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
5
,
No
.
2
,
A
p
r
il 2
0
2
6
,
p
p
.
10
50
~
10
61
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
2
.
p
p
10
50
-
10
61
1050
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Struct
ured
data
c
o
llection a
nd
de
e
p learning
for
ret
i
na
l O
CT
ima
g
e
-
to
-
te
x
t
t
ra
nsla
tion: a co
mpr
ehensiv
e f
ra
me
wo
rk
Uda
y
M
a
nd
e
1
,
Sh
a
f
i P
a
t
ha
n
1
,
P
a
nk
a
j
Cha
nd
re
1
,
Sh
a
rv
a
ri
M
a
nd
e
2
1
D
e
p
a
r
t
m
e
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
M
I
T
S
c
h
o
o
l
o
f
C
o
m
p
u
t
i
n
g
,
M
I
T
A
r
t
D
e
s
i
g
n
a
n
d
T
e
c
h
n
o
l
o
g
y
U
n
i
v
e
r
s
i
t
y
,
P
u
n
e
,
I
n
d
i
a
2
G
o
v
e
r
n
e
me
n
t
M
e
d
i
c
a
l
C
o
l
l
e
g
e
,
S
a
t
a
r
a
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
2
,
2
0
2
4
R
ev
is
ed
J
an
6
,
2
0
2
6
Acc
ep
ted
J
an
2
5
,
2
0
2
6
Th
is
p
a
p
e
r
p
re
se
n
ts
a
c
o
m
p
r
e
h
e
n
siv
e
fra
m
e
wo
rk
f
o
r
str
u
c
t
u
re
d
d
a
ta
c
o
ll
e
c
ti
o
n
a
n
d
d
e
e
p
lea
rn
in
g
(
DL)
-
b
a
se
d
tran
sla
ti
o
n
o
f
re
t
in
a
l
o
p
ti
c
a
l
c
o
h
e
re
n
c
e
to
m
o
g
ra
p
h
y
(
OCT
)
i
m
a
g
e
s
in
to
d
iag
n
o
stic
tex
t.
Th
e
su
g
g
e
ste
d
a
p
p
ro
a
c
h
g
u
a
ra
n
tee
s
h
i
g
h
-
q
u
a
li
ty
OCT
d
a
ta
f
o
r
m
o
d
e
l
train
i
n
g
th
ro
u
g
h
th
e
u
se
o
f
so
p
h
isti
c
a
ted
ima
g
e
p
ro
c
e
ss
in
g
m
e
th
o
d
s
li
k
e
e
d
g
e
d
e
tec
t
io
n
,
n
o
ise
su
p
p
re
ss
io
n
,
a
n
d
c
o
n
tras
t
im
p
ro
v
e
m
e
n
t.
Th
e
stu
d
y
u
ti
li
z
e
s
8
4
,
4
8
4
re
ti
n
a
l
ima
g
e
s
fro
m
t
h
e
OCT
d
a
tas
e
t
a
v
a
il
a
b
le
o
n
Ka
g
g
le.
T
h
e
re
se
a
rc
h
u
ti
li
z
e
s
v
a
rio
u
s
p
re
p
ro
c
e
ss
in
g
tec
h
n
iq
u
e
s,
su
c
h
a
s
m
e
d
ian
a
n
d
G
a
u
ss
ian
fil
terin
g
,
a
lo
n
g
wi
th
d
a
ta
a
u
g
m
e
n
tatio
n
stra
teg
ies
li
k
e
tran
sla
ti
o
n
,
r
o
t
a
ti
o
n
,
a
n
d
sc
a
li
n
g
,
to
m
it
ig
a
te
c
las
s
imb
a
la
n
c
e
s
a
n
d
imp
ro
v
e
m
o
d
e
l
p
e
rfo
r
m
a
n
c
e
.
Th
e
sy
ste
m
a
u
to
m
a
ti
c
a
ll
y
id
e
n
ti
fies
a
n
d
c
a
teg
o
rize
s
re
ti
n
a
l
d
ise
a
se
s
su
c
h
a
s
d
ru
se
n
,
d
iab
e
ti
c
m
a
c
u
lar
e
d
e
m
a
(DME
),
a
n
d
c
h
o
ro
i
d
a
l
n
e
o
v
a
sc
u
lariz
a
ti
o
n
(CNV
)
b
y
in
te
g
ra
ti
n
g
fe
a
tu
re
e
x
t
ra
c
ti
o
n
a
n
d
se
lec
ti
o
n
with
DL
te
c
h
n
iq
u
e
s.
