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.
1
4
,
No
.
5
,
Octo
b
er
2
0
2
5
,
p
p
.
4
1
3
7
~
4
1
5
0
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
14
.i
5
.
p
p
4
1
3
7
-
4
1
5
0
4137
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
M
ulticlas
s insta
n
ce segm
entatio
n
optimiza
tion
for fe
tal hea
rt
ima
g
e object
i
nte
rpretation
H
a
di Sy
a
pu
t
ra
1,
4
,
Siti
Nurma
ini
2
,
Ra
diy
a
t
i U
m
i P
a
r
t
a
n
3
,
M
uh
a
m
m
a
d T
a
ufik
Ro
s
eno
1,
4
1
D
o
c
t
o
r
a
l
P
r
o
g
r
a
m i
n
En
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
a
s Sr
i
w
i
j
a
y
a
,
I
n
d
r
a
l
a
y
a
,
I
n
d
o
n
e
si
a
2
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
R
e
s
e
a
r
c
h
G
r
o
u
p
,
F
a
c
u
l
t
y
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
U
n
i
v
e
r
si
t
a
s
S
r
i
w
i
j
a
y
a
,
P
a
l
e
m
b
a
n
g
,
I
n
d
o
n
e
s
i
a
3
D
e
p
a
r
t
me
n
t
o
f
I
n
t
e
r
n
a
l
M
e
d
i
c
i
n
e
,
F
a
c
u
l
t
y
o
f
M
e
d
i
c
i
n
e
,
U
n
i
v
e
r
si
t
a
s Sr
i
w
i
j
a
y
a
,
I
n
d
r
a
l
a
y
a
,
I
n
d
o
n
e
si
a
4
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
S
t
u
d
y
P
r
o
g
r
a
m,
F
a
c
u
l
t
y
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
,
U
n
i
v
e
r
s
i
t
a
s S
u
ma
t
e
r
a
S
e
l
a
t
a
n
,
P
a
l
e
mb
a
n
g
,
I
n
d
o
n
e
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
1
8
,
2
0
2
4
R
ev
is
ed
J
u
n
1
7
,
2
0
2
5
Acc
ep
ted
Ju
l
1
0
,
2
0
2
5
Th
is
re
se
a
rc
h
a
ims
to
d
e
v
e
l
o
p
a
m
u
lt
i
-
c
las
s
in
sta
n
c
e
se
g
m
e
n
tatio
n
m
o
d
e
l
fo
r
se
g
m
e
n
ti
n
g
,
d
e
tec
ti
n
g
,
a
n
d
c
las
sify
in
g
o
b
jec
ts
in
fe
tal
h
e
a
rt
u
lt
ra
so
u
n
d
ima
g
e
s
d
e
riv
e
d
fro
m
fe
tal
h
e
a
rt
u
lt
ra
so
u
n
d
v
id
e
o
s.
P
re
v
i
o
u
s
st
u
d
ies
h
a
v
e
p
e
rfo
rm
e
d
o
b
jec
t
d
e
tec
ti
o
n
o
n
fe
t
a
l
h
e
a
rt
ima
g
e
s,
id
e
n
t
ify
i
n
g
n
i
n
e
a
n
a
to
m
ica
l
c
las
se
s.
F
u
rth
e
r,
th
e
se
st
u
d
ies
h
a
v
e
c
o
n
d
u
c
ted
in
sta
n
c
e
se
g
m
e
n
tatio
n
o
n
fe
tal
h
e
a
rt
ima
g
e
s
f
o
r
si
x
a
n
a
to
m
ica
l
c
las
se
s.
Th
is
re
se
a
rc
h
se
e
k
s
t
o
e
x
p
a
n
d
th
e
sc
o
p
e
b
y
i
n
c
re
a
sin
g
th
e
n
u
m
b
e
r
o
f
c
las
se
s
to
ten
,
e
n
c
o
m
p
a
ss
in
g
fo
u
r
m
a
in
c
h
a
m
b
e
rs
left
a
tri
u
m
(LA),
rig
h
t
a
tri
u
m
(RA),
left
v
e
n
tri
c
le
(
LV),
rig
h
t
v
e
n
tri
c
l
e
(RV);
fo
u
r
v
a
lv
e
s
tri
c
u
sp
id
v
a
lv
e
(TV),
p
u
lmo
n
a
ry
v
a
lv
e
(P
V),
m
it
ra
l
v
a
lv
e
(M
V)
,
a
n
d
a
o
rti
c
v
a
l
v
e
(AV
);
o
n
e
a
o
rta
(A
o
),
a
n
d
t
h
e
sp
in
e
.
B
y
d
e
v
e
lo
p
in
g
a
n
i
n
sta
n
c
e
se
g
m
e
n
ta
ti
o
n
m
e
th
o
d
f
o
r
se
g
m
e
n
ti
n
g
te
n
a
n
a
to
m
ica
l
stru
c
tu
re
s
o
f
t
h
e
fe
tal
h
e
a
rt,
t
h
is
re
se
a
rc
h
a
ims
t
o
m
a
k
e
a
sig
n
ifi
c
a
n
t
c
o
n
tri
b
u
ti
o
n
t
o
imp
ro
v
in
g
m
e
d
ica
l
ima
g
e
a
n
a
l
y
sis
i
n
h
e
a
lt
h
c
a
re
.
It
a
lso
a
ims
to
p
a
v
e
t
h
e
wa
y
fo
r
fu
rt
h
e
r
re
se
a
rc
h
o
n
fe
tal
h
e
a
rt
d
is
e
a
se
s
u
sin
g
AI
.
Th
e
in
sta
n
c
e
se
g
m
e
n
tatio
n
a
p
p
r
o
a
c
h
is
e
x
p
e
c
ted
t
o
e
n
h
a
n
c
e
th
e
a
c
c
u
ra
c
y
o
f
se
g
m
e
n
ti
n
g
fe
tal
h
e
a
rt
ima
g
e
s
a
n
d
a
ll
o
w
fo
r
m
o
re
e
fficie
n
t
i
d
e
n
ti
f
i
c
a
ti
o
n
a
n
d
lab
e
li
n
g
o
f
e
a
c
h
a
n
a
t
o
m
ica
l
stru
c
t
u
re
in
th
e
fe
tal
h
e
a
rt.
K
ey
w
o
r
d
s
:
Ar
tific
ial
in
tellig
en
ce
Fetal
h
ea
r
t
I
n
s
tan
ce
s
eg
m
en
tatio
n
Mu
lticlas
s
R
esNet
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
:
Sit
i N
u
r
m
ain
i
I
n
tellig
en
t Sy
s
tem
R
esear
ch
Gr
o
u
p
,
Facu
lty
o
f
C
o
m
p
u
ter
Sci
en
ce
,
Un
iv
er
s
itas
Sriwijay
a
Palem
b
an
g
,
I
n
d
o
n
esia
E
m
ail:
s
iti_
n
u
r
m
ain
i@
u
n
s
r
i.a
c.
id
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
ap
id
d
ev
elo
p
m
en
t
o
f
AI
tech
n
o
lo
g
y
h
as
b
ec
o
m
e
an
in
t
eg
r
al
p
ar
t
o
f
m
o
d
er
n
s
o
ciety
.
T
h
is
is
d
u
e
to
th
e
ca
p
ab
ilit
y
o
f
AI
to
r
atio
n
alize
an
d
tak
e
ac
tio
n
s
o
r
s
o
lu
tio
n
s
th
at
h
av
e
th
e
h
ig
h
est
p
r
o
b
ab
ilit
y
o
f
ac
h
iev
in
g
s
et
g
o
als
[
1
]
.
I
n
r
ec
en
t
y
ea
r
s
,
AI
h
as
b
ee
n
wid
ely
ap
p
lied
ac
r
o
s
s
v
ar
io
u
s
s
ec
to
r
s
,
in
clu
d
in
g
g
o
v
er
n
m
en
t
[
2
]
,
in
f
r
astru
ct
u
r
e
[
3
]
,
ag
r
ic
u
ltu
r
e
[
4
]
,
an
d
h
ea
lth
ca
r
e
[
5
]
.
B
y
lev
er
ag
i
n
g
th
is
tech
n
o
lo
g
y
,
co
m
p
an
ies
an
d
o
r
g
an
izatio
n
s
ca
n
in
teg
r
ate
v
ast
am
o
u
n
ts
o
f
d
ata
to
p
r
o
ce
s
s
in
f
o
r
m
atio
n
a
n
d
m
ak
e
d
ec
is
io
n
s
.
T
o
s
u
p
p
o
r
t
d
ec
is
io
n
-
m
a
k
in
g
p
r
o
ce
s
s
es,
an
AI
-
b
ased
ap
p
r
o
a
ch
in
d
ev
el
o
p
in
g
m
o
d
els
u
s
in
g
m
ac
h
in
e
lear
n
in
g
(
ML
)
alg
o
r
ith
m
s
is
n
ec
ess
ar
y
.
Var
io
u
s
AI
m
eth
o
d
o
lo
g
ies
h
av
e
b
ee
n
d
ev
el
o
p
ed
,
o
n
e
o
f
wh
ich
is
ML
.
M
L
o
p
er
ates
b
y
u
tili
zin
g
n
e
u
r
al
n
etwo
r
k
s
to
p
r
o
ce
s
s
d
ata
with
th
e
aim
o
f
g
e
n
er
atin
g
k
n
o
wl
ed
g
e
th
at
s
u
p
p
o
r
ts
o
r
g
an
izatio
n
al
o
r
in
d
iv
id
u
al
ac
tiv
ities
.
I
n
th
e
p
r
o
ce
s
s
,
ML
ex
tr
ac
ts
k
ey
f
ea
tu
r
es
f
r
o
m
d
ata
f
o
r
m
o
d
el
f
o
r
m
atio
n
[
6
]
–
[
8
]
.
I
n
th
e
h
ea
lth
ca
r
e
f
ield
,
ML
h
as
b
ee
n
ex
ten
s
iv
ely
u
s
ed
to
aid
m
ed
ical
p
r
o
f
ess
io
n
als
in
d
ec
is
io
n
-
m
ak
in
g
.
R
esear
ch
b
y
Pu
llag
u
r
a
et
a
l
.
[
9
]
u
tili
ze
d
ML
to
en
h
an
ce
th
e
ac
c
u
r
ac
y
o
f
f
etal
h
ea
r
t
d
is
ea
s
e
id
en
tific
atio
n
.
C
an
ad
illa
et
a
l
.
[
1
0
]
co
n
d
u
cte
d
r
esear
ch
em
p
l
o
y
in
g
ML
to
im
p
r
o
v
e
t
h
e
ev
al
u
atio
n
o
f
f
etal
h
ea
r
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.
1
4
,
No
.
5
,
Octo
b
er
2
0
2
5
:
4
1
3
7
-
4
1
5
0
4138
f
u
n
ctio
n
b
y
o
p
tim
izin
g
im
ag
e
ac
q
u
is
itio
n
an
d
m
ea
s
u
r
e
m
e
n
ts
,
th
er
eb
y
aid
in
g
in
p
r
en
ata
l
d
iag
n
o
s
is
o
f
f
etal
h
ea
r
t
r
em
o
d
elin
g
an
d
ab
n
o
r
m
alities
.
Ho
o
d
b
h
o
y
et
a
l.
[
1
1
]
s
tu
d
ied
th
e
ac
cu
r
ac
y
o
f
ML
al
g
o
r
ith
m
tech
n
iq
u
es
in
id
en
tify
in
g
h
ig
h
-
r
is
k
f
etu
s
es
th
r
o
u
g
h
ca
r
d
io
to
c
o
g
r
ap
h
y
.
C
ö
m
er
t
an
d
Ko
ca
m
az
[
1
2
]
u
s
ed
ML
as
a
m
o
n
ito
r
in
g
tech
n
iq
u
e
th
at
p
r
o
v
id
es
cr
u
cial
an
d
v
ital
in
f
o
r
m
atio
n
ab
o
u
t
f
etal
s
tatu
s
d
u
r
i
n
g
an
tep
ar
tu
m
an
d
in
tr
ap
ar
tu
m
p
er
io
d
s
,
as
well
as
class
if
y
in
g
f
etal
h
ea
r
t
r
ate
s
ig
n
als.
Ho
wev
er
,
p
r
ev
io
u
s
s
tu
d
ies
h
av
e
s
h
o
wn
th
at
ML
m
eth
o
d
s
h
av
e
lim
itatio
n
s
in
an
aly
zin
g
s
tr
u
ctu
r
ed
an
d
lim
ited
d
ata.
Ad
d
itio
n
ally
,
M
L
m
eth
o
d
s
in
v
o
lv
e
m
o
r
e
co
m
p
le
x
s
tag
es,
s
u
ch
as
m
an
u
al
im
ag
e
au
g
m
en
tati
o
n
,
wh
ich
ca
n
b
e
tim
e
-
c
o
n
s
u
m
in
g
to
p
r
o
d
u
ce
ac
tio
n
ab
le
in
f
o
r
m
atio
n
f
o
r
d
ec
is
io
n
-
m
ak
in
g
a
n
d
ac
tio
n
s
[
1
3
]
.
