tili
ze
s
th
e
Ad
am
o
p
tim
izer
with
a
l
ea
r
n
in
g
r
ate
o
f
1
e
-
4
,
r
u
n
n
in
g
o
v
er
1
0
0
ep
o
ch
s
.
Fo
r
th
e
s
eg
m
en
tatio
n
task
,
a
co
m
b
in
atio
n
o
f
b
in
ar
y
c
r
o
s
s
-
en
tr
o
p
y
an
d
Dic
e
lo
s
s
f
u
n
ctio
n
s
is
u
s
ed
,
w
h
ile
DE
T
R
em
p
lo
y
s
its
s
tan
d
ar
d
lo
s
s
f
u
n
ctio
n
s
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
B
C
_
F
r
a
me_
N
et:
a
h
yb
r
id
d
e
ep
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
d
etec
tio
n
o
f red
b
lo
o
d
…
(
Mu
h
a
mma
d
S
h
a
mee
m
P
.
)
1493
in
clu
d
in
g
t
h
e
Hu
n
g
ar
ian
lo
s
s
f
o
r
ef
f
ec
tiv
e
o
b
ject
alig
n
m
en
t
.
T
h
r
o
u
g
h
o
u
t
tr
ain
i
n
g
,
k
ey
p
e
r
f
o
r
m
a
n
ce
m
etr
ics
s
u
ch
as
th
e
Dice
co
ef
f
icien
t,
I
o
U,
m
AP,
an
d
r
ec
all
—
ar
e
tr
ac
k
ed
an
d
lo
g
g
e
d
.
Fin
al
m
o
d
el
ev
alu
atio
n
is
co
n
d
u
cte
d
o
n
th
e
test
s
et,
an
d
p
e
r
f
o
r
m
an
ce
o
v
er
tim
e
is
v
is
u
alize
d
u
s
in
g
m
etr
ic
p
lo
ts
,
en
s
u
r
i
n
g
a
co
m
p
r
eh
e
n
s
iv
e
an
d
r
e
p
r
o
d
u
ci
b
le
ass
ess
m
en
t
o
f
th
e
s
y
s
te
m
'
s
ca
p
ab
ilit
y
to
ac
cu
r
ately
d
etec
t
an
d
s
eg
m
en
t
m
alar
ia
-
in
f
ec
ted
R
B
C
s
.
4
.
2
.
Resul
t
s
T
h
e
p
er
f
o
r
m
an
ce
e
v
alu
atio
n
i
s
ca
r
r
ied
o
u
t
b
y
test
in
g
th
e
t
r
ain
ed
C
B
AM
-
UNe
t
an
d
DE
T
R
h
y
b
r
id
m
o
d
el
o
n
a
d
e
d
icate
d
test
s
et
co
n
s
is
tin
g
o
f
m
alar
ia
b
lo
o
d
s
m
ea
r
im
ag
es.
Fo
r
t
h
e
s
eg
m
en
tatio
n
task
,
th
e
g
en
er
ated
m
ask
s
ar
e
c
o
m
p
ar
ed
with
g
r
o
u
n
d
tr
u
th
an
n
o
tat
io
n
s
u
s
in
g
th
e
Dice
co
ef
f
icie
n
t
an
d
I
o
U,
wh
ich
r
ef
lect
h
o
w
ac
cu
r
ately
th
e
m
o
d
el
o
u
tlin
es
RBC
s
.
I
n
ter
m
s
o
f
o
b
ject
d
etec
tio
n
,
th
e
DE
T
R
m
o
d
u
le
p
r
o
v
i
d
es
b
o
u
n
d
in
g
b
o
x
es
(
as
s
h
o
wn
i
n
Fig
u
r
e
7
)
a
n
d
ass
o
ciate
d
class
p
r
ed
ictio
n
s
,
wh
ich
a
r
e
ass
es
s
ed
b
ased
o
n
p
r
ec
is
io
n
,
r
ec
all,
an
d
m
AP,
ca
lcu
lated
at
a
n
I
o
U
th
r
esh
o
ld
o
f
0
.
5
.
T
o
e
v
alu
ate
e
f
f
ici
en
cy
,
t
h
e
a
v
er
ag
e
in
f
er
en
ce
tim
e
p
er
im
ag
e
is
also
m
ea
s
u
r
ed
.
All
ev
alu
atio
n
m
etr
ics
ar
e
co
m
p
u
ted
ac
r
o
s
s
th
e
en
tire
test
s
et
to
p
r
o
v
id
e
a
th
o
r
o
u
g
h
an
al
y
s
is
o
f
th
e
m
o
d
el’
s
ef
f
ec
tiv
en
ess
in
b
o
th
s
eg
m
en
tatio
n
a
n
d
d
etec
tio
n
task
s
.
Fig
u
r
e
7
.
