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
.
1
,
Feb
r
u
ar
y
2
0
2
6
,
p
p
.
841
~
851
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
1
.
p
p
8
4
1
-
8
5
1
841
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Ada
ptive f
ea
ture
fusio
n net
wo
rk f
o
r f
e
tal hea
d seg
m
entatio
n in
ultras
o
und ima
g
es
Vim
a
la
Na
g
a
bo
t
u
1
,
P
a
v
a
n K
um
a
r
Reddy
Sa
na
2
,
B
.
L
a
k
s
hm
i Bh
a
v
a
ni
1
,
Do
na
pa
t
i Srik
a
nth
3
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
(
D
a
t
a
S
c
i
e
n
c
e
)
,
M
a
l
l
a
R
e
d
d
y
U
n
i
v
e
r
s
i
t
y
,
H
y
d
e
r
a
b
a
d
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
D
a
t
a
S
c
i
e
n
c
e
,
B
V
R
a
j
u
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
N
a
r
s
a
p
u
r
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
V
a
a
g
e
sw
a
r
i
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
,
K
a
r
i
mn
a
g
a
r
,
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
Au
g
1
3
,
2
0
2
5
R
ev
is
ed
Dec
2
7
,
2
0
2
5
Acc
ep
ted
J
an
1
0
,
2
0
2
6
Th
e
m
e
a
su
re
m
e
n
t
o
f
fe
tal
b
io
m
e
tri
c
s
fro
m
u
lt
ra
so
u
n
d
ima
g
e
s
p
lay
s
a
v
it
a
l
ro
le i
n
a
ss
e
ss
in
g
p
o
te
n
ti
a
l
d
e
v
e
l
o
p
m
e
n
t
d
u
ri
n
g
p
re
g
n
a
n
c
y
.
Ho
we
v
e
r,
e
x
isti
n
g
fe
tal
se
g
m
e
n
tatio
n
m
e
th
o
d
s
fa
il
e
d
t
o
a
c
c
u
ra
tely
se
g
m
e
n
t
a
n
d
a
ss
e
s
th
e
h
e
a
d
c
ircu
m
fe
re
n
c
e
th
a
t
g
iv
e
s
i
n
a
c
c
u
ra
te
se
g
m
e
n
tatio
n
re
su
lt
s.
To
o
v
e
rc
o
m
e
th
is
li
m
it
a
ti
o
n
,
a
fe
a
tu
re
fe
e
d
b
a
c
k
a
n
d
g
l
o
b
a
l
fe
a
tu
re
wi
th
a
d
a
p
ti
v
e
fe
a
tu
re
fu
si
o
n
n
e
two
r
k
(
F
G
A
–
Ne
t)
m
o
d
e
l
is
p
r
o
p
o
se
d
to
e
n
h
a
n
c
e
fe
tal
h
e
a
d
se
g
m
e
n
tatio
n
(F
HS).
It
i
n
v
o
lv
e
s
f
o
u
r
k
e
y
c
o
m
p
o
n
e
n
ts
f
o
r
fe
a
tu
re
e
x
trac
ti
o
n
,
f
u
sio
n
,
a
n
d
c
o
rre
c
ti
o
n
,
re
sp
e
c
ti
v
e
l
y
.
Th
e
a
d
a
p
ti
v
e
fe
a
tu
re
f
u
sio
n
m
o
d
u
le
(AF
F
M
)
a
n
d
c
o
rre
c
ti
o
n
m
a
p
in
te
g
ra
te
th
e
lo
c
a
l
fe
a
tu
re
s
a
n
d
g
l
o
b
a
l
fe
a
tu
re
s
a
n
d
re
fin
e
th
e
fe
a
tu
re
s
to
e
n
h
a
n
c
e
a
c
c
u
ra
te
F
HS
fro
m
t
h
e
u
lt
ra
so
u
n
d
ima
g
e
s
e
fficie
n
tl
y
.
In
it
iall
y
,
u
lt
ra
so
u
n
d
ima
g
e
s
a
re
o
b
tai
n
e
d
fro
m
t
h
e
two
p
u
b
li
c
ly
a
v
a
il
a
b
le
d
a
tas
e
ts
a
n
d
p
re
p
ro
c
e
ss
e
d
u
sin
g
n
o
rm
a
li
z
a
ti
o
n
a
n
d
d
a
ta
a
u
g
m
e
n
tatio
n
tec
h
n
iq
u
e
s.
F
in
a
ll
y
,
p
re
p
r
o
c
e
ss
e
d
ima
g
e
s
a
re
fe
d
t
o
F
HS
b
y
p
ro
p
o
se
d
F
G
A
-
Ne
t
u
ti
li
z
in
g
Eff
icie
n
tNe
t
-
B0
a
s
th
e
b
a
c
k
b
o
n
e
n
e
two
rk
fo
r
e
fficie
n
t
fe
a
tu
re
e
x
trac
ti
o
n
.
E
x
p
e
rime
n
tal
re
su
lt
s
o
f
p
ro
p
o
se
d
F
G
A
-
Ne
t
a
re
e
v
a
lu
a
ted
u
si
n
g
th
e
d
ice
c
o
e
fficie
n
t
(DC)
o
f
9
5
.
7
8
%
a
n
d
9
8
.
9
5
%
f
o
r
F
H
-
PS
-
Ao
P
a
n
d
HC
-
18
d
a
tas
e
ts,
wh
ich
sh
o
ws
b
e
tt
e
r
re
su
lt
s
th
a
n
t
h
e
e
x
isti
n
g
se
g
m
e
n
tatio
n
a
p
p
ro
a
c
h
e
s li
k
e
in
v
e
rted
b
o
tt
len
e
c
k
p
a
tch
e
x
p
a
n
d
in
g
(IBP
E)
m
e
th
o
d
.
K
ey
w
o
r
d
s
:
Ad
ap
tiv
e
f
ea
tu
r
e
f
u
s
io
n
m
o
d
u
le
Data
au
g
m
en
tatio
n
E
f
f
icien
tNet
-
B0
Fetal
h
ea
d
s
eg
m
en
tatio
n
FGA
-
Net
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
:
Vim
ala
Nag
ab
o
tu
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
(
Data
Sci
en
ce
)
,
Ma
lla
R
ed
d
y
Un
i
v
er
s
ity
Ma
is
am
m
ag
u
d
a,
Du
lap
ally
,
H
y
d
er
ab
a
d
,
I
n
d
ia
E
m
ail: N
ag
ab
o
tu
.
v
im
ala@ma
llar
ed
d
y
u
n
iv
er
s
ity
.
ac
.
in
1.
I
NT
RO
D
UCT
I
O
N
I
n
m
ed
ical
d
iag
n
o
s
tics
f
o
r
f
et
al
g
r
o
wth
an
al
y
s
is
an
d
m
o
n
ito
r
in
g
,
th
e
b
io
m
etr
ic
m
ea
s
u
r
em
en
t
o
f
f
etal
h
ea
d
cir
cu
m
f
e
r
en
ce
(
HC
)
p
lay
s
an
im
p
o
r
tan
t
r
o
le.
T
h
is
HC
f
r
o
m
u
ltra
s
o
u
n
d
im
ag
es
ef
f
icien
tly
ass
is
ts
to
m
ea
s
u
r
e
th
e
d
u
e
d
ate,
g
estatio
n
al
ag
e,
a
n
d
f
etal
weig
h
t
d
u
r
in
g
p
r
e
g
n
an
c
y
,
wh
ich
a
r
e
m
ajo
r
ly
d
ep
e
n
d
en
t
o
n
th
is
HC
m
ea
s
u
r
em
en
t
[
1
]
.
Fo
r
ass
ess
in
g
th
e
n
eu
r
o
lo
g
ical
d
ev
elo
p
m
en
t
an
d
ab
n
o
r
m
alities
in
f
etu
s
g
r
o
wth
in
clin
ical
p
r
ac
tice,
p
r
en
atal
u
ltra
s
o
u
n
d
p
lay
s
a
cr
u
cial
r
o
le
in
th
is
ass
ess
m
en
t.
Gen
er
al
ly
,
th
is
ass
ess
m
en
t
in
v
o
lv
es
th
r
ee
s
tag
es:
in
itially
,
an
u
ltra
s
o
u
n
d
p
r
o
b
e
s
ca
n
s
th
e
f
etal
b
r
ain
a
n
d
ac
q
u
ir
es
t
h
e
3
D
d
ata
d
ir
ec
tly
d
u
r
in
g
th
e
s
ca
n
n
in
g
p
r
o
ce
s
s
.
Af
ter
th
at,
an
u
ltra
s
o
u
n
d
s
p
e
cialist
r
ec
o
g
n
izes
th
e
2
D
f
etal
b
r
ain
m
id
s
ag
ittal
p
lan
e
im
ag
e
m
a
n
u
ally
ac
co
r
d
in
g
t
o
an
at
o
m
ical
k
n
o
wled
g
e
an
d
ac
q
u
i
r
es
a
3
D
u
ltra
s
o
u
n
d
im
ag
e.
At
last
,
th
e
p
h
y
s
ician
m
a
n
u
ally
id
e
n
tif
ies
an
d
s
eg
m
en
ts
th
e
tar
g
et
r
e
g
io
n
s
in
th
e
2
D
im
ag
e
[
2
]
,
[
3
]
.
I
n
r
ec
e
n
t
d
ec
a
d
es,
d
if
f
u
s
io
n
-
weig
h
ted
m
a
g
n
etic
r
eso
n
an
ce
im
a
g
in
g
(
DM
R
I
)
h
as
b
ee
n
m
o
r
e
u
s
ed
f
o
r
ev
a
lu
atin
g
f
etal
b
r
ain
d
ev
elo
p
m
e
n
t
in
u
ter
o
.
Ho
w
ev
er
,
th
e
p
r
o
ce
s
s
o
f
d
ata
ac
q
u
is
itio
n
f
r
o
m
DM
R
I
is
d
i
f
f
icu
lt
to
s
eg
m
en
t
an
d
an
aly
ze
[
4
]
,
[
5
]
.
A
d
d
itio
n
ally
,
tr
an
s
v
ag
i
n
al
u
ltra
s
o
n
o
g
r
ap
h
y
(
T
VS)
is
a
co
m
m
o
n
ly
u
s
ed
tech
n
iq
u
e
b
y
d
o
cto
r
s
a
n
d
p
h
y
s
ician
s
to
m
o
n
ito
r
th
e
d
e
v
elo
p
m
e
n
t
o
f
th
e
e
m
b
r
y
o
[
6
]
.
B
io
lo
g
ical
in
d
icato
r
s
lik
e
cr
o
wn
-
r
u
m
p
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
841
-
8
5
1
842
len
g
th
(
C
R
L
)
,
g
estatio
n
al
s
ac
ar
ea
(
GSA)
,
an
d
y
o
lk
s
ac
d
i
am
eter
(
YSD)
f
r
o
m
T
VS
im
a
g
es
ar
e
u
s
ed
b
y
th
e
d
o
cto
r
s
to
ass
es
s
th
e
g
r
o
wth
a
n
d
d
e
v
elo
p
m
e
n
t
o
f
th
e
f
etu
s
.
Ho
wev
er
,
th
e
ass
ess
m
en
t
o
f
HC
an
d
ev
alu
atio
n
o
f
th
e
af
o
r
em
en
tio
n
ed
in
d
icato
r
s
b
y
T
VS
tech
n
iq
u
e
ar
e
o
b
tain
ed
m
an
u
ally
b
y
th
e
p
h
y
s
ician
s
,
wh
ich
co
n
s
u
m
es
m
o
r
e
tim
e.
T
o
ad
d
r
ess
th
ese
is
s
u
es,
f
ast
m
ag
n
etic
r
eso
n
an
ce
im
a
g
in
g
(
MRI)
ac
q
u
is
itio
n
m
eth
o
d
s
ar
e
u
s
ed
to
ac
q
u
ir
e
2
D
s
lices
f
o
r
f
etal
h
e
ad
s
eg
m
en
tatio
n
(
FHS)
[
7
]
.
I
n
v
ar
io
u
s
ap
p
licatio
n
s
,
b
r
ain
ex
tr
ac
tio
n
f
r
o
m
MRI
s
lices
is
th
e
p
r
im
ar
y
s
tep
,
w
h
ich
in
clu
d
es
s
lice
-
lev
el
m
o
tio
n
co
r
r
ec
tio
n
,
s
lice
-
to
-
v
o
l
u
m
e
r
ec
o
n
s
tr
u
ctio
n
,
an
d
m
o
n
ito
r
in
g
th
e
m
o
tio
n
s
o
f
t
h
e
f
etal
h
ea
d
[
8
]
.
Ho
we
v
er
,
au
t
o
m
ated
f
etal
b
r
ain
s
eg
m
en
tatio
n
f
ac
es
lim
itatio
n
s
d
u
e
to
v
a
r
io
u
s
b
r
ain
s
h
ap
es,
s
tr
u
ctu
r
e,
an
d
s
ize
ac
r
o
s
s
g
estatio
n
al
ag
e,
as
well
as
im
ag
e
d
is
to
r
tio
n
s
an
d
in
ten
s
ity
n
o
n
-
u
n
if
o
r
m
ity
.
Als
o
,
co
n
tr
ast
o
f
th
e
im
ag
e
v
ar
ies
f
o
r
d
is
tin
ct
f
etal
MRI
s
eq
u
en
c
es,
s
u
ch
as
DM
R
I
-
b
ased
d
ata
[
9
]
.
T
h
u
s
,
au
to
m
atic
s
eg
m
en
tatio
n
is
r
eq
u
ir
e
d
in
m
ed
ical
im
a
g
es
an
aly
s
is
f
o
r
p
er
s
o
n
alize
d
m
ed
icin
e
an
d
to
s
tu
d
y
an
ato
m
ical
d
ev
elo
p
m
e
n
t
in
h
ea
lth
y
p
o
p
u
latio
n
s
as
well
as
p
ath
o
lo
g
y
[
1
0
]
.
Hen
ce
,
ar
tific
ial
in
tellig
en
ce
(
AI
)
alg
o
r
ith
m
s
ar
e
u
s
ed
in
m
e
d
ical
i
m
ag
e
s
eg
m
e
n
tatio
n
d
u
e
to
th
e
ir
ab
ilit
y
to
s
eg
m
en
t
o
r
p
r
o
ce
s
s
d
ata
ef
f
icien
tly
with
o
u
t
an
y
m
a
n
u
al
in
ter
v
e
n
tio
n
.
Dee
p
lear
n
i
n
g
(
DL
)
m
o
d
els
s
ig
n
if
ican
tly
en
h
an
ce
d
th
e
im
ag
e
p
r
o
ce
s
s
in
g
in
th
e
m
ed
ical
d
o
m
ai
n
th
at
in
v
o
lv
es
an
aly
s
is
o
f
u
ltra
s
o
u
n
d
f
etal
im
a
g
es
s
eg
m
en
tatio
n
to
ass
ess
f
etal
g
r
o
wth
[
1
1
]
.
R
ec
en
tly
,
DL
-
b
ased
s
eg
m
en
tatio
n
ap
p
r
o
ac
h
e
s
lik
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
[
1
2
]
,
t
r
an
s
f
o
r
m
er
s
,
an
d
o
th
er
d
ee
p
n
e
u
r
al
n
etwo
r
k
(
DNN)
ac
h
ie
v
ed
b
etter
p
er
f
o
r
m
an
ce
in
s
eg
m
en
tatio
n
r
elate
d
task
s
,
wh
ich
h
as
th
e
ab
ilit
y
to
h
an
d
le
lar
g
e
d
atasets
w
ith
p
ix
el
an
n
o
tatio
n
s
ef
f
ec
tiv
ely
[
1
3
]
‒
[
1
5
]
.