T
o
ad
d
r
ess
th
e
c
h
allen
g
es
o
f
t
r
ad
itio
n
al
ML
m
eth
o
d
s
,
s
ev
er
al
s
tu
d
ies
h
av
e
ad
o
p
ted
a
d
ee
p
lear
n
i
n
g
(
DL
)
[
1
4
]
a
p
p
r
o
ac
h
f
o
r
a
n
a
ly
zin
g
an
d
p
r
ed
ictin
g
m
ed
ic
al
ex
am
in
atio
n
o
u
tco
m
es,
esp
ec
ially
in
im
ag
e
class
if
icatio
n
an
d
o
b
ject
d
etec
tio
n
to
s
u
p
p
o
r
t
f
etal
ec
h
o
ca
r
d
io
g
r
a
p
h
y
ex
am
in
atio
n
s
.
B
y
p
r
o
ce
s
s
in
g
lar
g
e
am
o
u
n
ts
o
f
d
ata,
DL
h
as
d
e
m
o
n
s
tr
ated
p
o
ten
tial
in
e
n
h
a
n
cin
g
ac
c
u
r
ac
y
an
d
ef
f
icien
c
y
in
m
e
d
ical
im
ag
e
an
aly
s
is
.
DL
m
eth
o
d
s
ar
e
f
r
e
q
u
en
tly
em
p
lo
y
ed
in
th
e
m
e
d
ical
f
ield
,
s
u
ch
as
in
f
etal
c
ar
d
io
g
r
a
p
h
y
im
a
g
e
d
etec
tio
n
[
1
5
]
.
On
e
o
f
th
e
p
r
i
m
ar
y
ad
v
a
n
tag
es
o
f
DL
tech
n
i
q
u
es
is
th
eir
ab
ilit
y
to
ex
tr
ac
t
s
ig
n
if
ican
t
in
s
ig
h
ts
,
p
atter
n
s
,
an
d
i
n
f
o
r
m
atio
n
f
r
o
m
im
ag
es
an
d
v
id
eo
s
.
T
h
is
is
ac
h
iev
ed
th
r
o
u
g
h
th
e
d
ev
elo
p
m
en
t
o
f
alg
o
r
ith
m
s
an
d
m
o
d
els
th
at
en
ab
le
m
ac
h
in
es
to
a
n
aly
ze
,
p
r
o
ce
s
s
,
a
n
d
m
a
k
e
d
ec
is
io
n
s
b
ased
o
n
v
is
u
al
d
ata
[
1
6
]
.
Mo
r
eo
v
er
,
DL
tech
n
iq
u
es c
an
id
en
tify
an
d
d
e
p
ict
in
d
iv
id
u
al
o
b
jects in
im
ag
es wh
ile
p
r
o
v
id
in
g
lab
els f
o
r
ea
ch
o
b
ject,
m
ak
in
g
th
em
a
p
p
lica
b
le
in
v
ar
io
u
s
f
ield
s
s
u
ch
as
o
b
ject
tr
ac
k
in
g
[
1
7
]
an
d
m
e
d
ical
im
ag
in
g
[
1
8
]
.
Ho
wev
er
,
th
ese
s
tu
d
ies
m
ain
ly
f
o
cu
s
o
n
th
e
class
if
icatio
n
o
f
m
ed
ical
im
ag
es
o
r
v
id
eo
s
b
y
co
m
p
ar
in
g
o
n
e
im
ag
e
o
b
ject
with
a
n
o
th
er
.
Ad
d
itio
n
ally
,
t
h
e
class
if
icatio
n
tech
n
i
q
u
e
in
DL
m
eth
o
d
s
ca
n
o
n
ly
i
d
en
tify
a
s
in
g
le
o
b
ject
with
in
a
n
im
ag
e
an
d
ca
te
g
o
r
ize
it
b
ased
o
n
th
at
o
b
ject.
T
o
o
v
er
co
m
e
th
e
lim
itatio
n
s
o
f
DL
class
if
icatio
n
tech
n
iq
u
es,
a
s
o
l
u
tio
n
is
r
eq
u
ir
ed
th
at
ca
n
d
ete
ct
m
u
ltip
le
o
b
jects
with
in
a
s
in
g
le
im
ag
e
o
r
v
id
e
o
[
1
9
]
.
I
n
ad
d
itio
n
to
class
if
icatio
n
an
d
d
etec
tio
n
ca
p
a
b
ilit
ies,
DL
m
eth
o
d
s
also
p
o
s
s
ess
t
h
e
ab
ilit
y
t
o
d
etec
t
m
u
ltip
le
o
b
jects
in
o
n
e
im
ag
e
an
d
v
id
eo
.
Fo
r
ex
am
p
le,
r
esear
ch
co
n
d
u
cted
b
y
Sap
itri
et
a
l.
[
2
0
]
u
tili
ze
d
DL
f
o
r
o
b
ject
d
etec
tio
n
in
f
etal
u
ltra
s
o
u
n
d
v
id
e
o
s
,
id
en
tify
i
n
g
an
ato
m
ical
s
u
b
s
tr
u
ctu
r
es
o
f
th
e
f
etal
h
ea
r
t,
in
clu
d
in
g
i)
f
o
u
r
m
ai
n
ch
am
b
er
s
:
lef
t
atr
iu
m
(
L
A)
,
r
ig
h
t
atr
iu
m
(
R
A)
,
lef
t
v
en
tr
icle
(
L
V)
,
r
ig
h
t
v
en
t
r
icle
(
R
V)
;
ii)
f
o
u
r
v
alv
es:
tr
icu
s
p
id
v
alv
e
(
T
V)
,
p
u
lm
o
n
ar
y
v
alv
e
(
PV)
,
m
itra
l
v
alv
e
(
MV
)
,
an
d
ao
r
tic
v
alv
e
(
AV)
;
an
d
iii)
o
n
e
ao
r
ta
(
A
o
)
.
Su
b
s
e
q
u
e
n
t
d
e
v
e
lo
p
m
e
n
ts
in
o
b
je
ct
d
e
tec
ti
o
n
[
2
1
]
,
[
2
2
]
h
a
v
e
e
n
a
b
l
ed
t
h
e
i
d
en
ti
f
ic
ati
o
n
a
n
d
ca
t
eg
o
r
iz
ati
o
n
o
f
e
v
e
r
y
p
i
x
el
i
n
a
n
im
ag
e
i
n
t
o
m
e
an
in
g
f
u
l
o
b
je
ct
ca
t
e
g
o
r
i
es
o
r
a
r
ea
s
,
k
n
o
wn
as
s
e
g
m
e
n
ta
ti
o
n
.
Seg
m
e
n
t
ati
o
n
te
c
h
n
iq
u
es
in
clu
d
e
s
e
m
a
n
ti
c
s
e
g
m
e
n
tat
io
n
a
n
d
i
n
s
ta
n
c
e
s
e
g
m
e
n
ta
ti
o
n
.
R
es
ea
r
c
h
b
y
R
ac
h
m
a
tu
lla
h
et
a
l
.
[
2
3
]
u
s
e
d
s
em
a
n
t
ic
s
eg
m
e
n
t
ati
o
n
m
et
h
o
d
s
t
o
d
e
v
e
lo
p
a
s
e
m
a
n
ti
c
m
o
d
el
th
a
t
d
et
ec
ts
o
b
je
cts
b
y
ass
ig
n
i
n
g
l
a
b
els
t
o
ea
c
h
p
i
x
el
i
n
an
i
m
a
g
e
,
en
s
u
r
i
n
g
t
h
a
t
p
i
x
els
w
it
h
t
h
e
s
a
m
e
la
b
el
h
a
v
e
t
h
e
s
am
e
im
ag
e.
Sim
p
l
y
p
u
t
,
s
em
a
n
ti
c
im
ag
e
s
e
g
m
e
n
tat
io
n
is
a
t
ec
h
n
i
q
u
e
u
s
ed
t
o
id
e
n
ti
f
y
s
p
e
ci
f
ic
o
b
je
ct
t
y
p
es
w
it
h
i
n
a
n
i
m
ag
e.
Ho
w
ev
er
,
s
em
a
n
tic
s
e
g
m
e
n
ta
t
io
n
t
ec
h
n
i
q
u
es
h
a
v
e
s
e
v
e
r
a
l
d
r
aw
b
a
ck
s
,
in
cl
u
d
in
g
th
e
i
n
a
b
il
it
y
t
o
d
is
t
in
g
u
is
h
b
et
we
en
i
n
d
iv
id
u
al
o
b
j
ec
ts
i
n
a
n
i
m
a
g
e
a
n
d
d
i
f
f
ic
u
l
ty
i
d
e
n
ti
f
y
i
n
g
i
n
d
i
v
i
d
u
al
o
b
j
ec
ts
wi
th
s
im
ila
r
te
x
t
u
r
es
[
2
3
]
,
[
2
4
]
.
I
n
co
n
tr
ast,
in
s
tan
c
e
s
eg
m
en
tatio
n
ca
n
p
r
o
v
id
e
u
n
iq
u
e
lab
els f
o
r
ea
c
h
in
d
i
v
id
u
al
o
b
ject
[
2
5
]
,
[
2
6
]
.
E
f
f
o
r
t
s
to
r
ec
o
g
n
i
ze
an
d
s
e
p
ar
a
te
ea
c
h
c
la
s
s
o
f
o
b
j
ec
ts
in
a
n
i
m
ag
e
r
e
ly
h
ea
v
ily
o
n
in
s
t
an
c
e
s
eg
m
en
ta
tio
n
,
wh
i
ch
in
tu
r
n
d
ep
en
d
s
o
n
th
e
b
a
ck
b
o
n
e
ar
c
h
it
ec
tu
r
e
[
2
7
]
.
T
h
e
b
ac
k
b
o
n
e
ar
ch
it
ec
tu
r
e
p
l
ay
s
a
cr
u
c
ia
l r
o
le
i
n
in
s
t
an
ce
s
eg
m
e
n
ta
tio
n
b
y
p
r
o
v
id
in
g
e
s
s
e
n
t
ia
l
f
ea
tu
r
e
in
f
o
r
m
a
t
io
n
o
f
th
e
ar
e
as
to
b
e
s
eg
m
en
ted
f
o
r
th
e
m
o
d
e
l
[
2
8
]
.
R
e
s
ea
r
ch
co
n
d
u
c
ted
b
y
Nu
r
m
a
in
i
et
a
l
.
[
2
9
]
,
h
a
s
u
ti
l
ize
d
th
e
u
s
e
o
f
R
e
s
Ne
t
a
s
th
e
m
a
in
s
t
r
u
c
tu
r
e
to
ac
h
iev
e
o
p
ti
m
a
l
i
n
s
tan
c
e
s
eg
m
en
ta
tio
n
.
T
h
e
ap
p
li
ca
tio
n
o
f
in
s
t
an
ce
s
eg
m
en
t
at
io
n
in
th
e
m
ed
i
ca
l
f
ie
ld
in
c
lu
d
e
s
au
to
m
at
in
g
th
e
s
eg
m
en
ta
tio
n
p
r
o
ce
s
s
a
n
d
im
p
r
o
v
in
g
d
e
tec
t
io
n
ac
c
u
r
a
cy
[
3
0
]
.
Fo
r
in
s
t
an
c
e,
an
in
s
ta
n
ce
s
eg
m
en
t
at
io
n
ap
p
r
o
a
ch
f
o
r
f
et
al
ec
h
o
ca
r
d
io
g
r
ap
h
y
ca
n
s
im
u
lt
an
eo
u
s
ly
s
ep
ar
a
t
e
th
e
f
o
u
r
s
tan
d
ar
d
h
ea
r
t
v
i
ew
s
an
d
d
e
te
ct
d
ef
e
ct
s
[
2
9
]
.
T
o
ac
cu
r
a
te
ly
d
e
te
ct
f
eta
l
h
e
ar
t
ab
n
o
r
m
a
li
ti
e
s
th
r
o
u
g
h
f
et
al
u
l
tr
a
s
o
u
n
d
,
al
l
h
ea
r
t
s
u
b
s
tr
u
ctu
r
e
s
m
u
s
t
b
e
r
ec
o
g
n
iz
ed
in
n
o
r
m
al
a
n
a
to
m
y
[
2
0
]
.
On
e
o
f
th
e
m
o
s
t
s
ig
n
if
i
ca
n
t
l
im
i
ta
t
io
n
s
as
s
o
ci
at
ed
wi
th
u
l
tr
aso
u
n
d
in
v
o
lv
e
s
in
te
r
p
er
s
o
n
al
v
ar
ia
b
i
li
t
y
,
m
ea
n
in
g
i
t
d
ep
en
d
s
o
n
th
e
ex
am
in
in
g
d
o
cto
r
's
s
k
il
l
s
an
d
th
e
p
a
t
ien
t
's
co
n
d
it
io
n
[
2
8
]
.
R
ef
er
r
in
g
to
r
e
s
e
ar
ch
b
y
Sap
it
r
i
et
a
l
.
[
2
0
]
,
wh
ich
ex
am
in
ed
an
at
o
m
i
ca
l
s
tr
u
ctu
r
e
d
et
ec
ti
o
n
in
f
et
al
h
e
ar
t
im
ag
e
s
,
a
s
we
ll
a
s
r
e
s
ea
r
ch
b
y
Nu
r
m
ain
i
et
a
l.
