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
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
6
0
3
~
3
6
1
3
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
6
0
3
-
3
6
1
3
3603
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Efficien
t
de
ep lea
rning
f
o
r auto
ma
t
ed cornea
l ul
cer
s
ev
erity
cla
ss
ificatio
n f
ro
m fluores
cei
n ima
g
es
Ro
dia
h
1
,
I
nd
a
h S
inthy
a
P
er
m
a
t
a
Sa
ri
2
,
M
a
t
riss
y
a
H
er
mita
3
,
Sa
rif
ud
din
M
a
de
nd
a
2
,
Dia
na
T
ri
Su
s
et
ia
nin
g
t
ia
s
4
1
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
c
s
,
F
a
c
u
l
t
y
o
f
I
n
d
u
st
r
y
Te
c
h
n
o
l
o
g
y
,
U
n
i
v
e
r
si
t
a
s G
u
n
a
d
a
r
ma
,
D
e
p
o
k
,
I
n
d
o
n
e
si
a
2
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
G
r
a
d
u
a
t
e
P
r
o
g
r
a
m
,
U
n
i
v
e
r
si
t
a
s
G
u
n
a
d
a
r
ma
,
D
e
p
o
k
,
I
n
d
o
n
e
si
a
3
D
e
p
a
r
t
me
n
t
o
f
M
e
d
i
c
a
l
Ed
u
c
a
t
i
o
n
,
F
a
c
u
l
t
y
o
f
M
e
d
i
c
i
n
e
,
U
n
i
v
e
r
s
i
t
a
s
G
u
n
a
d
a
r
ma
,
D
e
p
o
k
,
I
n
d
o
n
e
si
a
4
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
y
st
e
ms
,
F
a
c
u
l
t
y
o
f
I
n
f
o
r
ma
t
i
o
n
S
y
st
e
ms
a
n
d
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
U
n
i
v
e
r
si
t
a
s
G
u
n
a
d
a
r
ma
,
D
e
p
o
k
,
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
J
u
l 2
3
,
2
0
2
5
R
ev
is
ed
Ma
y
1
3
,
2
0
2
6
Acc
ep
ted
Ma
y
2
4
,
2
0
2
6
Co
rn
e
a
l
u
lce
rs
c
a
n
c
a
u
se
p
e
r
m
a
n
e
n
t
v
isi
o
n
l
o
ss
if
n
o
t
d
iag
n
o
se
d
a
n
d
m
a
n
a
g
e
d
p
r
o
m
p
tl
y
,
p
a
rti
c
u
larl
y
i
n
se
tt
in
g
s
wit
h
l
imited
a
c
c
e
s
s
to
o
p
h
th
a
lmo
l
o
g
y
se
rv
ice
s.
Th
is
st
u
d
y
a
ims
to
d
e
v
e
lo
p
a
n
a
u
to
m
a
ted
d
e
e
p
lea
rn
in
g
a
p
p
r
o
a
c
h
fo
r
c
las
sify
i
n
g
c
o
rn
e
a
l
u
lce
r
se
v
e
rit
y
fr
o
m
flu
o
re
sc
e
in
slit
-
lam
p
ima
g
e
s.
An
Eff
icie
n
tNe
tV2
-
S
–
b
a
se
d
m
o
d
e
l
is
e
m
p
l
o
y
e
d
,
in
c
o
rp
o
ra
ti
n
g
c
o
rn
e
a
l
a
re
a
m
a
sk
in
g
to
su
p
p
re
ss
n
o
n
-
re
lev
a
n
t
re
g
i
o
n
s
a
n
d
c
las
s
d
istri
b
u
ti
o
n
–
b
a
se
d
a
u
g
m
e
n
tatio
n
to
a
d
d
re
ss
d
a
ta
imb
a
lan
c
e
.
T
o
imp
ro
v
e
e
v
a
lu
a
ti
o
n
re
li
a
b
il
it
y
,
a
lea
k
a
g
e
-
a
wa
re
d
a
ta
sp
li
tt
i
n
g
stra
teg
y
i
s
a
p
p
li
e
d
b
e
fo
re
a
n
d
a
fter
a
u
g
m
e
n
tati
o
n
.
E
x
p
e
rime
n
tal
re
su
lt
s
s
h
o
w
t
h
a
t
th
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
a
c
h
ie
v
e
s
a
m
a
x
imu
m
v
a
li
d
a
ti
o
n
a
c
c
u
ra
c
y
o
f
9
5
.
9
3
%
u
n
d
e
r
n
o
n
-
lea
k
a
g
e
c
o
n
d
it
i
o
n
s
fo
r
t
h
e
c
a
teg
o
ry
c
las
sifica
ti
o
n
sc
e
n
a
ri
o
,
w
h
il
e
m
a
in
tain
in
g
h
i
g
h
train
i
n
g
e
ffici
e
n
c
y
.
Th
e
se
re
su
lt
s
d
e
m
o
n
stra
t
e
th
a
t
th
e
p
ro
p
o
se
d
m
e
th
o
d
p
ro
v
id
e
s
a
ro
b
u
st
a
n
d
e
fficie
n
t
s
o
lu
ti
o
n
f
o
r
a
u
to
m
a
ted
c
o
rn
e
a
l
u
lce
r
se
v
e
rit
y
a
ss
e
ss
m
e
n
t
a
n
d
h
a
s
th
e
p
o
ten
t
ial
to
su
p
p
o
rt
c
li
n
ica
l
d
e
c
isio
n
-
m
a
k
i
n
g
i
n
o
p
h
th
a
lmic
p
r
a
c
ti
c
e
.
K
ey
w
o
r
d
s
:
Au
to
m
ated
d
iag
n
o
s
is
C
o
r
n
ea
l u
lcer
E
f
f
icien
tNetV2
Flu
o
r
escein
im
ag
in
g
Ocu
lar
im
ag
in
g
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
:
R
o
d
iah
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atics,
Facu
lty
o
f
I
n
d
u
s
tr
y
T
ec
h
n
o
lo
g
y
,
Un
iv
er
s
itas
Gu
n
ad
ar
m
a
Ma
r
g
o
n
d
a
R
ay
a
1
0
0
,
Po
n
d
o
k
C
in
a,
Dep
o
k
,
I
n
d
o
n
esia
E
m
ail: r
o
d
iah
@
s
taf
f
.
g
u
n
a
d
ar
m
a.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
A
co
r
n
ea
l
u
lcer
is
a
p
ath
o
lo
g
ic
al
co
n
d
itio
n
in
v
o
lv
in
g
d
am
ag
e
to
th
e
co
r
n
ea
l tis
s
u
e
f
r
o
m
th
e
ep
ith
elial
lay
er
to
th
e
s
tr
o
m
a
,
an
d
m
ay
r
esu
lt in
p
er
m
an
en
t b
lin
d
n
ess
if
n
o
t m
an
ag
ed
ap
p
r
o
p
r
iately
[
1
]
.
Acc
o
r
d
in
g
to
th
e
W
o
r
ld
Hea
lth
Or
g
an
izatio
n
(
W
HO)
,
th
e
an
n
u
al
in
cid
en
ce
o
f
co
r
n
ea
l
u
lcer
s
r
ea
ch
es
ap
p
r
o
x
im
ately
7
1
0
ca
s
es
p
er
1
0
0
,
0
0
0
p
o
p
u
latio
n
,
wit
h
v
is
io
n
lo
s
s
ca
u
s
ed
b
y
o
c
u
lar
tr
au
m
a
af
f
ec
tin
g
1
.
5
–
2
m
illi
o
n
in
d
iv
id
u
als
wo
r
ld
wid
e,
p
a
r
ticu
lar
ly
i
n
d
e
v
elo
p
in
g
co
u
n
tr
ies
[
2
]
.
Ad
v
a
n
ce
d
m
a
n
if
estatio
n
s
s
u
ch
as
co
r
n
ea
l
o
p
ac
ity
ar
e
co
m
m
o
n
l
y
o
b
s
er
v
ed
am
o
n
g
p
o
p
u
latio
n
s
with
lo
w
ed
u
c
atio
n
al
b
ac
k
g
r
o
u
n
d
s
a
n
d
h
ig
h
-
r
is
k
o
cc
u
p
atio
n
s
,
in
clu
d
in
g
f
a
r
m
er
s
,
f
is
h
er
m
e
n
,
an
d
m
an
u
al
la
b
o
r
er
s
,
wh
o
ar
e
f
r
eq
u
en
tl
y
ex
p
o
s
ed
to
ey
e
tr
au
m
a
wh
ile
h
av
in
g
lim
ited
ac
ce
s
s
to
tim
ely
an
d
h
ig
h
-
q
u
ality
o
p
h
t
h
alm
ic
c
ar
e
[
3
]
.
T
h
ese
co
n
d
itio
n
s
h
ig
h
lig
h
t
p
e
r
s
is
ten
t
ch
allen
g
es r
elate
d
to
d
elay
ed
d
iag
n
o
s
is
an
d
ac
c
u
r
ate
s
ev
er
it
y
ass
ess
m
en
t o
f
co
r
n
ea
l u
lcer
s
.
At
p
r
esen
t,
c
o
r
n
ea
l
u
lcer
s
ev
e
r
ity
ass
ess
m
en
t
p
r
ed
o
m
in
an
tly
r
elies
o
n
m
a
n
u
al
s
lit
-
lam
p
ex
am
in
atio
n
with
f
lu
o
r
escein
s
tain
in
g
[
4
]
.
Alth
o
u
g
h
wid
ely
u
s
ed
in
clin
i
ca
l
p
r
ac
tice,
th
is
m
eth
o
d
r
e
q
u
ir
es
h
ig
h
ex
p
er
tis
e,
is
tim
e
-
co
n
s
u
m
in
g
,
a
n
d
is
in
h
er
en
tly
s
u
b
jectiv
e,
lea
d
in
g
t
o
i
n
ter
-
o
b
s
er
v
er
v
a
r
iab
ilit
y
an
d
p
o
ten
tial
d
iag
n
o
s
tic
d
elay
s
[
5
]
.
Su
c
h
lim
itatio
n
s
in
cr
ea
s
e
th
e
r
is
k
o
f
ir
r
e
v
er
s
ib
le
co
r
n
ea
l
d
am
a
g
e
an
d
v
is
io
n
lo
s
s
,
u
n
d
er
s
co
r
in
g
th
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
0
3
-
3
6
1
3
3604
g
r
o
win
g
n
ee
d
f
o
r
a
u
to
m
ate
d
,
f
ast,
an
d
r
eliab
le
d
ia
g
n
o
s
tic
to
o
ls
to
s
u
p
p
o
r
t
clin
ica
l
d
ec
is
io
n
-
m
ak
in
g
,
p
ar
ticu
lar
ly
in
r
eso
u
r
ce
-
lim
ite
d
s
ettin
g
s
[
6
]
.
R
e
c
e
n
t
ad
v
an
c
e
s
in
d
e
e
p
le
a
r
n
i
n
g
h
av
e
d
e
m
o
n
s
tr
a
t
e
d
p
r
o
m
i
s
i
n
g
p
e
r
f
o
r
m
a
n
c
e
in
au
t
o
m
a
te
d
co
r
n
e
a
l
u
l
c
e
r
c
l
a
s
s
i
f
i
c
a
t
io
n
u
s
in
g
f
lu
o
r
e
s
c
e
in
im
a
g
e
s
.
V
i
s
i
o
n
tr
a
n
s
f
o
r
m
e
r
(
V
i
T
)
-
b
a
s
e
d
a
p
p
r
o
a
ch
e
s
h
a
v
e
a
ch
i
e
v
e
d
h
i
g
h
a
c
cu
r
a
cy
ac
r
o
s
s
m
u
l
t
i
p
le
c
l
a
s
s
i
f
i
c
a
t
i
o
n
s
c
en
a
r
io
s
,
in
c
l
u
d
in
g
c
a
t
e
g
o
r
y
,
ty
p
e
,
an
d
g
r
ad
e
c
l
a
s
s
i
f
i
ca
t
i
o
n
[
7
]
.
C
o
n
v
o
l
u
t
io
n
a
l
n
eu
r
a
l
n
e
tw
o
r
k
(
C
N
N
)
-
b
a
s
ed
m
e
t
h
o
d
s
a
n
d
h
y
b
r
id
a
r
ch
i
t
e
c
tu
r
e
s
co
m
b
i
n
in
g
d
e
ep
f
e
a
t
u
r
e
e
x
t
r
a
c
to
r
s
w
i
th
c
l
a
s
s
i
c
a
l
c
l
a
s
s
i
f
i
e
r
s
,
s
u
ch
a
s
De
n
s
e
N
e
t1
2
1
w
i
t
h
s
u
p
p
o
r
t
v
e
c
to
r
m
a
ch
i
n
e
(
S
V
M
)
,
h
av
e
a
l
s
o
r
e
p
o
r
t
ed
s
t
r
o
n
g
c
l
a
s
s
i
f
ic
a
t
i
o
n
p
e
r
f
o
r
m
a
n
ce
o
n
p
u
b
l
i
c
d
at
a
s
e
t
s
[
8
]
.
O
th
e
r
s
t
u
d
i
e
s
h
a
v
e
e
x
p
l
o
r
e
d
v
ar
i
o
u
s
C
N
N
a
r
c
h
i
t
ec
t
u
r
e
s
f
o
r
u
l
c
er
ty
p
e
c
l
a
s
s
i
f
i
c
a
t
io
n
[
9
]
,
in
f
e
c
t
io
u
s
k
e
r
a
t
i
t
i
s
d
i
ag
n
o
s
i
s
[
1
0
]
,
a
n
d
b
in
a
r
y
s
e
v
e
r
i
t
y
c
l
a
s
s
i
f
i
c
a
t
i
o
n
[
1
1
]
.
W
h
i
le
th
e
s
e
ap
p
r
o
a
ch
e
s
s
h
o
w
en
co
u
r
a
g
in
g
r
e
s
u
l
t
s
,
t
h
ey
o
f
t
en
s
u
f
f
e
r
f
r
o
m
h
i
g
h
m
o
d
e
l
c
o
m
p
l
ex
i
t
y
,
l
i
m
i
t
ed
e
v
a
l
u
a
t
i
o
n
o
f
in
f
er
en
c
e
e
f
f
ic
i
e
n
cy
,
n
o
n
-
e
n
d
-
to
-
en
d
p
i
p
el
i
n
e
s
,
o
r
r
e
s
t
r
ic
t
e
d
c
l
a
s
s
i
f
i
c
a
t
i
o
n
s
c
o
p
e
.
Desp
ite
th
eir
ef
f
ec
tiv
e
n
ess
,
e
x
is
tin
g
s
tu
d
ies
s
till
ex
h
ib
it
s
e
v
er
al
m
eth
o
d
o
l
o
g
ical
g
a
p
s
.
T
r
an
s
f
o
r
m
er
-
b
ased
m
o
d
els
ten
d
to
b
e
co
m
p
u
tatio
n
ally
ex
p
e
n
s
iv
e
an
d
la
ck
an
aly
s
is
o
f
d
ep
lo
y
m
e
n
t
f
e
asib
ilit
y
in
clin
ical
en
v
ir
o
n
m
en
ts
[
7
]
.
Hy
b
r
id
C
NN
–
SVM
ap
p
r
o
ac
h
es
r
ely
o
n
s
ep
ar
ated
f
ea
tu
r
e
e
x
tr
ac
tio
n
an
d
class
if
icatio
n
s
tag
es,
wh
ich
m
ay
r
ed
u
ce
ad
a
p
tab
ilit
y
to
n
ew
d
atasets
[
8
]
.
Mo
r
eo
v
er
,
s
ev
er
al
C
NN
-
b
ase
d
s
tu
d
ies
f
o
cu
s
o
n
a
s
in
g
le
class
if
icatio
n
s
ce
n
ar
io
o
r
b
in
a
r
y
s
ev
er
ity
le
v
els
a
n
d
d
o
n
o
t
ex
p
licitly
ad
d
r
ess
v
is
u
al
n
o
is
e
f
r
o
m
non
-
c
o
r
n
ea
l
r
eg
io
n
s
,
s
u
c
h
as
th
e
s
cler
a
an
d
e
y
elash
es,
d
u
e
t
o
th
e
ab
s
en
ce
o
f
c
o
r
n
ea
l
a
r
ea
m
ask
in
g
[
9
]
,
[
1
1
]
.