Ho
wev
e
r
,
th
ese
D
L
m
o
d
els
also
h
av
e
d
r
awb
ac
k
s
s
u
ch
as
tim
e
co
m
p
lex
ity
,
l
ab
o
r
in
ten
s
iv
e,
an
d
ex
p
en
s
iv
e
to
ac
q
u
i
r
e
lar
g
e
s
ca
le
p
ix
el
an
n
o
tated
d
ataset
[
1
6
]
,
[
1
7
]
.
M
o
r
eo
v
er
,
th
e
g
e
n
er
ali
za
tio
n
p
er
f
o
r
m
an
ce
o
f
ex
is
tin
g
FHS
m
eth
o
d
s
is
li
m
ited
d
u
e
to
v
ar
iatio
n
s
in
q
u
a
n
tity
an
d
q
u
ality
o
f
d
is
tin
ct
d
ataset.
Qiu
et
a
l.
[
1
8
]
d
ev
elo
p
e
d
a
s
eg
m
e
n
tatio
n
m
eth
o
d
t
o
id
en
tif
y
th
e
p
u
b
ic
s
y
m
p
h
y
s
is
‐
f
etal
h
ea
d
s
tan
d
ar
d
p
la
n
e
(
PS
FHSP
)
f
r
o
m
in
tr
ap
ar
tu
m
u
ltra
s
o
u
n
d
im
a
g
e
s
b
ased
o
n
an
e
f
f
icien
t
lig
h
tweig
h
t
n
etwo
r
k
.
R
elev
an
t
f
ea
tu
r
es
wer
e
ex
tr
ac
ted
u
s
in
g
R
esNet
-
1
8
,
wh
ich
u
s
e
r
esid
u
al
b
lo
ck
s
to
im
p
r
o
v
e
f
ea
tu
r
e
ex
tr
ac
tio
n
b
y
p
r
e
v
en
tin
g
v
an
is
h
in
g
g
r
ad
ien
t
is
s
u
es.
T
ask
s
p
ec
if
ic
lay
er
s
we
r
e
u
s
ed
f
o
r
ac
cu
r
ate
class
if
icatio
n
an
d
to
id
en
tif
y
th
e
co
r
r
ec
t
u
ltra
s
o
u
n
d
p
lan
e.
C
ai
et
a
l.
[
1
9
]
p
r
esen
ted
a
s
eg
m
en
tatio
n
m
o
d
el
f
o
r
f
etal
h
e
a
d
an
d
p
u
b
ic
s
y
m
p
h
y
s
is
u
s
in
g
u
ltra
s
o
u
n
d
im
ag
es.
T
h
e
p
r
esen
te
d
s
eg
m
e
n
tatio
n
m
o
d
el
u
tili
ze
d
a
U
-
Ne
t
-
lik
e
ap
p
r
o
ac
h
with
an
in
v
er
t
ed
b
o
ttlen
ec
k
p
atch
ex
p
an
d
i
n
g
(
I
B
PE)
m
o
d
u
le
to
ef
f
icien
tly
ca
p
tu
r
e
b
o
th
lo
ca
l
an
d
g
lo
b
al
s
em
an
tic
f
ea
tu
r
es.
Du
b
ey
et
a
l.
[
2
0
]
d
ev
elo
p
e
d
a
f
etal
u
ltra
s
o
u
n
d
s
eg
m
en
tatio
n
m
o
d
el
b
ased
o
n
h
ier
ar
ch
ical
d
en
s
ity
r
e
g
r
ess
io
n
(
HDR)
with
d
ee
p
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
(
DC
NN)
.
An
ad
v
an
tag
e
o
f
t
h
e
d
ev
el
o
p
ed
FHS
m
o
d
el
wa
s
its
u
s
e
o
f
ellip
s
e
f
itti
n
g
to
ev
alu
ate
f
etal
h
ea
d
ci
r
cu
m
f
er
e
n
ce
,
wh
ich
h
elp
s
th
e
s
eg
m
en
tatio
n
p
r
o
ce
s
s
ef
f
icien
tly
.
C
h
en
et
a
l
.
[
2
1
]
s
u
g
g
ested
a
s
eg
m
en
tatio
n
m
o
d
el
b
ased
o
n
th
e
f
etal
h
ea
d
-
p
u
b
ic
s
y
m
p
h
y
s
is
s
eg
m
en
tatio
n
n
etwo
r
k
(
FH
-
PS
SNet)
f
o
r
th
e
esti
m
atio
n
o
f
au
to
m
atic
an
g
le
o
f
p
r
o
g
r
ess
io
n
(
Ao
P)
with
d
ir
ec
tio
n
g
u
i
d
an
ce
.
An
ad
v
a
n
tag
e
o
f
th
e
FH
-
PS
SNet
m
o
d
el
with
d
i
r
ec
tio
n
s
tr
ateg
y
was
th
at
it
h
elp
ed
id
en
tify
f
etal
h
ea
d
an
d
f
o
cu
s
o
n
th
e
p
o
s
itio
n
o
f
th
e
p
u
b
ic
s
y
m
p
h
y
s
is
ef
f
ec
t
iv
ely
.
C
h
en
et
a
l.
[
2
2
]
d
esig
n
ed
a
d
u
al
p
ath
b
o
u
n
d
ar
y
g
u
id
ed
r
esid
u
al
n
etwo
r
k
(
DB
R
N)
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
f
o
r
au
to
m
ated
f
etal
h
e
ad
-
p
u
b
ic
s
y
m
p
h
y
s
is
(
FH
-
PS
)
s
eg
m
en
tatio
n
.
T
h
e
d
esig
n
ed
DB
R
N
m
o
d
el
in
v
o
l
v
es
a
m
u
lti
-
s
ca
le
weig
h
ted
m
o
d
u
le
to
c
o
llect
g
lo
b
al
c
o
n
te
x
t
in
f
o
r
m
atio
n
an
d
en
h
an
ce
d
b
o
u
n
d
ar
y
m
o
d
u
le
to
ac
q
u
ir
e
p
r
ec
is
e
b
o
u
n
d
ar
y
in
f
o
r
m
atio
n
ab
o
u
t th
e
f
etal
h
ea
d
.
Alth
o
u
g
h
v
a
r
io
u
s
DL
b
ased
m
o
d
els
h
av
e
b
ee
n
u
tili
ze
d
f
o
r
FHS,
th
ey
f
ac
e
s
ev
er
al
ch
allen
g
es
s
u
ch
as
p
o
o
r
g
e
n
er
aliza
tio
n
ac
r
o
s
s
d
at
asets
,
s
en
s
itiv
ity
to
n
o
is
e,
a
n
d
d
u
e
to
v
ar
iatio
n
s
in
f
etal
h
ea
d
s
h
ap
e,
s
ize,
an
d
o
r
ien
tatio
n
.
Mo
r
e
o
v
er
,
m
o
s
t
o
f
s
eg
m
en
tatio
n
m
o
d
els
r
ely
o
n
f
ix
ed
an
ato
m
ical
ass
u
m
p
tio
n
s
an
d
ellip
s
e
f
itti
n
g
,
wh
ich
ar
e
i
n
ef
f
ec
tiv
e
f
o
r
d
if
f
e
r
en
t
f
etal
p
o
s
itio
n
s
an
d
lo
w
-
q
u
ality
u
ltra
s
o
u
n
d
im
ag
es.
T
h
u
s
,
to
o
v
er
co
m
e
th
ese
lim
itatio
n
s
,
p
r
o
p
o
s
ed
s
eg
m
en
t
atio
n
m
o
d
el
is
u
s
ed
f
o
r
FHS f
r
o
m
u
ltra
s
o
u
n
d
im
ag
es e
f
f
icien
tly
.
T
h
e
m
ajo
r
c
o
n
tr
ib
u
tio
n
s
o
f
t
h
is
r
esear
ch
ar
e:
i)
I
n
teg
r
atio
n
o
f
E
f
f
icien
tNet
-
B
0
as
a
b
ac
k
b
o
n
e:
t
h
e
p
r
o
p
o
s
e
d
f
ea
tu
r
e
f
ee
d
b
ac
k
a
n
d
g
lo
b
a
l
f
ea
tu
r
e
with
ad
ap
tiv
e
f
ea
t
u
r
e
f
u
s
io
n
n
et
wo
r
k
(
FGA
–
Net)
b
ased
s
e
g
m
en
tatio
n
m
o
d
el
in
tr
o
d
u
ce
s
th
e
u
s
e
o
f
E
f
f
icien
tNet
-
B
0
m
o
d
el
in
FH
S,
wh
ich
en
ab
le
th
e
e
f
f
ec
tiv
e
ex
tr
ac
tio
n
o
f
b
o
th
lo
w
-
lev
el
a
n
d
h
ig
h
-
lev
el
f
ea
tu
r
es a
cr
o
s
s
m
u
ltip
le
s
ca
les in
th
e
u
ltra
s
o
u
n
d
im
a
g
es.
ii)
Mu
lti
-
s
ca
le
co
n
tex
t
-
awa
r
e
s
e
g
m
en
tatio
n
:
t
h
e
m
u
lti
-
s
ca
le
f
ea
tu
r
e
f
ee
d
b
ac
k
(
MSFF)
m
o
d
u
le
in
th
e
p
r
o
p
o
s
ed
FGA
-
Net
m
o
d
el
is
p
r
o
p
o
s
ed
to
ca
p
tu
r
e
b
o
th
g
lo
b
al
s
em
an
tic
co
n
te
x
t
an
d
f
in
e
-
g
r
ain
e
d
s
tr
u
ctu
r
al
d
etails
o
f
f
etal
h
ea
d
ef
f
icien
tly
.
T
h
e
h
ig
h
to
lo
w
lev
el
f
ea
tu
r
e
f
u
s
io
n
(
HL
F)
m
o
d
u
le
in
FGA
-
Net
h
elp
s
to
s
eg
m
en
t th
e
f
etal
h
ea
d
s
o
f
v
a
r
io
u
s
s
izes a
n
d
s
h
a
p
es in
u
ltra
s
o
u
n
d
im
ag
es p
r
ec
i
s
ely
.
iii)
Ad
ap
tiv
e
f
ea
tu
r
e
f
u
s
io
n
m
o
d
u
le
(
AFFM)
:
an
A
FF
M
with
i
n
FGA
-
Net
i
s
u
s
ed
to
i
ter
ativ
ely
r
ef
in
e
an
d
f
u
s
e
th
e
b
o
th
lo
w
lev
el
an
d
h
ig
h
-
lev
el
f
ea
tu
r
es.
T
h
is
m
o
d
u
le
en
h
a
n
ce
d
ac
cu
r
ate
s
eg
m
en
tatio
n
b
y
ef
f
ec
tiv
ely
elim
in
atin
g
n
o
is
e
a
n
d
h
ig
h
lig
h
tin
g
th
e
r
elev
a
n
t stru
ctu
r
es in
co
m
p
lex
u
ltra
s
o
u
n
d
im
ag
es.
T
h
e
r
em
ai
n
in
g
p
ar
t
o
f
th
is
m
a
n
u
s
cr
ip
t
is
o
r
g
an
ize
d
as
f
o
llo
ws:
s
ec
tio
n
2
d
is
cu
s
s
es
th
e
m
eth
o
d
o
lo
g
y
.
Sectio
n
3
ex
p
lai
n
s
th
e
p
r
o
p
o
s
ed
f
ea
tu
r
e
f
ee
d
b
ac
k
an
d
g
lo
b
al
f
ea
tu
r
e
with
a
d
ap
tiv
e
f
ea
tu
r
e
f
u
s
io
n
(
AFF)
n
etwo
r
k
-
b
ased
s
eg
m
en
tatio
n
m
eth
o
d
.
Sectio
n
4
p
r
esen
ts
ex
p
er
im
en
tal
r
esu
lts
an
d
d
is
c
u
s
s
io
n
s
.
Sectio
n
5
co
n
clu
d
es th
e
p
ap
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
A
d
a
p
tive
fea
tu
r
e
fu
s
io
n
n
etw
o
r
k
fo
r
feta
l h
ea
d
s
eg
men
ta
tio
n
in
u
ltr
a
s
o
u
n
d
ima
g
e
s
(
V
ima
la
N
a
g
a
b
o
tu
)
843
2.
M
E
T
H
O
D
T
h
e
aim
o
f
th
is
r
esear
ch
is
to
d
ev
elo
p
a
n
ef
f
ec
tiv
e
s
eg
m
e
n
tatio
n
m
eth
o
d
f
o
r
FHS
,
wh
ic
h
in
v
o
lv
es
th
r
ee
p
h
ases
:
d
ata
ac
q
u
is
itio
n
,
p
r
ep
r
o
ce
s
s
in
g
,
an
d
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
.
Fig
u
r
e
1
r
ep
r
e
s
en
ts
th
e
p
r
o
p
o
s
ed
FHS
f
r
am
ewo
r
k
u
s
in
g
a
DL
-
b
ased
s
eg
m
en
tatio
n
m
eth
o
d
.
T
h
e
p
r
o
ce
s
s
o
f
FHS
is
:
in
iti
ally
,
th
e
u
ltra
s
o
u
n
d
im
ag
es
o
f
f
etal
a
r
e
o
b
tain
ed
f
r
o
m
two
b
e
n
ch
m
ar
k
d
atasets
.
T
h
en
,
th
e
ac
q
u
ir
ed
im
ag
es
a
r
e
p
r
e
p
r
o
ce
s
s
ed
b
y
tech
n
iq
u
e
lik
e
n
o
r
m
aliza
tio
n
,
r
esizin
g
,
a
n
d
d
ata
a
u
g
m
e
n
tatio
n
to
en
h
a
n
ce
th
e
s
eg
m
en
ta
tio
n
p
r
o
ce
s
s
.
Af
te
r
p
r
ep
r
o
ce
s
s
in
g
,
th
e
e
n
h
an
ce
d
i
m
ag
es
ar
e
u
s
ed
f
o
r
FHS
b
y
t
h
e
p
r
o
p
o
s
ed
DL
ap
p
r
o
ac
h
.
T
h
e
o
v
er
all
p
r
o
ce
s
s
is
b
r
ief
ly
ex
p
lain
ed
in
th
is
s
ec
tio
n
.
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
FHS
f
r
am
ewo
r
k
u
s
in
g
a
DL
f
ea
t
u
r
e
f
u
s
io
n
-
b
ased
s
eg
m
e
n
tatio
n
m
e
th
o
d
2
.
1
.
Da
t
a
s
et
s
I
n
th
is
r
esear
ch
,
th
e
u
ltra
s
o
u
n
d
f
ea
t
h
ea
d
i
m
ag
es
u
s
ed
f
o
r
s
eg
m
en
tatio
n
ar
e
ac
q
u
ir
ed
f
r
o
m
tw
o
b
en
ch
m
ar
k
d
atasets
:
th
e
FH
-
PS
-
Ao
P
[
2
2
]
an
d
HC
-
1
8
[
2
3
]
d
atasets
r
esp
ec
tiv
ely
.
T
h
ese
tw
o
d
atasets
p
r
o
v
id
e
a
wid
e
r
an
g
e
o
f
f
etal
h
ea
d
u
ltra
s
o
u
n
d
im
ag
es,
wh
ic
h
en
s
u
r
e
r
o
b
u
s
tn
ess
in
tr
ain
in
g
an
d
ev
alu
atio
n
f
o
r
s
eg
m
en
tatio
n
task
s
.
Mo
r
eo
v
e
r
,
th
ey
i
n
clu
d
e
a
n
n
o
tatio
n
s
f
r
o
m
th
e
ex
p
er
ts
th
at
en
s
u
r
e
h
ig
h
-
q
u
ality
g
r
o
u
n
d
tr
u
th
f
o
r
ac
cu
r
ate
s
eg
m
e
n
tatio
n
o
f
f
etal
h
ea
d
s
tr
u
ct
u
r
es
wh
ich
is
cr
u
cial
f
o
r
p
r
e
n
atal
d
iag
n
o
s
tics
.