Th
e
re
se
a
rc
h
h
ig
h
li
g
h
ts
th
e
imp
o
r
tan
c
e
o
f
e
ffe
c
ti
v
e
d
a
ta
h
a
n
d
li
n
g
a
n
d
m
o
d
e
l
sc
a
lab
il
it
y
to
a
d
d
re
ss
th
e
in
c
re
a
sin
g
n
e
e
d
fo
r
a
u
t
o
m
a
ted
d
iag
n
o
sti
c
to
o
ls
in
o
p
h
th
a
lmo
l
o
g
y
.
Th
is
fra
m
e
wo
rk
a
ims
t
o
s
u
p
p
o
rt
o
p
h
t
h
a
lmo
l
o
g
ists
i
n
m
a
n
a
g
in
g
th
e
in
c
re
a
sin
g
in
c
i
d
e
n
c
e
o
f
d
ia
b
e
ti
c
re
ti
n
o
p
a
t
h
y
(DR)
a
n
d
o
th
e
r
re
ti
n
a
l
c
o
n
d
it
i
o
n
s
b
y
e
n
h
a
n
c
i
n
g
th
e
e
fficie
n
c
y
o
f
re
ti
n
a
l
ima
g
e
a
n
a
ly
sis
,
th
e
re
b
y
imp
ro
v
in
g
p
a
ti
e
n
t
re
su
l
ts
th
ro
u
g
h
e
a
rly
d
e
tec
ti
o
n
a
n
d
trea
tme
n
t.
K
ey
w
o
r
d
s
:
Au
to
m
ated
d
iag
n
o
s
is
Dee
p
lear
n
in
g
Diab
etic
m
ac
u
lar
ed
em
a
I
m
ag
e
p
r
ep
r
o
ce
s
s
in
g
R
etin
al
OC
T
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
:
Ud
ay
Ma
n
d
e
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
M
I
T
Sch
o
o
l o
f
C
o
m
p
u
tin
g
MI
T
Ar
t D
esig
n
an
d
T
ec
h
n
o
l
o
g
y
Un
iv
er
s
ity
L
o
n
i K
alb
h
o
r
,
Pu
n
e,
I
n
d
ia
E
m
ail: u
d
ay
.
m
a
n
d
e@
m
itu
n
iv
e
r
s
ity
.
ed
u
.
in
1.
I
NT
RO
D
UCT
I
O
N
Op
tical
co
h
er
e
n
ce
to
m
o
g
r
a
p
h
y
(
OC
T
)
is
a
n
o
n
-
i
n
v
asiv
e
im
ag
in
g
tec
h
n
iq
u
e
th
at
g
en
e
r
ates
cr
o
s
s
-
s
ec
tio
n
al,
h
ig
h
-
r
eso
lu
tio
n
im
ag
es
o
f
th
e
r
etin
a.
I
t
is
ess
en
tial
to
o
p
h
th
alm
o
lo
g
y
s
in
c
e
it
en
ab
les
d
o
cto
r
s
to
in
v
esti
g
ate
th
e
r
etin
a
’
s
lay
er
s
an
d
id
en
tify
v
ar
io
u
s
ey
e
d
is
o
r
d
er
s
,
in
clu
d
in
g
g
lau
co
m
a,
d
iab
etic
r
etin
o
p
ath
y
(
DR
)
,
an
d
ag
e
-
r
elate
d
m
ac
u
lar
d
eg
en
er
atio
n
(
AM
D)
[
1
]
.
OC
T
’
s
ab
ilit
y
to
id
en
tify
m
in
u
te
a
lter
atio
n
s
in
r
etin
al
s
tr
u
ctu
r
e,
wh
ich
ca
n
s
ig
n
if
ica
n
tly
af
f
ec
t
p
atien
t
o
u
tco
m
es,
m
ak
es
it
ess
en
tial
to
o
l
f
o
r
t
h
e
ea
r
ly
i
d
en
tific
atio
n
an
d
tr
ac
k
i
n
g
o
f
ey
e
d
is
ea
s
es
[
2
]
.
Desp
ite
b
ein
g
a
s
u
cc
ess
f
u
l
tech
n
iq
u
e,
OC
T
p
ictu
r
e
in
ter
p
r
etatio
n
is
tim
e
-
co
n
s
u
m
in
g
an
d
r
eq
u
ir
es sp
ec
ialized
k
n
o
wled
g
e,
wh
ich
co
u
ld
d
elay
d
iag
n
o
s
is
an
d
tr
ea
tm
en
t
[
3
]
,
[
4
]
.
T
h
e
lar
g
e
v
o
lu
m
e
an
d
co
m
p
lex
ity
o
f
d
ata
p
r
o
d
u
ce
d
b
y
OC
T
s
ca
n
s
r
eq
u
i
r
e
th
e
u
s
e
o
f
s
o
p
h
i
s
ticated
au
to
m
ated
an
aly
s
is
to
o
ls
,
p
r
o
m
p
tin
g
in
v
esti
g
atio
n
in
to
m
ac
h
in
e
lear
n
i
n
g
(
ML
)
a
n
d
im
ag
e
-
to
-
tex
t
c
o
n
v
er
s
io
n
m
eth
o
d
s
.