[
2
8
]
,
w
h
i
ch
f
o
cu
s
ed
o
n
in
s
t
an
c
e
s
eg
m
en
t
a
tio
n
f
o
r
th
e
f
o
u
r
m
a
in
ch
am
b
er
s
o
f
th
e
f
e
ta
l
h
ea
r
t
an
d
h
ea
r
t
d
i
s
ea
s
e
d
et
ec
tio
n
,
th
i
s
s
tu
d
y
ex
p
an
d
s
it
s
s
co
p
e
t
o
in
c
lu
d
e
ad
d
i
tio
n
al
an
a
to
m
ic
al
o
b
jec
t
s
,
n
a
m
e
ly
th
e
s
p
in
e
.
T
h
e
ad
d
it
io
n
o
f
th
e
s
p
in
e
i
s
c
r
u
c
ia
l
f
o
r
m
ed
ic
al
p
r
ac
ti
tio
n
er
s
in
id
en
ti
f
y
in
g
th
e
f
o
u
r
-
ch
a
m
b
er
v
ie
w
(
A4
C
)
o
f
th
e
f
e
ta
l
h
e
ar
t
i
n
im
ag
e
s
[
3
1
]
.
T
h
er
ef
o
r
e
,
th
e
co
n
tr
ib
u
t
io
n
s
o
f
th
i
s
s
tu
d
y
ar
e
th
e
in
cl
u
s
io
n
o
f
ten
an
ato
m
ic
al
o
b
je
ct
s
o
f
th
e
f
e
ta
l
h
ea
r
t,
n
am
ely
L
A,
R
A,
L
V,
R
V,
T
V,
P
V,
M
V,
AV,
Ao
,
a
n
d
s
p
in
e,
an
d
th
e
d
e
v
e
lo
p
m
en
t
o
f
a
DL
a
p
p
r
o
a
ch
u
s
in
g
in
s
tan
ce
s
eg
m
en
ta
t
io
n
m
eth
o
d
s
f
o
r
th
e
s
e
t
en
an
ato
m
ic
al
s
tr
u
c
tu
r
e
s
.
B
y
d
ev
elo
p
in
g
a
s
am
p
le
s
eg
m
en
ta
tio
n
m
et
h
o
d
f
o
r
t
en
f
eta
l
h
e
ar
t
an
a
to
m
y
o
b
je
ct
s
an
d
ap
p
ly
in
g
h
y
p
er
p
ar
am
et
er
tu
n
in
g
to
f
in
d
th
e
o
p
ti
m
a
l
s
et
t
in
g
s
[
3
2
]
,
[
3
3
]
,
th
i
s
s
tu
d
y
a
im
s
t
o
s
ig
n
if
i
ca
n
t
ly
i
m
p
r
o
v
e
m
ed
i
ca
l
im
ag
e
an
aly
s
is
in
th
e
h
ea
lt
h
ca
r
e
f
ie
ld
an
d
p
av
e
th
e
way
f
o
r
f
u
t
u
r
e
r
e
s
e
ar
ch
in
d
e
te
ct
in
g
f
e
ta
l
h
ea
r
t
d
i
s
e
a
s
e.
T
h
i
s
ap
p
r
o
ac
h
p
r
o
m
i
s
e
s
ac
cu
r
ac
y
in
s
eg
m
en
tin
g
th
e
f
e
ta
l h
ea
r
t.
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
Mu
lticla
s
s
in
s
ta
n
ce
s
eg
men
ta
t
io
n
o
p
timiz
a
tio
n
fo
r
feta
l h
e
a
r
t ima
g
e
o
b
ject
in
terp
r
eta
tio
n
(
Ha
d
i S
ya
p
u
tr
a
)
4139
T
h
e
s
eg
m
en
tatio
n
o
f
th
ese
ten
an
ato
m
ical
s
tr
u
ctu
r
es
was
ch
o
s
en
b
ased
o
n
clin
ical
co
n
s
id
er
atio
n
s
as
ea
ch
h
as
an
im
p
o
r
tan
t
r
o
le
i
n
th
e
d
iag
n
o
s
is
o
f
co
n
g
e
n
ital
h
ea
r
t
d
ef
ec
ts
.
T
h
e
f
o
u
r
m
a
in
h
ea
r
t
ch
am
b
er
s
(
L
A,
R
A,
L
V,
R
V)
an
d
th
e
f
o
u
r
v
alv
es
(
T
V,
PV,
MV
,
A
V)
ar
e
th
e
s
tr
u
ctu
r
es
m
o
s
t
f
r
eq
u
en
tly
u
s
ed
in
th
e
f
u
n
ctio
n
al
ass
ess
m
en
t
o
f
th
e
f
etal
h
ea
r
t
v
ia
u
ltra
s
o
n
o
g
r
a
p
h
y
.
T
h
e
s
tr
u
ctu
r
e
o
f
th
e
Ao
is
im
p
o
r
tan
t
in
id
en
tify
in
g
b
lo
o
d
o
u
tf
lo
w,
wh
ile
th
e
s
p
in
e
h
elp
s
to
en
s
u
r
e
co
r
r
ec
t
an
ato
m
ical
o
r
ie
n
tatio
n
in
th
e
A4
C
.
Acc
u
r
ate
s
eg
m
en
tatio
n
o
f
th
e
s
e
s
tr
u
ctu
r
es
allo
ws
ea
r
ly
id
e
n
tific
atio
n
o
f
v
a
r
io
u
s
ab
n
o
r
m
alities
s
u
ch
as
s
ep
tal
d
ef
ec
ts
,
v
alv
e
s
ten
o
s
is
,
an
d
a
b
n
o
r
m
al
p
o
s
itio
n
in
g
o
f
th
e
h
ea
r
t o
r
o
th
e
r
o
r
g
an
s
.
2.
M
AT
E
R
I
AL
A
ND
M
E
T
H
O
D
Dete
ctin
g
th
e
n
o
r
m
al
f
etal
h
e
ar
t
an
ato
m
y
f
r
o
m
A4
C
v
id
eo
b
etwe
en
1
4
an
d
2
8
wee
k
s
o
f
g
estatio
n
al
ag
e
is
a
co
m
p
lex
task
.
Seg
m
e
n
tatio
n
aim
s
to
d
elin
ea
te
ca
r
d
i
ac
s
tr
u
ctu
r
es
u
s
in
g
co
n
to
u
r
b
o
u
n
d
ar
ies;
h
o
wev
e
r
,
th
is
m
eth
o
d
is
lim
ited
in
ca
p
tu
r
in
g
th
e
s
p
atial
r
elatio
n
s
h
ip
s
am
o
n
g
c
o
m
p
o
n
en
ts
.
As
illu
s
tr
a
ted
in
Fig
u
r
e
1
,
th
e
wo
r
k
f
lo
w
b
eg
in
s
with
th
e
e
x
tr
ac
tio
n
a
n
d
s
elec
tio
n
o
f
v
i
d
eo
f
r
am
es
b
ased
o
n
th
e
A4
C
p
er
s
p
ec
tiv
e.
T
h
e
s
elec
ted
f
r
am
es
ar
e
r
ef
in
ed
th
r
o
u
g
h
cr
o
p
p
in
g
,
f
ilter
in
g
,
an
d
r
esizin
g
,
f
o
llo
wed
b
y
m
an
u
al
an
n
o
tatio
n
o
f
f
etal
h
ea
r
t
an
at
o
m
y
g
u
id
e
d
b
y
ex
p
er
t
k
n
o
wled
g
e.
T
h
e
d
ataset
is
th
en
d
iv
id
e
d
in
t
o
tr
ain
i
n
g
an
d
test
in
g
s
ets.
T
h
e
m
o
d
el
co
n
f
ig
u
r
atio
n
i
n
clu
d
es
h
y
p
e
r
p
ar
am
eter
tu
n
in
g
as
w
ell
as
r
ef
in
em
en
t
o
f
an
c
h
o
r
b
o
x
es
an
d
p
r
ed
ictio
n
lay
er
s
with
in
th
e
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
(
R
PN)
.
T
h
e
m
o
d
el
is
tr
ain
ed
iter
ativ
ely
u
s
in
g
v
ar
io
u
s
co
n
f
ig
u
r
atio
n
s
.
I
ts
p
er
f
o
r
m
a
n
ce
is
ev
al
u
ated
u
s
in
g
m
ea
n
av
e
r
ag
e
p
r
ec
is
io
n
(
m
AP)
,
wh
ich
r
ef
lects
th
e
ac
cu
r
ac
y
o
f
o
b
ject
d
etec
tio
n
a
cr
o
s
s
d
if
f
er
en
t r
ec
all
lev
els in
m
ed
ical
im
ag
e
an
aly
s
is
.
Fig
u
r
e
1
.
T
h
e
f
lo
wch
a
r
t o
f
t
h
e
AI
-
b
ased
m
o
d
els an
d
ex
p
er
im
en
tal
m
eth
o
d
s
ap
p
lied
2
.
1
.
Da
t
a
a
cquis
it
io
n
T
h
e
in
itial
p
h
ase
o
f
th
is
s
tu
d
y
b
eg
an
with
th
e
ac
q
u
is
itio
n
o
f
f
etal
ec
h
o
ca
r
d
i
o
g
r
a
p
h
y
v
id
eo
s
o
b
tain
ed
f
r
o
m
au
th
o
r
ized
o
n
lin
e
s
o
u
r
ce
s
[
3
4
]
.
T
h
ese
v
id
eo
s
d
is
p
lay
t
h
e
f
etal
h
ea
r
t
f
r
o
m
th
e
A
4
C
p
er
s
p
ec
tiv
e
an
d
ar
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.
1
4
,
No
.
5
,
Octo
b
er
2
0
2
5
:
4
1
3
7
-
4
1
5
0
4140
p
r
o
v
id
e
d
in
.
m
p
4
f
o
r
m
at,
with
a
f
ile
s
ize
o
f
1
3
.
7
MB,
a
d
u
r
atio
n
o
f
1
7
8
s
ec
o
n
d
s
,
an
d
a
f
r
am
e
r
ate
o
f
3
0
f
p
s
.
T
h
e
en
tire
v
id
e
o
was
c
o
n
v
er
te
d
in
to
two
-
d
im
e
n
s
io
n
al
im
a
g
e
s
with
a
r
eso
lu
tio
n
o
f
1
2
8
0
×7
2
0
p
ix
els
th
r
o
u
g
h
a
f
r
am
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
.
2
.
2
.
Da
t
a
pre
-
pro
ce
s
s
ing
A
f
t
e
r
th
e
f
r
a
m
e
e
x
tr
a
c
t
io
n
p
r
o
c
e
s
s
,
a
p
r
e
p
r
o
c
e
s
s
i
n
g
s
t
a
g
e
w
a
s
c
a
r
r
i
e
d
o
u
t
t
o
f
i
l
t
e
r
an
d
s
e
l
e
c
t
r
e
l
e
v
an
t
i
m
a
g
e
s
,
e
n
s
u
r
i
n
g
t
h
e
q
u
a
l
i
ty
o
f
t
h
e
d
a
ta
u
s
ed
f
o
r
m
o
d
e
l
t
r
a
in
i
n
g
.
T
h
i
s
s
t
a
g
e
c
o
n
s
i
s
t
s
o
f
t
h
r
e
e
m
a
i
n
s
t
e
p
s
:
f
i
l
t
e
r
in
g
,
c
r
o
p
p
i
n
g
,
a
n
d
r
e
s
i
z
in
g
.
F
i
l
t
e
r
in
g
w
a
s
p
e
r
f
o
r
m
e
d
t
o
r
e
t
a
i
n
o
n
l
y
th
e
im
a
g
e
s
th
a
t
c
l
e
ar
l
y
d
e
p
i
c
t
f
e
t
al
h
e
a
r
t
s
tr
u
c
tu
r
e
s
[
3
5
]
.
C
r
o
p
p
i
n
g
w
a
s
ap
p
l
ie
d
to
f
o
cu
s
o
n
th
e
r
e
g
io
n
s
c
o
n
t
a
i
n
i
n
g
th
e
f
e
t
a
l
h
e
ar
t
;
i
n
s
o
m
e
c
a
s
e
s
,
m
u
l
t
i
p
l
e
cr
o
p
s
w
e
r
e
t
ak
e
n
f
r
o
m
a
s
i
n
g
l
e
im
a
g
e
i
f
i
t
c
o
n
t
a
in
e
d
m
o
r
e
th
a
n
o
n
e
f
e
t
a
l
h
ea
r
t
o
b
j
e
c
t
.
F
i
n
a
l
ly
,
r
e
s
i
z
i
n
g
w
a
s
p
e
r
f
o
r
m
e
d
t
o
s
t
an
d
ar
d
i
z
e
t
h
e
i
m
a
g
e
d
i
m
e
n
s
i
o
n
s
,
w
i
t
h
a
l
l
i
m
a
g
e
s
r
e
s
i
z
e
d
to
4
0
0
×
3
0
0
p
i
x
e
l
s
.
2
.
3
.
Da
t
a
la
belin
g
Su
b
s
eq
u
en
tly
,
t
h
e
s
elec
ted
n
o
r
m
al
f
etal
h
ea
r
t
im
ag
es
wer
e
m
an
u
ally
a
n
n
o
tated
b
y
f
etal
c
ar
d
io
lo
g
y
ex
p
er
ts
u
s
in
g
a
s
p
ec
ialized
g
r
ap
h
ical
an
n
o
tatio
n
to
o
l,
n
am
el
y
th
e
m
ak
esen
s
e
ap
p
licatio
n
[
3
6
]
.