T
ab
le
1
r
ep
r
esen
ts
d
etail
d
escr
ip
tio
n
o
f
two
b
en
c
h
m
ar
k
d
atasets
.
T
ab
le
1
.
Data
s
et
d
escr
ip
tio
n
F
e
a
t
u
r
e
s
FH
-
PS
-
A
o
P
d
a
t
a
s
et
HC
-
1
8
d
a
t
a
se
t
S
o
u
r
c
e
/
i
n
st
i
t
u
t
i
o
n
P
e
r
i
n
a
t
a
l
t
r
a
n
sp
e
r
i
n
e
a
l
u
l
t
r
a
s
o
u
n
d
v
i
d
e
o
s
R
a
d
b
o
u
d
u
n
i
v
e
r
s
i
t
y
m
e
d
i
c
a
l
c
e
n
t
e
r
,
N
e
t
h
e
r
l
a
n
d
s
To
t
a
l
sam
p
l
e
s
5
1
0
0
i
ma
g
e
s
1
3
5
4
i
ma
g
e
s
S
u
b
j
e
c
t
s
/
p
a
t
i
e
n
t
s
M
u
l
t
i
p
l
e
p
r
e
g
n
a
n
t
w
o
m
e
n
(
g
e
st
a
t
i
o
n
a
l
a
g
e
v
a
r
i
e
s)
5
5
1
p
r
e
g
n
a
n
t
w
o
me
n
Tr
a
i
n
i
n
g
sam
p
l
e
s
4
0
0
0
9
9
9
V
a
l
i
d
a
t
i
o
n
sam
p
l
e
s
4
0
0
N
o
t
s
p
e
c
i
f
i
e
d
Te
st
i
n
g
s
a
m
p
l
e
s
7
0
0
3
5
5
I
mag
e
r
e
s
o
l
u
t
i
o
n
r
a
n
g
e
8
0
0
×
6
0
0
t
o
1
0
2
4
×
7
6
8
p
i
x
e
l
s
(
a
p
p
r
o
x
.
)
6
4
0
×
4
8
0
t
o
8
0
0
×
6
0
0
p
i
x
e
l
s (a
p
p
r
o
x
.
)
U
l
t
r
a
s
o
u
n
d
m
a
c
h
i
n
e
t
y
p
e
s
H
i
g
h
-
f
r
e
q
u
e
n
c
y
t
r
a
n
sp
e
r
i
n
e
a
l
p
r
o
b
e
s;
G
E
V
o
l
u
s
o
n
E
8
P
h
i
l
i
p
s a
n
d
G
E
u
l
t
r
a
s
o
u
n
d
mac
h
i
n
e
s
A
n
n
o
t
a
t
i
o
n
q
u
a
l
i
t
y
V
e
r
i
f
i
e
d
b
y
t
w
o
s
e
n
i
o
r
r
a
d
i
o
l
o
g
i
st
s
G
o
l
d
-
st
a
n
d
a
r
d
a
n
n
o
t
a
t
i
o
n
s
b
y
me
d
i
c
a
l
e
x
p
e
r
t
s
F
o
r
mat
P
N
G
/
JP
EG
u
l
t
r
a
s
o
u
n
d
f
r
a
mes
e
x
t
r
a
c
t
e
d
f
r
o
m
v
i
d
e
o
D
I
C
O
M
o
r
c
o
n
v
e
r
t
e
d
g
r
a
y
sc
a
l
e
P
N
G
2
.
1
.
1
.
F
H
-
PS
-
Ao
P
da
t
a
s
et
T
h
is
FH
-
PS
-
Ao
P
d
atase
t
is
an
o
p
en
-
ac
ce
s
s
d
ataset
wid
ely
u
s
ed
in
FHS
an
d
HC
m
ea
s
u
r
e
m
en
t.
T
h
is
d
ataset
in
clu
d
es
5
,
1
0
0
d
ata
s
am
p
les,
wh
ich
ar
e
ex
tr
ac
ted
f
r
o
m
p
er
in
atal
tr
a
n
s
p
er
in
ea
l
u
ltra
s
o
u
n
d
v
id
eo
s
.
T
h
ese
v
id
eo
s
ar
e
c
ar
ef
u
lly
ac
q
u
ir
ed
b
y
a
p
r
o
f
icien
t
s
o
n
o
g
r
ap
h
er
an
d
co
n
f
ir
m
ed
b
y
2
ex
p
er
i
en
ce
d
r
a
d
io
lo
g
is
ts
.
At
last
,
th
e
ac
q
u
ir
es
im
ag
es
f
r
o
m
th
e
v
id
e
o
s
ar
e
ca
teg
o
r
ized
in
to
3
s
ets
s
u
ch
as
tr
ain
in
g
(
4
,
0
0
0
)
,
test
in
g
(
7
0
0
)
,
an
d
v
alid
atio
n
(
4
0
0
)
,
r
esp
ec
tiv
ely
.
2
.
1
.
2
.
H
C
-
1
8
da
t
a
s
et
T
h
e
HC
-
1
8
d
ataset
at
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/th
an
h
b
n
h
p
h
a
n
/h
c1
8
-
g
r
a
n
d
-
c
h
allen
g
e
is
a
p
u
b
licly
a
v
ailab
le
d
ataset
th
at
is
g
en
er
ally
u
s
ed
to
v
alid
ate
t
h
e
s
eg
m
en
tatio
n
p
r
o
ce
s
s
o
f
m
o
d
els.
T
h
is
HC
-
1
8
d
ataset
in
v
o
lv
es
1
3
5
4
u
ltra
s
o
u
n
d
im
ag
es
ac
q
u
ir
ed
f
r
o
m
5
5
1
p
r
eg
n
an
t
wo
m
en
,
wh
ich
ar
e
f
u
r
th
er
d
iv
id
ed
in
t
o
tr
ain
in
g
(
9
9
9
)
an
d
test
in
g
(
3
5
5
)
.
T
h
ese
im
ag
es a
r
e
p
ass
ed
to
th
e
n
ex
t
p
h
ase,
th
e
p
r
e
-
p
r
o
ce
s
s
in
g
s
tag
e.
2
.
2
.
P
re
-
pro
ce
s
s
ing
R
aw
f
etal
im
ag
es
ac
q
u
ir
ed
f
o
r
th
is
s
tu
d
y
ar
e
p
r
o
ce
s
s
ed
in
a
p
r
e
-
p
r
o
ce
s
s
in
g
s
tag
e
to
en
s
u
r
e
th
e
d
ata
is
in
a
u
s
ef
u
l
f
o
r
m
at
f
o
r
ef
f
ec
tiv
e
s
eg
m
en
tatio
n
.
T
h
e
s
eg
m
en
t
atio
n
tech
n
i
q
u
es,
s
u
ch
as
n
o
r
m
aliza
tio
n
an
d
d
ata
au
g
m
en
tatio
n
,
ar
e
u
s
ed
in
t
h
is
r
esear
ch
to
en
h
an
ce
t
h
e
im
ag
es.
T
h
e
p
r
o
ce
s
s
es
o
f
th
ese
tech
n
iq
u
es
ar
e
d
is
cu
s
s
ed
as f
o
llo
ws:
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
841
-
8
5
1
844
2
.
2
.
1
.
No
rma
liza
t
io
n
Gen
er
ally
,
u
ltra
s
o
u
n
d
im
a
g
es
h
av
e
a
p
ix
el
i
n
ten
s
ity
r
an
g
e
o
f
0
to
2
5
5
.
Ho
wev
er
,
in
t
h
e
ac
q
u
ir
ed
u
ltra
s
o
u
n
d
im
a
g
e,
p
ix
els
v
a
r
y
b
etwe
en
im
ag
es,
wh
ich
af
f
ec
ts
th
e
ac
cu
r
ac
y
o
f
th
e
s
eg
m
en
t
atio
n
p
r
o
ce
s
s
.
T
h
u
s
,
a
m
in
-
m
ax
n
o
r
m
aliza
tio
n
[
2
4
]
is
ap
p
lied
to
r
escale
th
e
p
ix
el
in
ten
s
ities
in
th
e
u
ltra
s
o
u
n
d
im
ag
es
with
in
th
e
r
an
g
e
o
f
0
a
n
d
1
.
2
.
2
.
2
.
Da
t
a
a
ug
m
ent
a
t
io
n
Data
au
g
m
en
tatio
n
tech
n
i
q
u
e
s
ar
e
em
p
lo
y
ed
f
o
llo
win
g
th
e
n
o
r
m
aliza
tio
n
o
f
u
ltra
s
o
u
n
d
im
ag
es
to
g
en
er
ate
ad
d
itio
n
al
tr
ain
in
g
d
ata
f
r
o
m
th
e
e
x
is
tin
g
d
ataset.
T
h
is
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
e
g
en
er
ates
n
ew
d
ata
b
y
m
o
d
if
y
i
n
g
th
e
ex
is
tin
g
d
ata
co
n
d
itio
n
s
an
d
f
etal
p
o
s
itio
n
s
.
Ho
r
izo
n
tal
f
lip
p
in
g
[
2
5
]
,
ce
n
ter
f
lip
p
in
g
,
r
o
tatio
n
,
an
d
ad
ju
s
tm
en
ts
to
co
n
tr
ast
an
d
b
r
ig
h
t
n
ess
ar
e
th
e
d
ata
au
g
m
en
tatio
n
tec
h
n
iq
u
es
u
s
ed
in
th
e
n
o
r
m
alize
d
im
a
g
es
to
in
cr
ea
s
e
th
e
d
ata
s
am
p
les.
Ad
d
itio
n
ally
,
b
r
ig
h
tn
ess
an
d
c
o
n
tr
ast
ad
ju
s
tm
en
ts
ar
e
u
tili
ze
d
to
h
an
d
le
v
ar
io
u
s
lig
h
tin
g
co
n
d
itio
n
s
in
t
h
e
n
o
r
m
alize
d
im
ag
es,
wh
ic
h
h
el
p
im
p
r
o
v
e
th
e
FHS
p
er
f
o
r
m
an
ce
.
T
ab
le
2
r
e
p
r
ese
n
ts
th
e
d
ata
au
g
m
e
n
tatio
n
tec
h
n
iq
u
e
u
s
ed
f
o
r
th
e
p
r
o
p
o
s
ed
FHS
f
r
am
ewo
r
k
.
T
h
ese
p
r
ep
r
o
ce
s
s
ed
im
ag
es a
r
e
th
en
p
ass
ed
to
th
e
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
m
o
d
el.
T
ab
le
2
.
Au
g
m
en
tatio
n
tec
h
n
i
q
u
es u
s
ed
in
p
r
e
-
p
r
o
ce
s
s
in
g
f
o
r
u
ltra
s
o
u
n
d
im
ag
es
A
u
g
m
e
n
t
a
t
i
o
n
t
e
c
h
n
i
q
u
e
V
a
l
u
e
s
F
l
i
p
p
i
n
g
C
e
n
t
e
r
f
l
i
p
=
−
10
°
t
o
+
10
°
H
o
r
i
z
o
n
t
a
l
f
l
i
p
=
−
20
°
t
o
+
20
°
R
o
t
a
t
i
n
g
R
a
n
d
o
m
l
y
r
o
t
a
t
e
d
u
p
t
o
10
°
t
o
15
°
B
r
i
g
h
t
n
e
ss
−
20%
to
+
20%
C
o
n
t
r
a
st
±
10%
to
±
20%
3.
T
H
E
P
RO
P
O
SE
D
F
G
A
-
NE
T
ARCH
I
T
E
CT
U
RE
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
ar
ch
it
ec
tu
r
e
p
r
o
v
id
es
a
r
o
b
u
s
t
f
r
am
e
wo
r
k
f
o
r
FHS
b
y
ef
f
ec
tiv
ely
p
r
o
ce
s
s
in
g
p
r
e
-
p
r
o
ce
s
s
ed
u
ltra
s
o
u
n
d
im
a
g
es.
T
h
e
m
ain
o
b
jectiv
e
o
f
th
e
p
r
o
p
o
s
ed
FGA
-
Net
is
to
u
tili
ze
m
u
ltip
le
f
ea
tu
r
es
at
v
ar
io
u
s
s
ca
les
b
y
c
o
m
b
in
i
n
g
b
o
th
lo
w
-
a
n
d
h
ig
h
-
lev
el
f
ea
tu
r
es
with
th
e
m
o
d
if
icati
o
n
o
f
th
e
f
ee
d
b
ac
k
m
o
d
u
le.
T
h
e
n
,
a
g
lo
b
al
f
ea
tu
r
e
m
ap
is
g
en
er
ated
alo
n
g
with
a
m
o
d
if
icatio
n
m
ap
to
d
etec
t
f
in
e
g
r
ain
ed
d
etails.
Fig
u
r
e
2
r
ep
r
esen
ts
th
e
p
r
o
p
o
s
ed
F
GA
-
Net
m
o
d
el
f
o
r
FHS
u
s
in
g
u
ltra
s
o
u
n
d
im
a
g
es.
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
ar
ch
itectu
r
e
co
m
p
r
is
es
o
f
f
o
u
r
co
m
p
o
n
en
ts
s
u
ch
as
B
ac
k
b
o
n
e
n
etwo
r
k
,
MSFF
m
o
d
u
le,
g
lo
b
al
f
ea
tu
r
e
m
o
d
u
l
e
(
GFM)
,
an
d
AFFM.
T
h
ese
co
m
p
o
n
e
n
ts
an
d
th
eir
p
r
o
ce
s
s
es a
r
e
ex
p
lain
ed
b
r
ief
ly
as
Fig
u
r
e
2
.
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
p
r
o
p
o
s
ed
FGA
-
Net
m
o
d
el
f
o
r
FHS u
s
in
g
u
ltra
s
o
u
n
d
im
ag
es
3
.
1
.
B
a
c
k
bo
ne
net
wo
rk
/f
e
a
t
ure
ex
t
ra
ct
io
n net
wo
r
k
I
n
th
e
p
r
e
-
p
r
o
ce
s
s
ed
u
ltra
s
o
u
n
d
im
a
g
es,
th
e
ap
p
ea
r
an
ce
o
f
f
in
e
-
g
r
ain
ed
d
etails
is
d
if
f
ic
u
lt
ex
tr
ac
t,
wh
ich
h
elp
s
in
ac
cu
r
ate
FHS.
T
h
u
s
,
a
DC
NN
is
u
tili
ze
d
to
c
ap
tu
r
e
d
ee
p
f
ea
tu
r
es
lik
e
b
o
th
lo
w
-
lev
el
f
ea
t
u
r
es
an
d
h
ig
h
-
le
v
el
f
ea
t
u
r
es,
f
o
r
le
ar
n
in
g
f
in
e
-
g
r
ain
e
d
d
etails
to
s
eg
m
en
t
th
e
f
etal
h
ea
d
p
r
ec
i
s
ely
.
Ho
wev
er
,
th
e
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
A
d
a
p
tive
fea
tu
r
e
fu
s
io
n
n
etw
o
r
k
fo
r
feta
l h
ea
d
s
eg
men
ta
tio
n
in
u
ltr
a
s
o
u
n
d
ima
g
e
s
(
V
ima
la
N
a
g
a
b
o
tu
)
845
p
r
o
p
o
s
ed
FGA
-
Net
h
av
e
th
e
R
esNET
-
5
0
as
a
b
ac
k
b
o
n
e
n
et
wo
r
k
,
d
u
e
t
o
its
d
ee
p
an
d
h
ea
v
y
ar
ch
itectu
r
e,
t
h
at
lead
to
o
v
er
f
it
o
n
lim
ited
f
etal
u
ltra
s
o
u
n
d
d
ata.