A
u
t
o
m
atica
lly
d
etec
t
ed
R
B
C
’
s
T
ab
le
1
p
r
esen
ts
th
e
p
er
f
o
r
m
an
ce
co
m
p
ar
is
o
n
o
f
th
e
C
B
A
M
-
UNe
t+DE
T
R
w
ith
UNe
t+
DE
T
R
,
an
d
DE
T
R
alo
n
e
u
s
ed
f
o
r
an
aly
z
in
g
m
alar
ia
b
lo
o
d
s
m
ea
r
im
ag
es.
T
h
e
co
m
b
in
e
d
C
B
AM
-
UNe
t+DE
T
R
m
o
d
el
ac
h
iev
es
th
e
b
est
r
esu
lts
,
with
s
eg
m
en
tatio
n
s
co
r
es o
f
0
.
9
7
I
o
U
an
d
0
.
9
8
Dice
c
o
ef
f
icien
t,
as
well
as
d
etec
tio
n
s
co
r
es
o
f
0
.
9
9
9
p
r
ec
is
io
n
,
0
.
9
9
8
r
ec
all,
an
d
0
.
9
9
5
m
AP.
I
ts
h
ig
h
ac
c
u
r
ac
y
ca
n
b
e
cr
ed
ite
d
to
th
e
in
teg
r
atio
n
o
f
C
B
AM
,
wh
ich
r
ef
in
es
b
o
th
s
p
atial
an
d
ch
an
n
el
-
wis
e
f
ea
tu
r
es,
an
d
th
e
s
y
n
er
g
y
o
f
s
im
u
ltan
eo
u
s
s
eg
m
en
tatio
n
an
d
d
etec
tio
n
tr
ain
i
n
g
th
at
en
h
an
ce
s
o
v
er
all
p
er
f
o
r
m
a
n
ce
.
I
n
co
n
tr
ast,
th
e
UNe
t+DE
T
R
co
n
f
ig
u
r
atio
n
y
ield
s
s
lig
h
tly
lo
wer
r
esu
lts
,
p
ar
tic
u
lar
ly
in
s
eg
m
en
tatio
n
ac
cu
r
ac
y
(
0
.
9
0
I
o
U
an
d
0
.
9
3
D
ice
)
an
d
d
etec
tio
n
(
0
.
9
6
p
r
ec
is
io
n
a
n
d
0
.
9
4
r
e
ca
ll),
lik
ely
d
u
e
to
th
e
a
b
s
e
n
ce
o
f
atten
tio
n
m
o
d
u
les
th
at
im
p
r
o
v
e
f
ea
tu
r
e
d
is
cr
im
in
atio
n
.
T
h
e
DE
T
R
-
o
n
ly
m
o
d
el,
wh
ich
is
n
o
t
d
esig
n
ed
f
o
r
s
eg
m
e
n
tatio
n
,
s
co
r
es
ze
r
o
in
I
o
U
a
n
d
Dice
b
u
t
s
till
p
er
f
o
r
m
s
ad
e
q
u
ately
in
d
etec
tio
n
,
ac
h
iev
in
g
0
.
9
3
p
r
ec
is
io
n
,
0
.
9
1
r
ec
all,
an
d
0
.
9
1
m
AP.
T
h
ese
f
in
d
in
g
s
h
ig
h
lig
h
t
th
e
b
en
e
f
it
o
f
co
m
b
in
in
g
s
eg
m
en
tat
io
n
n
etwo
r
k
s
with
atten
tio
n
m
ec
h
an
is
m
s
an
d
tr
an
s
f
o
r
m
er
-
b
ased
d
etec
to
r
s
f
o
r
m
o
r
e
ac
c
u
r
ate
an
d
r
o
b
u
s
t a
n
aly
s
is
in
m
ed
ical
im
ag
e
p
r
o
ce
s
s
in
g
.
T
ab
le
1
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
C
B
AM
-
UNe
t+DE
T
R
with
UNe
t+DE
T
R
an
d
DE
T
R
M
o
d
e
l
S
e
g
m
e
n
t
a
t
i
o
n
I
o
U
D
i
c
e
c
o
e
f
f
i
c
i
e
n
t
D
e
t
e
c
t
i
o
n
p
r
e
c
i
si
o
n
D
e
t
e
c
t
i
o
n
r
e
c
a
l
l
mA
P
(
d
e
t
e
c
t
i
o
n
)
C
B
A
M
-
U
N
e
t
+
D
ETR
0
.
9
7
0
.
9
8
0
.
9
9
9
0
.
9
9
8
0
.
9
9
5
U
N
e
t
+
D
ET
R
0
.
9
0
0
.
9
3
0
.
9
6
0
.
9
4
0
.
9
3
D
ETR
0
.
0
0
(
N
o
s
e
g
me
n
t
e
r
)
0
.
0
0
0
.
9
3
0
.
9
1
0
.