Ad
d
itio
n
ally
,
lim
ited
atten
tio
n
h
as
b
ee
n
p
aid
to
d
ata
lea
k
ag
e
is
s
u
es
ar
is
in
g
f
r
o
m
im
p
r
o
p
er
d
ata
s
p
litt
in
g
s
tr
ateg
ies,
wh
ich
m
ay
co
m
p
r
o
m
is
e
ev
alu
atio
n
v
alid
ity
a
n
d
g
en
er
aliza
tio
n
p
er
f
o
r
m
a
n
ce
[
1
2
]
.
T
o
ad
d
r
ess
th
ese
lim
itatio
n
s
,
th
is
s
tu
d
y
p
r
o
p
o
s
es
an
a
u
to
m
ated
co
r
n
ea
l
u
lcer
s
ev
er
ity
cla
s
s
if
icatio
n
f
r
am
ewo
r
k
b
ased
o
n
th
e
E
f
f
i
cien
tNetV2
-
S
ar
ch
itectu
r
e
u
s
in
g
f
l
u
o
r
escein
s
lit
-
lam
p
im
a
g
es.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
in
teg
r
ates
s
elec
tiv
e
co
r
n
ea
l
ar
ea
m
ask
in
g
t
o
s
u
p
p
r
ess
n
o
n
-
r
elev
a
n
t
v
is
u
al
in
f
o
r
m
atio
n
an
d
class
d
is
tr
ib
u
tio
n
–
b
ased
au
g
m
en
tat
io
n
to
m
itig
ate
d
ata
im
b
alan
ce
.
Fu
r
th
er
m
o
r
e,
a
leak
a
g
e
-
a
war
e
d
ata
s
p
litt
in
g
s
tr
ateg
y
is
s
y
s
tem
atica
lly
d
esig
n
ed
b
ef
o
r
e
an
d
af
ter
a
u
g
m
en
tatio
n
to
e
n
s
u
r
e
r
eliab
le
m
o
d
el
ev
alu
atio
n
.
B
y
p
r
o
v
id
i
n
g
an
ef
f
icien
t
en
d
-
to
-
en
d
p
ip
elin
e
ca
p
ab
le
o
f
h
a
n
d
lin
g
m
u
ltip
le
class
if
icatio
n
s
ce
n
ar
io
s
:
ca
teg
o
r
y
,
ty
p
e,
a
n
d
g
r
ad
e
,
th
is
s
tu
d
y
aim
s
to
en
h
a
n
ce
r
o
b
u
s
tn
es
s
,
g
en
er
aliza
tio
n
,
an
d
p
r
ac
ti
ca
l
ap
p
licab
ilit
y
o
f
au
to
m
ated
co
r
n
ea
l u
lcer
d
iag
n
o
s
is
s
y
s
tem
s
.
2.
M
E
T
H
O
D
T
h
is
s
tu
d
y
em
p
l
o
y
s
a
d
ee
p
lea
r
n
in
g
-
b
ased
ap
p
r
o
ac
h
to
a
u
to
m
atica
lly
class
if
y
th
e
s
ev
er
ity
o
f
c
o
r
n
ea
l
u
lcer
s
u
s
in
g
f
lu
o
r
escein
im
a
g
es.
T
h
e
o
v
e
r
all
m
eth
o
d
o
l
o
g
y
is
o
r
g
an
ized
in
to
s
ev
er
al
s
eq
u
en
tial
s
tag
es,
as
illu
s
tr
ated
in
Fig
u
r
e
1
.
T
h
e
p
r
o
ce
s
s
b
eg
in
s
with
im
ag
e
a
cq
u
is
itio
n
an
d
p
r
e
p
r
o
ce
s
s
in
g
to
im
p
r
o
v
e
im
ag
e
q
u
ality
an
d
p
r
ep
ar
e
t
h
e
d
ata
s
et
f
o
r
an
aly
s
is
.
T
h
e
n
ex
t
s
tag
e
in
v
o
lv
es
m
o
d
el
d
ev
elo
p
m
en
t
u
s
in
g
a
d
ee
p
lear
n
in
g
ar
ch
itectu
r
e
f
o
r
c
o
r
n
ea
l
u
lcer
s
ev
e
r
ity
class
if
icatio
n
.
Fin
ally
,
th
e
t
r
ain
ed
m
o
d
el
is
ev
alu
ated
u
s
in
g
p
er
f
o
r
m
an
ce
m
etr
ics
to
ass
ess
its
clas
s
if
icatio
n
ac
cu
r
ac
y
a
n
d
r
eliab
ilit
y
.
T
h
ese
s
tag
es
p
r
o
v
id
e
a
s
y
s
tem
atic
an
d
r
ep
r
o
d
u
cib
le
f
r
a
m
ewo
r
k
f
o
r
th
e
p
r
o
p
o
s
ed
r
esear
ch
m
eth
o
d
o
lo
g
y
.
Fig
u
r
e
1
.
R
esear
ch
m
eth
o
d
o
l
o
g
y
f
o
r
co
r
n
ea
l u
lcer
s
ev
e
r
ity
cl
ass
if
icatio
n
u
s
in
g
d
ee
p
lear
n
i
n
g
-
b
ased
ap
p
r
o
ac
h
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
E
fficien
t d
ee
p
lea
r
n
in
g
f
o
r
a
u
t
o
ma
ted
co
r
n
ea
l
u
lcer
s
ev
erit
y
cla
s
s
ifica
tio
n
fr
o
m
flu
o
r
escein
ima
g
e
s
(
R
o
d
ia
h
)
3605
2
.
1
.
Co
rnea
l ulce
r
da
t
a
s
et
T
h
e
f
lu
o
r
escein
ey
e
im
a
g
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
is
a
s
ec
o
n
d
ar
y
d
ataset.
T
h
is
d
a
taset
wa
s
o
b
tain
ed
f
r
o
m
th
e
r
esear
c
h
[
1
3
]
.
T
h
is
d
ataset
co
n
s
is
ted
o
f
7
1
2
f
lu
o
r
escein
-
s
tain
ed
ey
e
im
ag
es
co
llected
f
r
o
m
p
atien
ts
with
v
ar
y
in
g
d
eg
r
ee
s
o
f
co
r
n
ea
l
u
lcer
s
at
th
e
Z
h
o
n
g
s
h
an
Op
h
th
alm
o
lo
g
y
C
en
ter
at
Su
n
Yat
-
s
en
Un
iv
er
s
ity
.
T
h
e
f
lu
o
r
escein
e
y
e
im
ag
e
d
ataset
is
d
iv
id
ed
i
n
to
th
r
ee
ca
teg
o
r
ies,
n
am
el
y
ca
teg
o
r
y
,
ty
p
e,
an
d
g
r
ad
e
[
1
3
]
.
T
h
e
f
o
llo
win
g
is
a
d
etailed
ex
p
lan
atio
n
o
f
th
e
th
r
ee
s
ce
n
ar
io
s
.
i)
T
h
e
ca
teg
o
r
y
s
ce
n
ar
io
r
e
p
r
ese
n
ts
th
e
g
en
er
al
ty
p
e
o
f
c
o
r
n
ea
l
u
lcer
an
d
is
d
iv
id
ed
in
to
t
h
r
ee
s
u
b
class
es:
ca
teg
o
r
y
0
,
ca
teg
o
r
y
1
,
a
n
d
ca
teg
o
r
y
2
.
C
ateg
o
r
y
0
(
p
o
i
n
t
-
lik
e
co
r
n
ea
l
u
lcer
)
r
ef
er
s
to
s
m
all
,
dot
-
s
h
ap
e
d
u
lcer
s
th
at
ar
e
g
e
n
er
ally
m
ild
an
d
co
m
m
o
n
ly
ap
p
ea
r
d
u
r
in
g
th
e
ea
r
ly
s
tag
e
o
f
in
f
lam
m
atio
n
o
r
h
ea
lin
g
.
C
ateg
o
r
y
1
(
p
o
in
t
-
f
lak
y
m
ix
ed
u
lcer
)
r
ep
r
esen
ts
a
co
m
b
in
atio
n
o
f
p
o
in
t
a
n
d
f
la
k
y
u
lcer
p
atter
n
s
ir
r
eg
u
lar
ly
d
is
tr
ib
u
te
d
ac
r
o
s
s
t
h
e
co
r
n
ea
,
in
d
icatin
g
a
m
o
r
e
s
ev
er
e
co
n
d
itio
n
.
C
ateg
o
r
y
2
(
f
lak
y
co
r
n
ea
l
u
lcer
)
r
ef
er
s
to
lar
g
er
f
lak
y
le
s
io
n
s
with
clea
r
b
o
u
n
d
ar
ies
th
at
m
ay
lead
to
co
r
n
ea
l
s
ca
r
r
in
g
an
d
v
is
io
n
lo
s
s
[
1
4
]
.
ii)
T
h
e
ty
p
e
s
ce
n
ar
io
p
r
o
v
id
es
a
m
o
r
e
d
etailed
d
escr
ip
tio
n
o
f
u
lcer
m
o
r
p
h
o
lo
g
y
b
ased
o
n
lesi
o
n
s
h
ap
e
an
d
s
ize.
T
h
is
s
ce
n
ar
io
c
o
n
s
is
ts
o
f
f
i
v
e
s
u
b
class
es:
ty
p
e
0
,
t
y
p
e
1
,
ty
p
e
2
,
t
y
p
e
3
,
a
n
d
ty
p
e
4
.
T
y
p
e
0
in
d
icate
s
a
n
o
r
m
al
c
o
r
n
ea
l
e
p
ith
eliu
m
with
o
u
t
u
lcer
atio
n
.
T
y
p
e
1
(
m
icr
o
p
u
n
ctate
)
r
e
p
r
esen
ts
s
m
all
s
ca
tter
ed
lesi
o
n
s
o
n
th
e
co
r
n
ea
l su
r
f
ac
e.
T
y
p
e
2
(
m
ac
r
o
p
u
n
c
tate)
in
d
icate
s
lar
g
er
p
u
n
ctate
lesi
o
n
s
,
wh
ile
t
y
p
e
3
(
co
alescen
t
m
ac
r
o
p
u
n
ctate
)
d
escr
ib
es
lar
g
e
r
lesi
o
n
s
th
at
m
er
g
e
o
r
ap
p
ea
r
clo
s
ely
g
r
o
u
p
ed
.
T
y
p
e
4
r
ef
er
s
to
u
lcer
p
atch
es
m
ea
s
u
r
in
g
≥
1
m
m
,
i
n
d
icatin
g
m
o
r
e
s
ev
er
e
c
o
r
n
ea
l
d
am
ag
e
[
1
3
]
.
iii)
T
h
e
g
r
ad
e
s
ce
n
a
r
io
d
escr
ib
es
th
e
s
ev
er
ity
an
d
s
p
atial
d
is
tr
ib
u
tio
n
o
f
c
o
r
n
ea
l
u
lcer
s
b
ased
o
n
th
e
af
f
ec
te
d
co
r
n
ea
l
q
u
ad
r
a
n
ts
[
1
5
]
.
T
h
is
s
ce
n
ar
io
is
d
iv
id
e
d
in
to
f
iv
e
s
u
b
class
es:
g
r
ad
e
0
,
g
r
a
d
e
1
,
g
r
a
d
e
2
,
g
r
ad
e
3
,
an
d
g
r
a
d
e
4
.
Gr
ad
e
0
in
d
icate
s
n
o
u
lcer
atio
n
,
wh
er
ea
s
g
r
ad
e
s
1
to
3
r
e
p
r
esen
t
u
lcer
in
v
o
lv
em
en
t
in
o
n
e,
two
,
an
d
th
r
ee
o
r
f
o
u
r
co
r
n
ea
l
q
u
ad
r
an
ts
,
r
esp
ec
tiv
ely
.
G
r
a
d
e
4
in
d
icate
s
u
lcer
s
lo
ca
ted
in
th
e
ce
n
tr
al
o
p
tical
zo
n
e
o
f
th
e
co
r
n
ea
,
wh
ich
is
th
e
m
o
s
t
cr
it
ical
r
eg
io
n
af
f
ec
tin
g
v
is
io
n
[
1
6
]
.
T
h
e
r
ep
r
esen
tatio
n
o
f
th
e
co
r
n
ea
l
q
u
ad
r
an
ts
u
s
ed
in
t
h
e
g
r
ad
e
s
ce
n
ar
io
is
illu
s
tr
ated
in
Fig
u
r
e
2
.
Fig
u
r
e
2
.
Qu
a
d
r
an
t
r
ep
r
esen
tat
io
n
in
co
r
n
ea
l u
lcer
s
[
1
0
]
2
.
2
.
P
re
pro
ce
s
s
ing
o
f
c
o
rnea
l ulce
rs
Pre
p
r
o
ce
s
s
in
g
is
th
e
in
itial
s
tag
e
o
f
d
ata
p
r
e
p
ar
atio
n
ap
p
lied
to
f
lu
o
r
escein
ey
e
im
ag
es
b
ef
o
r
e
th
ey
ar
e
u
s
ed
f
o
r
tr
ai
n
in
g
m
a
ch
in
e
lear
n
in
g
m
o
d
els.
T
h
is
s
tag
e
aim
s
to
im
p
r
o
v
e
d
ata
c
o
n
s
is
ten
cy
,
r
e
d
u
ce
ir
r
elev
an
t
v
is
u
al
in
f
o
r
m
atio
n
,
an
d
en
s
u
r
e
t
h
at
th
e
im
ag
es
a
r
e
s
u
itab
le
f
o
r
ef
f
ec
tiv
e
f
ea
tu
r
e
lear
n
in
g
.
I
n
th
is
s
tu
d
y
,
p
r
e
p
r
o
ce
s
s
in
g
co
n
s
is
ts
o
f
d
ata
o
r
g
an
izatio
n
,
co
r
n
ea
l
ar
ea
m
ask
in
g
,
d
ata
a
u
g
m
e
n
tatio
n
,
im
ag
e
r
esizin
g
,
an
d
n
o
r
m
aliza
tio
n
.
i)
Data
o
r
g
a
n
izatio
n
:
c
o
r
n
ea
l
u
l
ce
r
im
ag
es
wer
e
o
r
g
an
ized
b
ased
o
n
th
eir
d
ir
ec
t
o
r
y
s
tr
u
ctu
r
e
an
d
f
ile
n
am
in
g
c
o
n
v
e
n
tio
n
s
to
en
s
u
r
e
s
y
s
tem
atic
d
ata
h
an
d
lin
g
.
T
h
e
im
ag
es
an
d
c
o
r
r
esp
o
n
d
i
n
g
lab
els
wer
e
in
teg
r
ated
i
n
to
a
d
ataf
r
am
e
u
s
in
g
th
e
Pan
d
as
lib
r
ar
y
[
1
7
]
,
i
n
clu
d
in
g
th
e
cr
ea
tio
n
o
f
im
ag
e
p
a
t
h
r
ef
er
en
ce
s
an
d
s
tan
d
ar
d
ized
s
ce
n
ar
io
lab
els.
T
h
is
s
tep
en
s
u
r
es
co
n
s
is
ten
cy
b
etwe
en
im
ag
e
d
ata
an
d
lab
els f
o
r
s
u
b
s
eq
u
e
n
t p
r
ep
r
o
ce
s
s
in
g
an
d
m
o
d
eli
n
g
s
tag
es
[
1
8
]
.
ii)
C
o
r
n
e
a
l
a
r
e
a
m
a
s
k
in
g
:
c
o
r
n
ea
l
m
a
s
k
in
g
w
a
s
p
e
r
f
o
r
m
ed
u
s
i
n
g
t
h
e
O
p
en
C
V
l
i
b
r
a
r
y
[
1
9
]
to
r
em
o
v
e
non
-
co
r
n
e
a
l
r
eg
i
o
n
s
,
s
u
c
h
a
s
t
h
e
s
c
l
er
a
a
n
d
s
u
r
r
o
u
n
d
in
g
b
ac
k
g
r
o
u
n
d
s
t
r
u
c
tu
r
e
s
.