Als
o
,
th
e
R
e
s
Net
-
5
0
m
o
d
el
f
ailed
to
ca
p
t
u
r
e
s
u
b
tle
v
ar
iatio
n
s
in
f
etal
h
ea
d
b
o
u
n
d
a
r
ies,
wh
ic
h
m
ak
e
t
h
e
FGA
-
Net
s
tr
u
g
g
le
to
s
eg
m
en
t
th
e
f
etal
h
ea
d
s
with
v
ar
y
in
g
s
izes
an
d
s
h
ap
es.
T
h
u
s
,
in
th
is
r
esear
c
h
,
th
e
E
f
f
icien
tNet
-
B
0
m
o
d
el
is
u
s
ed
as
th
e
b
ac
k
b
o
n
e
n
et
wo
r
k
f
o
r
ex
t
r
ac
tin
g
f
ea
tu
r
es
in
th
e
p
r
o
p
o
s
ed
FGA
-
Net.
T
h
e
E
f
f
icien
tNet
-
B
0
o
v
er
co
m
es
th
ese
lim
itatio
n
s
with
its
lig
h
tweig
h
t
an
d
b
alan
ce
d
s
ca
lin
g
s
tr
ateg
y
,
e
n
ab
lin
g
b
etter
g
en
er
aliza
tio
n
,
f
in
er
b
o
u
n
d
ar
y
lo
ca
lizatio
n
,
an
d
im
p
r
o
v
ed
s
eg
m
en
tatio
n
o
f
f
etal
h
ea
d
s
i
n
d
iv
er
s
e
u
ltra
s
o
u
n
d
s
ettin
g
s
T
h
is
b
ac
k
b
o
n
e
n
etwo
r
k
in
th
e
p
r
o
p
o
s
ed
FGA
-
Net
s
ca
les
th
e
d
ep
th
,
wid
th
,
an
d
r
eso
lu
tio
n
o
f
th
e
p
r
e
p
r
o
ce
s
s
ed
u
ltra
s
o
u
n
d
im
a
g
es
u
s
in
g
th
e
co
m
p
o
u
n
d
s
ca
lin
g
m
eth
o
d
f
o
r
ef
f
icien
t
f
e
atu
r
e
ex
tr
ac
tio
n
.
T
h
e
s
ca
lin
g
tec
h
n
i
q
u
e
in
E
f
f
icien
tNet
-
B
0
ex
tr
a
cts
b
o
th
g
lo
b
al
an
d
lo
ca
l
f
ea
tu
r
es
to
im
p
r
o
v
e
f
et
al
h
ea
d
b
o
u
n
d
a
r
y
d
etec
tio
n
.
Un
lik
e
r
esid
u
al
n
etwo
r
k
(
R
e
s
Net)
,
v
is
u
al
g
r
a
p
h
g
eo
m
etr
y
(
VGG)
,
an
d
Den
s
eNe
t
m
o
d
els,
th
e
E
f
f
icie
n
tNet
-
B
0
m
o
d
el
ex
t
r
ac
ts
th
e
f
ea
tu
r
es
ef
f
icien
tly
with
f
ewe
r
p
ar
am
eter
s
.
Als
o
,
th
is
n
etwo
r
k
ef
f
icien
tly
h
a
n
d
les
th
e
s
p
ec
if
ic
n
o
is
e
in
u
ltra
s
o
u
n
d
im
ag
es,
k
n
o
wn
as
s
p
ec
k
le
n
o
is
e
th
at
m
ak
es th
e
m
o
d
el
m
o
r
e
r
o
b
u
s
t f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
.
3
.
2
.
M
SFF
mo
d
ule (
H
L
F
mo
du
le)
T
h
en
,
t
h
e
f
ea
tu
r
es
ex
tr
ac
ted
f
r
o
m
th
e
E
f
f
icien
tNet
-
B
0
m
o
d
el
ar
e
f
e
d
to
th
e
MSFF
m
o
d
u
le,
wh
ich
p
r
im
ar
ily
in
v
o
lv
es
o
f
2
co
m
p
o
n
en
ts
s
u
ch
as
th
e
co
r
r
ec
tio
n
m
ap
f
ee
d
b
ac
k
m
o
d
u
le
an
d
H
L
F
m
o
d
u
le
.
I
n
th
is
m
o
d
u
le,
t
h
er
e
ar
e
th
r
ee
HL
F
m
o
d
u
les
u
tili
ze
d
f
o
r
g
en
er
a
tin
g
f
o
u
r
d
if
f
er
en
t
s
ca
le
o
f
f
ea
tu
r
es,
wh
ich
ar
e
r
ep
r
esen
ted
as
F1
,
F2
,
F3
,
an
d
F4
,
r
esp
ec
tiv
ely
.
T
h
ese
f
ea
t
u
r
es
ar
e
u
s
ed
as
in
p
u
ts
to
g
en
er
ate
a
f
in
al
f
ea
tu
r
e
m
o
d
if
icatio
n
m
a
p
,
wh
ich
is
d
en
o
ted
as
SF
,
th
en
it
co
m
p
en
s
ates
an
d
g
en
er
ates
f
o
u
r
s
ca
les
o
f
o
u
tp
u
t
m
ap
s
,
wh
ich
ar
e
in
d
icate
d
as
S1
,
S2
,
S3
,
an
d
S4
.
B
esid
es
it
s
m
u
lti
-
s
em
an
tic
in
f
o
r
m
atio
n
f
ea
tu
r
es
an
d
h
ig
h
r
eso
lu
tio
n
,
MSFF m
o
d
u
le
p
r
o
v
id
es p
r
ec
is
e
s
eg
m
en
tatio
n
r
es
u
lts
.
I
n
MSFF,
th
e
HL
F
m
o
d
u
le
is
a
cr
o
s
s
-
f
ea
tu
r
e
m
o
d
u
le,
w
h
ich
co
m
b
in
es
f
ea
tu
r
es
ex
tr
a
cted
f
r
o
m
v
ar
io
u
s
s
ca
les
to
m
itig
ate
b
ac
k
g
r
o
u
n
d
n
o
is
e
b
etwe
en
t
h
e
f
e
atu
r
es.
T
h
is
f
ea
t
u
r
e
f
u
s
io
n
m
o
d
u
le
c
o
m
p
e
n
s
ates
f
o
r
m
is
s
in
g
p
ar
ts
in
th
e
f
ea
tu
r
es
to
en
h
an
ce
th
e
s
eg
m
en
tatio
n
o
f
th
e
f
etal
h
ea
d
ef
f
icien
tly
.
At
f
ir
s
t,
th
e
d
ee
p
f
ea
tu
r
es
d
er
iv
ed
f
r
o
m
th
e
b
ac
k
b
o
n
e
m
o
d
el
ar
e
f
e
d
in
to
th
e
HL
F
m
o
d
u
le,
an
d
a
m
atr
ix
m
u
ltip
licatio
n
o
p
er
atio
n
is
p
e
r
f
o
r
m
ed
to
r
e
m
o
v
e
th
e
b
ac
k
g
r
o
u
n
d
n
o
is
e
p
r
esen
t
in
th
e
f
ea
tu
r
es.
T
h
es
e
f
u
s
ed
f
ea
tu
r
es
ar
e
f
u
r
th
er
u
s
ed
to
c
o
r
r
ec
t
th
e
ac
t
u
al
ex
tr
ac
ted
f
ea
tu
r
es
b
y
th
e
c
o
r
r
ec
tio
n
m
ap
f
ee
d
b
ac
k
m
o
d
u
l
e.
T
h
e
o
u
t
p
u
t o
f
t
h
e
HL
F c
o
n
tain
s
b
o
th
FH a
n
d
FL,
wh
ich
ar
e
m
at
h
em
atica
lly
ex
p
r
ess
ed
in
(
1
)
an
d
(
2
)
.
=
(
(
)
)
+
(
(
(
)
)
)
×
(
(
)
)
(
1
)
=
(
(
)
)
+
(
(
(
)
)
)
×
(
(
)
)
(
2
)
W
h
e
r
e
d
e
n
o
t
e
s
t
h
e
c
o
n
v
o
l
u
t
io
n
a
l
b
l
o
c
k
;
r
e
p
r
e
s
e
n
t
s
t
h
e
u
p
s
a
m
p
l
i
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
b
l
o
ck
.
H
i
g
h
-
l
e
v
e
l
f
e
a
t
u
r
e
i
s
u
p
s
a
m
p
l
e
d
a
n
d
c
o
m
b
i
n
e
d
b
y
m
a
t
r
i
x
m
u
l
t
ip
l
i
c
a
t
i
o
n
.
A
f
t
e
r
a
c
o
n
v
o
l
u
t
io
n
o
p
e
r
a
t
i
o
n
,
t
h
e
c
o
n
v
o
l
u
t
i
o
n
a
l
b
l
o
c
k
,
w
h
i
c
h
c
o
m
p
r
i
s
e
s
o
f
c
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
e
r
s
,
b
a
t
c
h
n
o
r
m
a
l
i
z
a
ti
o
n
,
a
n
d
t
h
e
r
e
c
t
i
f
i
e
d
l
i
n
e
a
r
u
n
i
t
(
R
e
L
U
)
a
c
t
i
v
at
i
o
n
f
u
n
c
t
i
o
n
.
S
im
i
l
a
r
l
y
,
t
h
e
l
o
w
-
le
v
e
l
f
e
at
u
r
e
is
c
o
m
b
i
n
e
d
t
h
r
o
u
g
h
t
h
e
m
a
t
r
i
x
m
u
l
t
i
p
li
c
a
t
i
o
n
i
n
t
h
e
c
o
n
v
o
l
u
t
i
o
n
a
l
b
l
o
c
k
.
B
o
t
h
a
n
d
i
n
t
h
e
H
L
F
m
o
d
u
l
e
c
o
n
ta
in
s
a
s
i
m
i
l
a
r
n
u
m
b
e
r
o
f
c
o
n
v
o
l
u
t
i
o
n
a
l
la
y
e
r
s
,
p
o
o
l
i
n
g
l
a
y
e
r
s
,
a
n
d
a
c
t
i
v
a
t
i
o
n
l
a
y
e
r
s
.
T
h
i
s
H
L
F
m
o
d
u
l
e
u
t
i
l
i
ze
s
t
h
e
f
e
a
t
u
r
e
s
t
o
p
r
o
v
i
d
e
r
i
c
h
i
n
f
o
r
m
a
t
i
o
n
f
r
o
m
l
o
w
l
e
v
e
l
f
e
a
t
u
r
e
s
t
o
t
h
e
a
n
d
l
e
s
s
b
a
c
k
g
r
o
u
n
d
n
o
i
s
e
f
r
o
m
t
o
t
h
e
f
e
a
t
u
r
e
s
,
r
e
s
p
e
c
ti
v
e
l
y
.
T
h
i
s
M
S
FF
m
o
d
u
l
e
u
t
i
l
i
z
es
s
e
v
e
r
a
l
H
L
F
m
o
d
u
l
e
s
t
o
g
r
a
d
u
a
l
l
y
i
m
p
r
o
v
e
t
h
e
o
p
t
i
m
a
l
f
e
a
t
u
r
es
a
t
v
a
r
i
o
u
s
l
e
v
e
ls
t
o
m
a
k
e
t
h
e
f
i
n
a
l
f
e
a
t
u
r
e
m
o
d
i
f
i
c
a
t
i
o
n
m
a
p
.
C
o
r
r
e
c
t
i
o
n
m
a
p
f
e
e
d
b
a
c
k
i
n
t
h
e
M
S
F
F
m
o
d
u
l
e
i
s
m
a
t
h
e
m
a
ti
c
a
l
l
y
f
o
r
m
u
l
a
t
e
d
a
s
i
n
(
3
)
.
W
h
e
r
e
d
e
n
o
t
e
s
t
h
e
o
u
t
p
u
t
c
o
r
r
e
c
t
e
d
a
n
d
r
e
f
i
n
e
f
ea
t
u
r
e
m
a
p
;
r
e
p
r
e
s
e
n
t
s
e
x
t
r
a
ct
ed
f
e
a
t
u
r
e
s
f
r
o
m
E
f
f
i
c
i
e
n
t
Ne
t
-
B
0
w
h
e
r
e
=
1
,
2
,
3
4
;
i
n
d
ic
a
t
e
s
t
h
e
f
i
n
a
l
f
e
a
t
u
r
e
m
o
d
i
f
ic
a
t
i
o
n
m
a
p
.
(
.
)
r
e
p
r
e
s
e
n
ts
d
o
w
n
s
a
m
p
l
i
n
g
o
f
t
o
t
h
e
s
a
m
e
s
i
z
e;
+
i
n
d
i
c
a
t
e
s
m
at
r
i
x
a
d
d
i
ti
o
n
o
p
e
r
a
t
o
r
.
=
+
(
)
(
3
)
3
.
3
.
G
l
o
ba
l f
ea
t
ure
m
o
du
le
I
n
th
e
GFM,
two
o
p
er
ato
r
s
ar
e
u
tili
ze
d
to
f
o
cu
s
o
n
f
etal
h
e
ad
b
o
u
n
d
a
r
y
in
f
o
r
m
atio
n
b
y
i
n
teg
r
atin
g
g
lo
b
al
f
ea
tu
r
es
f
r
o
m
th
e
g
en
er
ated
m
u
lti
-
s
ca
le
f
ea
t
u
r
e
m
ap
s
.
Op
er
ato
r
s
lik
e
elem
en
t
-
by
-
elem
e
n
t
m
u
ltip
licatio
n
an
d
co
n
ca
ten
ati
o
n
f
u
n
ctio
n
o
f
m
atr
i
x
s
p
atial
d
im
en
s
io
n
al
s
p
licin
g
ar
e
t
h
e
f
u
s
io
n
o
p
e
r
ato
r
s
u
s
ed
to
f
u
s
e
g
lo
b
al
f
ea
tu
r
es.
I
n
itial
ly
,
in
th
e
GFM
,
th
e
ex
tr
ac
ted
f
ea
tu
r
es,
s
u
ch
as
F1
,
F2
,
F3
,
an
d
F4
at
v
a
r
io
u
s
s
ca
les,
ar
e
u
p
s
am
p
led
to
a
u
n
if
o
r
m
s
ize,
a
n
d
th
e
n
th
ese
r
esized
f
ea
tu
r
es
ar
e
f
u
s
ed
b
y
th
e
m
u
ltip
licatio
n
o
p
er
ato
r
.
Sin
ce
th
e
n
etwo
r
k
la
y
er
s
in
th
e
b
ac
k
b
o
n
e
n
etwo
r
k
(
E
f
f
icien
tNet
-
B
0
)
in
c
r
ea
s
es
at
last
f
o
r
ex
tr
ac
tin
g
h
ig
h
lev
el
f
ea
tu
r
es,
th
e
e
x
tr
ac
tio
n
is
g
r
ad
u
ally
ch
an
g
es
f
r
o
m
lo
w
-
lev
el
to
h
ig
h
lev
el
s
e
m
an
tic
in
f
o
r
m
atio
n
.
Simu
ltan
eo
u
s
ly
,
elem
en
t
-
by
-
e
lem
en
t
m
u
ltip
licatio
n
o
p
e
r
ato
r
m
ak
es
th
e
lo
w
-
lev
el
f
ea
tu
r
e
s
to
f
o
cu
s
o
n
h
ig
h
lev
el
f
ea
tu
r
es
with
s
em
an
tic
i
n
f
o
r
m
atio
n
as
th
e
tar
g
et,
ef
f
i
cien
tly
m
itig
atin
g
b
ac
k
g
r
o
u
n
d
in
ter
f
e
r
en
ce
.
T
h
e
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
841
-
8
5
1
846
f
in
al
g
lo
b
al
in
f
o
r
m
atio
n
g
r
ap
h
is
o
b
tain
ed
b
y
u
p
s
am
p
lin
g
th
e
f
u
s
ed
f
ea
tu
r
es,
wh
ich
ar
e
th
en
s
p
liced
i
n
s
p
atial
ch
an
n
el
d
im
e
n
s
io
n
th
at
h
elp
s
to
av
o
id
lo
s
s
o
f
f
i
n
e
-
g
r
ain
e
d
d
etails
af
ter
th
e
elem
en
t
-
by
-
elem
e
n
t
m
u
ltip
licatio
n
o
p
e
r
ato
r
.