9
1
T
ab
le
2
o
u
tlin
es
a
co
m
p
ar
ati
v
e
o
v
e
r
v
iew
o
f
th
e
p
r
o
p
o
s
ed
C
B
AM
-
UNe
t+
DE
T
R
o
b
ject
d
etec
tio
n
m
o
d
el
with
s
tate
-
of
-
th
e
-
ar
t
o
b
ject
d
etec
tio
n
m
o
d
els,
ev
alu
ated
s
p
ec
if
ically
f
o
r
th
eir
ef
f
ec
tiv
en
ess
in
id
en
tify
in
g
R
B
C
s
in
m
alar
ia
b
lo
o
d
s
m
ea
r
im
ag
es.
E
ac
h
m
o
d
el
is
ass
ess
ed
u
s
in
g
s
tan
d
ar
d
p
er
f
o
r
m
a
n
ce
m
etr
ics,
in
clu
d
in
g
d
etec
tio
n
p
r
ec
is
io
n
,
r
ec
all,
an
d
m
AP
at
an
I
o
U
th
r
esh
o
ld
o
f
0
.
5
,
an
d
in
f
er
en
ce
tim
e
p
er
im
ag
e.
Stan
d
ar
d
p
er
f
o
r
m
an
ce
p
ar
am
eter
s
,
s
u
ch
as
d
etec
tio
n
p
r
ec
is
io
n
,
r
ec
all,
m
AP
at
an
I
o
U
th
r
esh
o
ld
o
f
0
.
5
,
an
d
in
f
er
en
ce
tim
e
p
er
im
a
g
e,
ar
e
u
s
ed
to
ev
alu
ate
ea
ch
m
o
d
el.
I
n
ad
d
itio
n
t
o
th
ese
m
etr
i
cs,
b
r
ief
n
o
tes
ar
e
p
r
o
v
id
e
d
f
o
r
ea
c
h
m
o
d
el
to
h
ig
h
lig
h
t
th
eir
m
ain
a
d
v
an
ta
g
e
s
an
d
lim
itatio
n
s
,
h
el
p
in
g
t
o
co
n
tex
tu
alize
th
eir
p
r
ac
tical
ap
p
licatio
n
s
in
d
iag
n
o
s
tic
wo
r
k
f
lo
ws.
Am
o
n
g
th
e
lis
ted
m
o
d
els,
th
e
C
B
AM
-
U
Net
co
m
b
in
ed
with
DE
T
R
ac
h
iev
es
th
e
h
ig
h
est
o
v
er
all
p
er
f
o
r
m
a
n
ce
,
r
ec
o
r
d
i
n
g
n
ea
r
ly
p
er
f
ec
t
d
etec
tio
n
m
etr
i
cs:
0
.
9
9
9
p
r
ec
is
io
n
,
0
.
9
9
8
r
ec
all,
a
n
d
a
0
.
9
9
5
m
AP.
W
h
ile
its
in
f
er
en
ce
tim
e
s
tan
d
s
at
8
5
m
illi
s
ec
o
n
d
s
p
er
i
m
ag
e,
s
lo
wer
th
a
n
s
o
m
e
alter
n
ativ
es,
it
co
m
p
en
s
ates
with
ex
ce
p
tio
n
al
ac
cu
r
ac
y
an
d
in
teg
r
ate
d
s
eg
m
en
t
atio
n
.
T
h
e
C
B
A
M
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
8
6
-
1
4
9
6
1494
m
o
d
u
le
en
h
an
ce
s
f
ea
tu
r
e
atte
n
tio
n
,
wh
ile
DE
T
R
p
r
o
v
id
es
r
o
b
u
s
t
o
b
ject
d
etec
tio
n
,
m
ak
in
g
th
is
h
y
b
r
id
m
o
d
el
id
ea
l
f
o
r
m
e
d
ical
im
ag
e
an
aly
s
is
wh
er
e
p
r
ec
is
io
n
is
a
to
p
p
r
i
o
r
ity
.
On
th
e
o
th
e
r
h
an
d
,
YOL
Ov
8
s
tan
d
s
o
u
t
f
o
r
its
s
p
ee
d
,
p
r
o
ce
s
s
in
g
im
a
g
e
s
in
ju
s
t
1
2
m
illi
s
ec
o
n
d
s
wh
ile
s
till
d
eliv
er
in
g
s
tr
o
n
g
d
etec
tio
n
ac
c
u
r
ac
y
(
0
.
9
8
3
p
r
ec
is
io
n
an
d
0
.
9
7
9
m
AP)
.
I
ts
ef
f
icien
cy
m
a
k
es
it
well
-
s
u
ited
f
o
r
r
ea
l
-
tim
e
u
s
e
ca
s
es,
th
o
u
g
h
it
lack
s
th
e
s
eg
m
en
tatio
n
ca
p
ab
ilit
ies
p
r
esen
t
in
th
e
C
B
AM
-
UNe
t+DE
T
R
ap
p
r
o
ac
h
.