A
b
i
n
a
r
y
m
a
s
k
im
a
g
e
i
n
P
N
G
f
o
r
m
a
t
w
a
s
u
s
e
d
t
o
d
e
f
in
e
th
e
c
o
r
n
ea
l
r
e
g
io
n
.
T
h
e
m
a
s
k
i
n
g
p
r
o
c
e
s
s
in
v
o
l
v
e
d
i
n
v
e
r
t
in
g
t
h
e
m
a
s
k
u
s
i
n
g
t
h
e
b
i
t
w
i
s
e
_
n
o
t
f
u
n
c
t
io
n
,
f
o
l
lo
w
e
d
b
y
ap
p
ly
i
n
g
th
e
b
i
t
w
i
s
e_
a
n
d
o
p
e
r
a
t
io
n
to
p
r
e
s
er
v
e
th
e
c
o
r
n
e
a
l
a
r
ea
wh
i
l
e
s
u
p
p
r
e
s
s
i
n
g
i
r
r
e
l
e
v
a
n
t
r
eg
i
o
n
s
.
A
s
a
r
e
s
u
l
t
,
th
e
o
u
t
p
u
t
i
m
ag
e
r
e
t
a
in
e
d
o
n
l
y
th
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
0
3
-
3
6
1
3
3606
c
o
r
n
e
a
l
r
e
g
io
n
a
s
t
h
e
p
r
i
m
ar
y
a
r
e
a
o
f
i
n
t
er
e
s
t
f
o
r
m
o
d
e
l
t
r
ai
n
i
n
g
.
An
ex
a
m
p
l
e
o
f
th
e
m
a
s
k
i
n
g
r
e
s
u
l
t
i
s
p
r
e
s
e
n
t
e
d
i
n
F
ig
u
r
e
3
.
iii)
Data
au
g
m
en
tatio
n
:
d
ata
au
g
m
en
tatio
n
was
ap
p
lied
af
ter
d
ataset
s
p
litt
in
g
to
in
cr
ea
s
e
d
ata
v
ar
iab
ilit
y
an
d
b
alan
ce
th
e
n
u
m
b
er
o
f
s
am
p
les
ac
r
o
s
s
clas
s
es.
T
h
e
au
g
m
e
n
tatio
n
p
r
o
ce
s
s
u
tili
ze
d
th
e
I
m
ag
eDa
taGe
n
er
ato
r
m
o
d
u
le
f
r
o
m
T
en
s
o
r
Flo
w
[
4
]
with
s
ev
er
al
tr
an
s
f
o
r
m
atio
n
s
,
i
n
clu
d
i
n
g
h
o
r
izo
n
tal
an
d
v
er
tical
f
lip
p
in
g
,
r
o
tatio
n
,
an
d
zo
o
m
in
g
.
Au
g
m
en
tatio
n
was
s
elec
t
iv
ely
ap
p
lied
to
m
i
n
o
r
ity
class
es
to
m
ain
tain
class
b
alan
ce
[
2
0
]
.
T
h
e
tr
ain
in
g
an
d
v
alid
atio
n
d
ata
r
atio
r
em
ain
ed
co
n
s
is
ten
t
at
8
0
:2
0
af
ter
au
g
m
en
tatio
n
.
A
s
u
m
m
ar
y
o
f
t
h
e
au
g
m
e
n
ted
d
ataset
d
is
tr
ib
u
t
io
n
is
p
r
esen
ted
in
T
ab
le
1
.
iv
)
I
m
ag
e
r
esizin
g
:
t
h
e
E
f
f
icien
t
NetV2
-
S
m
o
d
el
h
as
a
n
in
p
u
t
im
ag
e
s
ize
o
f
2
2
4
×
2
2
4
p
ix
e
ls
[
2
1
]
.
I
n
th
e
o
r
ig
in
al
d
ataset,
th
e
f
lu
o
r
escein
ey
e
im
ag
e
was
2
5
9
2
×
1
7
2
8
p
ix
els.
I
n
th
is
s
tu
d
y
,
th
e
i
m
ag
e
s
ize
was
ad
ju
s
ted
to
2
2
4
×2
2
4
p
i
x
els.
T
h
is
s
tep
is
im
p
o
r
tan
t
to
en
s
u
r
e
th
at
th
e
im
ag
e
m
atch
es
th
e
s
p
ec
if
icatio
n
s
r
eq
u
ir
ed
b
y
t
h
e
m
o
d
el
with
o
u
t
o
m
itti
n
g
im
p
o
r
tan
t d
etails in
t
h
e
im
ag
e.
v)
Data
n
o
r
m
aliza
tio
n
:
i
t
is
th
e
l
ast
s
tag
e
o
f
th
e
s
u
b
-
s
tag
e
o
f
d
ata
p
r
e
p
r
o
ce
s
s
in
g
ca
r
r
ied
o
u
t
in
th
is
s
tu
d
y
.
T
h
e
p
u
r
p
o
s
e
o
f
n
o
r
m
aliza
tio
n
is
to
h
elp
th
e
m
o
d
el
t
r
ain
in
g
p
r
o
ce
s
s
r
u
n
f
aster
an
d
m
o
r
e
e
f
f
ec
tiv
ely
.
E
ac
h
p
ix
el
o
n
th
e
in
p
u
t
im
ag
e
h
as
a
v
alu
e
f
r
o
m
th
e
r
an
g
e
o
f
0
–
2
5
5
.
T
h
er
ef
o
r
e,
n
o
r
m
aliza
tio
n
is
d
o
n
e
o
n
ea
ch
g
r
ay
s
ca
le
im
ag
e
b
y
d
iv
i
d
in
g
t
h
e
p
ix
el
v
al
u
e
b
y
2
5
5
,
r
esu
ltin
g
in
a
n
ew
v
alu
e
with
a
r
a
n
g
e
o
f
0
–
1
[
2
2
]
.
Fig
u
r
e
3
.
E
x
am
p
le
o
f
co
r
n
ea
l
ar
ea
m
ask
in
g
a
p
p
lied
to
f
lu
o
r
e
s
ce
in
im
ag
es
T
ab
le
1
.
Su
m
m
a
r
y
o
f
d
ataset
d
is
tr
ib
u
tio
n
af
ter
au
g
m
en
tatio
n
S
c
e
n
a
r
i
o
O
r
i
g
i
n
a
l
t
r
a
i
n
/
v
al
A
d
d
e
d
t
r
a
i
n
/
v
al
F
i
n
a
l
t
r
a
i
n
/
v
al
C
a
t
e
g
o
r
y
5
6
9
/
1
4
3
2
8
9
/
7
3
8
5
8
/
2
1
6
Ty
p
e
5
6
9
/
1
4
3
1
,
6
2
1
/
4
0
7
2
,
1
9
0
/
5
5
0
G
r
a
d
e
5
6
9
/
1
4
3
5
2
1
/
1
3
2
1
,
0
9
0
/
2
7
5
2
.
3
.
Da
t
a
s
pli
t
t
ing
T
h
e
d
ata
is
d
iv
id
ed
in
to
two
,
n
am
ely
tr
ain
in
g
d
ata
(
tr
ain
_
d
f
)
an
d
v
alid
atio
n
(
v
al
_
d
f
)
with
a
r
atio
o
f
8
:2
.
T
h
is
m
ea
n
s
th
at
as
m
u
ch
as
8
0
%
is
f
o
r
tr
ain
in
g
d
ata
a
n
d
2
0
%
is
f
o
r
v
alid
atio
n
d
ata.
T
h
e
d
iv
is
io
n
o
f
d
ata
is
d
iv
id
ed
in
to
two
class
es
b
e
ca
u
s
e
o
f
th
e
s
m
all
am
o
u
n
t
o
f
d
ata
an
d
th
e
r
e
is
a
n
im
b
alan
c
e
in
t
h
e
a
m
o
u
n
t
o
f
d
ata
b
etwe
en
th
e
class
es.
E
ac
h
s
ce
n
ar
io
im
p
lem
en
ts
.
c
o
p
y
(
)
t
o
d
u
p
licate
th
e
in
itial
d
ataf
r
a
m
e.
E
ac
h
d
ataf
r
am
e
will
co
n
tain
o
n
ly
th
e
f
ile
n
am
e,
im
ag
e
d
ir
ec
to
r
y
,
an
d
lab
el
n
am
e
attr
ib
u
tes
th
at
ar
e
r
elev
an
t
to
th
at
s
ce
n
ar
io
.
T
h
e
d
ataf
r
am
e
o
f
ea
ch
s
ce
n
ar
io
is
d
ef
in
ed
as
d
f
_
ca
t
f
o
r
ca
teg
o
r
ies,
d
f
_
ty
p
e
f
o
r
th
e
ty
p
e
,
an
d
d
f
_
g
r
ad
e
f
o
r
g
r
ad
e.
T
h
u
s
,
th
e
d
ata
f
o
r
ea
c
h
s
ce
n
ar
io
ca
n
b
e
m
an
a
g
ed
s
ep
ar
ately
,
m
ak
in
g
it
ea
s
ier
to
an
aly
ze
an
d
tr
ain
s
p
ec
if
ic
m
o
d
els
o
n
ea
ch
d
ata
ch
ar
ac
ter
is
tic.
Nex
t,
ea
ch
s
ce
n
ar
io
d
ataf
r
am
e
will
b
e
d
iv
id
ed
in
to
tr
ain
in
g
an
d
v
alid
atio
n
d
ata.
T
h
e
d
ata
s
h
ar
in
g
r
atio
is
8
0
%
f
o
r
tr
ain
in
g
d
ata
an
d
2
0
%
f
o
r
v
alid
atio
n
d
ata.
Data
s
h
ar
in
g
is
ca
r
r
ied
o
u
t
u
s
in
g
th
e
f
u
n
ctio
n
s
in
th
e
Scik
itLe
ar
n
Mo
d
u
le,
n
am
ely
tr
ain
_
test
_
s
p
lit()
[
2
3
]
.
Var
iab
le
Data
f
r
am
e
d
if
f
er
en
tiated
b
y
d
ata
ty
p
e
a
n
d
s
ce
n
ar
io
ty
p
e.
2
.
3
.
1
.
Da
t
a
s
pli
t
t
ing
t
o
prev
e
nt
da
t
a
lea
k
a
g
e
I
n
p
r
o
ce
s
s
in
g
m
e
d
ical
im
ag
es,
esp
ec
ially
ey
e
im
a
g
es
with
f
lu
o
r
escein
,
d
ata
lea
k
a
g
e
is
v
er
y
im
p
o
r
tan
t
t
o
r
e
p
air
.
T
h
is
is
b
e
ca
u
s
e
m
ed
ical
im
a
g
es
ar
e
o
b
t
ain
ed
f
r
o
m
d
if
f
e
r
en
t
p
atien
ts
,
s
o
th
at
later
it
will
af
f
ec
t
f
o
llo
w
-
u
p
r
esear
ch
,
n
a
m
ely
d
iag
n
o
s
is
[
2
4
]
.
Data
le
ak
ag
e
o
cc
u
r
s
wh
en
im
ag
e
a
u
g
m
en
tatio
n
is
d
o
n
e
b
ef
o
r
e
d
ata
s
h
ar
i
n
g
.
Data
lea
k
ag
e
ca
n
b
e
av
o
id
ed
b
y
s
h
ar
in
g
d
ata
f
ir
s
t
b
e
f
o
r
e
th
e
a
u
g
m
en
tatio
n
s
tag
e
t
o
en
s
u
r
e
th
at
th
e
v
alid
atio
n
d
ata
d
o
es
n
o
t
co
n
tain
im
ag
e
s
o
f
th
e
s
am
e
p
atien
ts
as
th
e
tr
ain
in
g
d
ata.
T
h
u
s
,
v
alid
atio
n
d
ata
will
p
r
o
v
id
e
m
o
r
e
ac
cu
r
ate
r
esu
lts
b
ec
au
s
e
it
co
n
s
is
ts
o
f
d
ata
th
at
t
h
e
m
o
d
el
h
as
n
ev
e
r
s
ee
n
an
d
s
tu
d
ied
.
T
h
er
e
f
o
r
e,
d
ata
s
h
ar
in
g
is
im
p
o
r
tan
t
to
d
o
af
ter
im
ag
e
m
ask
in
g
an
d
b
ef
o
r
e
au
g
m
e
n
tatio
n
in
d
ata
p
r
ep
r
o
ce
s
s
in
g
[
2
5
]
.
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
E
fficien
t d
ee
p
lea
r
n
in
g
f
o
r
a
u
t
o
ma
ted
co
r
n
ea
l
u
lcer
s
ev
erit
y
cla
s
s
ifica
tio
n
fr
o
m
flu
o
r
escein
ima
g
e
s
(
R
o
d
ia
h
)
3607
2
.
3
.
2
.
Da
t
a
s
pli
t
t
ing
un
der
d
a
t
a
lea
k
a
g
e
co
nd
it
io
ns
Data
leak
ag
e
n
ee
d
s
t
o
b
e
test
ed
b
y
cr
ea
tin
g
c
o
n
d
itio
n
s
wh
er
e
d
ata
leak
ag
e
o
cc
u
r
s
to
co
m
p
ar
e
t
h
e
r
esu
lts
o
f
d
ata
leak
ag
e
an
d
n
o
t.
Usi
n
g
th
e
im
ag
e
d
ata
g
en
er
ato
r
(
I
DG)
in
th
e
d
ata
p
r
e
p
r
o
ce
s
s
in
g
in
ea
ch
au
g
m
en
tatio
n
p
r
o
ce
s
s
will
r
es
u
lt
in
d
if
f
er
en
t
im
ag
e
v
ar
iatio
n
s
,
th
u
s
av
o
id
in
g
d
u
p
licatio
n
[
2
6
]
.
I
n
th
is
s
tu
d
y
,
d
ata
leak
ag
e
ca
n
b
e
ca
r
r
ied
o
u
t
b
y
r
e
-
a
r
r
an
g
i
n
g
tr
ain
in
g
an
d
v
alid
atio
n
d
ata
th
at
h
av
e
b
e
en
au
g
m
e
n
ted
an
d
s
ep
ar
ated
p
r
e
v
io
u
s
ly
.
Data
a
u
g
m
en
tatio
n
is
d
o
n
e
o
n
ce
,
a
n
d
th
e
r
esu
lts
ca
n
b
e
c
o
m
p
a
r
ed
.
I
f
th
e
d
ata
h
as
b
ee
n
co
m
b
in
ed
,
th
e
n
th
e
d
ata
will
b
e
r
ed
is
tr
ib
u
ted
u
s
in
g
th
e
s
tr
ati
f
ied
s
am
p
lin
g
m
eth
o
d
[
2
7
]
wit
h
th
e
s
am
e
r
atio
o
f
8
0
:2
0
.
T
h
is
ca
n
ca
u
s
e
th
e
im
ag
e
d
ata
o
f
e
y
e
u
lcer
s
with
f
lu
o
r
escein
f
r
o
m
t
h
e
s
am
e
p
atien
t
to
b
e
in
b
o
th
tr
ain
in
g
a
n
d
v
alid
atio
n
d
ata.
T
h
e
r
esu
lts
o
f
th
e
m
o
d
el
p
er
f
o
r
m
an
ce
e
v
alu
atio
n
u
s
in
g
d
ata
leak
ag
e
ca
n
b
e
u
s
ed
to
co
m
p
ar
e
with
th
e
r
esu
lts
o
f
m
o
d
el
p
er
f
o
r
m
a
n
ce
ev
alu
atio
n
with
o
u
t d
ata
leak
a
g
e
co
n
d
itio
n
s
.
2
.
4
.
P
r
o
po
s
ed
m
o
del a
rc
hite
ct
ure
f
o
r
a
uto
m
a
t
ed
cla
s
s
if
ica
t
io
n o
f
c
o
rnea
l ulce
r
s
ev
er
it
y
Fig
u
r
e
4
illu
s
tr
ates
th
e
p
r
o
p
o
s
ed
E
f
f
icien
tNetV2
-
S
ar
ch
itectu
r
e
f
o
r
co
r
n
ea
l
u
lcer
s
ev
er
ity
class
if
icatio
n
,
co
n
s
is
tin
g
o
f
th
e
p
r
e
-
tr
ain
e
d
m
o
d
el
an
d
th
e
f
i
n
e
-
tu
n
ed
m
o
d
el.