3
.
4
.
Ada
ptiv
e
f
ea
t
ure
f
us
io
n m
o
du
le
A
n
AFFM
is
u
s
ed
to
co
m
b
i
n
e
g
lo
b
al
an
d
r
ef
in
ed
f
ea
tu
r
e
m
ap
s
ad
ap
tiv
ely
to
p
r
o
d
u
c
e
th
e
f
in
al
s
eg
m
en
tatio
n
d
etec
tio
n
r
esu
lts
.
Glo
b
al
f
ea
tu
r
e
m
ap
an
d
f
in
al
r
ef
in
ed
f
ea
tu
r
e
m
ap
f
r
o
m
MSFF
m
o
d
u
le
ar
e
f
ed
in
t
o
th
e
AFFM
f
o
r
f
ea
tu
r
e
f
u
s
io
n
.
T
h
e
AFF
m
o
d
u
le
u
tili
ze
s
lear
n
ab
le
weig
h
ts
,
s
u
ch
as
an
d
,
to
f
u
s
e
th
e
two
f
ea
tu
r
e
m
ap
s
,
g
e
n
er
atin
g
an
ac
cu
r
ate
s
eg
m
en
tatio
n
m
ap
.
T
h
is
d
y
n
am
ic
f
u
s
io
n
ad
ju
s
ts
th
e
weig
h
t
o
f
ea
ch
m
a
p
ac
co
r
d
in
g
t
o
th
eir
r
elev
an
ce
to
t
h
e
s
eg
m
en
ta
tio
n
task
.
T
h
e
s
eg
m
en
tatio
n
o
f
th
e
f
etal
h
ea
d
is
p
er
f
o
r
m
ed
in
t
h
is
AFF
m
o
d
u
le
af
ter
f
u
s
in
g
th
e
two
d
if
f
er
e
n
t
s
et
o
f
f
ea
tu
r
es.
T
h
e
m
ath
e
m
a
tical
ex
p
r
ess
io
n
f
o
r
th
e
f
in
al
s
eg
m
en
tatio
n
p
r
e
d
ictio
n
is
g
iv
en
i
n
(
4
)
.
=
×
+
×
(
4
)
Af
ter
p
er
f
o
r
m
i
n
g
f
ea
tu
r
e
ex
tr
a
ctio
n
u
s
in
g
E
f
f
icien
tNet
-
B
0
,
f
ea
tu
r
e
f
u
s
io
n
o
f
lo
w
lev
el
an
d
h
ig
h
-
lev
el
f
ea
tu
r
e
s
in
th
e
MSFF
m
o
d
u
le,
an
d
g
l
o
b
al
f
ea
t
u
r
es
f
u
s
io
n
in
th
e
A
FF
,
th
e
m
o
d
el
p
r
o
v
i
d
es
a
r
ef
i
n
ed
f
ea
tu
r
e
m
ap
at
v
ar
io
u
s
s
ca
les.
T
h
e
f
in
al
s
eg
m
en
tatio
n
p
r
ed
ictio
n
is
m
ad
e
b
y
co
m
b
in
in
g
an
d
th
r
o
u
g
h
a
d
ap
tiv
e
f
u
s
io
n
.
T
h
is
co
m
b
in
ed
m
ap
r
ep
r
esen
ts
th
e
s
eg
m
en
ted
r
eg
io
n
s
co
r
r
e
s
p
o
n
d
in
g
to
th
e
f
etal
h
ea
d
.
T
h
e
weig
h
ts
ω
g
an
d
ω
F a
r
e
d
y
n
a
m
ically
s
tu
d
ied
to
o
p
tim
ize
th
e
co
n
tr
ib
u
tio
n
o
f
e
ac
h
m
ap
to
war
d
s
th
e
f
in
al
s
eg
m
en
tatio
n
r
esu
lt.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
E
x
p
e
r
i
m
e
n
t
a
l
r
e
s
u
lt
s
o
f
t
h
e
p
r
o
p
o
s
e
d
F
G
A
-
N
et
-
b
a
s
e
d
F
H
S
m
e
t
h
o
d
u
t
i
l
i
z
i
n
g
t
w
o
b
e
n
c
h
m
ar
k
d
a
t
a
s
e
ts
a
r
e
i
l
l
u
s
t
r
at
e
d
i
n
t
h
i
s
s
e
ct
i
o
n
.
T
h
e
p
r
o
p
o
s
e
d
s
e
g
m
e
n
t
at
i
o
n
m
o
d
e
l
is
s
i
m
u
l
a
te
d
u
s
i
n
g
P
y
t
h
o
n
3
.
9
s
o
f
t
w
a
r
e
w
it
h
t
h
e
s
y
s
te
m
c
o
n
f
i
g
u
r
a
t
i
o
n
t
o
o
l
o
n
W
i
n
d
o
ws
1
0
,
1
6
G
B
R
A
M
,
an
d
a
n
i
5
p
r
o
c
e
s
s
o
r
.
P
e
r
f
o
r
m
a
n
c
e
e
v
a
l
u
a
t
i
o
n
o
f
t
h
e
F
G
A
-
Ne
t
m
o
d
e
l
is
e
v
a
l
u
a
te
d
a
g
a
i
n
s
t
s
t
a
t
e
-
of
-
t
h
e
-
a
r
t
a
p
p
r
o
a
c
h
e
s
,
a
s
r
e
p
r
es
e
n
t
e
d
i
n
s
e
c
ti
o
n
4
.
1
.
T
h
e
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
is
o
f
t
h
e
p
r
o
p
o
s
e
d
s
e
g
m
e
n
t
a
t
i
o
n
m
et
h
o
d
w
it
h
e
x
is
t
i
n
g
F
H
S
m
e
t
h
o
d
s
is
il
l
u
s
t
r
a
t
e
d
i
n
s
e
c
ti
o
n
4
.
2
.
T
a
b
l
e
3
i
l
l
u
s
t
r
a
t
es
t
h
e
p
a
r
a
m
et
e
r
s
e
t
ti
n
g
s
o
f
F
GA
-
N
e
t
m
o
d
e
l
a
n
d
E
f
f
i
c
ie
n
t
N
e
t
b
as
e
d
b
a
c
k
b
o
n
e
n
e
t
w
o
r
k
.
T
ab
le
3
.
Par
am
eter
s
ettin
g
s
o
f
th
e
p
r
o
p
o
s
ed
FGA
-
Net
m
o
d
el
M
e
t
h
o
d
s
P
a
r
a
me
t
e
r
s
V
a
l
u
e
s
Ef
f
i
c
i
e
n
t
N
e
t
-
B0
B
a
t
c
h
si
z
e
32
Ep
o
c
h
s
1
0
0
I
n
i
t
i
a
l
l
e
a
r
n
i
n
g
r
a
t
e
0
.
0
0
1
O
p
t
i
mi
z
e
r
A
d
a
m
Lo
ss f
u
n
c
t
i
o
n
B
i
n
a
r
y
c
r
o
ss
-
e
n
t
r
o
p
y
F
G
A
-
N
e
t
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
0
0
1
O
p
t
i
mi
z
e
r
A
d
a
m
S
t
e
p
S
i
z
e
30
G
a
mm
a
0
.
1
Lo
ss f
u
n
c
t
i
o
n
W
e
i
g
h
t
e
d
b
i
n
a
r
y
c
r
o
ss
e
n
t
r
o
p
y
E
v
a
l
u
a
t
i
o
n
m
e
t
r
ic
s
s
u
c
h
a
s
d
ic
e
c
o
e
f
f
i
c
i
e
n
t
(
D
C
)
,
h
a
n
d
o
f
f
d
i
s
t
a
n
c
e
(
H
D
)
,
a
n
d
m
e
a
n
i
n
t
er
s
e
c
t
i
o
n
o
f
u
n
i
o
n
(
m
I
o
U
)
a
r
e
u
s
e
d
f
o
r
e
v
a
l
u
a
ti
n
g
e
f
f
e
c
ti
v
e
n
e
s
s
o
f
F
GA
-
N
e
t
a
p
p
r
o
a
c
h
,
w
h
ic
h
a
r
e
m
at
h
e
m
a
t
i
ca
l
l
y
r
e
p
r
e
s
e
n
t
e
d
b
y
(
5
)
a
n
d
(
6
)
.
W
h
e
r
e
r
e
p
r
e
s
e
n
t
s
p
r
e
d
i
ct
e
d
r
e
s
u
l
t
,
d
e
n
o
t
e
s
g
r
o
u
n
d
t
r
u
t
h
;
|
⋅
|
in
d
i
c
a
t
e
s
n
u
m
b
e
r
o
f
p
i
x
e
l
s
.
W
h
e
r
e
d
e
n
o
t
e
s
s
u
b
s
e
t
o
f
o
r
,
a
n
d
(
,
)
i
n
d
i
c
a
t
es
d
i
s
t
a
n
ce
b
e
t
w
e
e
n
p
o
i
n
t
a
n
d
p
o
i
n
t
.
=
2
×
|
∩
|
|
|
+
|
|
(
5
)
=
(
(
,
)
,
,
(
,
)
)
(
6
)
T
h
e
mIoU
ev
al
u
ates
th
e
av
er
ag
e
I
o
U
ac
r
o
s
s
s
ev
er
al
s
eg
m
en
ted
ar
ea
s
,
s
u
ch
as
th
e
f
etal
h
ea
d
an
d
b
ac
k
g
r
o
u
n
d
m
ea
s
u
r
em
e
n
ts
.
T
h
is
p
r
o
d
u
ce
s
an
o
v
er
all
e
s
tim
atio
n
f
o
r
s
eg
m
e
n
tatio
n
p
er
f
o
r
m
ed
b
y
th
e
au
to
m
ated
s
eg
m
en
tatio
n
alg
o
r
ith
m
s
th
at
alig
n
s
with
th
e
g
r
o
u
n
d
tr
u
th
a
n
n
o
tatio
n
s
.
A
h
ig
h
er
m
I
o
U
v
alu
e
r
ep
r
esen
ts
b
etter
s
eg
m
en
tatio
n
p
er
f
o
r
m
an
ce
.
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
A
d
a
p
tive
fea
tu
r
e
fu
s
io
n
n
etw
o
r
k
fo
r
feta
l h
ea
d
s
eg
men
ta
tio
n
in
u
ltr
a
s
o
u
n
d
ima
g
e
s
(
V
ima
la
N
a
g
a
b
o
tu
)
847
4
.
1
.
P
er
f
o
r
m
a
nce
an
a
ly
s
is
Qu
an
titativ
e
an
d
q
u
alitativ
e
a
n
aly
s
is
o
f
FGA
-
Net
s
eg
m
en
tatio
n
m
eth
o
d
,
u
s
in
g
two
p
u
b
licl
y
av
ailab
le
d
atasets
s
u
ch
as th
e
F
H
-
PS
-
A
o
P a
n
d
th
e
HC
-
1
8
d
atasets
,
ar
e
d
ep
icted
in
th
is
s
ec
t
io
n
.
Stat
e
-
of
-
th
e
-
ar
t m
eth
o
d
s
s
u
ch
as
UNe
t
-
b
ased
s
eg
m
en
tatio
n
m
o
d
els,
T
r
an
s
f
o
r
m
er
b
ased
m
o
d
els,
en
co
d
e
r
-
d
ec
o
d
er
-
b
ased
m
o
d
els,
th
r
esh
o
ld
-
b
ased
m
o
d
els,
r
e
g
io
n
-
b
ased
s
eg
m
en
tatio
n
m
o
d
els,
an
d
clu
s
ter
-
b
ased
m
o
d
els
ar
e
u
s
ed
f
o
r
ev
alu
atio
n
.
I
n
th
is
s
ec
tio
n
,
t
h
e
r
ec
e
n
t
tr
an
s
f
o
r
m
er
-
b
ased
s
eg
m
en
tatio
n
m
o
d
els
ar
e
a
ls
o
co
n
s
id
er
ed
f
o
r
p
er
f
o
r
m
an
ce
e
v
alu
atio
n
with
p
r
o
p
o
s
ed
m
eth
o
d
s
u
ch
as
s
eg
m
en
tatio
n
tr
an
s
f
o
r
m
er
(
Seg
Fo
r
m
er
)
,
b
id
ir
ec
tio
n
al
to
k
en
tr
a
n
s
f
o
r
m
er
(
B
i
-
Fo
r
m
er
)
an
d
m
o
b
ile
v
is
io
n
t
r
an
s
f
o
r
m
er
(
Mo
b
ileViT
)
r
esp
ec
tiv
ely
.
T
h
e
ab
o
v
e
-
m
en
tio
n
ed
p
er
f
o
r
m
an
ce
m
ea
s
u
r
es
ar
e
u
s
ed
f
o
r
th
e
ev
al
u
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
ag
ai
n
s
t
v
ar
io
u
s
s
eg
m
en
tatio
n
m
eth
o
d
s
.
T
ab
le
4
r
ep
r
esen
ts
th
e
p
er
f
o
r
m
an
c
e
o
f
th
e
FGA
-
Net
m
eth
o
d
u
s
in
g
th
e
FH
-
PS
-
Ao
P
d
ataset.
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
ac
h
iev
ed
a
DC
o
f
9
5
.
7
8
%
an
d
an
HD
o
f
2
.
5
4
m
m
,
wh
ich
r
e
p
r
esen
ts
ac
cu
r
at
e
s
eg
m
en
tatio
n
o
f
th
e
f
etal
h
ea
d
u
s
in
g
u
ltra
s
o
u
n
d
im
a
g
es.
Per
f
o
r
m
an
ce
an
al
y
s
is
o
f
th
e
p
r
o
p
o
s
ed
FGA
-
Net
-
b
ased
FH
S
u
s
in
g
HC
-
1
8
d
ataset
i
s
r
ep
r
esen
ted
in
T
ab
le
5
.
T
h
e
ex
is
tin
g
DL
m
et
h
o
d
s
u
s
ed
f
o
r
s
eg
m
en
tatio
n
,
wh
ich
ar
e
DL
-
b
ased
ap
p
r
o
ac
h
es
s
u
ch
as
2
D
-
UNe
t,
3D
-
UNe
t,
an
d
V
-
Net.
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
ac
h
iev
ed
a
D
C
o
f
9
8
.
9
5
%
an
d
an
HD
o
f
1
.
1
9
m
m
,
wh
ich
r
e
p
r
esen
ts
ac
cu
r
ate
s
eg
m
en
tatio
n
o
f
th
e
f
etal
h
ea
d
u
s
in
g
u
ltra
s
o
u
n
d
im
ag
es.
Per
f
o
r
m
an
ce
an
aly
s
is
o
f
th
e
p
r
o
p
o
s
ed
FGA
-
Net
-
b
ased
FHS
u
s
in
g
th
e
FH
-
PS
-
Ao
P
d
ataset
is
r
ep
r
esen
ted
i
n
T
ab
le
6
.
State
o
f
t
h
e
ar
t
m
eth
o
d
s
u
s
ed
f
o
r
s
eg
m
en
tatio
n
s
u
ch
as
clu
s
ter
-
b
ased
m
eth
o
d
s
,
th
r
esh
o
ld
-
b
ased
m
eth
o
d
s
,
a
n
d
r
eg
io
n
-
b
ased
ap
p
r
o
ac
h
es.
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
ac
h
iev
ed
a
DC
o
f
9
5
.
7
8
%
an
d
an
HD
o
f
2
.