Fas
ter
r
eg
io
n
-
b
ased
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
(
Fas
ter
R
-
C
NN)
an
d
R
etin
aNe
t
also
p
er
f
o
r
m
well
in
ter
m
s
o
f
ac
cu
r
ac
y
,
b
u
t
th
eir
s
lo
wer
p
r
o
ce
s
s
in
g
tim
es (
1
1
0
m
s
an
d
9
5
m
s
,
r
esp
ec
tiv
e
ly
)
m
ay
r
e
d
u
ce
t
h
eir
p
r
ac
ticality
in
tim
e
-
s
en
s
itiv
e
d
iag
n
o
s
tic
s
y
s
tem
s
.
E
f
f
icien
t
Det
-
D3
o
f
f
e
r
s
a
b
alan
ce
d
p
er
f
o
r
m
an
ce
,
co
m
b
in
in
g
a
r
ea
s
o
n
ab
le
p
r
o
ce
s
s
in
g
tim
e
o
f
6
5
m
illi
s
ec
o
n
d
s
with
s
o
lid
d
etec
tio
n
m
etr
ics,
in
cl
u
d
in
g
a
0
.
9
6
8
m
AP.
I
t
s
er
v
es
as
a
m
i
d
d
le
-
g
r
o
u
n
d
s
o
lu
tio
n
f
o
r
s
ce
n
a
r
io
s
r
eq
u
ir
in
g
b
o
th
ac
cu
r
ac
y
a
n
d
s
p
ee
d
.
Me
an
wh
ile,
th
e
o
r
ig
in
al
DE
T
R
m
o
d
e
l,
wh
ile
d
eliv
e
r
in
g
ac
ce
p
tab
le
d
etec
tio
n
r
esu
lts
(
0
.
9
3
0
p
r
ec
is
io
n
,
0
.
9
0
5
m
AP)
,
f
alls
s
h
o
r
t
d
u
e
to
its
s
lo
wer
in
f
er
e
n
ce
s
p
ee
d
(
1
2
5
m
s
)
a
n
d
lack
o
f
s
eg
m
en
tatio
n
ca
p
ab
ilit
ies.
C
o
llectiv
el
y
,
th
ese
co
m
p
ar
is
o
n
s
p
r
o
v
id
e
clea
r
g
u
id
an
ce
o
n
ch
o
o
s
in
g
t
h
e
m
o
s
t
ap
p
r
o
p
r
iate
m
o
d
el
d
ep
en
d
in
g
o
n
wh
eth
er
ac
cu
r
ac
y
,
s
p
ee
d
,
o
r
co
m
p
r
eh
e
n
s
iv
e
f
u
n
ctio
n
ality
is
th
e
p
r
im
ar
y
r
eq
u
ir
em
e
n
t.
T
ab
le
2
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
C
B
AM
-
UNe
t+DE
T
R
with
an
o
th
er
o
b
ject
d
etec
tio
n
alg
o
r
ith
m
M
o
d
e
l
D
e
t
e
c
t
i
o
n
p
r
e
c
i
si
o
n
D
e
t
e
c
t
i
o
n
r
e
c
a
l
l
mA
P
(
I
o
U
@
0
.
5
)
I
n
f
e
r
e
n
c
e
t
i
me
(
ms/i
mg
)
C
B
A
M
-
U
N
e
t
+
D
ETR
0
.
9
9
9
0
.
9
9
8
0
.
9
9
5
85
Y
O
LO
v
8
0
.
9
8
3
0
.
9
8
1
0
.
9
7
9
12
F
a
st
e
r
R
-
C
N
N
0
.
9
7
0
0
.
9
6
5
0
.
9
6
1
1
1
0
R
e
t
i
n
a
N
e
t
0
.
9
5
8
0
.
9
5
1
0
.
9
4
5
95
Ef
f
i
c
i
e
n
t
D
e
t
-
D3
0
.
9
7
5
0
.
9
6
3
0
.
9
6
8
65
D
ETR
(
V
a
n
i
l
l
a
)
0
.
9
3
0
0
.
9
1
0
0
.
9
0
5
1
2
5
T
ab
le
3
co
m
p
ar
es
d
if
f
e
r
en
t
v
ar
ian
ts
o
f
th
e
UNe
t+DE
T
R
m
o
d
el
th
at
in
co
r
p
o
r
ate
v
ar
io
u
s
atten
tio
n
m
ec
h
an
is
m
s
to
en
h
an
ce
th
eir
p
er
f
o
r
m
an
ce
in
s
eg
m
en
tin
g
a
n
d
d
etec
tin
g
m
alar
ia
-
i
n
f
ec
ted
R
B
C
s
.