Fig
u
r
e
4
(
a)
p
r
esen
ts
p
r
e
-
tr
ain
ed
E
f
f
icien
tNetV2
-
S
ar
ch
itectu
r
e
u
s
ed
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
w
h
ile
Fig
u
r
e
4
(
b
)
s
h
o
ws
th
e
f
in
e
-
tu
n
ed
ar
ch
itectu
r
e
p
r
o
p
o
s
ed
f
o
r
co
r
n
ea
l
u
lcer
s
ev
er
ity
class
if
icatio
n
.
T
h
e
p
r
e
-
tr
ain
ed
E
f
f
icien
tNetV2
-
S
m
o
d
e
l
u
s
ed
in
th
is
s
tu
d
y
was
o
b
tain
ed
f
r
o
m
th
e
tf
.
k
er
as
.
ap
p
licatio
n
s
m
o
d
u
le
an
d
h
ad
b
ee
n
p
r
e
v
io
u
s
ly
tr
ain
e
d
o
n
th
e
I
m
ag
eNe
t
d
ataset
co
n
tain
in
g
1
,
0
0
0
class
es.
Flu
o
r
escein
im
ag
es
with
a
r
eso
lu
tio
n
o
f
2
2
4
×2
2
4
p
ix
els
wer
e
u
s
ed
as
in
p
u
t,
wh
ile
th
e
ex
tr
ac
ted
f
ea
tu
r
es
wer
e
p
r
o
ce
s
s
ed
u
s
in
g
a
Glo
b
alAv
e
r
ag
ePo
o
lin
g
2
D
la
y
er
[
2
8
]
,
f
o
llo
w
ed
b
y
a
d
en
s
e
lay
er
with
R
eL
U
ac
tiv
atio
n
an
d
a
f
i
n
al
s
o
f
tm
ax
class
if
icatio
n
lay
e
r
[
2
9
]
.
T
h
e
p
r
o
p
o
s
ed
ar
ch
itect
u
r
e
was d
esig
n
ed
as
an
ef
f
icien
t
en
d
-
to
-
e
n
d
m
o
d
e
l
with
r
elativ
ely
lo
w
co
m
p
u
t
atio
n
al
co
m
p
lex
ity
an
d
p
o
ten
tial
in
teg
r
atio
n
in
to
clin
ical
d
ec
is
io
n
-
s
u
p
p
o
r
t
s
y
s
t
em
s
f
o
r
ea
r
ly
c
o
r
n
ea
l
u
lcer
d
iag
n
o
s
is
[
3
0
]
.
E
f
f
icien
tNet
V2
-
S
was
s
elec
te
d
b
ec
au
s
e
it
is
a
lig
h
tweig
h
t
an
d
m
o
d
er
n
ar
ch
itectu
r
e
th
at
r
em
ain
s
r
elativ
ely
u
n
d
er
ex
p
l
o
r
ed
f
o
r
f
lu
o
r
escein
im
ag
e
class
if
icatio
n
in
o
p
h
th
alm
o
lo
g
y
,
wh
ile
th
e
s
elec
tiv
e
f
in
e
-
tu
n
in
g
s
tr
ateg
y
was
in
ten
d
ed
to
im
p
r
o
v
e
m
o
d
el
g
e
n
er
aliza
tio
n
o
n
lim
ited
d
atasets
.
(
a)
(
b
)
Fig
u
r
e
4
.
Pro
p
o
s
ed
E
f
f
icien
tNetV2
-
S
ar
ch
itectu
r
e
f
o
r
co
r
n
ea
l u
lcer
s
ev
er
ity
class
if
icatio
n
(
a)
p
r
e
-
tr
ai
n
ed
m
o
d
el
an
d
(
b
)
f
in
e
-
t
u
n
ed
m
o
d
el
2
.
5
.
Cla
s
s
if
ica
t
io
n
m
o
del t
ra
ini
ng
I
n
th
is
s
tu
d
y
,
th
e
m
o
d
el
was
tr
ain
ed
f
o
r
5
0
ep
o
ch
s
u
s
in
g
e
ar
ly
s
to
p
p
in
g
an
d
ad
a
p
tiv
e
lear
n
in
g
r
ate
s
tr
ateg
ies
to
im
p
r
o
v
e
co
n
v
e
r
g
en
ce
ef
f
icien
cy
.
T
h
e
ea
r
ly
s
to
p
p
in
g
m
ec
h
an
is
m
was
co
n
f
ig
u
r
ed
with
a
p
atien
ce
v
alu
e
o
f
1
0
ep
o
c
h
s
,
m
ea
n
in
g
t
h
at
tr
ain
in
g
was
au
to
m
atica
lly
ter
m
in
ated
if
n
o
p
er
f
o
r
m
an
ce
im
p
r
o
v
em
e
n
t
was
o
b
s
er
v
ed
f
o
r
1
0
co
n
s
ec
u
tiv
e
ep
o
c
h
s
.
I
n
a
d
d
itio
n
,
th
e
le
ar
n
in
g
r
ate
was
ad
ap
tiv
ely
ad
ju
s
ted
u
s
in
g
th
e
R
ed
u
ce
L
R
On
Plateau
s
tr
ateg
y
b
ased
o
n
th
e
v
alid
atio
n
lo
s
s
m
etr
ic,
with
a
p
atien
ce
v
alu
e
o
f
5
ep
o
ch
s
an
d
a
r
ed
u
ctio
n
f
ac
to
r
o
f
0
.
2
.
T
h
e
m
in
im
u
m
lear
n
i
n
g
r
ate
was
l
im
ited
to
0
.
0
0
0
0
0
1
to
p
r
ev
e
n
t
ex
ce
s
s
iv
ely
s
lo
w
tr
ain
in
g
[
3
1
]
.
Fo
r
th
e
f
in
al
p
er
f
o
r
m
an
ce
ev
alu
atio
n
,
t
h
e
v
alid
atio
n
d
ataf
r
am
e
(
v
al_
d
f
)
was
a
ls
o
u
s
ed
as
th
e
test
d
ataf
r
am
e
(
test
_
d
f
)
.
T
h
ese
s
tr
ateg
ies
wer
e
im
p
lem
en
ted
to
m
ain
tain
tr
ain
in
g
s
tab
ilit
y
an
d
im
p
r
o
v
e
th
e
m
o
d
el’
s
g
en
e
r
aliza
tio
n
ca
p
a
b
ilit
y
f
o
r
c
o
r
n
ea
l u
lce
r
class
if
icatio
n
.
2
.
6
.
E
v
a
lua
t
i
o
n
m
et
rics
T
o
c
o
m
p
r
eh
en
s
iv
el
y
e
v
al
u
ate
t
h
e
p
e
r
f
o
r
m
a
n
ce
o
f
t
h
e
p
r
o
p
o
s
e
d
c
o
r
n
ea
l
u
lc
e
r
class
if
ica
ti
o
n
m
o
d
el
,
m
u
lti
p
l
e
e
v
al
u
at
io
n
m
etr
ics
a
r
e
e
m
p
l
o
y
e
d
i
n
a
d
d
iti
o
n
t
o
o
v
er
all
ac
c
u
r
ac
y
.
I
n
m
ed
ic
al
i
m
ag
e
a
n
al
y
s
is
tas
k
s
,
r
el
y
i
n
g
s
o
l
el
y
o
n
a
cc
u
r
ac
y
is
w
id
e
ly
r
ec
o
g
n
iz
ed
as
i
n
s
u
f
f
ic
ie
n
t d
u
e
t
o
class
i
m
b
ala
n
ce
an
d
t
h
e
h
i
g
h
cl
in
ica
l c
o
s
t
o
f
m
is
class
if
ica
ti
o
n
,
p
a
r
ti
c
u
la
r
l
y
i
n
d
is
e
ase
d
e
tec
ti
o
n
s
ce
n
ar
i
o
s
[
3
2
]
,
[
3
3
]
.
T
h
er
e
f
o
r
e
,
t
h
is
s
t
u
d
y
e
v
al
u
ates
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
0
3
-
3
6
1
3
3608
m
o
d
el
p
er
f
o
r
m
an
ce
u
s
i
n
g
ac
cu
r
a
cy
,
s
e
n
s
it
iv
it
y
(
r
e
ca
l
l)
,
s
p
ec
if
icit
y
,
a
n
d
t
h
e
a
r
ea
u
n
d
e
r
t
h
e
c
u
r
v
e
(
AUC)
,
w
h
ic
h
ar
e
co
m
m
o
n
l
y
r
ec
o
m
m
e
n
d
e
d
m
et
r
i
cs i
n
m
e
d
ic
al
a
r
ti
f
ic
ial
i
n
t
elli
g
e
n
ce
e
v
al
u
ati
o
n
f
r
a
m
ew
o
r
k
s
[
3
4
]
,
[
3
5
]
.
Acc
u
r
ac
y
m
ea
s
u
r
es
th
e
o
v
e
r
all
co
r
r
ec
tn
ess
o
f
th
e
class
i
f
icatio
n
,
wh
ile
s
en
s
itiv
ity
r
ef
lects
th
e
m
o
d
el’
s
ab
ilit
y
to
co
r
r
ec
tly
id
en
tify
p
o
s
itiv
e
u
lcer
ca
s
es,
w
h
ich
is
cr
itical
f
o
r
m
in
im
izin
g
m
is
s
ed
d
iag
n
o
s
es
an
d
d
elay
e
d
tr
ea
tm
en
t
[
3
6
]
.
Sp
ec
if
icity
ev
alu
ates
th
e
ab
ilit
y
o
f
th
e
m
o
d
el
to
co
r
r
ec
tly
id
e
n
tify
n
eg
ativ
e
ca
s
es,
th
er
eb
y
r
ed
u
ci
n
g
f
alse
-
p
o
s
itiv
e
p
r
e
d
ictio
n
s
th
at
m
ay
lea
d
to
u
n
n
ec
ess
ar
y
clin
ical
in
t
er
v
en
tio
n
s
[
3
7
]
.
I
n
ad
d
itio
n
,
t
h
e
AUC
m
etr
ic
p
r
o
v
id
es
a
th
r
esh
o
ld
-
i
n
d
ep
e
n
d
e
n
t
ass
ess
m
en
t
o
f
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
b
y
q
u
an
tify
in
g
th
e
tr
ad
e
-
o
f
f
b
et
wee
n
s
en
s
itiv
ity
an
d
s
p
ec
if
ici
ty
,
m
ak
in
g
it
p
a
r
ticu
lar
ly
s
u
it
ab
le
f
o
r
ev
alu
atin
g
d
iag
n
o
s
tic
r
eliab
ilit
y
in
o
p
h
th
alm
ic
im
ag
in
g
ap
p
licatio
n
s
[
3
8
]
,
[
3
9
]
.
Fo
r
m
u
lti
-
class
class
if
icatio
n
s
ce
n
ar
io
s
(
ca
teg
o
r
y
,
ty
p
e,
an
d
g
r
ad
e
)
,
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
ar
e
co
m
p
u
ted
u
s
in
g
a
m
ac
r
o
-
a
v
er
ag
in
g
s
tr
ateg
y
to
en
s
u
r
e
b
alan
ce
d
p
er
f
o
r
m
an
ce
ev
alu
atio
n
ac
r
o
s
s
all
class
es,
r
eg
ar
d
less
o
f
class
d
is
tr
ib
u
tio
n
[
4
0
]
.
C
o
llectiv
ely
,
th
ese
m
etr
ics
o
f
f
er
a
m
o
r
e
clin
ically
m
ea
n
in
g
f
u
l
ass
ess
m
en
t
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
e
l
an
d
f
ac
ilit
ate
f
air
co
m
p
ar
is
o
n
with
p
r
ev
io
u
s
s
tu
d
ies in
au
to
m
ated
c
o
r
n
ea
l
u
lcer
class
if
icatio
n
.
3.
RE
SU
L
T
AND
DI
SCUS
SI
O
N
T
h
e
d
a
t
a
p
r
e
p
r
o
c
e
s
s
i
n
g
s
t
a
g
e
is
c
a
r
r
i
e
d
o
u
t
t
o
i
n
c
r
ea
s
e
t
h
e
v
ar
i
e
t
y
o
f
t
r
a
i
n
i
n
g
d
a
t
a
a
n
d
p
r
e
p
ar
e
t
h
e
d
a
t
a
s
o
t
h
a
t
t
h
e
t
r
a
i
n
i
n
g
p
r
o
c
es
s
c
an
r
u
n
e
f
f
e
c
t
i
v
e
l
y
.
T
h
i
s
s
t
a
g
e
s
ta
r
t
s
f
r
o
m
l
o
a
d
i
n
g
d
a
t
a
i
n
t
o
t
h
e
d
a
t
a
f
r
a
m
e
,
m
as
k
i
n
g
t
h
e
c
o
r
n
e
a
l
a
r
e
a
,
a
n
d
a
d
d
i
n
g
d
at
a
w
i
t
h
a
u
g
m
e
n
ta
t
i
o
n
.
I
t
c
o
n
ti
n
u
e
s
w
it
h
a
d
j
u
s
ti
n
g
t
h
e
i
m
a
g
e
s
iz
e
a
n
d
n
o
r
m
a
l
i
z
i
n
g
.
3
.
1
.
Resul
t
s
o
f
t
he
ima
g
e
m
a
s
k
ing
s
t
a
g
e
T
h
e
m
ask
in
g
p
r
o
ce
s
s
was
ca
r
r
ied
o
u
t
to
elim
in
ate
n
o
is
e
o
u
t
s
id
e
th
e
co
r
n
ea
l
ar
ea
u
s
in
g
th
e
Op
en
C
V
m
o
d
u
le.
Fig
u
r
e
5
illu
s
tr
ates
th
e
co
r
n
ea
l
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
,
s
h
o
win
g
th
e
c
o
m
p
a
r
is
o
n
b
etwe
en
th
e
o
r
ig
in
al
f
lu
o
r
escein
im
ag
e
a
n
d
th
e
m
ask
ed
c
o
r
n
ea
l
r
eg
io
n
.
Fig
u
r
e
5
(
a
)
s
h
o
ws
th
e
o
r
ig
in
al
f
lu
o
r
escein
im
ag
e,
wh
ile
Fig
u
r
e
5
(
b
)
p
r
esen
ts
th
e
m
ask
in
g
r
esu
lt
in
wh
ich
r
eg
io
n
s
o
u
ts
id
e
th
e
co
r
n
ea
ar
e
s
u
p
p
r
ess
ed
u
s
in
g
b
lack
p
ix
els,
allo
win
g
th
e
c
o
r
n
ea
l
ar
ea
to
b
ec
o
m
e
th
e
p
r
im
a
r
y
f
o
c
u
s
f
o
r
m
o
d
el
tr
ai
n
in
g
.
A
f
ter
m
ask
in
g
,
d
ata
au
g
m
en
tatio
n
was
ap
p
lied
to
g
en
er
ate
ad
d
itio
n
al
im
ag
e
v
a
r
iatio
n
s
an
d
b
alan
ce
th
e
n
u
m
b
e
r
o
f
s
am
p
les
in
ea
ch
class
.
T
h
e
au
g
m
en
tatio
n
p
r
o
c
ess
wa
s
p
er
f
o
r
m
ed
u
s
in
g
I
m
ag
eDa
taGe
n
er
ato
r
with
s
ev
er
al
tr
an
s
f
o
r
m
atio
n
s
,
in
clu
d
in
g
h
o
r
izo
n
tal
an
d
v
e
r
tical
f
lip
p
in
g
,
r
o
tatio
n
,
an
d
zo
o
m
in
g
.
T
h
e
d
ataset
was
d
iv
id
ed
in
to
tr
ain
in
g
an
d
v
alid
atio
n
s
ets
u
s
in
g
an
8
0
:2
0
r
atio
ac
r
o
s
s
th
e
ca
teg
o
r
y
,
ty
p
e,
an
d
g
r
ad
e
class
if
icatio
n
s
ce
n
ar
io
s
.