T
h
e
an
n
o
tatio
n
p
r
o
ce
s
s
was
co
n
d
u
cted
i
n
d
iv
i
d
u
ally
f
o
r
ea
ch
im
a
g
e,
g
u
id
e
d
b
y
ex
p
er
t
k
n
o
wled
g
e
o
f
f
et
al
ca
r
d
iac
a
n
ato
m
y
.
T
h
e
an
n
o
tated
o
b
jects
i
n
clu
d
e
d
:
L
A,
R
A,
L
V,
R
V,
T
V,
PV,
MV
,
AV,
Ao
,
an
d
s
p
in
e.
T
h
e
an
n
o
tatio
n
r
esu
lts
wer
e
s
av
ed
in
J
SON
f
o
r
m
at
an
d
s
er
v
ed
as th
e
g
r
o
u
n
d
tr
u
th
f
o
r
m
o
d
el
tr
ai
n
in
g
.
2
.
4
.
Da
t
a
s
pli
t
t
ing
Fo
llo
win
g
th
e
an
n
o
tatio
n
p
r
o
ce
s
s
,
th
e
d
ataset
was
d
iv
id
ed
in
to
two
p
r
im
a
r
y
s
u
b
s
ets:
tr
ain
in
g
d
ata
an
d
v
alid
atio
n
d
ata,
u
s
in
g
an
8
0
:2
0
s
p
lit
r
atio
.
T
h
e
s
p
litt
in
g
was
p
er
f
o
r
m
ed
r
a
n
d
o
m
ly
wh
i
le
en
s
u
r
in
g
th
at
th
e
class
d
is
tr
ib
u
tio
n
r
em
ain
ed
b
a
lan
ce
d
ac
r
o
s
s
b
o
th
s
u
b
s
ets.
T
h
is
ap
p
r
o
ac
h
allo
ws
th
e
m
o
d
e
l
to
lear
n
f
r
o
m
th
e
m
ajo
r
ity
o
f
th
e
av
ailab
le
d
at
a
wh
ile
r
eser
v
i
n
g
a
p
o
r
tio
n
f
o
r
ev
al
u
atin
g
its
g
en
er
aliza
ti
o
n
p
er
f
o
r
m
an
ce
o
n
u
n
s
ee
n
s
am
p
les.
Su
c
h
a
s
tr
ateg
y
is
c
o
m
m
o
n
ly
e
m
p
lo
y
e
d
in
DL
wo
r
k
f
l
o
ws
to
p
r
ev
e
n
t
o
v
e
r
f
itti
n
g
an
d
e
n
s
u
r
e
an
u
n
b
iased
p
er
f
o
r
m
an
ce
ass
ess
m
en
t.
2
.
5
.
Co
nfi
g
ura
t
io
n
Prio
r
to
tr
ain
in
g
,
a
h
y
p
er
p
ar
a
m
eter
tu
n
in
g
p
r
o
ce
s
s
was
co
n
d
u
cted
,
in
clu
d
in
g
th
e
co
n
f
ig
u
r
atio
n
o
f
an
ch
o
r
b
o
x
es,
lear
n
in
g
r
ate,
b
atch
s
ize,
an
d
n
u
m
b
e
r
o
f
e
p
o
ch
s
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
was
d
ev
elo
p
ed
a
n
d
tr
ain
ed
o
n
a
co
m
p
u
ter
e
q
u
ip
p
ed
with
an
I
n
tel
C
o
r
e
i3
-
4
1
7
0
C
PU
@
3
.
7
0
GHz
(
4
C
PUs
)
,
8
GB
o
f
R
AM
,
an
d
an
Nv
id
ia
GeFo
r
ce
GT
X
1
0
5
0
T
i
GPU
f
ea
tu
r
in
g
7
6
8
C
UDA
co
r
es,
a
GPU
clo
ck
s
p
ee
d
o
f
1
3
9
2
/1
5
0
6
MH
z,
4
GB
o
f
GDDR5
m
em
o
r
y
,
a
n
d
a
m
em
o
r
y
b
an
d
wid
t
h
o
f
1
1
2
.
1
GB
/s
.
T
h
e
p
r
o
g
r
a
m
m
in
g
lan
g
u
a
g
e
u
s
ed
was
Py
th
o
n
3
.
6
.
1
3
,
with
T
en
s
o
r
Flo
w
1
.
1
4
.
0
,
Ker
as 2
.
3
.
1
,
a
n
d
Pro
to
b
u
f
3
.
1
9
.
6
lib
r
a
r
ies.
2
.
6
.
I
ns
t
a
nce
s
eg
m
ent
a
t
i
o
n
I
n
th
e
s
u
b
s
eq
u
en
t
s
tag
e,
th
e
Ma
s
k
r
eg
io
n
-
b
ased
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
Ma
s
k
R
-
C
NN
)
[
3
7
]
in
s
tan
ce
s
eg
m
en
tatio
n
m
o
d
el
is
em
p
lo
y
ed
.
T
h
is
m
o
d
el
c
o
n
s
is
ts
o
f
s
ev
er
al
k
ey
co
m
p
o
n
en
ts
:
a
b
ac
k
b
o
n
e
n
etwo
r
k
(
R
esNet5
0
)
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
a
R
PN
f
o
r
g
en
e
r
atin
g
ca
n
d
id
ate
o
b
ject
r
e
g
io
n
s
,
r
eg
io
n
o
f
in
ter
est
(
R
OI
)
alig
n
s
f
o
r
alig
n
i
n
g
p
r
o
p
o
s
ed
r
eg
io
n
s
with
th
e
f
ea
tu
r
e
m
ap
s
,
f
u
lly
c
o
n
n
ec
ted
la
y
er
s
f
o
r
b
o
u
n
d
in
g
b
o
x
class
if
icatio
n
an
d
r
eg
r
ess
io
n
;
an
d
a
f
u
lly
co
n
v
o
lu
tio
n
al
n
et
wo
r
k
(
FC
N)
f
o
r
g
en
er
atin
g
b
i
n
ar
y
m
ask
s
o
f
ea
ch
d
etec
ted
o
b
ject.
T
h
e
m
o
d
el
is
s
p
ec
if
ically
d
esig
n
ed
to
p
er
f
o
r
m
s
eg
m
e
n
tatio
n
o
f
th
e
n
o
r
m
al
f
etal
h
ea
r
t
an
ato
m
y
b
ased
o
n
th
e
A4
C
v
iew.
Ma
s
k
R
-
C
NN
was
s
el
ec
ted
d
u
e
to
its
ab
ilit
y
to
p
er
f
o
r
m
b
o
t
h
o
b
ject
d
etec
tio
n
an
d
i
n
s
tan
ce
-
lev
el
s
eg
m
en
tatio
n
with
h
i
g
h
ac
cu
r
a
cy
.
Ma
s
k
R
-
C
NN
o
f
f
er
s
b
etter
p
er
f
o
r
m
an
ce
o
n
m
ed
ical
im
ag
in
g
d
atasets
with
lim
ited
d
ata
an
d
c
o
m
p
lex
o
b
ject
b
o
u
n
d
ar
ies.
No
s
tr
u
ct
u
r
al
m
o
d
if
icatio
n
s
wer
e
m
ad
e
to
th
e
o
r
ig
in
al
ar
ch
itec
tu
r
e,
b
u
t
m
o
d
el
p
er
f
o
r
m
a
n
ce
was
o
p
tim
ized
th
r
o
u
g
h
h
y
p
er
p
ar
am
eter
tu
n
i
n
g
s
p
ec
if
ic
to
f
etal
h
ea
r
t im
ag
e
c
h
ar
ac
ter
is
tics
.
2
.
7
.
E
v
a
lua
t
i
o
n m
et
rics
T
o
ev
alu
ate
th
e
o
v
er
all
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el,
a
s
p
ec
if
i
c
m
etr
ic
ca
lled
m
AP
was
u
s
ed
.
m
AP
is
a
wid
ely
em
p
lo
y
e
d
m
etr
ic
f
o
r
a
s
s
es
s
in
g
th
e
q
u
ality
o
f
o
b
ject
d
etec
to
r
s
.
T
h
is
m
etr
ic
m
ea
s
u
r
es
th
e
ac
cu
r
ac
y
o
f
th
e
m
o
d
el
in
d
etec
tin
g
o
b
jec
ts
b
y
ca
lcu
latin
g
th
e
av
er
a
g
e
p
r
ec
is
io
n
(
AP)
f
o
r
ea
c
h
cla
s
s
.
m
AP
p
r
o
v
id
es
v
alu
ab
le
in
s
ig
h
ts
in
to
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
DL
m
o
d
el
in
th
e
task
o
f
d
etec
tin
g
f
etal
h
ea
r
t
o
b
jects.
T
o
o
b
tai
n
th
e
m
AP
v
al
u
e
[
1
7
]
,
th
e
AP
i
s
f
ir
s
t
ca
lcu
lated
b
y
co
m
b
in
in
g
p
r
ec
is
io
n
an
d
r
ec
all
at
v
ar
io
u
s
th
r
esh
o
ld
lev
els.
T
h
e
eq
u
atio
n
s
f
o
r
AP a
n
d
m
A
P a
r
e
p
r
o
v
i
d
ed
in
(
1
)
an
d
(
2
)
,
r
esp
ec
tiv
ely
:
=
∑
(
×
∆
)
(
1
)
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
Mu
lticla
s
s
in
s
ta
n
ce
s
eg
men
ta
t
io
n
o
p
timiz
a
tio
n
fo
r
feta
l h
e
a
r
t ima
g
e
o
b
ject
in
terp
r
eta
tio
n
(
Ha
d
i S
ya
p
u
tr
a
)
4141
w
h
er
e
p
r
ec
is
io
n
at
r
ec
all
p
o
i
n
t
k
is
th
e
p
r
ec
is
io
n
v
alu
e
at
a
s
p
ec
if
ic
r
ec
all;
a
n
d
∆
is
th
e
ch
an
g
e
in
r
ec
all
b
etwe
en
two
ad
jace
n
t
r
e
ca
ll p
o
in
ts
.
=
1
∑
=
1
(
2
)
w
h
er
e
N
is
th
e
n
u
m
b
er
o
f
class
es o
r
o
b
jects; an
d
is
th
e
AP
f
o
r
th
e
i
-
th
class
.
T
o
ca
lcu
late
p
r
ec
is
io
n
an
d
r
e
ca
ll,
u
s
e
(
3
)
an
d
(
4
)
.
Pre
cisi
o
n
m
ea
s
u
r
es
h
o
w
m
an
y
o
f
th
e
p
r
ed
icted
p
o
s
itiv
e
ca
s
es
ar
e
tr
u
ly
p
o
s
itiv
e,
an
d
it
d
ec
r
ea
s
es
wh
en
th
er
e
ar
e
m
an
y
f
alse
p
o
s
itiv
es.
R
ec
all
in
d
icate
s
h
o
w
m
an
y
ac
tu
al
p
o
s
itiv
e
ca
s
es
ar
e
co
r
r
ec
tly
d
etec
ted
,
an
d
it
d
e
cr
ea
s
es
with
h
ig
h
f
alse
n
eg
ativ
es.
T
o
g
eth
er
,
th
ese
v
alu
es
d
eter
m
in
e
AP,
wh
ich
is
th
en
av
er
ag
ed
to
co
m
p
u
te
m
AP,
g
iv
in
g
a
r
o
b
u
s
t
o
v
er
all
m
ea
s
u
r
e
o
f
o
b
ject
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
.
=
+
(
3
)
=
+
(
4
)
w
h
er
e
P is
p
r
ec
is
io
n
;
R
is
r
ec
all,
T
P
is
tr
u
e
p
o
s
itiv
e;
an
d
F
P
is
f
alse p
o
s
itiv
e.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
P
re
-
pro
ce
s
s
ing
o
f
no
rma
l f
et
a
l hea
rt
im
a
g
e
da
t
a
Fo
llo
win
g
th
e
p
r
e
p
r
o
ce
s
s
in
g
t
o
o
b
tain
f
etal
h
ea
r
t
im
ag
es,
th
e
p
r
o
ce
s
s
in
v
o
lv
e
d
co
n
v
er
tin
g
u
ltra
s
o
u
n
d
v
id
eo
s
in
to
s
till
im
ag
es,
r
esu
ltin
g
in
a
to
tal
o
f
3
5
7
im
a
g
es.
T
h
ese
im
ag
es
in
clu
d
e
th
o
s
e
s
h
o
win
g
f
etal
h
ea
r
t
o
b
jects,
with
s
o
m
e
im
ag
es
co
n
tain
in
g
o
n
e
,
two
,
o
r
th
r
ee
f
e
tal
h
ea
r
t
o
b
jects.
Ad
d
itio
n
ally
,
th
er
e
ar
e
im
ag
es
th
at
d
o
n
o
t
s
h
o
w
an
y
f
etal
h
ea
r
t
o
b
jects
an
d
th
o
s
e
wh
er
e
th
e
f
etal
h
ea
r
t
o
b
jects
ar
e
o
u
t
o
f
f
o
cu
s
o
r
b
lu
r
r
e
d
,
as
illu
s
tr
ated
in
Fig
u
r
e
2
.