I
t
h
ig
h
lig
h
ts
k
ey
m
etr
ics
s
u
c
h
as
s
e
g
m
e
n
t
atio
n
I
o
U,
Dice
c
o
ef
f
icien
t,
d
etec
tio
n
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
m
AP
at
an
I
o
U
th
r
esh
o
ld
o
f
0
.
5
.
E
ac
h
m
o
d
e
l
v
ar
ian
t
em
p
lo
y
s
a
d
is
tin
ct
atten
tio
n
m
o
d
u
le
d
esig
n
ed
to
r
ef
in
e
f
ea
tu
r
es
in
d
if
f
er
en
t
way
s
,
with
r
em
ar
k
s
s
u
m
m
ar
izin
g
h
o
w
th
ese
m
o
d
u
les
im
p
ac
t
th
e
o
v
er
all
r
es
u
lts
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el,
wh
ic
h
co
m
b
in
es
C
B
AM
with
UNe
t
an
d
DE
T
R
,
ac
h
iev
es
th
e
h
ig
h
est
s
co
r
es
ac
r
o
s
s
all
m
etr
ics,
in
clu
d
in
g
a
s
eg
m
en
tatio
n
I
o
U
o
f
0
.
9
7
,
Dice
c
o
ef
f
icien
t
o
f
0
.
9
8
,
a
n
d
d
etec
tio
n
p
r
ec
is
io
n
n
ea
r
in
g
p
er
f
ec
t
at
0
.
9
9
9
.
T
h
is
s
u
p
e
r
io
r
p
er
f
o
r
m
a
n
ce
is
attr
ib
u
ted
to
C
B
AM
’
s
ab
ilit
y
to
f
o
cu
s
atten
tio
n
o
n
b
o
t
h
ch
an
n
el
-
wis
e
an
d
s
p
atial
f
ea
tu
r
es,
en
ab
lin
g
m
o
r
e
p
r
ec
is
e
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
b
etter
o
v
er
all
s
eg
m
e
n
ta
tio
n
an
d
d
etec
tio
n
r
esu
lts
.
T
h
e
in
clu
s
io
n
o
f
d
u
al
atten
tio
n
ty
p
es
h
elp
s
th
e
m
o
d
el
ex
ce
l
in
b
o
th
lo
ca
lizatio
n
an
d
class
if
icatio
n
task
s
with
in
m
alar
ia
s
m
ea
r
im
ag
es.
Oth
er
v
a
r
ian
ts
s
h
o
w
s
lig
h
tly
lo
wer
p
er
f
o
r
m
an
ce
b
u
t
s
t
ill
d
em
o
n
s
tr
ate
th
e
v
alu
e
o
f
in
c
o
r
p
o
r
atin
g
atten
ti
o
n
m
ec
h
a
n
is
m
s
.
Fo
r
ex
a
m
p
le,
t
h
e
UNe
t+DE
T
R
m
o
d
el
with
SE
atten
tio
n
s
co
r
es
wel
l,
im
p
r
o
v
i
n
g
c
h
an
n
el
-
wis
e
f
ea
tu
r
e
r
ef
in
e
m
en
t
a
n
d
ac
h
ie
v
in
g
a
s
eg
m
en
tatio
n
I
o
U
o
f
0
.
9
4
an
d
d
etec
tio
n
p
r
ec
is
io
n
o
f
0
.
9
9
.
T
h
e
ef
f
icien
t
ch
an
n
el
atten
tio
n
(
E
C
A)
v
ar
i
an
t
o
f
f
e
r
s
a
lig
h
tweig
h
t
alter
n
ativ
e
with
b
alan
ce
d
s
p
ee
d
an
d
ac
cu
r
ac
y
,
wh
ile
th
e
b
o
ttlen
ec
k
att
en
tio
n
m
o
d
u
le
(
B
AM
)
v
ar
ian
t
f
o
cu
s
es
m
o
r
e
o
n
s
em
an
tic
co
n
tex
t
b
u
t
p
er
f
o
r
m
s
s
lig
h
tly
lo
wer
o
v
er
all.
T
h
e
m
o
d
el
with
o
u
t
an
y
atten
tio
n
m
o
d
u
les
p
er
f
o
r
m
s
th
e
wea
k
est,
u
n
d
er
s
co
r
i
n
g
th
e
im
p
o
r
tan
ce
o
f
atten
tio
n
m
ec
h
a
n
is
m
s
in
en
h
an
cin
g
f
ea
t
u
r
e
r
e
p
r
esen
t
atio
n
f
o
r
ac
cu
r
ate
s
eg
m
en
tatio
n
an
d
d
etec
tio
n
.
T
ab
le
3
.