5
4
m
m
,
wh
ich
r
ep
r
esen
ts
ac
cu
r
ate
s
eg
m
en
tatio
n
o
f
th
e
f
etal
h
ea
d
u
s
in
g
u
ltra
s
o
u
n
d
im
ag
es.
P
e
r
f
o
r
m
a
n
c
e
a
n
a
l
y
s
is
o
f
p
r
o
p
o
s
e
d
F
GA
-
Net
-
b
as
e
d
F
HS
u
s
i
n
g
H
C
-
1
8
d
a
t
a
s
et
i
s
r
e
p
r
es
e
n
t
e
d
in
T
a
b
l
e
7
.
State
o
f
th
e
ar
t
ap
p
r
o
ac
h
es
u
s
ed
f
o
r
s
eg
m
en
tatio
n
lik
e
clu
s
ter
-
b
ased
m
eth
o
d
s
,
th
r
esh
o
ld
-
b
ased
ap
p
r
o
ac
h
es,
an
d
r
eg
i
o
n
-
b
ased
m
eth
o
d
s
.
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
ac
h
ie
v
ed
a
DC
o
f
9
8
.
9
5
% a
n
d
an
HD
o
f
1
.
1
9
m
m
,
wh
ich
r
ep
r
esen
ts
ac
cu
r
ate
s
eg
m
en
tatio
n
o
f
f
etal
h
ea
d
u
s
in
g
u
ltra
s
o
u
n
d
im
ag
es.
T
ab
le
4
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
with
DL
-
b
ased
m
eth
o
d
s
o
n
F
H
-
PS
-
Ao
P d
ataset
M
e
t
h
o
d
s
M
e
t
r
i
c
e
s
D
C
(
%)
H
D
(
mm
)
M
I
o
U
(
%)
2D
-
U
N
e
t
9
1
.
7
5
3
.
5
6
9
2
.
8
0
3D
-
U
N
e
t
9
2
.
4
6
3
.
1
8
9
3
.
9
1
V
-
N
e
t
S
e
g
F
o
r
mer
Bi
-
F
o
r
m
e
r
M
o
b
i
l
e
V
i
T
9
3
.
2
4
9
4
.
3
2
9
4
.
6
8
9
3
.
8
5
2
.
8
9
2
.
7
3
2
.
6
3
2
.
7
8
9
5
.
5
6
9
6
.
0
1
9
6
.
2
8
9
5
.
3
4
P
r
o
p
o
se
d
F
G
A
-
N
e
t
me
t
h
o
d
9
5
.
7
8
2
.
5
4
9
6
.
7
2
T
ab
le
5
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
with
DL
-
b
ased
m
eth
o
d
s
o
n
HC
-
1
8
d
ataset
M
e
t
h
o
d
s
M
e
t
r
i
c
e
s
D
C
(
%)
H
D
(
mm
)
M
I
o
U
(
%)
2D
-
U
N
e
t
9
5
.
4
4
2
.
1
1
9
4
.
1
9
3D
-
U
N
e
t
9
5
.
8
6
1
.
8
2
9
6
.
6
3
V
-
N
e
t
S
e
g
F
o
r
mer
Bi
-
F
o
mr
e
r
M
o
b
i
l
e
V
i
T
9
6
.
9
2
9
7
.
4
5
9
7
.
8
3
9
6
.
5
4
1
.
5
7
1
.
4
3
1
.
3
2
1
.
6
1
9
7
.
2
4
9
7
.
6
8
9
8
.
1
2
9
6
.
9
4
P
r
o
p
o
se
d
F
G
A
-
N
e
t
me
t
h
o
d
9
8
.
9
5
1
.
1
9
9
8
.
9
6
T
ab
le
6
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
with
tr
ad
itio
n
al
m
eth
o
d
s
o
n
FH
-
PS
-
Ao
P d
ataset
M
e
t
h
o
d
s
M
e
t
r
i
c
e
s
D
C
(
%)
H
D
(
mm
)
M
I
o
U
(
%)
C
l
u
st
e
r
i
n
g
me
t
h
o
d
s
8
9
.
7
5
3
.
1
5
9
1
.
2
9
Th
r
e
s
h
o
l
d
m
e
t
h
o
d
s
9
0
.
9
4
2
.
9
6
9
2
.
6
4
R
e
g
i
o
n
me
t
h
o
d
s
9
1
.
8
5
2
.
7
3
9
3
.
7
6
P
r
o
p
o
se
d
F
G
A
-
N
e
t
me
t
h
o
d
9
5
.
7
8
2
.
5
4
9
6
.
7
2
T
ab
le
7
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
with
tr
ad
itio
n
al
m
eth
o
d
s
o
n
HC
-
1
8
d
ataset
M
e
t
h
o
d
s
M
e
t
r
i
c
e
s
D
C
(
%)
H
D
(
mm
)
M
I
o
U
(
%)
C
l
u
st
e
r
i
n
g
me
t
h
o
d
s
8
9
.
7
8
2
.
0
3
9
2
.
6
3
Th
r
e
s
h
o
l
d
m
e
t
h
o
d
s
9
3
.
3
7
1
.
9
1
9
3
.
7
1
R
e
g
i
o
n
me
t
h
o
d
s
9
6
.
5
1
1
.
6
4
9
4
.
5
9
P
r
o
p
o
se
d
F
G
A
-
N
e
t
me
t
h
o
d
9
8
.
9
5
1
.
1
9
9
8
.
9
6
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
841
-
8
5
1
848
C
r
o
s
s
-
v
alid
atio
n
is
p
er
f
o
r
m
ed
to
e
v
alu
ate
th
e
p
er
f
o
r
m
a
n
ce
o
f
a
p
r
o
p
o
s
ed
s
eg
m
e
n
tatio
n
m
o
d
el
to
en
s
u
r
e
g
e
n
er
aliza
tio
n
a
b
ilit
y
f
o
r
u
n
s
ee
n
d
ata.
T
ab
le
8
r
ep
r
esen
ts
th
e
cr
o
s
s
-
v
alid
atio
n
r
esu
lts
o
f
p
r
o
p
o
s
ed
m
eth
o
d
u
s
in
g
v
ar
io
u
s
k
-
f
o
ld
s
.
Fro
m
th
e
r
esu
lts
,
th
e
k
=5
f
o
ld
s
co
n
f
ig
u
r
atio
n
o
f
p
r
o
p
o
s
e
d
m
eth
o
d
ac
h
iev
ed
b
etter
r
esu
lts
b
ec
au
s
e
it
p
r
o
v
i
d
es
a
b
alan
ce
d
tr
ad
e
-
o
f
f
b
etwe
en
tr
ain
in
g
d
ata
a
n
d
v
alid
atio
n
r
eliab
ilit
y
.
W
h
en
th
e
d
ata
is
s
p
lit
ted
in
to
k
=2
an
d
k
=3
f
o
ld
s
lead
to
u
n
d
er
f
itti
n
g
d
u
e
to
lim
ited
tr
ain
in
g
d
ata,
wh
ile
h
ig
h
er
f
o
ld
s
s
u
ch
as
k
=
7
a
n
d
k
=9
s
p
lits
r
ed
u
ce
th
e
v
alid
atio
n
s
et
s
ize
a
n
d
r
esu
lts
in
o
v
er
f
itti
n
g
.
Hen
ce
,
k
=5
y
ield
s
m
o
r
e
r
o
b
u
s
t
p
er
f
o
r
m
a
n
ce
wh
ich
r
ef
l
ec
ts
h
ig
h
er
DC
an
d
M
I
o
U
an
d
lo
wer
HD
th
at
e
n
s
u
r
e
th
e
g
e
n
er
aliza
tio
n
ab
ilit
y
f
o
r
u
n
s
e
en
d
ata
u
s
in
g
th
e
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
m
o
d
el.
T
h
e
p
e
r
f
o
r
m
a
n
ce
o
f
p
r
o
p
o
s
e
d
FGA
-
N
et
m
o
d
el
in
f
eta
l se
g
m
e
n
ta
ti
o
n
m
et
h
o
d
is
e
v
al
u
at
ed
b
a
s
ed
o
n
t
h
e
co
m
p
u
tati
o
n
a
l
co
m
p
le
x
i
ty
a
n
d
s
tatis
tic
al
tes
ts
w
h
i
c
h
is
r
e
p
r
ese
n
t
ed
i
n
T
a
b
l
e
9
.
F
r
o
m
co
m
p
u
tati
o
n
a
l
an
d
s
tatis
t
ic
al
a
n
al
y
s
is
i
n
F
HS,
t
h
e
FGA
-
N
et
m
o
d
el
a
ch
ie
v
e
d
l
ess
ti
m
e
b
e
ca
u
s
e
o
f
its
e
f
f
ici
e
n
t
f
ea
t
u
r
e
f
u
s
i
o
n
b
y
AFF
M
t
h
at
r
e
d
u
c
ed
r
ed
u
n
d
an
t
p
r
o
ce
s
s
in
g
an
d
e
n
h
a
n
c
e
d
f
aste
r
co
n
v
e
r
g
e
n
c
e.
M
o
r
e
o
v
e
r
,
t
h
e
p
r
o
p
o
s
ed
s
eg
m
e
n
t
ati
o
n
m
o
d
e
l
a
ch
ie
v
e
d
p
-
v
al
u
e
<0
.
0
5
b
ase
d
o
n
t
h
e
D
C
p
a
r
a
m
e
te
r
b
ec
a
u
s
e
i
t
d
i
r
ec
tl
y
r
e
f
le
cts
h
o
w
w
ell
th
e
p
r
e
d
i
cte
d
s
eg
m
e
n
t
ati
o
n
al
ig
n
s
wit
h
t
h
e
g
r
o
u
n
d
t
r
u
t
h
.
T
h
e
o
b
t
ai
n
e
d
p
-
v
al
u
e
i
n
d
ic
at
es
th
e
p
e
r
f
o
r
m
an
ce
im
p
r
o
v
e
m
e
n
t
o
v
e
r
e
x
is
t
in
g
m
e
th
o
d
s
is
s
ta
tis
ti
ca
ll
y
s
i
g
n
if
i
ca
n
t
.
T
h
e
o
p
ti
m
i
ze
d
a
r
c
h
i
tec
tu
r
e
i
m
p
r
o
v
es
b
o
th
s
p
ee
d
an
d
c
o
n
s
is
te
n
cy
,
le
ad
in
g
t
o
n
a
r
r
o
w
co
n
f
i
d
e
n
c
e
in
te
r
v
als
a
n
d
r
elia
b
l
e
s
e
g
m
en
tat
io
n
p
er
f
o
r
m
a
n
c
e
i
n
FHS
.
T
ab
le
8
.
C
r
o
s
s
v
alid
atio
n
o
f
p
r
o
p
o
s
ed
FGA
-
Net
m
o
d
el
u
s
in
g
d
if
f
er
en
t
k
-
f
o
l
d
s
K
-
f
o
l
d
s
D
a
t
a
s
e
t
D
C
(
%)
H
D
(
mm
)
M
I
o
U
(
%)
k
=
2
FH
-
PS
-
A
o
P
d
a
t
a
se
t
9
4
.
8
3
2
.
6
8
9
5
.
8
9
k
=
3
9
5
.
1
6
2
.
6
1
9
6
.
2
1
k
=
5
9
5
.
7
8
2
.
5
4
9
6
.
7
2
k
=
7
9
5
.
4
3
2
.
5
8
9
6
.
4
8
k
=
9
9
5
.
3
4
2
.
5
9
9
6
.
3
3
k
=
2
HC
-
1
8
d
a
t
a
se
t
9
7
.
9
1
1
.
3
4
9
8
.
1
2
k
=
3
9
8
.
1
7
1
.
2
8
9
8
.
3
5
k
=
5
9
8
.
9
5
1
.
1
9
9
8
.
9
6
k
=
7
9
8
.
6
2
1
.
2
3
9
8
.
7
1
k
=
9
9
8
.
4
5
1
.
2
6
9
8
.
5
2
T
ab
le
9
.
C
o
m
p
u
tatio
n
al
co
m
p
l
ex
ity
an
d
s
tatis
tical
an
aly
s
is
o
f
p
r
o
p
o
s
ed
FGA
-
Net
m
o
d
el
M
e
t
h
o
d
s
C
o
m
p
u
t
a
t
i
o
n
a
l
t
i
m
e
(
s)
p
-
v
a
l
u
e
C
o
n
f
i
d
e
n
c
e
i
n
t
e
r
v
a
l
(
C
I
)
2D
-
U
N
e
t
2
.
3
8
0
.
0
4
1
±
0
.
1
3
2
3D
-
U
N
e
t
2
.
8
5
0
.
0
3
4
±
0
.
1
2
8
V
-
N
e
t
3
.
1
2
0
.
0
2
9
±
0
.
1
1
7
S
e
g
F
o
r
mer
2
.
7
4
0
.
0
2
5
±
0
.
1
0
9
B
i
F
o
r
mer
2
.
5
6
0
.
0
1
9
±
0
.
1
0
3
M
o
b
i
l
e
V
i
T
2
.
6
3
0
.
0
2
1
±
0
.
1
0
7
P
r
o
p
o
se
d
F
G
A
-
N
e
t
2
.
1
2
0
.
0
1
2
±
0
.
0
9
5
4
.
2
.
Co
m
pa
ra
t
iv
e
a
na
ly
s
is
C
o
m
p
ar
ativ
e
ev
alu
atio
n
o
f
FGA
-
Net
b
ased
FHS
m
eth
o
d
with
ex
is
tin
g
v
ar
io
u
s
s
eg
m
en
tatio
n
ap
p
r
o
ac
h
es
b
y
u
s
in
g
two
p
u
b
licly
av
ailab
le
d
atasets
s
u
ch
as
F
H
-
PS
-
Ao
P
an
d
H
C
-
1
8
,
is
d
ep
icted
in
th
is
s
ec
tio
n
.
T
h
e
ex
is
tin
g
f
etal
s
e
g
m
en
tatio
n
m
o
d
els,
s
u
ch
as
I
B
PE
[
1
7
]
an
d
HDR
-
DC
N
N
[
1
8
]
,
ar
e
u
s
ed
f
o
r
co
m
p
ar
ativ
e
a
n
aly
s
is
with
th
e
p
r
o
p
o
s
ed
FHS
m
eth
o
d
.
T
h
e
ab
o
v
e
-
m
e
n
tio
n
ed
p
er
f
o
r
m
a
n
c
e
m
ea
s
u
r
es
ar
e
u
s
ed
f
o
r
ev
alu
atio
n
th
e
p
r
o
p
o
s
ed
m
eth
o
d
ag
ain
s
t
v
ar
io
u
s
s
eg
m
en
tatio
n
m
eth
o
d
s
.
T
ab
le
s
1
0
an
d
1
1
r
ep
r
esen
ts
co
m
p
ar
ativ
e
an
aly
s
is
o
f
FGA
-
Net
m
eth
o
d
with
ex
is
tin
g
s
e
g
m
en
tatio
n
m
o
d
els
u
s
in
g
FH
-
PS
-
Ao
P
an
d
HC
-
18
d
atasets
r
esp
ec
tiv
ely
.
T
a
b
l
e
1
0
.
C
o
m
p
a
r
a
ti
v
e
a
n
a
l
y
s
is
o
f
p
r
o
p
o
s
e
d
F
G
A
-
N
e
t
b
as
e
d
f
e
t
a
l
s
e
g
m
e
n
t
a
ti
o
n
m
e
t
h
o
d
i
n
FH
-
PS
-
A
o
P
d
a
t
as
e
t
M
e
t
h
o
d
s
M
e
t
r
i
c
e
s
D
C
(
%)
H
D
(
mm
)
M
I
o
U
(
%)
I
B
P
E
[
1
7
]
9
0
.
2
0
1
0
.
8
1
-
P
r
o
p
o
se
d
F
G
A
-
N
e
t
me
t
h
o
d
9
5
.
7
8
2
.
5
4
9
6
.