Fo
r
im
ag
es
co
n
tain
in
g
m
u
ltip
le
f
etal
h
ea
r
t
o
b
jects
o
r
wh
e
r
e
o
th
er
tex
t
o
r
o
b
jects
ar
e
p
r
esen
t
in
th
e
im
ag
e
,
cr
o
p
p
in
g
is
p
er
f
o
r
m
ed
to
e
n
s
u
r
e
th
at
t
h
e
d
ata
u
s
ed
f
o
r
th
e
i
n
s
tan
ce
s
eg
m
en
tatio
n
m
o
d
el
m
ee
ts
th
e
s
p
ec
if
ic
r
eq
u
ir
e
m
en
ts
.
T
h
is
p
r
o
ce
s
s
alig
n
s
with
th
e
s
tep
s
o
u
tlin
ed
in
m
eth
o
d
s
ec
tio
n
.
T
h
e
o
u
tp
u
t
o
f
th
e
v
id
eo
e
x
tr
ac
tio
n
p
r
o
ce
s
s
a
n
d
th
e
r
esu
ltin
g
im
ag
es a
r
e
s
u
m
m
ar
ized
in
T
a
b
le
1
.
T
ab
le
1
.
Vid
eo
ex
tr
ac
tio
n
No
I
mag
e
t
y
p
e
N
u
mb
e
r
o
f
e
x
t
r
a
c
t
e
d
i
m
a
g
e
s
1.
I
mag
e
s s
h
o
w
i
n
g
f
e
t
a
l
h
e
a
r
t
o
b
j
e
c
t
s
1
1
4
2.
I
mag
e
sh
o
w
i
n
g
m
u
l
t
i
p
l
e
f
e
t
a
l
h
e
a
r
t
o
b
j
e
c
t
s
50
3.
I
mag
e
s s
h
o
w
i
n
g
f
e
t
a
l
h
e
a
r
t
o
b
j
e
c
t
s
b
u
t
o
u
t
o
f
f
o
c
u
s
1
0
5
4.
I
mag
e
s n
o
t
s
h
o
w
i
n
g
a
n
y
f
e
t
a
l
h
e
a
r
t
o
b
j
e
c
t
s
88
To
t
a
l
3
5
7
T
ab
le
1
p
r
esen
ts
th
e
r
esu
lts
o
f
im
ag
e
ex
tr
ac
tio
n
f
r
o
m
f
etal
h
e
ar
t e
x
am
in
atio
n
v
id
e
o
s
,
ca
teg
o
r
ized
in
to
f
o
u
r
m
ain
g
r
o
u
p
s
b
ased
o
n
th
e
q
u
ality
an
d
p
r
esen
ce
o
f
f
etal
h
ea
r
t
s
tr
u
ct
u
r
es.
A
to
tal
o
f
3
5
7
im
ag
es
with
a
r
eso
lu
tio
n
o
f
1
2
8
0
×
7
2
0
p
ix
el
s
wer
e
o
b
tain
ed
.
Mo
s
t
o
f
th
e
im
ag
es
co
n
tain
f
etal
h
ea
r
t
o
b
jects
with
v
ar
y
in
g
lev
els
o
f
clar
ity
an
d
o
b
ject
co
u
n
t,
wh
ile
o
th
er
s
lack
r
elev
an
t
f
ea
tu
r
es
f
o
r
f
u
r
th
er
an
al
y
s
is
.
T
h
is
clas
s
if
icatio
n
s
u
p
p
o
r
ts
th
e
s
elec
tio
n
o
f
s
u
itab
le
im
ag
es f
o
r
t
h
e
an
n
o
tatio
n
an
d
m
o
d
el
tr
ain
in
g
s
tag
es.
Vis
u
ally
,
Fig
u
r
e
2
illu
s
tr
ates
f
o
u
r
m
ai
n
ca
teg
o
r
ies
r
esu
lti
n
g
f
r
o
m
th
e
im
a
g
e
e
x
tr
ac
tio
n
p
r
o
ce
s
s
.
Fig
u
r
e
2
(
a)
d
is
p
lay
s
im
ag
es
th
at
d
o
n
o
t
d
is
p
lay
an
y
f
etal
h
ea
r
t
o
b
ject,
Fig
u
r
e
2
(
b
)
p
r
esen
ts
im
ag
es
th
at
co
n
tain
a
f
etal
h
ea
r
t
o
b
ject
b
u
t
ar
e
o
u
t
o
f
f
o
cu
s
,
Fig
u
r
e
2
(
c)
s
h
o
ws
im
ag
es
th
at
clea
r
ly
s
h
o
w
a
s
in
g
le
f
etal
h
ea
r
t
o
b
ject,
an
d
Fig
u
r
e
2
(
d
)
p
r
esen
ts
im
ag
es
th
at
d
is
p
lay
m
u
ltip
le
f
etal
h
ea
r
t
o
b
jects
wi
th
in
a
s
in
g
le
f
r
am
e.
T
h
ese
ca
teg
o
r
ies
ar
e
d
er
i
v
ed
f
r
o
m
th
e
v
id
eo
-
to
-
im
ag
e
co
n
v
er
s
io
n
p
r
o
ce
s
s
an
d
will
s
u
b
s
eq
u
en
tly
u
n
d
e
r
g
o
p
r
ep
r
o
ce
s
s
in
g
as p
ar
t o
f
t
h
e
d
a
taset p
r
ep
ar
atio
n
f
o
r
tr
ain
i
n
g
t
h
e
s
eg
m
en
tatio
n
m
o
d
el.
Fo
llo
win
g
th
e
cr
o
p
p
i
n
g
an
d
s
elec
tio
n
p
r
o
ce
s
s
f
o
r
im
ag
es
d
is
p
lay
in
g
f
etal
h
ea
r
t
o
b
jects
,
th
e
to
tal
n
u
m
b
er
o
f
im
a
g
es
was
r
ed
u
c
ed
to
1
7
6
,
wh
ich
alig
n
s
with
th
e
r
eq
u
ir
e
m
en
ts
f
o
r
th
e
in
s
tan
ce
s
eg
m
en
tatio
n
m
o
d
el,
as
s
h
o
wn
in
Fig
u
r
e
3
.
Af
ter
o
b
tain
in
g
th
e
f
etal
h
ea
r
t im
ag
es,
th
e
n
ex
t
s
tep
in
v
o
l
v
ed
s
ca
lin
g
th
e
im
ag
es
to
en
s
u
r
e
u
n
if
o
r
m
s
ize
ac
r
o
s
s
th
e
d
ataset.
T
h
e
s
ca
lin
g
p
r
o
ce
s
s
was
co
n
d
u
cted
as
d
escr
ib
ed
in
th
e
m
eth
o
d
s
ec
tio
n
,
with
im
ag
es
r
esized
t
o
4
0
0
×3
0
0
p
ix
els.
Fo
llo
win
g
t
h
is
,
all
n
o
r
m
al
f
etal
h
ea
r
t
im
ag
es
wer
e
an
n
o
tated
with
ten
lab
els
co
r
r
esp
o
n
d
in
g
to
th
e
an
at
o
m
ical
f
ea
tu
r
es
o
f
th
e
f
etal
h
ea
r
t.
T
h
is
an
n
o
tatio
n
was
p
er
f
o
r
m
ed
u
s
in
g
p
o
ly
g
o
n
p
o
in
ts
o
n
th
e
f
etal
h
ea
r
t
o
b
ject
im
ag
es.
T
h
e
an
n
o
tatio
n
p
r
o
ce
s
s
is
illu
s
tr
ated
in
Fig
u
r
e
4
.
T
h
e
f
in
al
an
n
o
tated
f
etal
h
ea
r
t
im
ag
es we
r
e
ex
p
o
r
ted
in
J
SON
f
ile
f
o
r
m
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.
1
4
,
No
.
5
,
Octo
b
er
2
0
2
5
:
4
1
3
7
-
4
1
5
0
4142
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
2
.
Fo
u
r
im
ag
e
ca
teg
o
r
i
es f
r
o
m
th
e
e
x
tr
ac
tio
n
p
r
o
ce
s
s
of
(
a)
n
o
t sh
o
win
g
an
y
f
etal
h
e
ar
t
o
b
jects
,
(
b
)
s
h
o
win
g
f
etal
h
ea
r
t o
b
jects b
u
t o
u
t
o
f
f
o
cu
s
,
(
c
)
s
h
o
win
g
a
f
etal
h
ea
r
t o
b
ject
,
an
d
(
d
)
s
h
o
win
g
m
u
ltip
le
f
etal
h
ea
r
t o
b
ject
Fig
u
r
e
3
.
Fetal
h
ea
r
t im
a
g
es f
r
o
m
th
e
A4
C
Fig
u
r
e
4
(
a
)
r
e
p
r
esen
ts
th
e
a
n
a
to
m
ical
lo
ca
tio
n
o
f
t
h
e
f
etal
h
ea
r
t
th
at
h
as
b
ee
n
d
eter
m
in
ed
b
ased
o
n
ex
p
er
t
d
esig
n
atio
n
,
b
u
t
h
as
n
o
t
g
o
n
e
th
r
o
u
g
h
th
e
AI
-
b
ased
m
o
d
elin
g
s
tag
e.
T
h
is
id
e
n
tific
atio
n
is
d
o
n
e
m
an
u
ally
b
y
th
e
r
ad
io
lo
g
is
t
o
r
s
p
ec
ialis
t
b
y
co
n
s
id
er
in
g
th
e
v
is
u
al
c
h
ar
ac
ter
is
tics
s
ee
n
o
n
th
e
u
ltra
s
o
u
n
d
im
ag
e.
T
h
e
lo
ca
tio
n
o
f
an
ato
m
ical
s
tr
u
ctu
r
es
in
th
is
im
ag
e
s
er
v
es
as
th
e
g
r
o
u
n
d
tr
u
t
h
,
wh
ich
b
ec
o
m
es
th
e
r
ef
er
en
ce
in
f
u
r
th
er
a
n
n
o
tati
o
n
an
d
m
o
d
elin
g
s
tag
es.
Me
an
wh
ile,
Fig
u
r
e
4
(
b
)
is
th
e
r
esu
lt
o
f
an
n
o
tatio
n
p
er
f
o
r
m
ed
u
s
in
g
an
n
o
tatio
n
t
o
o
ls
,
wh
er
e
ea
c
h
f
etal
h
ea
r
t
s
tr
u
ctu
r
e
h
as
b
ee
n
lab
ele
d
with
a
co
lo
r
m
ask
a
n
d
b
o
u
n
d
in
g
b
o
x
to
d
is
tin
g
u
is
h
s
p
ec
if
ic
ar
ea
s
.
T
h
is
an
n
o
tatio
n
is
an
im
p
o
r
tan
t
p
ar
t
o
f
p
r
ep
a
r
in
g
th
e
d
ataset
f
o
r
tr
ain
in
g
AI
-
b
ased
s
eg
m
en
tatio
n
m
o
d
els.
(
a)
(
b
)
Fig
u
r
e
4
.
An
n
o
tatio
n
o
f
f
etal
h
ea
r
t im
ag
es (
a)
o
r
ig
in
al
im
ag
e
with
m
an
u
al
id
en
tific
atio
n
an
d
(
b
)
an
n
o
tated
im
ag
e
with
co
lo
r
m
ask
s
an
d
b
o
u
n
d
in
g
b
o
x
es
Af
ter
th
e
an
n
o
tatio
n
p
h
ase
is
co
m
p
lete,
t
h
e
J
SON
an
n
o
tat
io
n
f
iles
ar
e
p
air
ed
with
th
e
an
n
o
tated
im
ag
es.
T
h
is
co
m
b
in
ed
d
ataset
is
th
en
u
s
ed
to
tr
ain
th
e
in
s
tan
ce
s
eg
m
en
tatio
n
m
o
d
el
f
o
r
f
etal
h
ea
r
t
im
ag
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
Mu
lticla
s
s
in
s
ta
n
ce
s
eg
men
ta
t
io
n
o
p
timiz
a
tio
n
fo
r
feta
l h
e
a
r
t ima
g
e
o
b
ject
in
terp
r
eta
tio
n
(
Ha
d
i S
ya
p
u
tr
a
)
4143
o
b
jects.
A
s
am
p
le
o
f
th
e
an
n
o
tatio
n
r
esu
lts
i
s
s
h
o
wn
in
F
ig
u
r
e
5
.
Fig
u
r
e
5
s
h
o
ws
th
e
r
esu
lts
o
f
g
r
o
u
n
d
tr
u
th
an
n
o
tatio
n
f
o
r
s
eg
m
en
tatio
n
o
f
an
ato
m
ical
s
tr
u
ctu
r
es in
f
etal
h
ea
r
t u
ltra
s
o
u
n
d
im
ag
es.
Fig
u
r
e
5
(
a)
d
is
p
lay
s
th
e
o
r
ig
in
al
u
ltra
s
o
u
n
d
im
ag
e
,
wh
ile
Fig
u
r
e
s
5
(
b
)
to
5
(
k
)
r
ep
r
esen
t
th
e
m
an
u
ally
an
n
o
tate
d
s
eg
m
en
tatio
n
o
f
v
ar
io
u
s
h
ea
r
t
s
tr
u
ct
u
r
es.
T
h
e
s
tr
u
ctu
r
es
s
h
o
wn
in
clu
d
e
Ao
,
AV,
L
A,
L
V,
MV
,
PV,
R
A,
R
V,
s
p
in
e
,
an
d
T
V.
T
h
e
m
ask
in
g
v
is
u
alize
d
in
Fig
u
r
e
s
5
(
b
)
to
5
(
k
)
s
h
o
ws
th
e
ar
ea
s
id
en
tifie
d
as
p
ar
t
o
f
ea
ch
a
n
ato
m
ical
s
tr
u
ctu
r
e
b
ased
o
n
th
e
g
r
o
u
n
d
tr
u
th
an
n
o
tatio
n
s
.