T
o
av
o
i
d
d
ata
leak
ag
e,
a
u
g
m
e
n
ted
im
ag
es
d
er
iv
ed
f
r
o
m
t
h
e
s
am
e
o
r
ig
in
al
im
ag
e
wer
e
ca
r
ef
u
lly
s
ep
ar
ated
b
etwe
en
t
h
e
tr
ain
in
g
an
d
v
alid
atio
n
s
ets.
T
h
is
s
tr
ateg
y
en
s
u
r
e
d
th
at
m
o
d
el
ev
alu
atio
n
was
co
n
d
u
cte
d
u
s
in
g
u
n
s
ee
n
d
ata
an
d
p
r
o
v
id
e
d
a
m
o
r
e
r
eliab
le
a
s
s
es
s
m
en
t o
f
m
o
d
el
p
er
f
o
r
m
an
ce
.
(
a)
(
b
)
Fig
u
r
e
5
.
C
o
r
n
ea
l
im
ag
es b
ef
o
r
e
an
d
a
f
ter
m
ask
in
g
(
a)
th
e
o
r
ig
in
al
f
lu
o
r
escein
im
a
g
e
an
d
(
b
)
th
e
m
ask
e
d
co
r
n
ea
l r
e
g
io
n
3
.
2
.
Cla
s
s
if
ica
t
io
n
m
o
del t
ra
ini
ng
re
s
ult
s
T
h
e
r
esu
lts
o
f
th
e
m
ac
h
in
e
lear
n
in
g
m
o
d
elin
g
s
tag
es
ar
e
d
iv
i
d
ed
in
to
th
r
ee
s
tag
es,
n
am
ely
t
h
e
r
esu
lts
o
f
m
o
d
el
in
itiatio
n
,
th
e
r
esu
lts
o
f
m
o
d
el
tr
ain
i
n
g
,
a
n
d
th
e
r
esu
lts
o
f
m
o
d
el
tr
ain
in
g
ev
a
lu
atio
n
.
T
h
is
s
tag
e
p
r
o
d
u
ce
d
th
r
ee
p
r
e
-
tr
ain
ed
C
NN
m
o
d
els
.
T
h
ese
m
o
d
els
w
er
e
ab
le
to
class
if
y
co
r
n
ea
l
u
lcer
s
b
ased
o
n
th
r
ee
s
ce
n
ar
io
s
,
n
am
ely
ca
teg
o
r
y
,
ty
p
e,
an
d
g
r
ad
e
.
3
.
2
.
1
.
M
o
del
ini
t
ia
t
io
n r
esu
lt
s
T
h
e
m
o
d
el
in
itializatio
n
in
th
i
s
s
tu
d
y
was
p
er
f
o
r
m
ed
b
y
co
m
b
in
in
g
th
e
p
r
e
-
tr
ain
e
d
E
f
f
ici
en
tNetV2
-
S
ar
ch
itectu
r
e
with
f
u
lly
c
o
n
n
e
cted
lay
er
s
f
o
r
ca
teg
o
r
y
,
ty
p
e
,
an
d
g
r
a
d
e
class
if
icatio
n
o
f
co
r
n
ea
l
u
lcer
s
.
T
h
e
E
f
f
icien
tNetV2
-
S
m
o
d
el
was
o
b
tain
ed
f
r
o
m
th
e
t
f
.
k
er
as.a
p
p
licatio
n
s
m
o
d
u
le
a
n
d
h
ad
b
ee
n
p
r
e
v
io
u
s
ly
tr
ai
n
ed
o
n
th
e
I
m
a
g
eNe
t
d
ataset.
T
h
e
ca
teg
o
r
y
class
if
icatio
n
m
o
d
el
co
n
s
is
ted
o
f
2
1
,
6
4
6
,
1
7
9
t
o
tal
p
ar
am
eter
s
,
wh
ile
th
e
ty
p
e
an
d
g
r
a
d
e
class
if
icatio
n
m
o
d
els ea
ch
co
n
tain
e
d
2
1
,
6
4
8
,
2
2
9
p
ar
am
eter
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
fficien
t d
ee
p
lea
r
n
in
g
f
o
r
a
u
t
o
ma
ted
co
r
n
ea
l
u
lcer
s
ev
erit
y
cla
s
s
ifica
tio
n
fr
o
m
flu
o
r
escein
ima
g
e
s
(
R
o
d
ia
h
)
3609
Mo
d
el
tr
ain
in
g
e
v
alu
atio
n
wa
s
co
n
d
u
cted
to
a
n
aly
ze
th
e
le
ar
n
in
g
p
e
r
f
o
r
m
an
ce
d
u
r
in
g
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
.
Fig
u
r
e
6
illu
s
tr
ates
t
h
e
tr
ain
in
g
ev
alu
atio
n
r
esu
lts
o
f
th
e
ca
teg
o
r
y
class
if
icatio
n
m
o
d
el
u
s
in
g
th
e
E
f
f
icien
tNetV2
-
S
ar
ch
itectu
r
e
o
v
er
5
0
ep
o
c
h
s
.
Fig
u
r
e
6
(
a
)
p
r
esen
ts
th
e
tr
ain
in
g
an
d
v
ali
d
atio
n
lo
s
s
cu
r
v
es,
wh
ile
Fig
u
r
e
6
(
b
)
s
h
o
ws
th
e
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
a
cy
cu
r
v
es
f
o
r
th
e
ca
teg
o
r
y
cla
s
s
if
icatio
n
s
ce
n
ar
io
u
s
in
g
th
e
E
f
f
icien
tNetV2
-
S
a
r
ch
itectu
r
e.
T
h
e
m
o
d
el
ac
h
ie
v
ed
th
e
h
ig
h
est
v
alid
atio
n
ac
c
u
r
ac
y
o
f
9
5
.
9
3
%
at
ep
o
ch
2
4
,
wh
ile
th
e
l
o
west
v
alid
atio
n
lo
s
s
o
f
0
.
1
2
9
2
was
o
b
s
er
v
ed
d
u
r
in
g
th
e
ea
r
l
y
tr
a
in
in
g
s
tag
e.
T
h
ese
r
esu
lts
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
m
o
d
el
was
ab
le
t
o
l
ea
r
n
r
ele
v
an
t
im
ag
e
f
ea
tu
r
es
ef
f
ec
tiv
ely
wh
ile
m
ain
tain
in
g
s
tab
le
v
alid
atio
n
p
er
f
o
r
m
an
ce
.
(
a)
(
b
)
Fig
u
r
e
6
.
E
v
alu
atio
n
r
esu
lts
o
f
th
e
ca
teg
o
r
y
class
if
icatio
n
m
o
d
el
tr
ain
in
g
(
a)
tr
ain
in
g
an
d
v
a
lid
atio
n
lo
s
s
cu
r
v
es a
n
d
(
b
)
t
r
ain
in
g
a
n
d
v
alid
atio
n
ac
cu
r
ac
y
cu
r
v
es
Ov
er
all,
th
e
tr
ain
in
g
p
r
o
ce
s
s
f
o
r
th
e
n
o
n
-
leak
ag
e
ca
te
g
o
r
y
s
ce
n
ar
i
o
r
eq
u
i
r
ed
ap
p
r
o
x
im
ately
3
3
m
in
u
tes.
I
n
th
is
s
ce
n
ar
io
,
th
e
m
o
d
el
ac
h
ie
v
ed
th
e
h
ig
h
est
v
alid
atio
n
ac
c
u
r
ac
y
o
f
9
5
.
9
3
%
at
e
p
o
ch
2
4
.
Un
d
er
leak
ag
e
co
n
d
itio
n
s
,
th
e
ca
teg
o
r
y
class
if
icatio
n
m
o
d
el
o
b
tain
ed
a
m
ax
im
u
m
v
alid
atio
n
ac
cu
r
ac
y
o
f
8
7
.
9
1
%
at
ep
o
ch
3
0
,
wh
ile
th
e
lo
west
v
alid
atio
n
lo
s
s
o
f
0
.
5
2
9
2
was
r
ec
o
r
d
ed
at
ep
o
ch
1
0
.
Me
an
wh
ile,
th
e
ty
p
e
class
if
icatio
n
s
ce
n
ar
io
with
leak
ag
e
d
ata
ac
h
iev
e
d
th
e
h
i
g
h
est
v
alid
atio
n
ac
cu
r
ac
y
o
f
9
7
.
9
9
%
at
ep
o
ch
3
6
,
with
th
e
lo
west
v
a
lid
atio
n
lo
s
s
o
f
0
.
1
0
7
0
o
b
tai
n
ed
at
e
p
o
ch
2
2
.
T
h
ese
r
esu
l
ts
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
p
r
ep
r
o
ce
s
s
in
g
an
d
d
ata
-
s
p
litt
in
g
s
tr
ateg
y
co
n
tr
ib
u
t
ed
to
s
tab
le
m
o
d
el
p
er
f
o
r
m
a
n
ce
an
d
r
ed
u
ce
d
th
e
im
p
ac
t o
f
d
ata
lea
k
ag
e
d
u
r
in
g
ev
alu
atio
n
.
T
h
e
s
u
m
m
ar
ize
d
tr
ain
in
g
r
esu
lts
ar
e
p
r
esen
te
d
in
T
ab
le
2
.
T
ab
le
2
.
Su
m
m
a
r
y
o
f
f
in
e
-
tu
n
ed
E
f
f
icien
tNetV2
-
S
tr
ain
in
g
r
esu
lts
S
c
e
n
a
r
i
o
D
a
t
a
c
o
n
d
i
t
i
o
n
B
e
st
v
a
l
i
d
a
t
i
o
n
a
c
c
u
r
a
c
y
(
%)
B
e
st
v
a
l
i
d
a
t
i
o
n
l
o
ss
Ep
o
c
h
C
a
t
e
g
o
r
y
N
o
n
-
l
e
a
k
a
g
e
9
5
.
9
3
0
.
9
1
9
3
24
C
a
t
e
g
o
r
y
Le
a
k
a
g
e
8
7
.
9
1
0
.
5
2
9
2
30
Ty
p
e
Le
a
k
a
g
e
9
7
.
9
9
0
.
1
0
7
0
36
3
.
3
.
Co
m
pa
riso
n wit
h
prev
io
us
s
t
ud
ie
s
a
nd
perf
o
rm
a
nce
dis
cu
s
s
io
n
T
ab
le
3
s
u
m
m
ar
izes
a
co
m
p
ar
is
o
n
b
etwe
en
th
is
s
tu
d
y
an
d
r
elev
an
t
p
r
ev
io
u
s
wo
r
k
s
o
n
co
r
n
ea
l
u
lcer
class
if
icatio
n
u
s
in
g
f
lu
o
r
escein
im
ag
es,
h
ig
h
lig
h
tin
g
d
if
f
e
r
en
ce
s
in
m
eth
o
d
o
lo
g
y
,
class
if
icatio
n
s
co
p
e,
an
d
p
er
f
o
r
m
an
ce
.
T
h
e
co
m
p
ar
is
o
n
f
o
cu
s
es
o
n
th
e
lear
n
in
g
ar
ch
itectu
r
es,
class
if
icatio
n
s
ce
n
ar
io
s
,
an
d
r
ep
o
r
ted
ev
alu
atio
n
r
esu
lts
to
p
r
o
v
id
e
a
clea
r
co
n
te
x
tu
al
u
n
d
er
s
tan
d
in
g
o
f
ex
is
tin
g
ap
p
r
o
ac
h
es.
T
h
is
an
aly
s
is
h
elp
s
p
o
s
itio
n
th
e
co
n
tr
ib
u
tio
n
s
o
f
th
is
s
tu
d
y
in
r
elatio
n
to
p
r
io
r
r
esear
ch
an
d
s
u
p
p
o
r
ts
a
m
o
r
e
co
m
p
r
eh
en
s
iv
e
in
ter
p
r
etatio
n
o
f
th
e
e
x
p
er
im
e
n
tal
f
in
d
in
g
s
.
As
s
u
m
m
ar
ized
in
T
ab
le
3
,
p
r
ev
io
u
s
s
tu
d
ies
h
av
e
s
h
o
w
n
th
at
d
ee
p
lear
n
in
g
–
b
ased
ap
p
r
o
ac
h
es
ar
e
ef
f
ec
tiv
e
f
o
r
co
r
n
ea
l
u
lcer
cla
s
s
if
icatio
n
u
s
in
g
f
lu
o
r
escein
i
m
ag
es.
T
r
an
s
f
o
r
m
er
-
b
ased
m
eth
o
d
s
[
7
]
ac
h
iev
e
d
h
ig
h
class
if
icatio
n
ac
cu
r
ac
y
b
u
t
g
e
n
er
ally
r
e
q
u
ir
e
h
ig
h
er
co
m
p
u
tatio
n
al
c
o
m
p
lex
i
ty
,
wh
ile
h
y
b
r
id
C
NN
-
clas
s
if
ier
ap
p
r
o
ac
h
es
[
8
]
m
ay
h
av
e
lim
ited
a
d
ap
tab
il
ity
d
u
e
to
th
eir
n
o
n
en
d
-
to
-
e
n
d
s
tr
u
ctu
r
e.
Oth
er
C
NN
-
b
ased
s
tu
d
ies
[
9
]
,
[
1
1
]
m
ain
ly
f
o
cu
s
ed
o
n
s
p
ec
if
ic
cla
s
s
if
icatio
n
s
ce
n
ar
io
s
,
lim
itin
g
th
eir
ap
p
licab
ilit
y
i
n
b
r
o
ad
e
r
clin
ical
s
ettin
g
s
.
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
0
3
-
3
6
1
3
3610
I
n
co
n
tr
ast,
th
is
s
tu
d
y
ad
d
r
ess
es
ca
teg
o
r
y
,
ty
p
e,
an
d
g
r
ad
e
class
if
icatio
n
u
s
in
g
a
u
n
if
ie
d
E
f
f
icien
tNetV2
-
S
f
r
am
ewo
r
k
.
I
n
ad
d
itio
n
,
t
h
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
ex
p
licitly
ev
alu
ates
m
o
d
el
p
e
r
f
o
r
m
an
ce
u
n
d
er
leak
a
g
e
an
d
n
o
n
-
leak
a
g
e
co
n
d
itio
n
s
to
im
p
r
o
v
e
ev
alu
atio
n
r
eliab
ilit
y
.
T
h
e
r
esu
lts
in
d
icate
th
at
co
r
n
ea
l
ar
ea
m
ask
in
g
an
d
class
d
is
tr
ib
u
tio
n
–
b
ased
au
g
m
en
tatio
n
co
n
tr
ib
u
te
to
im
p
r
o
v
ed
m
o
d
el
g
en
e
r
aliza
tio
n
,
p
ar
ticu
lar
ly
u
n
d
er
n
o
n
-
leak
a
g
e
s
ettin
g
s
.
Desp
ite
h
av
in
g
lo
wer
m
o
d
el
co
m
p
lex
ity
th
an
tr
an
s
f
o
r
m
er
-
b
ased
ap
p
r
o
ac
h
es,
E
f
f
icien
tNetV2
-
S
ac
h
iev
ed
c
o
m
p
etitiv
e
v
alid
ati
o
n
p
er
f
o
r
m
a
n
ce
,
s
u
g
g
esti
n
g
a
b
alan
ce
d
tr
ad
e
-
o
f
f
b
etwe
en
ac
cu
r
ac
y
,
ef
f
icien
cy
,
an
d
r
o
b
u
s
tn
ess
f
o
r
au
to
m
ated
co
r
n
ea
l
u
lcer
d
ia
g
n
o
s
is
ap
p
licatio
n
s
.
Alth
o
u
g
h
ex
ter
n
al
d
ataset
v
alid
atio
n
w
as
n
o
t
co
n
d
u
cte
d
,
th
e
leak
ag
e
-
awa
r
e
ev
alu
atio
n
s
tr
ateg
y
an
d
co
m
p
ar
is
o
n
wit
h
p
r
ev
io
u
s
s
tu
d
ies p
r
o
v
id
e
i
n
d
ir
ec
t e
v
id
en
ce
o
f
th
e
m
o
d
el’
s
g
e
n
er
aliza
tio
n
ca
p
a
b
ilit
y
.
T
ab
le
3
.