C
o
m
p
a
r
is
o
n
o
f
p
r
o
p
o
s
ed
m
o
d
el
with
v
a
r
iab
le
atten
t
io
n
m
ec
h
a
n
is
m
M
o
d
e
l
v
a
r
i
a
n
t
A
t
t
e
n
t
i
o
n
mo
d
u
l
e
S
e
g
m
e
n
t
a
t
i
o
n
I
o
U
D
i
c
e
c
o
e
f
f
i
c
i
e
n
t
D
e
t
e
c
t
i
o
n
p
r
e
c
i
si
o
n
D
e
t
e
c
t
i
o
n
r
e
c
a
l
l
mA
P
(
I
o
U
@
0
.
5
)
P
r
o
p
o
se
d
(
C
B
A
M
-
U
N
e
t
+
D
E
TR
)
C
B
A
M
0
.
9
7
0
.
9
8
0
.
9
9
9
0
.
9
9
8
0
.
9
9
5
U
N
e
t
+
D
ET
R
+
S
E
S
E
(
sq
u
e
e
z
e
-
e
x
c
i
t
e
)
0
.
9
4
0
.
9
5
0
.
9
9
0
0
.
9
8
7
0
.
9
8
3
U
N
e
t
+
D
ET
R
+
EC
A
EC
A
(
e
f
f
i
c
i
e
n
t
C
A
)
0
.
9
3
0
.
9
4
0
.
9
8
5
0
.
9
8
0
0
.
9
7
8
U
N
e
t
+
D
ET
R
+
B
A
M
B
A
M
0
.
9
1
0
.
9
2
0
.
9
8
1
0
.
9
7
5
0
.
9
7
0
U
N
e
t
+
D
ET
R
(
N
o
a
t
t
e
n
t
i
o
n
)
N
o
n
e
0
.
9
0
0
.
9
1
0
.
9
6
0
0
.
9
4
0
0
.
9
3
0
5.
CO
NCLU
SI
O
N
T
h
e
d
ev
elo
p
m
en
t
o
f
au
to
m
ated
tech
n
o
lo
g
ies
f
o
r
m
ala
r
ia
d
iag
n
o
s
is
is
cr
u
cial
f
o
r
o
v
er
c
o
m
in
g
th
e
s
h
o
r
tco
m
in
g
s
o
f
tr
a
d
itio
n
al
m
icr
o
s
co
p
y
,
wh
ic
h
r
elies
s
i
g
n
if
ican
tly
o
n
e
x
p
er
t
an
aly
s
is
an
d
is
p
r
o
n
e
to
in
co
n
s
is
ten
cy
.
T
h
is
wo
r
k
in
tr
o
d
u
ce
s
R
B
C
_
Fra
m
e_
Net,
a
h
y
b
r
id
d
ee
p
lear
n
in
g
f
r
am
ew
o
r
k
th
at
ef
f
ec
tiv
ely
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
B
C
_
F
r
a
me_
N
et:
a
h
yb
r
id
d
e
ep
lea
r
n
in
g
fr
a
mewo
r
k
fo
r
d
etec
tio
n
o
f red
b
lo
o
d
…
(
Mu
h
a
mma
d
S
h
a
mee
m
P
.
)
1495
co
m
b
in
es
th
e
s
eg
m
en
tatio
n
s
t
r
en
g
th
s
o
f
C
B
AM
-
UNe
t
with
th
e
d
etec
tio
n
ca
p
a
b
ilit
ies
o
f
D
E
T
R
to
ac
cu
r
ately
id
en
tify
an
d
lo
ca
lize
R
B
C
s
in
m
alar
ia
-
in
f
ec
ted
b
lo
o
d
s
m
ea
r
s
.
T
h
e
u
s
e
o
f
atten
tio
n
m
ec
h
an
is
m
s
s
ig
n
if
ican
tly
en
h
an
ce
s
th
e
m
o
d
el'
s
ab
ilit
y
t
o
r
ef
in
e
cr
itical
f
ea
tu
r
es,
allo
win
g
it
to
co
n
ce
n
t
r
ate
o
n
d
iag
n
o
s
tically
r
elev
an
t
ar
ea
s
wh
ile
m
in
im
izin
g
b
ac
k
g
r
o
u
n
d
n
o
is
e
an
d
v
is
u
al
ar
tifa
cts.
Ou
r
ex
p
er
im
en
ts
co
n
f
ir
m
th
e
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
o
f
R
B
C
_
Fra
m
e_
Net,
wh
ich
o
u
tp
e
r
f
o
r
m
s
cu
r
r
e
n
t
s
tate
-
of
-
th
e
-
ar
t
m
eth
o
d
s
.
T
h
e
m
o
d
el
ac
h
iev
ed
h
ig
h
s
eg
m
e
n
tatio
n
ac
c
u
r
ac
y
(
I
o
U:
0
.
9
7
a
n
d
d
ice
:
0
.
9
8
)
an
d
ex
ce
p
tio
n
al
d
etec
tio
n
m
etr
ics
(
p
r
ec
is
io
n
:
0
.