T
h
is
im
ag
e
is
g
en
er
a
ted
f
r
o
m
an
n
o
tated
d
ata
in
J
SON
f
o
r
m
at
im
p
o
r
ted
in
to
Py
th
o
n
co
d
e
an
d
v
is
u
alize
d
u
s
in
g
im
a
g
e
p
r
o
ce
s
s
in
g
lib
r
ar
ies
s
u
ch
as
Op
en
C
V
o
r
Ma
t
p
lo
tlib
.
T
h
e
p
r
o
ce
s
s
in
v
o
lv
es m
ap
p
in
g
th
e
J
SON
d
ata
in
to
an
ar
r
ay
o
f
b
in
ar
y
im
a
g
es f
o
r
ea
ch
an
ato
m
ical
s
tr
u
ctu
r
e,
th
en
v
is
u
alize
d
ag
ain
s
t a
b
lu
e
b
ac
k
g
r
o
u
n
d
to
c
lar
if
y
th
e
s
eg
m
en
te
d
p
ar
ts
.
(a
)
(b
)
(c
)
(d
)
(e
)
(f)
(g
)
(h
)
(i)
(j)
(k
)
Fig
u
r
e
5
.
Gr
o
u
n
d
tr
u
th
o
f
a
n
n
o
tatio
n
r
esu
lts
(
a)
o
r
ig
in
al
i
m
a
g
e
,
(
b
)
Ao
,
(
c
)
AV
,
(
d
)
L
A
,
(
e)
L
V
,
(
f
)
MV
,
(
g
)
PV
,
(
h
)
R
A
,
(
i)
R
V
,
(
j)
s
p
in
e
,
an
d
(
k
)
T
V
3
.
2
.
Sp
litt
ing
da
t
a
T
h
is
s
tu
d
y
u
tili
ze
s
a
d
ata
s
et
t
h
at
is
d
iv
id
ed
in
to
two
p
ar
ts
:
th
e
tr
ain
in
g
s
et
an
d
th
e
v
alid
atio
n
s
et.
T
h
e
tr
ain
in
g
s
et
is
u
s
ed
t
o
tr
ain
th
e
m
o
d
el
,
wh
ile
th
e
v
alid
atio
n
s
et
is
u
s
ed
t
o
ev
alu
ate
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
o
n
d
ata
th
at
was
n
o
t
s
ee
n
d
u
r
i
n
g
th
e
tr
ai
n
in
g
p
r
o
ce
s
s
.
Ou
t
o
f
th
e
to
tal
1
7
6
im
ag
es,
th
e
d
ataset
is
s
p
lit in
to
1
4
0
im
a
g
es f
o
r
t
h
e
tr
ain
in
g
s
et
an
d
3
6
im
ag
es f
o
r
t
h
e
v
alid
atio
n
s
et,
as sh
o
wn
i
n
T
ab
le
2
.
T
ab
le
2
.
Data
s
et
s
p
lit f
o
r
tr
ain
i
n
g
an
d
v
alid
atio
n
s
ets
No
D
a
t
a
N
u
mb
e
r
o
f
i
m
a
g
e
s
1.
Tr
a
i
n
i
n
g
d
a
t
a
1
4
0
2.
V
a
l
i
d
a
t
i
o
n
d
a
t
a
36
3
.
3
.
M
o
del seg
m
ent
a
t
io
n desi
g
n
T
h
is
s
tu
d
y
em
p
lo
y
s
th
e
Ma
s
k
R
-
C
NN
m
eth
o
d
,
o
p
tim
izin
g
th
e
m
o
d
el
b
y
f
in
e
-
tu
n
in
g
th
e
h
y
p
er
p
ar
am
eter
s
s
p
ec
if
ic
to
Ma
s
k
R
-
C
NN.
T
h
e
h
y
p
e
r
p
ar
am
eter
s
u
s
ed
ar
e
lis
ted
in
T
ab
le
3
.
T
h
ese
h
y
p
er
p
ar
am
eter
s
ar
e
cr
itical
f
o
r
im
p
r
o
v
in
g
ac
cu
r
ac
y
with
th
e
im
ag
e
d
ata.
T
h
e
s
elec
ted
h
y
p
er
p
ar
am
eter
s
r
esu
lt
in
2
4
m
o
d
el
co
m
b
in
atio
n
s
.
A
cc
o
r
d
in
g
to
T
a
b
le
4
,
th
e
p
ar
a
m
eter
s
in
clu
d
e
im
ag
e
s
ize,
lear
n
in
g
r
ate,
lear
n
i
n
g
m
o
m
en
tu
m
,
with
8
ep
o
c
h
s
an
d
5
0
0
s
tep
s
p
er
ep
o
c
h
,
a
R
esNet
-
5
0
b
ac
k
b
o
n
e
ar
c
h
itectu
r
e,
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
(
SGD
)
o
p
tim
izer
,
a
n
d
a
b
atch
s
ize
o
f
1
.
T
ab
le
3
.
Hy
p
er
p
ar
a
m
eter
s
u
s
ed
f
o
r
Ma
s
k
R
-
C
NN
m
o
d
el
tr
ai
n
in
g
No
C
o
n
f
i
g
u
r
a
t
i
o
n
s
H
y
p
e
r
p
a
r
a
me
t
e
r
1.
I
mag
e
si
z
e
[
6
4
,
1
2
8
,
2
5
6
,
5
1
2
]
2.
Le
a
r
n
i
n
g
r
a
t
e
[
0
.
0
1
,
0
.
0
0
1
,
0
.
0
0
0
1
]
3.
Le
a
r
n
i
n
g
mo
me
n
t
u
m
[
0
.
7
,
0
.
9
]
T
ab
le
4
p
r
esen
ts
2
4
co
m
b
in
atio
n
s
o
f
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
(
C
NN
)
ar
c
h
itectu
r
e
m
o
d
els
ex
p
er
im
en
ted
to
ev
alu
ate
th
e
im
p
ac
t
o
f
v
a
r
io
u
s
h
y
p
e
r
p
ar
a
m
eter
s
o
n
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
T
h
e
v
ar
ied
p
ar
am
eter
s
in
clu
d
e
in
p
u
t
im
a
g
e
s
ize,
lear
n
in
g
m
o
m
en
tu
m
,
an
d
lear
n
in
g
r
ate.
All
m
o
d
els
u
s
e
th
e
R
e
s
Net
-
50
b
ac
k
b
o
n
e,
o
p
tim
ized
with
th
e
SGD
alg
o
r
ith
m
,
an
d
tr
ain
ed
f
o
r
8
ep
o
c
h
s
with
5
0
0
s
tep
s
p
e
r
ep
o
ch
,
a
n
d
with
a
b
atch
s
ize
o
f
1
.
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.
1
4
,
No
.
5
,
Octo
b
er
2
0
2
5
:
4
1
3
7
-
4
1
5
0
4144
T
ab
le
4
.
Mo
d
el
ex
p
e
r
im
en
tati
o
n
s
with
R
esNet
-
5
0
b
ac
k
b
o
n
e
an
d
SGD
o
p
tim
izer
M
o
d
e
l
S
i
z
e
Le
a
r
n
i
n
g
mo
me
n
t
u
m
Le
a
r
n
i
n
g
r
a
t
e
R
5
0
_
sg
d
_
1
(
6
4
,
6
4
)
0
.
7
0
.
0
1
R
5
0
_
sg
d
_
2
(
6
4
,
6
4
)
0
.
9
0
.
0
1
R
5
0
_
sg
d
_
3
(
6
4
,
6
4
)
0
.
7
0
.
0
0
1
R
5
0
_
sg
d
_
4
(
6
4
,
6
4
)
0
.
9
0
.
0
0
1
R
5
0
_
sg
d
_
5
(
6
4
,
6
4
)
0
.
7
0
.
0
0
0
1
R
5
0
_
sg
d
_
6
(
6
4
,
6
4
)
0
.
9
0
.
0
0
0
1
R
5
0
_
sg
d
_
7
(
1
2
8
,
1
2
8
)
0
.
7
0
.
0
1
R
5
0
_
sg
d
_
8
(
1
2
8
,
1
2
8
)
0
.
9
0
.
0
1
R
5
0
_
sg
d
_
9
(
1
2
8
,
1
2
8
)
0
.
7
0
.
0
0
1
R
5
0
_
sg
d
_
1
0
(
1
2
8
,
1
2
8
)
0
.
9
0
.
0
0
1
R
5
0
_
sg
d
_
1
1
(
1
2
8
,
1
2
8
)
0
.
7
0
.
0
0
0
1
R
5
0
_
sg
d
_
1
2
(
1
2
8
,
1
2
8
)
0
.
9
0
.
0
0
0
1
R
5
0
_
sg
d
_
1
3
(
2
5
6
,
2
5
6
)
0
.
7
0
.
0
1
R
5
0
_
sg
d
_
1
4
(
2
5
6
,
2
5
6
)
0
.
9
0
.
0
1
R
5
0
_
sg
d
_
1
5
(
2
5
6
,
2
5
6
)
0
.
7
0
.
0
0
1
R
5
0
_
sg
d
_
1
6
(
2
5
6
,
2
5
6
)
0
.
9
0
.
0
0
1
R
5
0
_
sg
d
_
1
7
(
2
5
6
,
2
5
6
)
0
.
7
0
.
0
0
0
1
R
5
0
_
sg
d
_
1
8
(
2
5
6
,
2
5
6
)
0
.
9
0
.
0
0
0
1
R
5
0
_
sg
d
_
1
9
(
5
1
2
,
5
1
2
)
0
.
7
0
.
0
1
R
5
0
_
sg
d
_
2
0
(
5
1
2
,
5
1
2
)
0
.
9
0
.
0
1
R
5
0
_
sg
d
_
2
1
(
5
1
2
,
5
1
2
)
0
.
7
0
.
0
0
1
R
5
0
_
sg
d
_
2
2
(
5
1
2
,
5
1
2
)
0
.
9
0
.
0
0
1
R
5
0
_
sg
d
_
2
3
(
5
1
2
,
5
1
2
)
0
.
7
0
.
0
0
0
1
R
5
0
_
sg
d
_
2
4
(
5
1
2
,
5
1
2
)
0
.
9
0
.
0
0
0
1
3
.
4
.
Resul
t
s
o
f
M
a
s
k
R
-
CN
N
m
o
del o
ptim
iza
t
io
n
T
h
e
ev
al
u
atio
n
o
f
th
e
Ma
s
k
R
-
C
NN
m
o
d
el
was
co
n
d
u
cte
d
b
y
ca
lcu
latin
g
th
e
m
AP,
wh
ich
r
e
f
lects
th
e
o
v
er
all
ac
cu
r
ac
y
o
f
th
e
m
o
d
el.
T
h
e
r
esu
lts
o
f
th
e
m
o
d
el
ev
alu
atio
n
f
r
o
m
v
a
r
io
u
s
ex
p
e
r
im
en
ts
ar
e
d
etailed
in
T
ab
le
5
.
T
h
is
tab
le
p
r
esen
t
s
th
e
m
AP
r
esu
lts
f
o
r
ea
c
h
ca
teg
o
r
y
with
in
th
e
tr
ain
in
g
d
ataset,
in
d
icatin
g
t
h
at
th
e
m
o
d
el
h
as su
cc
ess
f
u
lly
lear
n
ed
to
r
ec
o
g
n
ize
all
class
es.
T
ab
le
5
.
AP a
n
d
m
AP f
o
r
tr
ai
n
in
g
d
ataset
o
r
i
g
in
al
M
o
d
e
l
A
P
(
I
o
U
)
=
5
0
mA
P
Ao
LA
LV
RV
RA
S
p
i
n
e
TV
MV
PV
AV
R
5
0
_
sg
d
_
1
0
.
0
0
0
0
.
2
7
8
0
.
2
4
4
0
.
2
7
8
0
.
2
7
8
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
1
0
7
8
R
5
0
_
sg
d
_
2
0
.
0
0
0
0
.
0
0
0
0
.
2
2
2
0
.
2
4
7
0
.
2
2
2
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
6
9
1
R
5
0
_
sg
d
_
3
0
.
0
0
0
0
.
0
5
6
0
.
1
8
8
0
.
2
2
2
0
.
2
2
2
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
6
8
8
R
5
0
_
sg
d
_
4
0
.
0
0
0
0
.
1
3
9
0
.
2
7
8
0
.
1
9
1
0
.
2
5
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
8
5
8
R
5
0
_
sg
d
_
5
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
R
5
0
_
sg
d
_
6
0
.
0
0
0
0
.
0
0
0
0
.
1
3
9
0
.
0
0
0
0
.
0
2
8
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
1
6
7
R
5
0
_
sg
d
_
7
0
.
1
3
9
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
2
2
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
1
4
7
3
R
5
0
_
sg
d
_
8
0
.
2
4
4
0
.
2
7
8
0
.
2
7
8
0
.
2
5
0
0
.
2
7
8
0
.
2
5
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
1
5
7
8
R
5
0
_
sg
d
_
9
0
.
0
0
0
0
.
1
9
4
0
.
2
7
8
0
.
2
5
0
0
.
2
7
8
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
1
R
5
0
_
sg
d
_
1
0
0
.