W
ith
th
e
h
elp
o
f
th
ese
tech
n
iq
u
es,
clin
ician
s
ca
n
m
ak
e
q
u
ick
er
,
an
d
b
etter
d
ec
is
io
n
s
[
5
]
,
[
6
]
.
Du
e
to
th
e
in
cr
ea
s
in
g
d
ep
en
d
en
ce
o
n
im
ag
in
g
in
m
ed
ical
d
iag
n
o
s
tics
,
th
er
e
is
a
r
is
in
g
d
em
an
d
f
o
r
au
to
m
ated
s
y
s
tem
s
ca
p
ab
le
o
f
ac
cu
r
ately
in
ter
p
r
etin
g
co
m
p
lex
m
ed
ical
im
ag
es
an
d
tr
an
s
f
o
r
m
in
g
th
em
in
to
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
S
tr
u
ctu
r
ed
d
a
ta
c
o
llectio
n
a
n
d
d
ee
p
lea
r
n
in
g
fo
r
r
etin
a
l O
C
T
ima
g
e
-
to
-
text
tr
a
n
s
la
tio
n
…
(
Ud
a
y
Ma
n
d
e
)
1051
v
alu
ab
le
in
s
ig
h
ts
[
7
]
,
[
8
]
.
I
n
r
etin
al
OC
T
,
im
ag
e
-
to
-
tex
t
tr
a
n
s
latio
n
r
ef
er
s
to
co
n
v
er
tin
g
v
is
u
al
d
ata
f
r
o
m
OC
T
im
ag
es
in
to
d
escr
ip
tiv
e
tex
t
th
at
h
ig
h
lig
h
ts
k
ey
f
in
d
in
g
s
,
in
clu
d
in
g
r
etin
al
ab
n
o
r
m
alities
,
m
ea
s
u
r
em
en
ts
o
f
r
etin
al
lay
er
s
,
o
r
in
d
icatio
n
s
o
f
d
is
ea
s
e
p
r
o
g
r
ess
io
n
[
9
]
,
[
1
0
]
.
T
h
is
au
to
m
atio
n
ca
n
s
ig
n
if
ican
tly
less
en
th
e
co
g
n
itiv
e
b
u
r
d
e
n
o
n
p
h
y
s
i
cian
s
,
s
tan
d
ar
d
ize
r
e
p
o
r
tin
g
p
r
o
ce
d
u
r
es,
a
n
d
d
ec
r
ea
s
e
h
u
m
a
n
m
is
tak
es.
Fu
r
th
er
m
o
r
e
,
au
to
m
ated
im
ag
e
-
to
-
tex
t
s
y
s
tem
s
ca
n
aid
in
t
elem
ed
icin
e
an
d
r
em
o
te
d
iag
n
o
s
tics
,
en
h
an
cin
g
ac
ce
s
s
to
q
u
ality
ey
e
ca
r
e,
p
ar
ticu
lar
ly
in
u
n
d
er
s
er
v
ed
r
e
g
io
n
s
[
1
1
]
,
[
1
2
]
.
I
n
teg
r
atin
g
d
ee
p
lear
n
in
g
(
DL
)
m
eth
o
d
s
in
to
th
is
p
r
o
ce
d
u
r
e
ca
n
en
h
a
n
ce
th
e
ac
cu
r
ac
y
,
s
p
ee
d
,
a
n
d
c
o
n
s
is
ten
cy
o
f
O
C
T
im
ag
e
an
aly
s
is
,
u
ltima
tely
r
esu
ltin
g
in
im
p
r
o
v
ed
p
atien
t o
u
tco
m
es.
Vis
io
n
is
im
p
o
r
tan
t
i
n
h
u
m
an
life
.
W
ith
o
u
t
v
is
io
n
life
will
b
e
m
is
er
ab
le
a
n
d
ca
n
’
t
g
o
a
h
ea
d
in
t
h
e
r
ig
h
t
d
i
r
ec
tio
n
.
E
y
e
is
an
im
p
o
r
tan
t
o
r
g
an
an
d
p
lay
s
an
im
p
o
r
tan
t
r
o
le
in
h
u
m
an
v
is
io
n
.
Ma
in
co
m
p
o
n
en
t
o
f
th
e
ey
e
is
th
e
ey
eb
all
wh
ich
is
m
ad
e
u
p
o
f
lan
ce
s
an
d
r
etin
a
[
1
3
]
.
I
m
ag
e
p
r
o
d
u
ctio
n
is
d
o
n
e
in
th
e
ey
eb
all
d
u
e
to
lig
h
t
r
ay
s
.