C
ateg
o
r
y
leak
ag
e
tr
ain
in
g
r
esu
lts
S
t
u
d
y
D
a
t
a
s
e
t
M
e
t
h
o
d
C
l
a
s
si
f
i
c
a
t
i
o
n
t
a
sk
K
e
y
l
i
m
i
t
a
t
i
o
n
s
A
l
a
k
u
ş
a
n
d
B
a
y
k
a
r
a
[
7
]
S
U
S
Te
c
h
-
S
Y
S
U
(
7
1
2
i
ma
g
e
s)
V
i
T
C
a
t
e
g
o
r
y
,
t
y
p
e
,
g
r
a
d
e
H
i
g
h
m
o
d
e
l
c
o
m
p
l
e
x
i
t
y
;
i
n
f
e
r
e
n
c
e
e
f
f
i
c
i
e
n
c
y
n
o
t
e
v
a
l
u
a
t
e
d
S
u
si
l
a
[
8
]
S
U
S
Te
c
h
-
S
Y
S
U
(
7
1
2
i
ma
g
e
s)
D
e
n
seN
e
t
1
2
1
+
S
V
M
C
a
t
e
g
o
r
y
,
t
y
p
e
,
g
r
a
d
e
N
o
n
e
n
d
-
to
-
e
n
d
p
i
p
e
l
i
n
e
;
l
i
m
i
t
e
d
a
d
a
p
t
a
b
i
l
i
t
y
S
e
v
l
i
[
9
]
S
U
S
Te
c
h
-
S
Y
S
U
(
7
1
2
i
ma
g
e
s)
C
N
N
-
b
a
s
e
d
mo
d
e
l
s
(
A
l
e
x
N
e
t
,
V
G
G
1
6
)
U
l
c
e
r
t
y
p
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
F
o
c
u
se
d
o
n
a
si
n
g
l
e
c
l
a
ssi
f
i
c
a
t
i
o
n
sce
n
a
r
i
o
H
u
e
t
a
l
.
[
1
0
]
C
l
i
n
i
c
a
l
d
a
t
a
s
e
t
(
2
7
5
7
i
m
a
g
e
s)
M
u
l
t
i
p
l
e
D
L
m
o
d
e
l
s i
n
c
l
.
Ef
f
i
c
i
e
n
t
N
e
t
V
2
M
K
e
r
a
t
i
t
i
s
d
i
a
g
n
o
s
i
s
Lo
w
e
r
a
c
c
u
r
a
c
y
;
n
o
t
f
o
c
u
se
d
o
n
sev
e
r
i
t
y
g
r
a
d
i
n
g
G
r
o
ss
e
t
a
l
.
[
1
1
]
S
U
S
Te
c
h
-
S
Y
S
U
(
7
1
2
i
ma
g
e
s)
C
N
N
(
V
G
G
,
R
e
sN
e
t
)
B
i
n
a
r
y
se
v
e
r
i
t
y
c
l
a
ss
i
f
i
c
a
t
i
o
n
Li
mi
t
e
d
t
o
t
w
o
s
e
v
e
r
i
t
y
l
e
v
e
l
s
Th
i
s
s
t
u
d
y
F
l
u
o
r
e
sc
e
i
n
s
l
i
t
-
l
a
mp
i
ma
g
e
s
Ef
f
i
c
i
e
n
t
N
e
t
V
2
-
S
+
c
o
r
n
e
a
l
mask
i
n
g
+
l
e
a
k
a
g
e
-
a
w
a
r
e
sp
l
i
t
t
i
n
g
C
a
t
e
g
o
r
y
,
t
y
p
e
,
g
r
a
d
e
F
o
c
u
se
d
o
n
i
m
a
g
e
-
b
a
se
d
f
e
a
t
u
r
e
s
o
n
l
y
3
.
4
.
Dis
cus
s
io
n o
n
g
ener
a
liza
t
io
n a
nd
ex
t
er
na
l v
a
lid
it
y
Alth
o
u
g
h
th
is
s
tu
d
y
p
r
im
ar
il
y
u
s
ed
in
ter
n
al
v
alid
atio
n
o
n
th
e
SUSTec
h
-
SYSU
f
lu
o
r
escein
im
ag
e
d
ataset,
s
ev
er
al
s
tr
ateg
ies
wer
e
im
p
lem
en
ted
to
im
p
r
o
v
e
m
o
d
el
g
en
er
aliza
tio
n
.
A
leak
ag
e
-
awa
r
e
d
ata
s
p
litt
in
g
s
tr
ateg
y
was
ap
p
lied
to
p
r
e
v
en
t
o
v
er
ly
o
p
tim
is
tic
p
er
f
o
r
m
an
ce
esti
m
atio
n
a
n
d
en
s
u
r
e
s
tr
ict
s
ep
ar
atio
n
b
etwe
en
tr
ain
in
g
an
d
v
alid
a
tio
n
d
ata.
I
n
a
d
d
itio
n
,
co
r
n
ea
l
ar
ea
m
ask
in
g
an
d
class
d
is
tr
ib
u
tio
n
–
b
ased
au
g
m
en
tatio
n
wer
e
u
s
ed
t
o
r
e
d
u
ce
d
ataset
-
s
p
ec
if
ic
b
ias
an
d
im
p
r
o
v
e
r
o
b
u
s
tn
ess
to
v
ar
iati
o
n
s
in
illu
m
in
atio
n
,
lesi
o
n
ap
p
ea
r
an
ce
,
an
d
im
a
g
in
g
co
n
d
itio
n
s
co
m
m
o
n
ly
f
o
u
n
d
in
clin
ical
p
r
ac
tice.
Du
e
to
th
e
lim
ited
av
ailab
ilit
y
o
f
p
u
b
licly
ac
ce
s
s
ib
le
f
lu
o
r
escein
d
atasets
with
co
n
s
is
ten
t
an
n
o
tatio
n
s
,
ex
ter
n
al
v
alid
atio
n
co
u
ld
n
o
t
b
e
p
er
f
o
r
m
e
d
.
Nev
er
th
ele
s
s
,
th
e
ex
p
er
im
en
tal
f
in
d
in
g
s
ar
e
co
n
s
is
ten
t
with
p
r
ev
io
u
s
s
tu
d
ies
[
7
]
–
[
1
1
]
,
as
s
u
m
m
ar
ized
in
T
ab
le
3
.
Fu
tu
r
e
wo
r
k
will
f
o
c
u
s
o
n
m
u
lti
-
ce
n
ter
an
d
cr
o
s
s
-
d
ataset
ev
alu
atio
n
s
to
f
u
r
th
e
r
ass
ess
m
o
d
el
g
e
n
er
aliza
tio
n
in
d
iv
er
s
e
clin
ical
s
ettin
g
s
.
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
p
r
esen
ts
an
en
d
-
to
-
en
d
d
ee
p
lea
r
n
in
g
f
r
am
ew
o
r
k
f
o
r
au
to
m
ated
co
r
n
ea
l
u
lcer
s
ev
er
ity
class
if
icatio
n
b
ased
o
n
th
e
E
f
f
icien
tNetV2
-
S
ar
ch
itectu
r
e
u
s
in
g
f
lu
o
r
escein
s
lit
-
lam
p
i
m
ag
es.
T
h
e
p
r
im
ar
y
tech
n
ical
co
n
tr
ib
u
tio
n
lies
in
th
e
in
teg
r
atio
n
o
f
co
r
n
ea
l
ar
ea
m
ask
in
g
an
d
leak
ag
e
-
awa
r
e
d
ata
s
p
litt
in
g
in
to
th
e
tr
ain
in
g
p
i
p
elin
e,
wh
ich
ef
f
ec
tiv
ely
en
h
a
n
ce
s
f
ea
tu
r
e
f
o
cu
s
an
d
m
itig
ates
ev
alu
atio
n
b
ias
co
m
m
o
n
l
y
p
r
esen
t
in
m
ed
ical
im
ag
e
class
if
icatio
n
task
s
.
E
x
p
er
im
en
tal
r
esu
lts
d
em
o
n
s
tr
ate
th
at
th
e
p
r
o
p
o
s
ed
s
tr
ateg
y
im
p
r
o
v
es
m
o
d
el
s
tab
ilit
y
an
d
g
en
er
aliza
tio
n
,
with
n
o
n
-
leak
a
g
e
tr
ain
in
g
ac
h
iev
in
g
a
p
ea
k
v
alid
atio
n
ac
cu
r
ac
y
o
f
9
5
.
9
3
%
in
th
e
c
ateg
o
r
y
class
if
icatio
n
s
ce
n
ar
io
,
wh
ile
leak
ag
e
co
n
d
itio
n
s
co
n
s
is
ten
tly
lead
to
p
er
f
o
r
m
an
ce
d
eg
r
a
d
atio
n
d
u
e
to
o
v
er
f
itti
n
g
.
Fro
m
an
a
r
tific
ial
in
tellig
en
ce
p
e
r
s
p
ec
tiv
e,
th
e
p
r
o
p
o
s
ed
f
r
am
ew
o
r
k
o
f
f
er
s
a
r
o
b
u
s
t
an
d
co
m
p
u
tatio
n
ally
ef
f
icien
t
s
o
lu
tio
n
th
at
b
alan
ce
s
m
o
d
el
co
m
p
lex
ity
an
d
p
r
ed
ictiv
e
p
er
f
o
r
m
an
ce
,
m
a
k
in
g
it
s
u
itab
le
f
o
r
r
ea
l
-
wo
r
l
d
d
ep
lo
y
m
en
t
in
in
tellig
en
t
clin
ical
d
e
cisi
o
n
s
u
p
p
o
r
t
s
y
s
tem
s
.
T
h
e
li
g
h
tweig
h
t
n
atu
r
e
o
f
E
f
f
icien
tNetV2
-
S
en
ab
les
f
ast
in
f
er
en
ce
wh
ile
m
ain
tain
in
g
h
ig
h
class
if
icatio
n
ac
cu
r
ac
y
ac
r
o
s
s
m
u
ltip
le
s
ev
er
ity
s
ce
n
ar
io
s
.
Fu
t
u
r
e
s
tu
d
ies
will
ex
ten
d
th
is
wo
r
k
b
y
ev
alu
atin
g
th
e
p
r
o
p
o
s
ed
m
o
d
el
o
n
ex
ter
n
al
a
n
d
m
u
lti
-
ce
n
ter
d
atasets
to
f
u
r
t
h
e
r
v
alid
ate
its
g
en
er
aliza
b
ilit
y
a
n
d
s
u
p
p
o
r
t scala
b
le
clin
ical
ad
o
p
tio
n
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
a
u
t
h
o
r
s
w
o
u
l
d
l
i
k
e
t
o
ex
p
r
e
s
s
t
h
ei
r
s
i
n
ce
r
e
g
r
a
t
it
u
d
e
t
o
t
h
e
M
i
n
is
t
r
y
o
f
E
d
u
c
a
t
i
o
n
,
C
u
l
t
u
r
e,
R
es
e
a
r
c
h
,
a
n
d
T
e
c
h
n
o
l
o
g
y
o
f
t
h
e
R
e
p
u
b
l
i
c
o
f
I
n
d
o
n
e
s
ia
(
K
em
d
i
k
t
i
s
ai
n
t
e
k
)
f
o
r
f
u
n
d
i
n
g
t
h
i
s
r
e
s
e
a
r
c
h
u
n
d
e
r
t
h
e
f
u
n
d
a
m
e
n
t
a
l
r
e
s
e
a
r
c
h
–
r
e
g
u
l
a
r
s
c
h
e
m
e
.
T
h
e
a
u
t
h
o
r
s
a
l
s
o
ac
k
n
o
w
l
e
d
g
e
t
h
e
R
es
e
a
r
c
h
I
n
s
t
it
u
t
i
o
n
s
o
f
U
n
i
v
e
r
s
i
t
a
s
G
u
n
a
d
a
r
m
a
f
o
r
t
h
e
i
r
a
d
m
i
n
is
t
r
a
t
i
v
e
s
u
p
p
o
r
t
,
r
es
e
a
r
c
h
f
a
c
il
i
ti
es
,
a
n
d
c
o
n
t
i
n
u
o
u
s
as
s
i
s
ta
n
c
e
t
h
r
o
u
g
h
o
u
t
t
h
e
s
t
u
d
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
fficien
t d
ee
p
lea
r
n
in
g
f
o
r
a
u
t
o
ma
ted
co
r
n
ea
l
u
lcer
s
ev
erit
y
cla
s
s
ifica
tio
n
fr
o
m
flu
o
r
escein
ima
g
e
s
(
R
o
d
ia
h
)
3611
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
was
f
u
n
d
ed
b
y
th
e
Min
is
tr
y
o
f
E
d
u
ca
tio
n
,
C
u
ltu
r
e,
R
esear
ch
,
an
d
T
ec
h
n
o
l
o
g
y
o
f
th
e
R
ep
u
b
lic
o
f
I
n
d
o
n
esia
(
Kem
d
ik
tis
ain
tek
)
u
n
d
er
th
e
f
u
n
d
am
en
tal
r
esear
ch
–
r
eg
u
lar
s
ch
e
m
e
,
as
s
tip
u
lated
in
th
e
d
ir
ec
to
r
o
f
r
esear
ch
d
ec
r
ee
n
o
.
0
4
1
9
/C
3
/DT
.
0
5
.
0
0
/2
0
2
5
d
ate
d
Ma
y
2
2
,
2
0
2
5
,
m
ain
co
n
tr
ac
t
n
o
.
1
2
4
/C
3
/DT
.
0
5
.
0
0
/PL/2
0
2
5
d
a
ted
Ma
y
2
8
,
2
0
2
5
,
s
u
b
co
n
tr
ac
t
n
o
.
1
0
9
7
4
/LL
3
/AL
.
0
4
/2
0
2
5
d
ated
J
u
n
e
4
,
2
0
2
5
,
an
d
s
u
b
co
n
tr
ac
t n
o
.
1
6
.
2
8
/LP/
UG/VI
/2
0
2
5
d
ated
J
u
n
e
5
,
2
0
2
5
.
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
R
o
d
iah
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
I
n
d
ah
Sin
th
y
a
Per
m
ata
Sar
i
✓
✓
✓
✓
✓
✓
✓
✓
Ma
tr
is
s
y
a
Her
m
ita
✓
✓
✓
✓
✓
Sar
if
u
d
d
in
Ma
d
en
d
a
✓
✓
✓
✓
✓
✓
D
i
a
n
a
T
r
i
S
u
s
e
t
i
a
n
i
n
g
t
i
a
s
✓
✓
✓
✓
✓
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.
I
NF
O
RM
E
D
CO
NS
E
N
T
T
h
is
s
tu
d
y
d
id
n
o
t in
v
o
lv
e
h
u
m
an
p
ar
ticip
an
ts
; th
er
e
f
o
r
e,
in
f
o
r
m
ed
co
n
s
en
t w
as n
o
t
r
eq
u
ir
ed
.
E
T
H
I
CAL AP
P
RO
V
AL
T
h
is
s
t
u
d
y
u
tili
ze
d
a
s
e
c
o
n
d
a
r
y
d
at
aset
o
f
f
l
u
o
r
esc
ei
n
s
li
t
-
l
am
p
i
m
a
g
es
a
n
d
d
i
d
n
o
t
i
n
v
o
lv
e
d
ir
ec
t
in
t
er
ac
t
io
n
wi
th
h
u
m
a
n
p
a
r
ti
ci
p
a
n
ts
o
r
v
e
r
te
b
r
at
e
a
n
im
al
s
u
b
j
ec
ts
.
T
h
e
r
e
f
o
r
e,
et
h
i
ca
l
ap
p
r
o
v
al
was
n
o
t
r
e
q
u
ir
e
d
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
is
th
e
p
u
b
licly
a
v
ailab
le
SUSTec
h
-
SYSU
co
r
n
ea
l
u
lcer
d
at
aset,
wh
ich
co
n
s
is
ts
o
f
f
lu
o
r
escein
-
s
tain
ed
s
lit
-
lam
p
im
a
g
es
co
llected
at
th
e
Z
h
o
n
g
s
h
a
n
Op
h
th
alm
o
lo
g
y
C
en
ter
,
Su
n
Yat
-
s
en
Un
iv
er
s
ity
.