9
9
9
,
r
ec
all:
0
.
9
9
8
,
an
d
m
AP:
0
.
9
9
5
)
,
u
n
d
er
s
co
r
i
n
g
its
r
ea
d
i
n
ess
f
o
r
clin
ical
u
s
e.
I
ts
r
esil
ien
c
e
to
v
ar
iatio
n
s
in
im
ag
e
q
u
ality
an
d
s
tain
in
g
f
u
r
th
er
s
u
p
p
o
r
ts
its
ap
p
licatio
n
in
r
ea
l
-
wo
r
ld
d
iag
n
o
s
tic
en
v
ir
o
n
m
en
ts
,
wh
er
e
s
tan
d
ar
d
izatio
n
is
o
f
ten
a
ch
all
en
g
e.
Desp
ite
t
h
ese
en
c
o
u
r
ag
i
n
g
r
esu
lts
,
th
er
e
i
s
r
o
o
m
f
o
r
e
n
h
an
ce
m
e
n
t.
Fu
tu
r
e
wo
r
k
will
f
o
cu
s
o
n
d
e
v
elo
p
in
g
lig
h
tweig
h
t
v
e
r
s
io
n
s
o
f
th
e
m
o
d
el
f
o
r
ed
g
e
c
o
m
p
u
tin
g
a
p
p
licatio
n
s
,
ex
ten
d
in
g
its
ca
p
ab
ilit
ies
to
s
u
p
p
o
r
t
m
u
lti
-
class
d
etec
tio
n
f
o
r
d
is
tin
g
u
is
h
in
g
m
alar
ia
s
p
ec
ies,
an
d
in
teg
r
atin
g
it
with
m
o
b
ile
m
icr
o
s
co
p
y
p
latf
o
r
m
s
f
o
r
f
ield
u
s
e.
I
n
cr
ea
s
in
g
th
e
v
ar
iety
o
f
s
am
p
les
in
th
e
tr
ain
i
n
g
d
ataset
will
also
aid
in
en
h
an
ci
n
g
g
e
n
er
aliza
b
ilit
y
.
I
n
co
n
clu
s
io
n
,
R
B
C
_
Fra
m
e_
Net
r
ep
r
esen
ts
a
m
ajo
r
ad
v
a
n
ce
m
en
t
in
au
to
m
ated
m
alar
ia
d
ia
g
n
o
s
es.
I
t
p
r
o
v
id
es
a
d
ep
en
d
a
b
le,
ef
f
icien
t,
an
d
s
ca
lab
le
s
o
lu
tio
n
t
h
at
b
r
id
g
es
th
e
g
a
p
b
etwe
en
AI
in
n
o
v
atio
n
a
n
d
cl
in
ical
p
r
ac
tice.
B
y
m
in
im
izin
g
r
elian
ce
o
n
m
an
u
al
ev
alu
ati
o
n
,
th
is
f
r
am
ew
o
r
k
h
as
th
e
p
o
ten
tial
to
b
o
o
s
t
d
iag
n
o
s
tic
ac
cu
r
ac
y
,
s
tr
ea
m
lin
e
s
cr
ee
n
in
g
wo
r
k
f
lo
ws,
an
d
im
p
r
o
v
e
h
ea
lt
h
o
u
tco
m
es
in
m
alar
ia
-
en
d
e
m
ic
r
eg
io
n
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
M
u
h
a
m
m
a
d
S
h
a
m
e
e
m
P
.
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
M
a
t
h
i
a
r
as
i
B
a
l
a
k
r
i
s
h
n
a
n
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
t
h
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
o
p
en
ly
av
ailab
le
in
Go
o
g
le
Dr
iv
e
at
h
ttp
s
://d
r
iv
e.
g
o
o
g
le.
co
m
/d
r
iv
e
/f
o
ld
er
s
/1
R
8
m
J
m
5
NYz
ev
Z
f
b
g
r
B
YT
8
QL
C
x
d
R
9
OOXE
h
?u
s
p
=sh
ar
in
g
.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
A
h
me
d
,
P
.
S
.
A
b
d
a
l
q
a
d
i
r
,
S
.
A
.
A
b
d
u
l
l
a
h
,
a
n
d
Y
.
H
a
r
u
n
a
,
“
M
2
A
N
ET:
mo
b
i
l
e
m
a
l
a
r
i
a
a
t
t
e
n
t
i
o
n
n
e
t
w
o
r
k
f
o
r
e
f
f
i
c
i
e
n
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
p
l
a
sm
o
d
i
u
m
p
a
r
a
s
i
t
e
s
i
n
b
l
o
o
d
c
e
l
l
s,”
I
n
t
e
l
i
g
e
n
c
i
a
A
rt
i
f
i
c
i
a
l
,
v
o
l
.
2
8
,
n
o
.