7
2
T
ab
le
1
1
.
C
o
m
p
ar
ativ
e
an
aly
s
is
o
f
p
r
o
p
o
s
ed
FGA
-
Net
b
ased
f
etal
s
eg
m
en
tatio
n
m
eth
o
d
in
HC
-
1
8
d
ataset
M
e
t
h
o
d
s
M
e
t
r
i
c
e
s
D
C
(
%)
H
D
(
mm
)
M
I
o
U
(
%)
I
B
P
E
[
1
7
]
9
4
.
1
8
8
.
1
2
-
HDR
-
D
C
N
N
[
1
8
]
9
8
.
8
6
1
.
2
2
9
8
.
8
7
P
r
o
p
o
se
d
F
G
A
-
N
e
t
me
t
h
o
d
9
8
.
9
5
1
.
1
9
9
8
.
9
6
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
A
d
a
p
tive
fea
tu
r
e
fu
s
io
n
n
etw
o
r
k
fo
r
feta
l h
ea
d
s
eg
men
ta
tio
n
in
u
ltr
a
s
o
u
n
d
ima
g
e
s
(
V
ima
la
N
a
g
a
b
o
tu
)
849
4
.
3
.
Dis
cus
s
io
n
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
ac
h
iev
e
d
b
etter
r
esu
lts
in
FH
S
u
s
in
g
u
ltra
s
o
u
n
d
im
ag
es.
T
h
e
E
f
f
iciNet
-
B
0
as
t
h
e
b
ac
k
b
o
n
e
o
f
th
e
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
m
o
d
el
ex
tr
ac
ts
m
u
lti
-
s
ca
le
f
ea
tu
r
es
at
v
ar
io
u
s
s
ca
les
u
s
in
g
th
e
c
o
m
p
o
u
n
d
s
ca
lin
g
tech
n
i
q
u
e.
T
h
e
AFF
m
o
d
u
le
in
FGA
-
N
et
f
u
s
es
th
e
g
lo
b
al
co
n
tex
t
f
r
o
m
g
l
o
b
al
f
ea
t
u
r
es
a
n
d
f
in
e
d
etails
f
r
o
m
t
h
e
MSFF
m
o
d
u
le,
wh
ic
h
h
elp
s
to
s
eg
m
en
t
th
e
f
etal
h
ea
d
ac
cu
r
ately
,
w
h
ich
v
ar
ies
in
s
h
ap
e,
p
o
s
itio
n
s
,
an
d
s
izes
a
cr
o
s
s
d
is
tin
ct
im
ag
e
r
eso
lu
ti
o
n
s
.
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
r
e
f
in
es
an
d
o
p
tim
izes
t
h
e
f
ea
tu
r
es
iter
ativ
ely
th
at
im
p
r
o
v
ed
th
e
ac
cu
r
ac
y
o
f
s
eg
m
en
tatio
n
p
r
o
ce
s
s
,
esp
ec
ially
in
n
o
is
y
an
d
p
ar
tially
o
cc
lu
d
ed
f
etal
h
ea
d
im
ag
es.
E
x
is
tin
g
m
o
d
els
s
u
ch
as
I
B
PE
[
1
7
]
an
d
HDR
-
DC
NN
[
1
8
]
ac
h
iev
ed
less
an
d
in
e
f
f
icien
t
r
esu
lts
in
FHS
d
u
e
to
lim
itatio
n
s
s
u
ch
as
v
ar
iatio
n
s
o
f
s
h
a
p
es
an
d
s
ize
o
f
th
e
f
etal
h
ea
d
,
o
cc
l
u
s
io
n
,
n
o
i
s
e,
an
d
p
o
o
r
im
a
g
e
q
u
ality
th
a
t
af
f
ec
t
in
ac
cu
r
ate
s
eg
m
en
tatio
n
r
esu
lts
.
Ho
wev
er
,
th
e
p
r
o
p
o
s
ed
FGA
-
Net
is
a
f
ea
tu
r
e
f
u
s
io
n
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
th
a
t
in
teg
r
ates
b
o
th
lo
w
-
lev
el
f
ea
tu
r
es
an
d
h
ig
h
-
lev
el
f
ea
tu
r
es
f
r
o
m
u
ltra
s
o
u
n
d
im
ag
es,
wh
ic
h
h
elp
s
it
ad
ap
t
to
v
ar
io
u
s
s
izes
an
d
s
h
a
p
es
o
f
t
h
e
f
etal
h
ea
d
,
wh
ich
r
esu
lts
in
ac
cu
r
ate
s
eg
m
en
tatio
n
.
Als
o
,
th
e
b
r
ig
h
tn
ess
an
d
co
n
tr
ast
en
h
an
ce
m
en
t
in
th
e
d
ata
au
g
m
e
n
tatio
n
tech
n
i
q
u
e
im
p
r
o
v
es
th
e
q
u
ality
o
f
u
ltra
s
o
u
n
d
im
a
g
es
ef
f
icien
tly
,
wh
ich
h
elp
s
in
p
r
e
cise
FHS
.
5.
CO
NCLU
SI
O
N
An
ac
cu
r
ate
FHS
m
o
d
el
is
ess
en
tial
to
ass
e
s
s
an
d
m
o
n
i
to
r
th
e
d
e
v
elo
p
m
e
n
t
o
f
f
etal
f
r
o
m
t
h
e
u
ltra
s
o
u
n
d
im
ag
es.
H
o
wev
er
,
th
e
e
x
is
tin
g
s
eg
m
e
n
tatio
n
m
eth
o
d
s
ar
e
f
ailed
to
ac
cu
r
atel
y
s
eg
m
en
t
th
e
f
etal
h
ea
d
d
u
e
to
v
ar
iatio
n
s
i
n
f
etal
h
ea
d
s
h
ap
e,
o
r
ie
n
tatio
n
,
g
estatio
n
al
ag
e,
an
d
im
ag
e
q
u
ality
.
T
o
o
v
e
r
co
m
e
th
ese
ch
allen
g
es,
FGA
-
Net
b
ased
s
eg
m
en
tatio
n
m
o
d
el
is
p
r
o
p
o
s
ed
in
th
is
r
esear
ch
f
o
r
FH
S
u
s
in
g
u
ltra
s
o
u
n
d
im
ag
es.
T
h
e
p
r
o
p
o
s
ed
FGA
-
Net
m
o
d
el
in
v
o
lv
es
a
f
ea
tu
r
e
f
ee
d
b
ac
k
m
ec
h
an
is
m
to
im
p
r
o
v
e
th
e
b
o
u
n
d
a
r
y
r
ef
in
em
en
t
o
f
f
etal
h
ea
d
b
y
r
eu
s
in
g
th
e
e
x
tr
ac
ted
h
i
g
h
-
le
v
el
co
n
tex
tu
al
i
n
f
o
r
m
atio
n
w
h
ich
en
h
a
n
ce
d
th
e
s
eg
m
en
tatio
n
ac
cu
r
ac
y
f
o
r
d
i
f
f
er
en
t
a
n
ato
m
ical
v
ar
iatio
n
s
.
Sp
ec
if
ically
,
th
e
AFF
m
o
d
u
le
in
th
e
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
m
o
d
el
d
y
n
am
ica
lly
in
teg
r
ates th
e
m
u
lti
-
s
ca
le
f
ea
tu
r
es,
to
en
s
u
r
e
f
in
e
g
r
ain
e
d
d
etails to
th
e
F
HS.
Hen
ce
,
th
e
r
eliab
ilit
y
o
f
p
r
o
p
o
s
ed
FGA
-
Net
m
o
d
el,
wh
ich
co
n
tr
ib
u
tes
f
o
r
ef
f
ec
tiv
e
an
d
ac
cu
r
ate
FHS.
E
x
p
er
im
en
tal
r
esu
lts
o
f
th
e
p
r
o
p
o
s
ed
FGA
-
Net
ar
e
e
v
alu
ate
d
u
s
in
g
DC
an
d
HD
m
etr
ics,
wh
ich
s
h
o
w
b
etter
r
esu
lts
th
an
ex
is
tin
g
s
eg
m
en
tatio
n
ap
p
r
o
ac
h
es
lik
e
I
B
PE.
Ho
wev
er
,
th
e
E
f
f
icien
tNet
-
B
0
b
ased
b
ac
k
b
o
n
e
n
etwo
r
k
with
d
o
wn
s
am
p
lin
g
lay
er
s
r
ed
u
ce
th
e
s
ize
o
f
th
e
f
ea
tu
r
e
m
ap
s
,
wh
ic
h
h
el
p
s
th
e
m
o
d
el
ca
p
tu
r
e
h
ig
h
-
lev
el
i
n
f
o
r
m
atio
n
b
u
t
als
o
ca
u
s
es
a
lo
s
s
o
f
ce
r
tain
f
in
e
d
etails.
T
h
o
u
g
h
t
h
e
u
p
s
am
p
lin
g
is
u
s
ed
later
to
r
ec
o
v
er
t
h
e
o
r
ig
i
n
al
im
ag
e
s
iz
e,
s
o
m
e
o
f
t
h
e
d
etailed
b
o
u
n
d
ar
y
in
f
o
r
m
atio
n
a
r
e
lo
s
t
in
th
e
p
r
o
ce
s
s
,
th
at
lead
s
to
im
p
r
ec
is
e
s
eg
m
en
tatio
n
e
d
g
es.
I
n
th
e
f
u
tu
r
e,
a
d
v
an
ce
d
DL
-
b
ased
b
ac
k
b
o
n
e
n
etwo
r
k
an
d
im
p
r
o
v
e
d
s
eg
m
en
tatio
n
m
eth
o
d
will b
e
u
s
ed
to
en
h
a
n
ce
FHS.
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
Vim
ala
Nag
ab
o
tu
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Pav
an
Ku
m
ar
R
ed
d
y
San
a
✓
✓
✓
✓
✓
✓
✓
B
.
L
ak
s
h
m
i Bh
av
an
i
✓
✓
✓
✓
✓
✓
✓
✓
Do
n
ap
ati
Srik
an
th
✓
✓
✓
✓
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.
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
841
-
8
5
1
850
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t th
e
f
in
d
i
n
g
s
o
f
th
is
s
tu
d
y
a
r
e
av
ailab
le
f
r
o
m
:
‒
T
h
e
co
r
r
esp
o
n
d
i
n
g
au
t
h
o
r
[
VN]
.
‒
Op
en
ly
a
v
ailab
le
in
[
Kag
g
le
]
at
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
a
tasets
/th
an
h
b
n
h
p
h
a
n
/h
c1
8
-
g
r
a
n
d
-
ch
allen
g
e]
,
r
ef
er
en
ce
[
2
3
]
RE
F
E
R
E
NC
E
S
[
1
]
A
.
A
.
H
e
k
a
l
,
H
.
M
.
A
mer,
H
.
E.
-
D
.
M
o
u
s
t
a
f
a
,
a
n
d
A
.
El
n
a
k
i
b
,
“
A
u
t
o
m
a
t
i
c
mea
s
u
r
e
me
n
t
o
f
h
e
a
d
c
i
r
c
u
mf
e
r
e
n
c
e
i
n
f
e
t
a
l
u
l
t
r
a
so
u
n
d
i
ma
g
e
s
u
s
i
n
g
a
sq
u
e
e
z
e
a
t
r
o
u
s
p
o
o
l
i
n
g
U
N
e
t
,
”
Bi
o
m
e
d
i
c
a
l
S
i
g
n
a
l
P
ro
c
e
ssi
n
g
a
n
d
C
o
n
t
ro
l
,
v
o
l
.
1
0
3
,
M
a
y
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
b
s
p
c
.
2
0
2
4
.
1
0
7
4
3
4
.
[
2
]
Q
.
W
a
n
g
,
D
.
Z
h
a
o
,
H
.
M
a
,
X
.
Y
a
n
g
,
a
n
d
B
.
Li
u
,
“
I
n
t
e
g
r
a
t
e
d
g
e
n
e
r
a
t
i
o
n
a
d
v
e
r
sari
a
l
a
n
d
s
e
mi
-
s
u
p
e
r
v
i
se
d
n
e
t
w
o
r
k
f
o
r
c
o
r
p
u
s
c
a
l
l
o
s
u
m
a
n
d
c
a
v
u
m
s
e
p
t
u
m
p
e
l
l
u
c
i
d
u
m
c
o
mp
l
e
x
se
g
me
n
t
a
t
i
o
n
i
n
f
e
t
a
l
b
r
a
i
n
u
l
t
r
a
s
o
u
n
d
v
i
a
p
r
o
g
r
e
ssi
v
e
t
r
a
i
n
i
n
g
,
”
A
p
p
l
i
e
d
S
o
f
t
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
7
1
,
M
a
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
s
o
c
.
2
0
2
5
.
1
1
2
7
6
7
.
[
3
]
A
.
U
u
s
e
t
a
l
.
,
“
S
c
a
n
n
e
r
-
b
a
se
d
r
e
a
l
-
t
i
me
t
h
r
e
e
-
d
i
me
n
si
o
n
a
l
b
r
a
i
n
+
b
o
d
y
s
l
i
c
e
-
to
-
v
o
l
u
m
e
r
e
c
o
n
st
r
u
c
t
i
o
n
f
o
r
T
2
-
w
e
i
g
h
t
e
d
0
.
5
5
-
T
l
o
w
-
f
i
e
l
d
f
e
t
a
l
ma
g
n
e
t
i
c
r
e
s
o
n
a
n
c
e
i
mag
i
n
g
,
”
P
e
d
i
a
t
r
i
c
Ra
d
i
o
l
o
g
y
,
v
o
l
.
5
5
,
n
o
.
3
,
p
p
.
5
5
6
–
5
6
9
,
J
a
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
2
4
7
-
0
2
5
-
0
6
1
6
5
-
x.
[
4
]
C
.
C
a
l
i
x
t
o
e
t
a
l
.
,
“
W
h
i
t
e
m
a
t
t
e
r
t
r
a
c
t
c
r
o
ss
i
n
g
a
n
d
b
o
t
t
l
e
n
e
c
k
r
e
g
i
o
n
s
i
n
t
h
e
f
e
t
a
l
b
r
a
i
n
,
”
H
u
m
a
n
Bra
i
n
M
a
p
p
i
n
g
,
v
o
l
.
4
6
,
n
o
.
1
,
Jan
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
2
/
h
b
m.
7
0
1
3
2
.
[
5
]
B
.
S
.
-
F
a
d
i
d
a
e
t
a
l
.
,
“
D
e
e
p
l
e
a
r
n
i
n
g
–
b
a
s
e
d
se
g
me
n
t
a
t
i
o
n
o
f
w
h
o
l
e
-
b
o
d
y
f
e
t
a
l
M
R
I
a
n
d
f
e
t
a
l
w
e
i
g
h
t
e
st
i
m
a
t
i
o
n
:
a
ssess
i
n
g
p
e
r
f
o
r
m
a
n
c
e
,
r
e
p
e
a
t
a
b
i
l
i
t
y
,
a
n
d
r
e
p
r
o
d
u
c
i
b
i
l
i
t
y
,
”
E
u
r
o
p
e
a
n
R
a
d
i
o
l
o
g
y
,
v
o
l
.
3
4
,
n
o
.
3
,
p
p
.
2
0
7
2
–
2
0
8
3
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
3
3
0
-
023
-
1
0
0
3
8
-
y.
[
6
]
L.
Li
u
,
D
.
Ta
n
g
,
X
.
Li
,
a
n
d
Y
.
O
u
y
a
n
g
,
“
A
u
t
o
ma
t
i
c
f
e
t
a
l
u
l
t
r
a
s
o
u
n
d
i
m
a
g
e
seg
m
e
n
t
a
t
i
o
n
o
f
f
i
r
st
t
r
i
m
e
st
e
r
f
o
r
m
e
a
su
r
i
n
g
b
i
o
m
e
t
r
i
c
p
a
r
a
m
e
t
e
r
s
b
a
se
d
o
n
d
e
e
p
l
e
a
r
n
i
n
g
,
”
M
u
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
3
,
n
o
.