T
h
e
d
ataset
is
ac
ce
s
s
ib
le
th
r
o
u
g
h
th
e
o
r
i
g
in
al
p
u
b
licatio
n
b
y
Den
g
et
a
l.
an
d
is
co
m
m
o
n
ly
u
s
ed
f
o
r
b
en
c
h
m
ar
k
in
g
au
to
m
ated
co
r
n
ea
l
u
lcer
class
if
icatio
n
an
d
g
r
ad
i
n
g
task
s
.
All
d
ata
u
tili
ze
d
in
th
is
r
esear
ch
ar
e
p
u
b
licly
av
ailab
l
e
an
d
ca
n
b
e
ac
ce
s
s
ed
f
o
r
r
esear
ch
p
u
r
p
o
s
es,
allo
win
g
r
e
p
r
o
d
u
cib
ilit
y
o
f
th
e
ex
p
er
im
en
tal
s
etu
p
.
Deta
ile
d
in
f
o
r
m
atio
n
r
e
g
ar
d
in
g
d
a
ta
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
,
co
r
n
ea
l
ar
ea
m
ask
in
g
p
r
o
ce
d
u
r
es,
d
ata
a
u
g
m
en
tatio
n
s
tr
ateg
ies,
an
d
d
ata
s
p
litt
in
g
co
n
f
ig
u
r
atio
n
s
im
p
lem
en
t
ed
in
th
is
s
tu
d
y
is
av
ailab
le
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
au
th
o
r
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
G
.
R
.
V
.
-
D
u
a
r
t
e
e
t
a
l
.
,
“
N
e
u
r
o
t
r
o
p
h
i
c
k
e
r
a
t
o
p
a
t
h
y
:
g
e
n
e
r
a
l
f
e
a
t
u
r
e
s
a
n
d
n
e
w
t
h
e
r
a
p
i
e
s
,
”
S
u
rv
e
y
o
f
O
p
h
t
h
a
l
m
o
l
o
g
y
,
v
o
l
.
6
9
,
n
o
.
5
,
p
p
.
7
8
9
–
8
0
4
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
s
u
r
v
o
p
h
t
h
a
l
.
2
0
2
4
.
0
4
.
0
0
4
.
[
2
]
F
.
S
t
a
p
l
e
t
o
n
,
“
T
h
e
e
p
i
d
e
mi
o
l
o
g
y
o
f
i
n
f
e
c
t
i
o
u
s
k
e
r
a
t
i
t
i
s
,
”
T
h
e
O
c
u
l
a
r
S
u
rf
a
c
e
,
v
o
l
.
2
8
,
n
o
.
9
,
p
p
.
3
5
1
–
3
6
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
t
o
s.
2
0
2
1
.
0
8
.
0
0
7
.
[
3
]
K
.
W
a
sl
a
n
a
n
d
R
.
O
k
t
a
l
i
a
n
i
,
“
A
c
a
s
e
r
e
p
o
r
t
:
t
h
e
s
t
r
a
t
e
g
y
o
f
c
o
r
n
e
a
l
u
l
c
e
r
s,”
J
a
m
b
i
Me
d
i
c
a
l
J
o
u
rn
a
l
:
J
u
rn
a
l
K
e
d
o
k
t
e
ra
n
d
a
n
K
e
se
h
a
t
a
n
,
v
o
l
.
1
1
,
n
o
.
1
,
p
p
.
6
8
–
7
3
,
2
0
2
3
.
[
4
]
D
.
M
a
n
a
w
o
n
g
sa
k
u
l
a
n
d
K
.
P
a
t
a
n
u
k
h
o
m,
“
A
se
g
me
n
t
a
t
i
o
n
-
b
a
se
d
a
u
t
o
ma
t
e
d
c
o
r
n
e
a
l
u
l
c
e
r
g
r
a
d
i
n
g
s
y
s
t
e
m
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
H
o
u
g
h
c
i
r
c
l
e
t
r
a
n
sf
o
r
m,
”
A
l
g
o
r
i
t
h
m
s
,
v
o
l
.
1
7
,
n
o
.
9
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
a
1
7
0
9
0
4
0
5
.
[
5
]
S
.
G
h
a
f
a
r
i
a
n
e
t
a
l
.
,
“
C
l
i
n
i
c
a
l
e
v
a
l
u
a
t
i
o
n
o
f
c
o
r
n
e
a
l
u
l
c
e
r
w
i
t
h
a
p
o
r
t
a
b
l
e
a
n
d
s
mart
p
h
o
n
e
-
a
t
t
a
c
h
a
b
l
e
sl
i
t
l
a
m
p
d
e
v
i
c
e
:
smar
t
e
y
e
c
a
mer
a
,
”
S
c
i
e
n
t
i
f
i
c
R
e
p
o
rt
s
,
v
o
l
.
1
5
,
n
o
.
1
,
Ja
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
0
2
5
-
8
7
8
2
0
-
z.
[
6
]
D
.
D
.
P
a
i
,
I
.
M
.
R
.
R
,
a
n
d
H
.
S
u
n
i
l
,
“
A
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
i
n
c
l
i
n
i
c
a
l
d
e
c
i
si
o
n
s
u
p
p
o
r
t
s
y
st
e
ms:
a
r
c
h
i
t
e
c
t
u
r
e
s,
a
p
p
l
i
c
a
t
i
o
n
s,
a
n
d
e
t
h
i
c
a
l
c
h
a
l
l
e
n
g
e
s
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
E
n
g
i
n
e
e
r
i
n
g
Re
s
e
a
r
c
h
a
n
d
R
e
v
i
e
w
s
,
v
o
l
.
1
4
,
n
o
.
1
,
p
p
.
2
6
–
3
1
,
2
0
2
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
0
3
-
3
6
1
3
3612
[
7
]
T
.
B
.
A
l
a
k
u
ş
a
n
d
M
.
B
a
y
k
a
r
a
,
“
C
l
a
s
s
i
f
i
c
a
t
i
o
n
a
n
d
d
e
t
e
r
m
i
n
a
t
i
o
n
o
f
s
e
v
e
r
i
t
y
o
f
c
o
r
n
e
a
l
u
l
c
e
r
u
s
i
n
g
v
i
s
i
o
n
t
r
a
n
s
f
o
r
m
e
r
b
a
s
e
d
o
n
t
h
e
a
n
a
l
y
s
i
s
o
f
p
u
b
l
i
c
i
m
a
g
e
d
a
t
a
s
e
t
o
f
f
l
u
o
r
e
s
c
e
i
n
-
s
t
a
i
n
e
d
c
o
r
n
e
a
s
,
”
D
i
a
g
n
o
s
t
i
c
s
,
v
o
l
.
1
4
,
n
o
.
8
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
s
t
i
c
s
1
4
0
8
0
7
8
6
.
[
8
]
S
.
J
.
G
.
S
u
s
i
l
a
,
“
A
d
v
a
n
c
e
d
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
c
o
r
n
e
a
l
u
l
c
e
r
u
si
n
g
D
e
n
seN
e
t
1
2
1
a
n
d
S
V
M
c
l
a
ss
i
f
i
e
r
,
”
J
o
u
rn
a
l
o
f
El
e
c
t
r
i
c
a
l
S
y
st
e
m
s
,
v
o
l
.
2
0
,
n
o
.
7
,
p
p
.
1
8
6
7
–
1
8
8
3
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
5
2
7
8
3
/
j
e
s
.
3
8
9
2
.
[
9
]
O
.
S
e
v
l
i
,
“
D
e
e
p
l
e
a
r
n
i
n
g
-
b
a
s
e
d
c
l
a
s
si
f
i
c
a
t
i
o
n
o
f
c
o
r
n
e
a
l
u
l
c
e
r
s
o
n
o
c
u
l
a
r
st
a
i
n
i
n
g
i
m
a
g
e
s,
”
Ac
t
a
I
n
f
o
l
o
g
i
c
a
,
v
o
l
.
7
,
n
o
.
2
,
p
p
.
2
8
1
–
2
9
2
,
A
u
g
.
2
0
2
3
,
d
o
i
:
1
0
.
2
6
6
5
0
/
a
c
i
n
.
1
1
7
3
4
6
5
.
[
1
0
]
S
.
H
u
e
t
a
l
.
,
“
A
u
t
o
ma
t
i
c
d
i
a
g
n
o
si
s
o
f
i
n
f
e
c
t
i
o
u
s
k
e
r
a
t
i
t
i
s
b
a
se
d
o
n
sl
i
t
l
a
m
p
i
ma
g
e
s
a
n
a
l
y
si
s,”
J
o
u
rn
a
l
o
f
Pe
rs
o
n
a
l
i
z
e
d
Me
d
i
c
i
n
e
,
v
o
l
.
1
3
,
n
o
.
3
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
j
p
m
1
3
0
3
0
5
1
9
.
[
1
1
]
J.
G
r
o
ss
,
J.
B
r
e
i
t
e
n
b
a
c
h
,
H
.
B
a
u
m
g
a
r
t
l
,
a
n
d
R
.
B
u
e
t
t
n
e
r
,
“
H
i
g
h
-
p
e
r
f
o
r
man
c
e
d
e
t
e
c
t
i
o
n
o
f
c
o
r
n
e
a
l
u
l
c
e
r
a
t
i
o
n
u
s
i
n
g
i
m
a
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
w
i
t
h
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
i
n
5
4
t
h
H
a
w
a
i
i
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
S
y
st
e
m
S
c
i
e
n
c
e
s
,
2
0
2
1
,
p
p
.
3
4
1
6
–
3
4
2
5
,
d
o
i
:
1
0
.
2
4
2
5
1
/
H
I
C
S
S
.
2
0
2
1
.
4
1
5
.
[
1
2
]
D
.
P
a
s
s
o
s
a
n
d
P
.
M
i
s
h
r
a
,
“
A
t
u
t
o
r
i
a
l
o
n
a
u
t
o
m
a
t
i
c
h
y
p
e
r
p
a
r
a
m
e
t
e
r
t
u
n
i
n
g
o
f
d
e
e
p
s
p
e
c
t
r
a
l
m
o
d
e
l
l
i
n
g
f
o
r
r
e
g
r
e
s
s
i
o
n
a
n
d
c
l
a
s
s
i
f
i
c
a
t
i
o
n
t
a
s
k
s
,
”
C
h
e
m
o
m
e
t
r
i
c
s
a
n
d
I
n
t
e
l
l
i
g
e
n
t
L
a
b
o
r
a
t
o
r
y
S
y
s
t
e
m
s
,
v
o
l
.
2
2
3
,
A
p
r
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
h
e
m
o
l
a
b
.
2
0
2
2
.
1
0
4
5
2
0
.
[
1
3
]
L.
D
e
n
g
,
J.
Ly
u
,
H
.
H
u
a
n
g
,
Y
.
D
e
n
g
,
J.
Y
u
a
n
,
a
n
d
X
.
Ta
n
g
,
“
T
h
e
S
U
S
Te
c
h
-
S
Y
S
U
d
a
t
a
set
f
o
r
a
u
t
o
ma
t
i
c
a
l
l
y
s
e
g
m
e
n
t
i
n
g
a
n
d
c
l
a
ss
i
f
y
i
n
g
c
o
r
n
e
a
l
u
l
c
e
r
s,”
S
c
i
e
n
t
i
f
i
c
D
a
t
a
,
v
o
l
.
7
,
n
o
.
1
,
J
a
n
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
7
-
0
2
0
-
0
3
6
0
-
7.
[
1
4
]
H
.
A
l
q
u
r
a
n
,
Y
.
A
l
-
I
ssa,
M
.
A
l
sal
a
t
i
e
,
W
.
A
.
M
u
s
t
a
f
a
,
I
.
A
.
Q
a
smi
e
h
,
a
n
d
A
.
Z
y
o
u
t
,
“
I
n
t
e
l
l
i
g
e
n
t
d
i
a
g
n
o
si
s
a
n
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
k
e
r
a
t
i
t
i
s
,
”
D
i
a
g
n
o
s
t
i
c
s
,
v
o
l
.
1
2
,
n
o
.
6
,
M
a
y
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
s
t
i
c
s
1
2
0
6
1
3
4
4
.
[
1
5
]
K
.
C
z
a
r
n
a
c
k
a
e
t
a
l
.
,
“
D
i
a
g
n
o
s
i
s a
n
d
t
r
e
a
t
m
e
n
t
o
f
c
o
r
n
e
a
l
u
l
c
e
r
s
i
n
t
h
e
c
o
u
r
s
e
o
f
sy
s
t
e
m
i
c
a
u
t
o
i
mm
u
n
e
d
i
se
a
ses
,
”
O
p
h
t
h
a
T
h
e
r
a
p
y
.
C
o
n
ser
v
a
t
i
v
e
T
re
a
t
m
e
n
t
,
v
o
l
.
1
0
,
n
o
.
1
,
p
p
.
4
3
–
4
7
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
2
4
2
9
2
/
0
1
.
O
T.
2
1
0
3
2
3
.
4
.
[
1
6
]
A
.
A
.
M
a
h
mo
u
d
,
M
.
S
.
A
.
-
R
a
h
m
a
n
,
T.
A
.
M
o
h
a
me
d
,
a
n
d
D
.
M
.
El
-
S
e
b
a
i
t
y
,
“
I
s
c
o
r
n
e
a
l
c
o
l
l
a
g
e
n
c
r
o
ss
-
l
i
n
k
i
n
g
b
e
n
e
f
i
c
i
a
l
a
s
a
n
a
d
j
u
n
c
t
t
o
t
h
e
c
o
n
v
e
n
t
i
o
n
a
l
t
r
e
a
t
me
n
t
o
f
b
a
c
t
e
r
i
a
l
k
e
r
a
t
i
t
i
s?
”
T
h
e
O
p
e
n
O
p
h
t
h
a
l
m
o
l
o
g
y
J
o
u
r
n
a
l
,
v
o
l
.
1
6
,
n
o
.
1
,
p
p
.
1
–
8
,
A
p
r
.
2
0
2
2
,
d
o
i
:
1
0
.
2
1
7
4
/
1
8
7
4
3
6
4
1
-
v16
-
e
2
2
0
3
1
5
0
.
[
1
7
]
C
.
R
.
H
a
r
r
i
s
e
t
a
l
.
,
“
A
r
r
a
y
p
r
o
g
r
a
m
m
i
n
g
w
i
t
h
N
u
mP
y
,
”
N
a
t
u
r
e
,
v
o
l
.
5
8
5
,
p
p
.
3
5
7
–
3
6
2
,
S
e
p
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
8
6
-
020
-
2649
-
2.
[
1
8
]
I
.
A
.
Q
a
sm
i
e
h
,
H
.
A
l
q
u
r
a
n
,
A
.
Z
y
o
u
t
,
Y
.
A
l
-
I
ssa,
W
.
A
.
M
u
s
t
a
f
a
,
a
n
d
M
.
A
l
sal
a
t
i
e
,
“
A
u
t
o
m
a
t
e
d
d
e
t
e
c
t
i
o
n
o
f
c
o
r
n
e
a
l
u
l
c
e
r
u
s
i
n
g
c
o
m
b
i
n
a
t
i
o
n
i
ma
g
e
p
r
o
c
e
s
si
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
,
”
D
i
a
g
n
o
s
t
i
c
s
,
v
o
l
.
1
2
,
n
o
.
1
2
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
st
i
c
s
1
2
1
2
3
2
0
4
.
[
1
9
]
T.
I
n
n
e
c
i
a
n
d
H
.
B
a
d
e
m
,
“
D
e
t
e
c
t
i
o
n
o
f
c
o
r
n
e
a
l
u
l
c
e
r
u
si
n
g
a
g
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m
-
b
a
se
d
i
m
a
g
e
sel
e
c
t
i
o
n
a
n
d
r
e
si
d
u
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
Bi
o
e
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
0
,
n
o
.
6
,
M
a
y
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
b
i
o
e
n
g
i
n
e
e
r
i
n
g
1
0
0
6
0
6
3
9
.
[
2
0
]
L.
L
v
,
M
.
P
e
n
g
,
X
.
W
a
n
g
,
a
n
d
Y
.