7
6
,
p
p
.
1
8
6
–
1
9
9
,
J
u
l
.
2
0
2
5
,
d
o
i
:
1
0
.
4
1
1
4
/
i
n
t
a
r
t
i
f
.
v
o
l
2
8
i
ss
7
6
p
p
1
8
6
-
1
9
9
.
[
2
]
S
.
S
a
r
k
a
r
,
R
.
S
h
a
r
ma
,
a
n
d
K
.
S
h
a
h
,
“
M
a
l
a
r
i
a
d
e
t
e
c
t
i
o
n
f
r
o
m
R
B
C
i
ma
g
e
s
u
si
n
g
s
h
a
l
l
o
w
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,”
2
0
2
0
,
a
r
X
i
v
:
2
0
1
0
.
1
1
5
2
1
.
[
3
]
S
.
B
o
i
t
a
n
d
R
.
P
a
t
i
l
,
“
A
n
e
f
f
i
c
i
e
n
t
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
f
o
r
m
a
l
a
r
i
a
p
a
r
a
si
t
e
d
e
t
e
c
t
i
o
n
i
n
m
i
c
r
o
sc
o
p
i
c
i
m
a
g
e
s,”
D
i
a
g
n
o
st
i
c
s
,
v
o
l
.
1
4
,
n
o
.
2
3
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
s
t
i
c
s1
4
2
3
2
7
3
8
.
[
4]
S
.
R
a
j
a
r
a
ma
n
e
t
a
l
.
,
“
P
r
e
-
t
r
a
i
n
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
a
s
f
e
a
t
u
r
e
e
x
t
r
a
c
t
o
r
s
t
o
w
a
r
d
i
mp
r
o
v
e
d
ma
l
a
r
i
a
p
a
r
a
s
i
t
e
d
e
t
e
c
t
i
o
n
i
n
t
h
i
n
b
l
o
o
d
sme
a
r
i
m
a
g
e
s,
”
P
e
e
rJ
,
v
o
l
.
6
,
A
p
r
.
2
0
1
8
,
d
o
i
:
1
0
.
7
7
1
7
/
p
e
e
r
j
.
4
5
6
8
.
[
5
]
J.
N
i
y
o
g
i
s
u
b
i
z
o
,
K
.
Z
h
a
o
,
J
.
M
e
n
g
,
Y
.
P
a
n
,
R
.
D
i
d
i
,
a
n
d
Y
.
W
e
i
,
“
A
t
t
e
n
t
i
o
n
-
g
u
i
d
e
d
r
e
s
i
d
u
a
l
U
-
N
e
t
w
i
t
h
S
E
c
o
n
n
e
c
t
i
o
n
a
n
d
A
S
P
P
f
o
r
w
a
t
e
r
sh
e
d
-
b
a
s
e
d
c
e
l
l
se
g
m
e
n
t
a
t
i
o
n
i
n
mi
c
r
o
s
c
o
p
y
i
ma
g
e
s,
”
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
a
t
i
o
n
a
l
B
i
o
l
o
g
y
,
v
o
l
.
3
2
,
n
o
.
2
,
p
p
.
2
2
5
–
2
3
7
,
F
e
b
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
8
9
/
c
m
b
.
2
0
2
3
.
0
4
4
6
.
[
6
]
R
.
L
i
u
e
t
a
l
.
,
“
A
I
D
M
A
N
:
a
n
A
I
-
b
a
se
d
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
s
y
st
e
m
f
o
r
m
a
l
a
r
i
a
d
i
a
g
n
o
si
s
f
r
o
m
smar
t
p
h
o
n
e
t
h
i
n
-
b
l
o
o
d
-
smea
r
i
m
a
g
e
s,
”
Pa
t
t
e
r
n
s
,
v
o
l
.
4
,
n
o
.
9
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
t
e
r
.
2
0
2
3
.
1
0
0
8
0
6
.
[
7
]
S
.
B
.
N
e
t
t
u
r
e
t
a
l
.
,
“
U
l
t
r
a
L
i
g
h
t
S
q
u
e
e
z
e
N
e
t
:
a
d
e
e
p
l
e
a
r
n
i
n
g
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
m
a
l
a
r
i
a
c
l
a
s
s
i
f
i
c
a
t
i
o
n
w
i
t
h
u
p
t
o
5
4
×
f
e
w
e
r
t
r
a
i
n
a
b
l
e
p
a
r
a
m
e
t
e
r
s
f
o
r
r
e
s
o
u
r
c
e
c
o
n
s
t
r
a
i
n
e
d
d
e
v
i
c
e
s
,
”
I
E
E
E
A
c
c
e
s
s
,
v
o
l
.
1
3
,
p
p
.
8
9
4
2
8
–
8
9
4
4
0
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
E
S
S
.
2
0
2
5
.
3
5
7
1
6
9
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.