0
2
8
0
.
2
7
8
0
.
2
7
8
0
.
2
5
0
0
.
2
7
8
0
.
1
6
3
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
1
2
7
5
R
5
0
_
sg
d
_
1
1
0
.
0
0
0
0
.
0
0
0
0
.
1
9
4
0
.
0
5
6
0
.
1
1
1
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
3
6
1
R
5
0
_
sg
d
_
1
2
0
.
0
0
0
0
.
1
3
9
0
.
2
2
2
0
.
2
5
0
0
.
2
5
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
8
6
1
R
5
0
_
sg
d
_
1
3
0
.
0
2
8
0
.
0
8
3
0
.
0
5
6
0
.
0
8
3
0
.
2
7
8
0
.
0
2
8
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
5
5
6
R
5
0
_
sg
d
_
1
4
0
.
0
0
0
0
.
2
5
0
0
.
2
5
0
0
.
2
2
2
0
.
1
9
4
0
.
1
6
7
0
.
0
3
7
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
1
1
2
R
5
0
_
sg
d
_
1
5
0
.
2
5
0
0
.
2
2
2
0
.
2
5
0
0
.
2
1
9
0
.
2
7
8
0
.
2
1
9
0
.
0
5
4
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
1
4
9
2
R
5
0
_
sg
d
_
1
6
0
.
2
7
8
0
.
2
5
0
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
5
0
0
.
1
5
1
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
1
7
6
3
R
5
0
_
sg
d
_
1
7
0
.
0
0
0
0
.
0
0
0
0
.
1
3
9
0
.
1
1
1
0
.
0
5
6
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
3
0
6
R
5
0
_
sg
d
_
1
8
0
.
0
5
6
0
.
2
2
2
0
.
2
1
6
0
.
2
2
2
0
.
2
2
2
0
.
0
5
6
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
9
9
4
R
5
0
_
sg
d
_
1
9
0
.
8
4
2
0
.
2
5
0
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
2
2
0
.
2
3
1
0
.
1
8
8
0
.
1
8
2
0
.
0
0
0
0
.
2
7
4
9
R
5
0
_
sg
d
_
2
0
0
.
5
6
5
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
5
0
0
.
2
6
7
0
.
2
5
0
.
1
5
5
0
.
0
4
2
0
.
2
6
4
1
R
5
0
_
sg
d
_
2
1
0
.
3
3
3
0
.
2
7
8
0
.
2
5
0
0
.
2
3
9
0
.
2
7
8
0
.
1
1
1
0
.
0
0
7
0
.
0
0
0
0
.
0
0
0
0
.
0
0
7
0
.
1
5
0
3
R
5
0
_
sg
d
_
2
2
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
7
8
0
.
2
2
2
0
.
1
6
5
0
.
0
4
2
0
.
1
5
7
0
.
1
2
5
0
.
2
1
0
1
R
5
0
_
sg
d
_
2
3
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
R
5
0
_
sg
d
_
2
4
0
.
2
2
2
0
.
1
5
9
0
.
0
9
9
0
.
0
2
8
0
.
2
2
2
0
.
1
3
4
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
0
0
0
.
0
8
6
4
T
ab
le
5
p
r
esen
ts
th
e
Ma
s
k
R
-
C
NN
m
o
d
el
ev
alu
atio
n
r
esu
lts
b
ased
o
n
AP
at
in
ter
s
ec
tio
n
o
v
er
u
n
i
o
n
(
I
o
U
)
=5
0
f
o
r
ea
c
h
f
etal
h
ea
r
t
an
ato
m
y
ca
teg
o
r
y
in
th
e
tr
ain
in
g
d
ataset
as
well
as
th
e
m
A
P
as
a
m
ea
s
u
r
e
o
f
o
v
er
all
m
o
d
el
p
er
f
o
r
m
an
ce
.
B
ased
o
n
th
e
r
esu
lts
o
b
tain
ed
,
m
o
d
els
R
5
0
_
s
g
d
_
1
9
an
d
R
5
0
_
s
g
d
_
2
0
s
h
o
wed
th
e
b
est
p
er
f
o
r
m
an
ce
with
m
AP
o
f
0
.
2
7
4
9
a
n
d
0
.
2
6
4
1
,
in
d
ica
tin
g
th
e
a
b
ilit
y
to
r
ec
o
g
n
ize
v
ar
io
u
s
an
at
o
m
ical
s
tr
u
ctu
r
es
m
o
r
e
ac
c
u
r
ately
th
an
o
th
er
m
o
d
els.
C
ar
d
iac
s
tr
u
ctu
r
es
s
u
ch
as
r
ig
h
t
R
V,
L
V,
R
A,
L
A,
an
d
A
V
ten
d
ed
to
h
av
e
h
i
g
h
er
AP
v
al
u
es,
in
d
icatin
g
th
at
th
e
m
o
d
el
s
wer
e
ab
le
to
r
ec
o
g
n
ize
th
es
e
p
ar
ts
b
etter
th
an
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
Mu
lticla
s
s
in
s
ta
n
ce
s
eg
men
ta
t
io
n
o
p
timiz
a
tio
n
fo
r
feta
l h
e
a
r
t ima
g
e
o
b
ject
in
terp
r
eta
tio
n
(
Ha
d
i S
ya
p
u
tr
a
)
4145
o
th
er
s
tr
u
ctu
r
es,
s
u
ch
as
T
V
o
r
PV,
wh
ich
h
ad
lo
wer
o
r
ev
en
ze
r
o
AP
v
alu
es.
T
h
e
ev
alu
atio
n
r
esu
lts
also
s
h
o
w
th
at
th
er
e
ar
e
s
o
m
e
m
o
d
els
with
AP
v
alu
e=
0
.
0
0
0
in
c
er
tain
ca
teg
o
r
ies,
in
d
icatin
g
th
at
th
e
m
o
d
el
f
ailed
to
d
etec
t
o
b
jects
o
f
th
at
class
in
th
e
tr
ain
in
g
d
ataset.
T
h
is
co
u
ld
b
e
d
u
e
to
v
ar
io
u
s
f
ac
to
r
s
,
s
u
ch
as
a
lim
ited
am
o
u
n
t
o
f
a
n
n
o
tatio
n
d
ata
o
r
th
e
co
m
p
lex
ity
o
f
a
n
ato
m
i
ca
l
s
tr
u
ctu
r
es
th
at
ar
e
d
if
f
icu
lt
f
o
r
th
e
m
o
d
el
to
r
ec
o
g
n
ize.
I
n
ad
d
itio
n
,
m
o
d
els
s
u
ch
as R
5
0
_
s
g
d
_
5
an
d
R
5
0
_
s
g
d
_
2
3
h
a
v
e
m
AP=0
,
in
d
icati
n
g
th
at
th
ey
d
id
n
o
t
s
u
cc
ess
f
u
lly
s
eg
m
en
t
an
y
o
b
jects
in
th
e
d
ataset.
Mo
d
els
with
h
ig
h
er
m
AP
s
h
o
w
b
et
ter
p
er
f
o
r
m
an
c
e
in
d
etec
tin
g
an
d
lab
elin
g
f
etal
h
ea
r
t
s
tr
u
ctu
r
es,
wh
ile
m
o
d
els
with
m
an
y
v
alu
es
o
f
0
.
0
0
0
o
r
m
AP=0
s
h
o
w
wea
k
n
ess
es in
th
e
lear
n
in
g
p
r
o
ce
s
s
f
r
o
m
th
e
av
ailab
le
d
ata.
Fig
u
r
e
6
d
is
p
lay
s
th
e
m
AP
f
o
r
v
a
r
io
u
s
R
esNet
-
5
0
m
o
d
els
tr
ain
ed
u
s
in
g
th
e
SGD
o
p
tim
izer
with
d
if
f
er
en
t
h
y
p
er
p
ar
a
m
eter
c
o
m
b
in
atio
n
s
.
m
AP
is
a
co
m
m
o
n
ly
u
s
ed
m
etr
ic
to
e
v
alu
ate
th
e
p
er
f
o
r
m
an
ce
o
f
o
b
ject
d
etec
tio
n
m
o
d
els,
with
h
ig
h
er
v
alu
es
in
d
icatin
g
b
etter
p
er
f
o
r
m
an
ce
.
Fro
m
th
e
g
r
a
p
h
,
it
is
ev
id
en
t
th
at
m
o
d
els
R
5
0
_
s
g
d
_
1
9
an
d
R
5
0
_
s
g
d
_
2
0
ac
h
iev
ed
th
e
b
est
r
es
u
lts
,
with
m
AP
v
alu
es
o
f
ap
p
r
o
x
im
ately
0
.
2
7
an
d
0
.
2
6
,
r
esp
ec
tiv
ely
.
T
h
is
s
u
g
g
e
s
ts
th
at
m
o
d
els
with
an
i
n
p
u
t
i
m
ag
e
s
ize
o
f
5
1
2
×
5
1
2
an
d
a
l
ea
r
n
in
g
r
ate
o
f
0
.
0
1
p
er
f
o
r
m
b
etter
in
d
etec
tin
g
o
b
jects
with
in
th
e
d
ataset
u
s
ed
.
Oth
er
m
o
d
els,
s
u
ch
as
R
5
0
_
s
g
d
_
1
,
R
5
0
_
s
g
d
_
7
,
an
d
R
5
0
_
s
g
d
_
1
5
,
also
s
h
o
wed
f
air
ly
g
o
o
d
p
er
f
o
r
m
an
ce
with
m
AP
v
alu
es
r
an
g
in
g
f
r
o
m
0
.
1
to
0
.
1
5
.
Ho
we
v
er
,
th
eir
p
er
f
o
r
m
an
ce
was
s
till
b
elo
w
th
at
o
f
m
o
d
els
R
5
0
_
s
g
d
_
1
9
an
d
R
5
0
_
s
g
d
_
2
0
.
So
m
e
m
o
d
els
ex
h
ib
ited
v
er
y
lo
w
o
r
ev
en
ze
r
o
p
er
f
o
r
m
a
n
c
e,
s
u
ch
as
R
5
0
_
s
g
d
_
5
an
d
R
5
0
_
s
g
d
_
2
3
.
T
h
is
m
ay
b
e
attr
ib
u
ted
to
s
u
b
o
p
tim
al
h
y
p
er
p
ar
am
eter
c
o
m
b
in
atio
n
s
f
o
r
th
e
d
ataset.
Ov
er
all,
th
ese
r
esu
lts
h
ig
h
lig
h
t
th
e
im
p
o
r
ta
n
ce
o
f
s
elec
tin
g
th
e
ap
p
r
o
p
r
iate
im
ag
e
in
p
u
t
s
ize
an
d
lear
n
in
g
r
ate
to
ac
h
iev
e
o
p
tim
al
p
er
f
o
r
m
an
ce
in
o
b
jec
t
d
etec
tio
n
m
o
d
els
u
s
in
g
th
e
R
esNet
-
5
0
ar
ch
itect
u
r
e
with
th
e
SGD
o
p
tim
izer
.
Alt
h
o
u
g
h
Ma
s
k
R
-
C
NN
is
a
w
ell
-
es
ta
b
lis
h
e
d
m
et
h
o
d
,
th
i
s
s
t
u
d
y
p
r
es
e
n
ts
a
n
o
v
el
ap
p
lica
ti
o
n
b
y
in
t
eg
r
ati
n
g
i
n
s
ta
n
ce
s
eg
m
e
n
t
a
tio
n
w
it
h
t
ar
g
e
te
d
h
y
p
er
p
ar
a
m
et
er
o
p
t
im
i
za
ti
o
n
tai
lo
r
ed
f
o
r
A
4
C
f
eta
l
h
ea
r
t
u
lt
r
as
o
u
n
d
i
m
a
g
es.
T
h
e
co
m
b
i
n
at
io
n
o
f
i
n
p
u
t
r
eso
lu
ti
o
n
t
u
n
i
n
g
,
le
a
r
n
in
g
r
at
e,
a
n
d
m
o
m
e
n
t
u
m
o
n
a
d
atas
et
wit
h
ten
a
n
at
o
m
ic
al
class
es
r
e
p
r
ese
n
ts
a
u
n
iq
u
e
c
o
n
tr
ib
u
t
io
n
,
as
p
r
e
v
i
o
u
s
s
tu
d
i
es
t
y
p
i
ca
ll
y
lim
i
ted
s
e
g
m
e
n
tat
io
n
to
f
ew
er
s
t
r
u
ct
u
r
es
o
r
d
i
d
n
o
t
p
e
r
f
o
r
m
s
y
s
t
em
ati
c
m
o
d
el
o
p
ti
m
i
za
ti
o
n
.
T
h
is
a
p
p
r
o
ac
h
ad
d
r
ess
e
s
th
e
c
o
m
p
l
ex
it
y
o
f
f
et
al
ca
r
d
i
ac
im
a
g
i
n
g
a
n
d
d
em
o
n
s
tr
at
es i
m
p
r
o
v
e
d
cl
ass
-
wis
e
r
e
co
g
n
it
io
n
i
n
a
cli
n
i
ca
l
ly
r
el
e
v
a
n
t
c
o
n
te
x
t
.
Fig
u
r
e
6
.