9
,
p
p
.
2
7
2
8
3
–
2
7
3
0
4
,
A
u
g
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
023
-
1
6
5
6
5
-
6.
[
7
]
B
.
Jafr
a
s
t
e
h
e
t
a
l
.
,
“
M
G
A
-
N
e
t
:
A
n
o
v
e
l
mas
k
-
g
u
i
d
e
d
a
t
t
e
n
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
p
r
e
c
i
si
o
n
n
e
o
n
a
t
a
l
b
r
a
i
n
i
ma
g
i
n
g
,
”
N
e
u
r
o
I
m
a
g
e
,
v
o
l
.
3
0
0
,
O
c
t
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
r
o
i
ma
g
e
.
2
0
2
4
.
1
2
0
8
7
2
.
[
8
]
L
.
F
i
d
o
n
e
t
a
l
.
,
“
A
d
e
m
p
s
t
e
r
-
S
h
a
f
e
r
a
p
p
r
o
a
c
h
t
o
t
r
u
s
t
w
o
r
t
h
y
A
I
w
i
t
h
a
p
p
l
i
c
a
t
i
o
n
t
o
f
e
t
a
l
b
r
a
i
n
M
R
I
S
e
g
m
e
n
t
a
t
i
o
n
,
”
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
A
n
a
l
y
s
i
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
4
6
,
n
o
.
5
,
p
p
.
3
7
8
4
–
3
7
9
5
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
T
P
A
M
I
.
2
0
2
3
.
3
3
4
6
3
3
0
.
[
9
]
R
.
F
a
g
h
i
h
p
i
r
a
y
e
s
h
,
D
.
K
a
r
i
m
i
,
D
.
Er
d
o
ğ
m
u
ş,
a
n
d
A
.
G
h
o
l
i
p
o
u
r
,
“
F
e
t
a
l
-
B
ET:
b
r
a
i
n
e
x
t
r
a
c
t
i
o
n
t
o
o
l
f
o
r
f
e
t
a
l
M
R
I
,
”
I
EEE
O
p
e
n
J
o
u
rn
a
l
o
f
En
g
i
n
e
e
ri
n
g
i
n
Me
d
i
c
i
n
e
a
n
d
Bi
o
l
o
g
y
,
v
o
l
.
5
,
p
p
.
5
5
1
–
5
6
2
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
O
JE
M
B
.
2
0
2
4
.
3
4
2
6
9
6
9
.
[
1
0
]
N
.
A
.
El
Jo
u
d
i
,
M
.
La
z
a
a
r
,
F
.
D
e
l
mo
t
t
e
,
H
.
A
l
l
a
o
u
i
,
a
n
d
O
.
M
a
h
b
o
u
b
,
“
F
i
n
e
-
t
u
n
e
d
S
e
g
F
o
r
m
e
r
f
o
r
e
n
h
a
n
c
e
d
f
e
t
a
l
h
e
a
d
seg
m
e
n
t
a
t
i
o
n
,
”
Pr
o
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
2
5
1
,
p
p
.
3
5
0
–
3
5
7
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
2
4
.
1
1
.
1
2
0
.
[
1
1
]
K
.
P
r
a
sa
d
,
P
.
K
.
P
a
t
n
a
i
k
,
a
n
d
A
.
A
g
r
a
w
a
l
,
“
P
e
r
f
o
r
ma
n
c
e
a
n
a
l
y
si
s
o
f
d
e
n
o
i
si
n
g
f
i
l
t
e
r
s
f
o
r
u
l
t
r
a
s
o
u
n
d
f
e
t
a
l
h
e
a
d
i
ma
g
e
s
,
”
S
N
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
6
,
n
o
.
2
,
Ja
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s4
2
9
7
9
-
0
2
4
-
0
3
5
9
4
-
7.
[
1
2
]
H
.
K
w
o
n
e
t
a
l
.
,
“
Th
e
r
o
l
e
o
f
c
o
r
t
i
c
a
l
s
t
r
u
c
t
u
r
a
l
v
a
r
i
a
n
c
e
i
n
d
e
e
p
l
e
a
r
n
i
n
g
-
b
a
s
e
d
p
r
e
d
i
c
t
i
o
n
o
f
f
e
t
a
l
b
r
a
i
n
a
g
e
,
”
Fro
n
t
i
e
rs
i
n
N
e
u
ro
s
c
i
e
n
c
e
,
v
o
l
.
1
8
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
3
3
8
9
/
f
n
i
n
s
.
2
0
2
4
.
1
4
1
1
3
3
4
.
[
1
3
]
J.
F
u
e
t
a
l
.
,
“
U
M
-
C
A
M
:
U
n
c
e
r
t
a
i
n
t
y
-
w
e
i
g
h
t
e
d
m
u
l
t
i
-
r
e
s
o
l
u
t
i
o
n
c
l
a
ss
a
c
t
i
v
a
t
i
o
n
m
a
p
s
f
o
r
w
e
a
k
l
y
-
s
u
p
e
r
v
i
se
d
seg
m
e
n
t
a
t
i
o
n
,
”
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
v
o
l
.
1
6
0
,
A
p
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
c
o
g
.
2
0
2
4
.
1
1
1
2
0
4
.
[
1
4
]
M
.
Zh
o
u
e
t
a
l
.
,
“
T
h
e
se
g
me
n
t
a
t
i
o
n
e
f
f
e
c
t
o
f
s
t
y
l
e
t
r
a
n
sf
e
r
o
n
f
e
t
a
l
h
e
a
d
u
l
t
r
a
so
u
n
d
i
ma
g
e
:
a
s
t
u
d
y
o
f
m
u
l
t
i
-
so
u
r
c
e
d
a
t
a
,
”
Me
d
i
c
a
l
& Bi
o
l
o
g
i
c
a
l
E
n
g
i
n
e
e
ri
n
g
&
C
o
m
p
u
t
i
n
g
,
v
o
l
.
6
1
,
n
o
.
5
,
p
p
.
1
0
1
7
–
1
0
3
1
,
M
a
y
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
5
1
7
-
0
2
2
-
0
2
7
4
7
-
1.
[
1
5
]
S
.
S
r
i
v
a
st
a
v
a
,
A
.
V
i
d
y
a
r
t
h
i
,
a
n
d
S
.
Ja
i
n
,
“
A
r
e
g
r
e
ss
i
v
e
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
-
b
a
s
e
d
d
e
e
p
a
t
t
e
n
t
i
o
n
m
o
d
e
l
f
o
r
s
e
g
me
n
t
a
t
i
o
n
o
f
f
e
t
a
l
h
e
a
d
i
n
2
D
-
u
l
t
r
a
so
u
n
d
i
m
a
g
e
s,
”
I
m
a
g
e
a
n
d
Vi
si
o
n
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
3
6
,
A
u
g
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
m
a
v
i
s.
2
0
2
3
.
1
0
4
7
2
5
.
[
1
6
]
P
.
M
o
n
k
a
m,
S
.
Ji
n
,
a
n
d
W
.
L
u
,
“
A
n
n
o
t
a
t
i
o
n
c
o
st
mi
n
i
m
i
z
a
t
i
o
n
f
o
r
u
l
t
r
a
s
o
u
n
d
i
m
a
g
e
s
e
g
m
e
n
t
a
t
i
o
n
u
si
n
g
c
r
o
s
s
-
d
o
m
a
i
n
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
I
EEE
J
o
u
r
n
a
l
o
f
B
i
o
m
e
d
i
c
a
l
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
p
p
.
1
–
1
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
J
B
H
I
.
2
0
2
3
.
3
2
3
6
9
8
9
.
[
1
7
]
P
.
N
i
s
h
a
P
r
i
y
a
a
n
d
S
.
A
n
i
l
a
,
“
F
e
t
a
l
h
e
a
d
b
i
o
me
t
r
i
c
s
me
a
s
u
r
e
me
n
t
s
u
si
n
g
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
a
n
d
mi
d
-
p
o
i
n
t
e
l
l
i
p
s
e
d
r
a
w
i
n
g
a
l
g
o
r
i
t
h
m
,
”
Mu
l
t
i
d
i
m
e
n
si
o
n
a
l
S
y
st
e
m
s
a
n
d
S
i
g
n
a
l
Pro
c
e
ss
i
n
g
,
v
o
l
.
3
4
,
n
o
.
4
,
p
p
.
7
4
9
–
7
6
6
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
5
-
023
-
0
0
8
8
2
-
y.
[
1
8
]
R
.
Q
i
u
,
M
.
Z
h
o
u
,
J.
B
a
i
,
Y
.
L
u
,
a
n
d
H
.
W
a
n
g
,
“
P
S
F
H
S
P
-
N
e
t
:
a
n
e
f
f
i
c
i
e
n
t
l
i
g
h
t
w
e
i
g
h
t
n
e
t
w
o
r
k
f
o
r
i
d
e
n
t
i
f
y
i
n
g
p
u
b
i
c
s
y
mp
h
y
si
s
-
f
e
t
a
l
h
e
a
d
st
a
n
d
a
r
d
p
l
a
n
e
f
r
o
m
i
n
t
r
a
p
a
r
t
u
m
u
l
t
r
a
so
u
n
d
i
ma
g
e
s,
”
M
e
d
i
c
a
l
&
B
i
o
l
o
g
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
&
C
o
m
p
u
t
i
n
g
,
v
o
l
.
6
2
,
n
o
.
1
0
,
p
p
.
2
9
7
5
–
2
9
8
6
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
5
1
7
-
0
2
4
-
0
3
1
1
1
-
1.
[
1
9
]
P
.
C
a
i
,
L.
J
i
a
n
g
,
Y
.
L
i
,
X
.
L
i
u
,
a
n
d
L.
L
a
n
,
“
P
u
b
i
c
s
y
m
p
h
y
s
i
s
-
f
e
t
a
l
h
e
a
d
s
e
g
me
n
t
a
t
i
o
n
n
e
t
w
o
r
k
u
si
n
g
B
i
F
o
r
mer
a
t
t
e
n
t
i
o
n
mec
h
a
n
i
sm
a
n
d
m
u
l
t
i
p
a
t
h
d
i
l
a
t
e
d
c
o
n
v
o
l
u
t
i
o
n
,
”
i
n
M
u
l
t
i
Me
d
i
a
M
o
d
e
l
i
n
g
,
S
p
r
i
n
g
e
r
N
a
t
u
r
e
S
i
n
g
a
p
o
r
e
,
2
0
2
5
,
p
p
.
2
4
3
–
256
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
96
-
2
0
6
4
-
7
_
1
8
.
[
2
0
]
G
.
D
u
b
e
y
,
S
.
S
r
i
v
a
st
a
v
a
,
A
.
K
.
J
a
y
sw
a
l
,
M
.
S
a
r
a
sw
a
t
,
P
.
S
i
n
g
h
,
a
n
d
M
.
M
e
m
o
r
i
a
,
“
F
e
t
a
l
u
l
t
r
a
so
u
n
d
s
e
g
me
n
t
a
t
i
o
n
a
n
d
mea
s
u
r
e
me
n
t
s
u
s
i
n
g
a
p
p
e
a
r
a
n
c
e
a
n
d
sh
a
p
e
p
r
i
o
r
b
a
se
d
d
e
n
si
t
y
r
e
g
r
e
ssi
o
n
w
i
t
h
d
e
e
p
C
N
N
a
n
d
r
o
b
u
st
e
l
l
i
p
se
f
i
t
t
i
n
g
,
”
J
o
u
r
n
a
l
o
f
I
m
a
g
i
n
g
I
n
f
o
rm
a
t
i
c
s
i
n
Me
d
i
c
i
n
e
,
v
o
l
.
3
7
,
n
o
.
1
,
p
p
.
2
4
7
–
2
6
7
,
Ja
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
2
7
8
-
0
2
3
-
0
0
9
0
8
-
8.
[
2
1
]
Z.
C
h
e
n
,
Z.
O
u
,
Y
.
L
u
,
a
n
d
J.
B
a
i
,
“
D
i
r
e
c
t
i
o
n
-
g
u
i
d
e
d
a
n
d
m
u
l
t
i
-
sc
a
l
e
f
e
a
t
u
r
e
scr
e
e
n
i
n
g
f
o
r
f
e
t
a
l
h
e
a
d
–
p
u
b
i
c
s
y
m
p
h
y
s
i
s
seg
m
e
n
t
a
t
i
o
n
a
n
d
a
n
g
l
e
o
f
p
r
o
g
r
e
ssi
o
n
c
a
l
c
u
l
a
t
i
o
n
,
”
E
x
p
e
rt
S
y
s
t
e
m
s
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
2
4
5
,
J
u
l
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
3
.
1
2
3
0
9
6
.
[
2
2
]
Z.
C
h
e
n
,
Y
.
L
u
,
S
.
L
o
n
g
,
V
.
M
.
C
a
m
p
e
l
l
o
,
J
.
B
a
i
,
a
n
d
K
.
L
e
k
a
d
i
r
,
“
F
e
t
a
l
h
e
a
d
a
n
d
p
u
b
i
c
sy
m
p
h
y
si
s
s
e
g
me
n
t
a
t
i
o
n
i
n
i
n
t
r
a
p
a
r
t
u
m
u
l
t
r
a
s
o
u
n
d
i
ma
g
e
u
s
i
n
g
a
d
u
a
l
-
p
a
t
h
b
o
u
n
d
a
r
y
-
g
u
i
d
e
d
r
e
s
i
d
u
a
l
n
e
t
w
o
r
k
,
”
I
EEE
J
o
u
r
n
a
l
o
f
B
i
o
m
e
d
i
c
a
l
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
8
,
n
o
.
8
,
p
p
.
4
6
4
8
–
4
6
5
9
,
A
u
g
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
J
B
H
I
.
2
0
2
4
.
3
3
9
9
7
6
2
.
[
2
3
]
B
.
P
h
a
n
,
“
H
C
1
8
g
r
a
n
d
c
h
a
l
l
e
n
g
e
,
”
K
a
g
g
l
e
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m/
d
a
t
a
s
e
t
s
/
t
h
a
n
h
b
n
h
p
h
a
n
/
h
c
1
8
-
g
r
a
n
d
-
c
h
a
l
l
e
n
g
e
[
2
4
]
P
.
S
a
n
j
u
,
“
E
n
h
a
n
c
i
n
g
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
i
n
I
o
T
s
y
s
t
e
m
s
:
A
h
y
b
r
i
d
m
e
t
a
h
e
u
r
i
s
t
i
c
s
-
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
w
i
t
h
e
n
s
e
m
b
l
e
o
f
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
J
o
u
r
n
a
l
o
f
E
n
g
i
n
e
e
r
i
n
g
R
e
s
e
a
r
c
h
,
v
o
l
.
1
1
,
n
o
.
4
,
p
p
.
3
5
6
–
3
6
1
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
e
r
.
2
0
2
3
.
1
0
0
1
2
2
.
[
2
5
]
V
.
N
a
g
a
b
o
t
u
a
n
d
A
.
N
a
mb
u
r
u
,
“
P
r
e
c
i
se
se
g
me
n
t
a
t
i
o
n
o
f
f
e
t
a
l
h
e
a
d
i
n
u
l
t
r
a
so
u
n
d
i
ma
g
e
s
u
si
n
g
i
m
p
r
o
v
e
d
U
‐
N
e
t
m
o
d
e
l
,
”
ET
R
I
J
o
u
rn
a
l
,
v
o
l
.
4
6
,
n
o
.
3
,
p
p
.
5
2
6
–
5
3
7
,
J
u
n
.
2
0
2
4
,
d
o
i
:
1
0
.
4
2
1
8
/
e
t
r
i
j
.
2
0
2
3
-
0
0
5
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.