W
u
,
“
M
u
l
t
i
-
s
c
a
l
e
i
n
f
o
r
ma
t
i
o
n
f
u
s
i
o
n
n
e
t
w
o
r
k
w
i
t
h
l
a
b
e
l
sm
o
o
t
h
i
n
g
st
r
a
t
e
g
y
f
o
r
c
o
r
n
e
a
l
u
l
c
e
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
i
n
s
l
i
t
l
a
m
p
i
ma
g
e
s,
”
Fr
o
n
t
i
e
rs
i
n
N
e
u
r
o
sc
i
e
n
c
e
,
v
o
l
.
1
6
,
N
o
v
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
8
9
/
f
n
i
n
s.
2
0
2
2
.
9
9
3
2
3
4
.
[
2
1
]
M
.
Ta
n
a
n
d
Q
.
V
.
L
e
,
“
Ef
f
i
c
i
e
n
t
N
e
t
V
2
:
sma
l
l
e
r
m
o
d
e
l
s
a
n
d
f
a
st
e
r
t
r
a
i
n
i
n
g
,
”
i
n
3
8
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
2
0
2
1
,
v
o
l
.
1
3
9
,
p
p
.
1
0
0
9
6
–
1
0
1
0
6
.
[
2
2
]
L.
A
l
z
u
b
a
i
d
i
e
t
a
l
.
,
“
A
s
u
r
v
e
y
o
n
d
e
e
p
l
e
a
r
n
i
n
g
t
o
o
l
s
d
e
a
l
i
n
g
w
i
t
h
d
a
t
a
sca
r
c
i
t
y
:
d
e
f
i
n
i
t
i
o
n
s,
c
h
a
l
l
e
n
g
e
s,
so
l
u
t
i
o
n
s,
t
i
p
s
,
a
n
d
a
p
p
l
i
c
a
t
i
o
n
s,
”
J
o
u
rn
a
l
o
f
Bi
g
D
a
t
a
,
v
o
l
.
1
0
,
n
o
.
1
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
0
5
3
7
-
0
2
3
-
0
0
7
2
7
-
2.
[
2
3
]
S
.
J.
P
a
n
a
n
d
Q
.
Y
a
n
g
,
“
A
su
r
v
e
y
o
n
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
I
EE
E
T
r
a
n
sa
c
t
i
o
n
s
o
n
K
n
o
w
l
e
d
g
e
a
n
d
D
a
t
a
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
2
,
n
o
.
1
0
,
p
p
.
1
3
4
5
–
1
3
5
9
,
O
c
t
.
2
0
1
0
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E.
2
0
0
9
.
1
9
1
.
[
2
4
]
Y
.
D
e
n
g
e
t
a
l
.
,
“
A
d
v
a
n
c
e
d
a
n
d
i
n
t
e
r
p
r
e
t
a
b
l
e
c
o
r
n
e
a
l
s
t
a
i
n
i
n
g
a
ss
e
ssm
e
n
t
t
h
r
o
u
g
h
f
i
n
e
g
r
a
i
n
e
d
k
n
o
w
l
e
d
g
e
d
i
s
t
i
l
l
a
t
i
o
n
,
”
n
p
j
D
i
g
i
t
a
l
Me
d
i
c
i
n
e
,
v
o
l
.
8
,
n
o
.
1
,
M
a
y
2
0
2
5
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
7
4
6
-
0
2
5
-
0
1
7
0
6
-
y.
[
2
5
]
F.
-
F
.
Li
e
t
a
l
.
,
“
I
n
t
e
g
r
a
t
i
n
g
p
r
i
o
r
k
n
o
w
l
e
d
g
e
w
i
t
h
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
o
p
t
i
m
i
z
e
d
q
u
a
l
i
t
y
c
o
n
t
r
o
l
i
n
c
o
r
n
e
a
l
i
ma
g
e
s:
a
m
u
l
t
i
c
e
n
t
e
r
st
u
d
y
,
”
C
o
m
p
u
t
e
r
Me
t
h
o
d
s
a
n
d
Pro
g
r
a
m
s i
n
Bi
o
m
e
d
i
c
i
n
e
,
v
o
l
.
2
6
7
,
J
u
l
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
m
p
b
.
2
0
2
5
.
1
0
8
8
1
4
.
[
2
6
]
S
.
M
e
h
r
t
a
b
a
r
e
t
a
l
.
,
“
E
t
h
i
c
a
l
c
o
n
si
d
e
r
a
t
i
o
n
s
i
n
p
a
t
i
e
n
t
p
r
i
v
a
c
y
a
n
d
d
a
t
a
h
a
n
d
l
i
n
g
f
o
r
A
I
i
n
c
a
r
d
i
o
v
a
sc
u
l
a
r
i
m
a
g
i
n
g
a
n
d
r
a
d
i
o
l
o
g
y
,
”
J
o
u
rn
a
l
o
f
D
i
g
i
t
a
l
I
m
a
g
i
n
g
,
v
o
l
.
3
9
,
p
p
.
2
0
0
4
–
2
0
2
5
,
2
0
2
6
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
0
2
7
8
-
025
-
0
1
6
5
6
-
7.
[
2
7
]
T
.
I
s
l
a
m
,
M
.
S
.
H
a
f
i
z
,
J
.
R
.
J
i
m
,
M
.
M
.
K
a
b
i
r
,
a
n
d
M
.
F
.
M
r
i
d
h
a
,
“
A
s
y
s
t
e
m
a
t
i
c
r
e
v
i
e
w
o
f
d
e
e
p
l
e
a
r
n
i
n
g
d
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
i
n
m
e
d
i
c
a
l
i
m
a
g
i
n
g
:
r
e
c
e
n
t
a
d
v
a
n
c
e
s
a
n
d
f
u
t
u
r
e
r
e
s
e
a
r
c
h
d
i
r
e
c
t
i
o
n
s
,
”
H
e
a
l
t
h
c
a
r
e
A
n
a
l
y
t
i
c
s
,
v
o
l
.
5
,
J
u
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
h
e
a
l
t
h
.
2
0
2
4
.
1
0
0
3
4
0
.
[
2
8
]
J.
M
a
,
Y
.
H
e
,
F
.
L
i
,
L
.
H
a
n
,
C
.
Y
o
u
,
a
n
d
B
.
W
a
n
g
,
“
S
e
g
me
n
t
a
n
y
t
h
i
n
g
i
n
me
d
i
c
a
l
i
m
a
g
e
s,
”
N
a
t
u
re
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
v
o
l
.
1
5
,
n
o
.
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
4
6
7
-
0
2
4
-
4
4
8
2
4
-
z.
[
2
9
]
E.
J.
To
p
o
l
,
“
H
i
g
h
-
p
e
r
f
o
r
ma
n
c
e
me
d
i
c
i
n
e
:
t
h
e
c
o
n
v
e
r
g
e
n
c
e
o
f
h
u
ma
n
a
n
d
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
,
”
N
a
t
u
re
Me
d
i
c
i
n
e
,
v
o
l
.
2
5
,
n
o
.
1
,
p
p
.
4
4
–
5
6
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
5
9
1
-
0
1
8
-
0
3
0
0
-
7.
[
3
0
]
D
.
S
.
W
.
Ti
n
g
e
t
a
l
.
,
“
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
e
e
p
l
e
a
r
n
i
n
g
i
n
o
p
h
t
h
a
l
mo
l
o
g
y
,
”
Bri
t
i
s
h
J
o
u
r
n
a
l
o
f
O
p
h
t
h
a
l
m
o
l
o
g
y
,
v
o
l
.
1
0
3
,
n
o
.
2
,
p
p
.
1
6
7
–
1
7
5
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
3
6
/
b
j
o
p
h
t
h
a
l
m
o
l
-
2
0
1
8
-
3
1
3
1
7
3
.
[
3
1
]
O
.
A
.
R
a
mw
a
l
a
e
t
a
l
.
,
“
Es
t
a
b
l
i
sh
i
n
g
a
v
a
l
i
d
a
t
i
o
n
i
n
f
r
a
s
t
r
u
c
t
u
r
e
f
o
r
i
ma
g
i
n
g
-
b
a
s
e
d
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
a
l
g
o
r
i
t
h
m
s
b
e
f
o
r
e
c
l
i
n
i
c
a
l
i
mp
l
e
m
e
n
t
a
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
t
h
e
Am
e
ri
c
a
n
C
o
l
l
e
g
e
o
f
R
a
d
i
o
l
o
g
y
,
v
o
l
.
2
1
,
n
o
.
1
0
,
p
p
.
1
5
6
9
–
1
5
7
4
,
O
c
t
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
a
c
r
.
2
0
2
4
.
0
4
.
0
2
7
.
[
3
2
]
D
.
C
.
L
e
p
c
h
a
e
t
a
l
.
,
“
D
e
e
p
l
e
a
r
n
i
n
g
i
n
me
d
i
c
a
l
i
m
a
g
e
a
n
a
l
y
si
s
:
a
c
o
m
p
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
f
a
l
g
o
r
i
t
h
ms,
t
r
e
n
d
s
,
a
p
p
l
i
c
a
t
i
o
n
s,
a
n
d
c
h
a
l
l
e
n
g
e
s,”
C
o
m
p
u
t
e
r
Mo
d
e
l
i
n
g
i
n
E
n
g
i
n
e
e
r
i
n
g
&
S
c
i
e
n
c
e
s
,
v
o
l
.
1
4
5
,
n
o
.
2
,
p
p
.
1
4
8
7
–
1
5
7
3
,
O
c
t
.
2
0
2
5
,
d
o
i
:
1
0
.
3
2
6
0
4
/
c
mes.
2
0
2
5
.
0
7
0
9
6
4
.
[
3
3
]
S
.
U
.
R
.
K
h
a
n
,
S
.
A
si
f
,
M
.
Zh
a
o
,
W
.
Z
o
u
,
Y
.
L
i
,
a
n
d
X
.
L
i
,
“
O
p
t
i
mi
z
e
d
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
f
o
r
c
o
m
p
r
e
h
e
n
s
i
v
e
m
e
d
i
c
a
l
i
m
a
g
e
a
n
a
l
y
si
s
a
c
r
o
ss m
u
l
t
i
p
l
e
mo
d
a
l
i
t
i
e
s,”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
6
1
9
,
n
o
.
9
,
F
e
b
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m.
2
0
2
4
.
1
2
9
1
8
2
.
[
3
4
]
G
.
U
z
o
e
c
h
i
n
a
a
n
d
T.
O
sa
j
i
u
b
a
,
“
D
e
e
p
l
e
a
r
n
i
n
g
i
n
c
l
i
n
i
c
a
l
d
i
a
g
n
o
s
t
i
c
s:
a
s
c
o
p
i
n
g
r
e
v
i
e
w
o
f
i
n
n
o
v
a
t
i
o
n
s
s
h
a
p
i
n
g
f
u
t
u
r
e
h
e
a
l
t
h
c
a
r
e
d
e
l
i
v
e
r
y
,
”
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
i
n
M
e
d
i
c
i
n
e
,
v
o
l
.
1
4
6
.
N
o
v
.
1
8
,
2
0
2
5
,
d
o
i
:
1
0
.
2
1
2
0
3
/
r
s
.
3
.
r
s
-
7
8
7
6
5
9
8
/
v
1
.
[
3
5
]
B
.
K
o
c
a
k
e
t
a
l
.
,
“
E
v
a
l
u
a
t
i
o
n
me
t
r
i
c
s
i
n
me
d
i
c
a
l
i
ma
g
i
n
g
A
I
:
f
u
n
d
a
me
n
t
a
l
s
,
p
i
t
f
a
l
l
s,
mi
s
a
p
p
l
i
c
a
t
i
o
n
s,
a
n
d
r
e
c
o
mm
e
n
d
a
t
i
o
n
s,
”
Eu
r
o
p
e
a
n
J
o
u
r
n
a
l
o
f
R
a
d
i
o
l
o
g
y
:
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
3
,
S
e
p
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
j
r
a
i
.
2
0
2
5
.
1
0
0
0
3
0
.
[
3
6
]
H
.
S
a
mm
a
,
A
.
S
.
B
.
S
a
ma,
a
n
d
Q
.
S
.
H
a
m
a
d
,
“
O
p
t
i
mi
z
e
d
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
f
o
r
me
d
i
c
a
l
i
ma
g
e
d
i
a
g
n
o
si
s
,
”
J
o
u
r
n
a
l
o
f
En
g
i
n
e
e
ri
n
g
Re
s
e
a
rc
h
,
v
o
l
.
1
3
,
n
o
.
4
,
p
p
.
3
1
4
7
–
3
1
5
5
,
D
e
c
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
e
r
.
2
0
2
4
.
1
1
.
0
0
3
.
[
3
7
]
H
.
L
u
,
Y
.
L
i
n
,
Z
.
Li
,
M
.
L.
Y
i
u
,
Y
.
G
a
o
,
a
n
d
S
.
U
d
d
i
n
,
“
T
o
w
a
r
d
f
a
i
r
m
e
d
i
c
a
l
a
d
v
i
c
e
:
a
d
d
r
e
ssi
n
g
a
n
d
m
i
t
i
g
a
t
i
n
g
b
i
a
s
i
n
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
-
b
a
se
d
h
e
a
l
t
h
c
a
r
e
a
p
p
l
i
c
a
t
i
o
n
s
,
”
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
i
n
Me
d
i
c
i
n
e
,
v
o
l
.
1
6
8
,
n
o
.
4
,
O
c
t
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
r
t
me
d
.
2
0
2
5
.
1
0
3
2
1
6
.
[
3
8
]
A
.
M
.
C
a
r
r
i
n
g
t
o
n
e
t
a
l
.
,
“
D
e
e
p
R
O
C
a
n
a
l
y
si
s
a
n
d
A
U
C
a
s
b
a
l
a
n
c
e
d
a
v
e
r
a
g
e
a
c
c
u
r
a
c
y
f
o
r
i
m
p
r
o
v
e
d
c
l
a
ss
i
f
i
e
r
se
l
e
c
t
i
o
n
,
a
u
d
i
t
a
n
d
e
x
p
l
a
n
a
t
i
o
n
,
”
I
EE
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
Pa
t
t
e
rn
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
5
,
n
o
.
1
,
p
p
.
3
2
9
–
3
4
1
,
Ja
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
2
2
.
3
1
4
5
3
9
2
.
[
3
9
]
Z.
L
a
n
,
Y
.
H
u
,
S
.
Y
a
n
g
,
M
.
W
u
,
Y
.
H
u
a
n
g
,
a
n
d
H
.
Z
h
a
n
g
,
“
D
e
e
p
l
e
a
r
n
i
n
g
i
n
o
p
h
t
h
a
l
mi
c
i
ma
g
e
a
n
a
l
y
si
s
:
a
t
a
s
k
-
d
r
i
v
e
n
r
e
v
i
e
w
o
f
seg
m
e
n
t
a
t
i
o
n
,
d
i
a
g
n
o
si
s
,
a
n
d
p
r
o
g
r
e
ss
p
r
e
d
i
c
t
i
o
n
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
6
6
7
,
F
e
b
.
2
0
2
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m
.
2
0
2
5
.
1
3
2
2
8
2
.
[
4
0
]
S
.
H
i
sari
a
,
P
.
S
h
a
r
m
a
,
R
.
G
u
p
t
a
,
a
n
d
K
.
S
u
ma
l
a
t
h
a
,
“
A
n
a
n
a
l
y
si
s
o
f
mu
l
t
i
-
c
r
i
t
e
r
i
a
p
e
r
f
o
r
m
a
n
c
e
i
n
d
e
e
p
l
e
a
r
n
i
n
g
-
b
a
s
e
d
med
i
c
a
l
i
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
:
a
c
o
mp
r
e
h
e
n
si
v
e
r
e
v
i
e
w
,
”
n
p
j
D
i
g
i
t
a
l
Me
d
i
c
i
n
e
,
v
o
l
.
7
,
n
o
.
1
.
p
p
.
1
–
1
2
,
M
a
r
.
2
1
,
2
0
2
4
,
d
o
i
:
1
0
.
2
1
2
0
3
/
r
s.
3
.
r
s
-
4
1
2
5
3
0
1
/
v
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.