R
esu
lts
an
d
an
aly
s
is
o
f
m
o
d
el
p
er
f
o
r
m
a
n
ce
T
h
e
Fig
u
r
e
7
illu
s
tr
ates
th
e
A
P
at
an
I
o
U
th
r
esh
o
ld
o
f
5
0
f
o
r
ea
ch
class
ac
r
o
s
s
v
ar
io
u
s
R
esNet
-
5
0
m
o
d
els
tr
ain
ed
with
th
e
SGD
o
p
tim
izer
.
E
ac
h
lin
e
in
th
e
g
r
ap
h
r
ep
r
esen
ts
a
class
,
wi
th
AP
v
alu
es
f
o
r
ea
ch
m
o
d
el
p
lo
tted
as
p
o
in
ts
alo
n
g
th
at
lin
e
.
T
h
e
an
aly
s
is
r
ev
ea
ls
th
at
th
e
class
Ao
d
em
o
n
s
tr
ates
s
ig
n
if
ican
t
p
er
f
o
r
m
an
ce
v
ar
iatio
n
ac
r
o
s
s
m
o
d
els,
with
s
o
m
e
m
o
d
els
s
u
ch
as
R
5
0
_
s
g
d
_
1
9
a
n
d
R
5
0
_
s
g
d
_
2
0
ac
h
iev
in
g
h
ig
h
AP
v
alu
es.
Oth
e
r
class
es,
in
clu
d
in
g
L
A,
L
V,
an
d
R
V,
also
s
h
o
w
n
o
ticea
b
le
v
ar
iati
o
n
in
p
er
f
o
r
m
an
ce
am
o
n
g
th
e
test
ed
m
o
d
els.
Mo
d
els
R
5
0
_
s
g
d
_
1
9
an
d
R
5
0
_
s
g
d
_
2
0
e
x
h
ib
it
m
o
r
e
co
n
s
is
ten
t
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
an
y
class
es
co
m
p
ar
ed
to
o
t
h
er
s
.
C
er
tain
class
e
s
lik
e
s
p
in
e
,
T
V,
MV
,
PV,
an
d
AV
f
r
eq
u
en
tly
s
h
o
w
lo
w
o
r
ev
en
ze
r
o
AP
v
alu
es
in
m
an
y
m
o
d
els,
in
d
icatin
g
th
at
d
etec
ti
o
n
f
o
r
th
ese
class
es
is
m
o
r
e
c
h
allen
g
in
g
.
Ov
er
all,
m
o
d
els
with
lar
g
er
in
p
u
t
im
ag
e
s
izes
an
d
lo
wer
lear
n
in
g
r
ates
ap
p
ea
r
to
d
eliv
er
b
etter
an
d
m
o
r
e
co
n
s
is
ten
t
r
esu
lts
ac
r
o
s
s
v
ar
io
u
s
class
es.
T
h
e
b
est
-
p
er
f
o
r
m
in
g
m
o
d
el
in
t
h
is
ev
alu
atio
n
is
R
5
0
_
s
g
d
_
1
9
,
wh
ic
h
d
em
o
n
s
tr
ates
th
e
h
ig
h
est
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
o
s
t
clas
s
es.
Ou
t
o
f
th
e
2
4
id
en
tifie
d
m
o
d
els,
n
am
ed
f
r
o
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.
1
4
,
No
.
5
,
Octo
b
er
2
0
2
5
:
4
1
3
7
-
4
1
5
0
4146
R
5
0
_
s
g
d
_
1
to
R
5
0
_
s
g
d
_
2
4
,
th
e
r
esear
ch
s
elec
ted
two
m
o
d
el
s
with
o
p
tim
al
d
etec
tio
n
p
er
f
o
r
m
an
ce
f
o
r
class
es
s
u
ch
as
Ao
,
L
A,
L
V,
R
V,
R
A,
T
V,
MV
,
PV,
AV,
an
d
s
p
in
e
.
T
h
e
ev
al
u
atio
n
,
b
ased
o
n
m
A
P
v
alu
es,
id
en
tifie
d
th
e
f
ir
s
t
o
p
tim
al
m
o
d
el
as
R
5
0
_
s
g
d
_
1
9
,
wh
ich
ac
h
ie
v
ed
t
h
e
h
ig
h
est
m
AP
o
f
0
.
2
7
4
9
,
a
lth
o
u
g
h
it
f
ailed
to
d
etec
t
th
e
AV
class
.
T
h
e
s
ec
o
n
d
m
o
d
el,
R
5
0
_
s
g
d
_
2
0
,
s
u
cc
e
s
s
f
u
lly
d
etec
ted
all
class
es
wi
t
h
a
m
AP
o
f
0
.
2
6
4
1
.
B
o
th
m
o
d
els
d
em
o
n
s
tr
ated
s
tr
o
n
g
o
v
er
all
p
er
f
o
r
m
an
ce
.
T
ab
le
6
p
r
esen
ts
th
ese
two
o
p
tim
al
m
o
d
els
b
ased
o
n
th
e
r
esu
lts
f
r
o
m
T
ab
le
5
.
T
h
ese
m
o
d
els
wer
e
s
elec
ted
d
u
e
to
th
eir
h
ig
h
m
AP
v
alu
es
an
d
co
n
s
is
ten
t
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
o
s
t f
etal
h
ea
r
t a
n
ato
m
ical
class
es.
Fig
u
r
e
7
.
C
o
m
p
a
r
is
o
n
o
f
AP v
alu
es b
etwe
en
o
b
ject
class
es o
f
f
etal
h
ea
r
t im
a
g
es
T
ab
l
e
6
.
Op
tim
al
p
a
r
am
eter
s
f
o
r
Ma
s
k
R
-
C
NN
m
o
d
els
M
e
t
h
o
d
M
o
d
e
l
P
a
r
a
me
t
e
r
s
M
a
s
k
R
-
C
N
N
R
5
0
_
sg
d
_
1
9
I
mag
e
si
z
e
:
5
1
2
×
5
1
2
,
l
e
a
r
n
i
n
g
m
o
me
n
t
u
m:
0
.
7
,
l
e
a
r
n
i
n
g
r
a
t
e
:
0
.
0
1
R
5
0
_
sg
d
_
2
0
I
mag
e
S
i
z
e
:
5
1
2
×
5
1
2
,
l
e
a
r
n
i
n
g
m
o
me
n
t
u
m:
0
.
9
,
l
e
a
r
n
i
n
g
r
a
t
e
:
0
.
0
1
T
h
e
r
esu
lts
o
f
in
s
tan
ce
s
eg
m
e
n
tatio
n
f
r
o
m
th
e
two
o
p
tim
al
m
o
d
els
ar
e
d
is
p
lay
ed
in
Fig
u
r
e
8
.
T
h
is
f
ig
u
r
e
s
h
o
ws
th
e
s
eg
m
en
tatio
n
r
esu
lts
f
o
r
ten
class
es
o
f
f
etal
h
ea
r
t
o
b
jects.
T
h
ese
s
eg
m
e
n
tatio
n
o
u
tp
u
ts
ar
e
ess
en
tial
to
ev
alu
ate
th
e
m
o
d
el’
s
ab
ilit
y
to
d
if
f
er
e
n
tiate
ea
ch
an
ato
m
ical
s
tr
u
ctu
r
e
ac
cu
r
ately
.
Fig
u
r
e
8
(
a
)
s
h
o
ws
th
e
s
eg
m
en
tatio
n
r
esu
lts
f
o
r
s
ev
er
al
an
ato
m
ical
s
tr
u
ctu
r
es
in
a
m
ed
ical
im
ag
e,
lik
ely
an
ec
h
o
ca
r
d
io
g
r
am
o
f
th
e
h
ea
r
t.
T
h
e
s
eg
m
e
n
tatio
n
s
u
cc
ess
f
u
l
ly
id
en
tifie
s
an
d
lab
els
s
ev
er
al
k
ey
p
ar
ts
o
f
th
e
im
ag
e
with
h
i
g
h
c
o
n
f
id
e
n
ce
le
v
els,
in
clu
d
in
g
:
R
V
with
a
co
n
f
id
en
ce
o
f
0
.
9
9
5
L
V
with
a
c
o
n
f
id
en
ce
o
f
0
.
9
9
9
AV
with
a
co
n
f
id
en
ce
o
f
0
.
9
8
5
MV
with
a
co
n
f
id
en
ce
o
f
0
.
9
7
5
R
A
with
a
co
n
f
id
e
n
ce
o
f
1
.
0
0
0
PV
with
a
co
n
f
id
en
ce
o
f
0
.
9
7
2
L
A
with
a
co
n
f
id
en
ce
o
f
0
.
9
9
9
T
V
with
a
co
n
f
id
en
ce
o
f
0
.
9
7
0
Ao
with
a
co
n
f
id
en
ce
o
f
0
.
9
9
5
s
p
in
e
with
a
co
n
f
id
en
ce
o
f
0
.
9
9
7
.
T
h
is
s
eg
m
en
tatio
n
d
em
o
n
s
tr
ates
th
at
th
e
m
o
d
el
h
as
v
er
y
h
ig
h
ac
cu
r
ac
y
in
id
e
n
tify
in
g
a
n
d
la
b
elin
g
v
ar
io
u
s
an
ato
m
ical
s
tr
u
ctu
r
es
with
in
th
e
m
ed
ical
im
ag
e.
E
ac
h
s
eg
m
en
t
is
clea
r
ly
d
elin
ea
ted
,
an
d
th
e
h
i
g
h
co
n
f
id
e
n
ce
v
alu
es
s
u
g
g
est
th
at
th
is
m
o
d
el
is
r
eliab
le
f
o
r
d
iag
n
o
s
tic
p
u
r
p
o
s
es
an
d
f
u
r
t
h
er
m
ed
ical
an
aly
s
is
.
T
h
ese
r
esu
lts
ar
e
h
ig
h
ly
f
av
o
r
ab
le
f
o
r
m
ed
ical
ap
p
licatio
n
s
,
p
ar
ticu
lar
ly
in
ass
is
tin
g
p
h
y
s
ician
s
with
th
e
id
en
tific
atio
n
a
n
d
an
aly
s
is
o
f
cr
itical
p
a
r
ts
o
f
ec
h
o
ca
r
d
io
g
r
ap
h
ic
im
ag
es.
Fig
u
r
e
8
(
b
)
p
r
esen
ts
t
h
e
s
eg
m
en
tatio
n
r
esu
lts
o
f
s
ev
er
al
k
ey
an
at
o
m
ical
s
tr
u
ctu
r
es
in
a
n
ec
h
o
ca
r
d
io
g
r
ap
h
ic
im
ag
e,
with
ex
tr
em
el
y
h
ig
h
co
n
f
id
en
ce
lev
els.
Deta
iled
ex
p
lan
atio
n
s
f
o
r
ea
ch
i
d
en
tifi
ed
s
tr
u
ctu
r
e
ar
e
as
f
o
llo
ws
.
R
V:
th
is
s
tr
u
ctu
r
e
i
s
id
en
tifie
d
with
a
co
n
f
id
e
n
c
e
o
f
0
.
9
9
8
,
in
d
icatin
g
th
at
th
e
m
o
d
el
is
h
ig
h
ly
co
n
f
id
en
t
i
n
its
id
en
tific
atio
n
.
L
V:
s
im
ilar
ly
,
th
e
LV
is
id
en
tifie
d
with
a
v
er
y
h
ig
h
co
n
f
id
en
ce
o
f
0
.
9
9
8
.
AV:
th
is
v
alv
e
is
id
en
tifie
d
with
a
co
n
f
id
en
ce
o
f
0
.
9
5
4
.
Alth
o
u
g
h
s
lig
h
tly
lo
wer
t
h
an
o
th
er
s
tr
u
ctu
r
es,
th
is
v
alu
e
r
em
ain
s
v
er
y
h
ig
h
.
MV
:
m
ar
k
ed
with
a
co
n
f
id
en
ce
o
f
0
.
9
9
3
,
i
n
d
icatin
g
n
ea
r
ly
p
er
f
ec
t
co
n
f
id
e
n
ce
in
id
en
tify
in
g
th
is
v
alv
e
.
R
A:
with
a
co
n
f
id
e
n
ce
o
f
0
.
9
9
9
,
th
e
RA
is
id
en
tifie
d
with
n
ea
r
ly
p
er
f
ec
t
c
o
n
f
id
e
n
ce
.
PV:
th
is
v
alv
e
is
id
en
tifie
d
with
a
co
n
f
id
en
ce
o
f
0
.
9
4
5
,
wh
ich
r
em
ain
s
with
in
a
h
ig
h
co
n
f
id
e
n
ce
r
an
g
e.
T
V:
with
a
co
n
f
id
en
ce
o
f
0
.
9
8
8
,
th
e
TV
is
s
eg
m
en
ted
with
v
er
y
g
o
o
d
ac
cu
r
ac
y
.
Ao
:
th
is
s
tr
u
ctu
r
e
is
m
ar
k
ed
with
a
co
n
f
id
en
ce
o
f
0
.
9
9
8
,
i
n
d
icatin
g
h
ig
h
ly
ac
cu
r
ate
id
e
n
tific
atio
n
.
Sp
in
e:
th
e
s
p
in
e
i
s
s
eg
m
en
ted
with
a
co
n
f
id
en
ce
o
f
0
.
9
9
6
,
s
h
o
wi
n
g
h
ig
h
co
n
f
id
en
ce
in
th
e
id
en
tific
atio
n
o
f
th
is
s
tr
u
ctu
r
e.
Ov
er
all,
th
is
Evaluation Warning : The document was created with Spire.PDF for Python.