T
h
e
r
ef
r
ac
tio
n
(
b
en
d
in
g
)
o
f
lig
h
t
b
y
th
e
c
o
r
n
e
a
an
d
th
e
len
s
is
wh
at
ca
u
s
e
s
f
o
cu
s
ed
p
ictu
r
es
to
d
ev
elo
p
o
n
t
h
e
p
h
o
to
r
ec
e
p
to
r
s
o
f
th
e
r
etin
a
as
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
e
co
r
n
ea
p
er
f
o
r
m
s
th
e
m
ajo
r
ity
o
f
th
e
r
eq
u
ir
ed
r
ef
r
ac
tio
n
,
a
r
o
le
th
at
is
r
ea
d
ily
u
n
d
er
s
to
o
d
wh
en
o
n
e
co
n
s
id
er
s
th
e
b
lu
r
r
y
,
o
u
t
-
of
-
f
o
cu
s
im
ag
es
th
at
ar
e
p
r
esen
t
wh
en
s
wim
m
in
g
u
n
d
er
wate
r
.
T
h
e
len
s
h
as
a
m
u
ch
lo
wer
r
ef
r
ac
tiv
e
p
o
wer
th
an
th
e
co
r
n
ea
,
b
u
t
b
ec
au
s
e
it
ca
n
b
e
a
d
ju
s
ted
,
it
m
ay
b
r
in
g
o
b
jects
at
d
if
f
er
en
t
d
is
tan
ce
s
f
r
o
m
t
h
e
o
b
s
er
v
er
i
n
to
f
in
e
f
o
cu
s
o
n
t
h
e
r
etin
al
s
u
r
f
ac
e.
Fig
u
r
e
1
.
I
m
ag
e
f
o
r
m
atio
n
p
r
o
ce
s
s
T
h
e
r
etin
a
is
v
er
y
im
p
o
r
tan
t
a
n
d
a
n
in
teg
r
al
p
a
r
t
in
th
e
im
a
g
e
f
o
r
m
atio
n
.
I
n
d
ia
is
m
o
v
in
g
t
o
war
d
s
th
e
d
iab
etic
ca
p
ital
o
f
th
e
wo
r
l
d
h
av
in
g
7
7
,
0
0
0
,
0
0
0
d
iab
etic
af
f
ec
ts
th
e
r
etin
a
o
f
h
u
m
an
s
in
lar
g
e
way
s
.
Du
e
to
th
at
r
etin
al
d
is
o
r
d
er
s
ar
e
g
en
er
ated
in
d
iab
etic
p
atien
ts
b
y
a
n
d
lar
g
e
.
T
h
e
d
ev
ice
av
ailab
le
f
o
r
ca
p
tu
r
in
g
r
etin
al
im
ag
es
is
OC
T
.
DR
is
o
n
e
o
f
th
e
b
ig
is
s
u
es
f
o
r
ey
e
s
u
r
g
eo
n
s
.
OC
T
g
iv
es
r
etin
al
im
ag
es
.
L
ar
g
e
v
o
lu
m
e
o
f
im
ag
es
is
g
en
er
ated
u
s
in
g
OC
T
.
I
t
b
ec
o
m
es
d
if
f
icu
lt
f
o
r
e
y
e
s
u
r
g
eo
n
s
to
g
o
th
r
o
u
g
h
ea
c
h
an
d
ev
e
r
y
im
a
g
e
in
d
etail.
So
,
it
is
tim
e
co
n
s
u
m
i
n
g
an
d
p
r
o
n
e
to
e
r
r
o
r
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
will
p
r
o
ce
s
s
th
e
i
m
ag
es
an
d
g
en
e
r
ate
p
r
ec
is
e
tex
tu
al
r
ep
o
r
ts
u
s
in
g
D
L
m
eth
o
d
s
.
T
h
is
r
ep
o
r
t w
ill h
e
lp
s
u
r
g
eo
n
s
f
o
r
c
o
r
r
ec
t d
ia
g
n
o
s
is
.
T
h
is
s
u
r
v
ey
s
ee
k
s
to
d
eliv
er
a
d
etailed
s
u
m
m
ar
y
o
f
r
ec
e
n
t
p
r
o
g
r
ess
in
tr
an
s
f
o
r
m
in
g
d
escr
ip
tiv
e
tex
t
f
r
o
m
r
etin
al
OC
T
im
ag
es
th
r
o
u
g
h
DL
tec
h
n
iq
u
es.
I
t
u
n
d
er
s
c
o
r
es
th
e
s
ig
n
if
ican
ce
o
f
g
ath
er
in
g
o
r
g
an
ized
d
ata
f
r
o
m
tr
u
s
two
r
th
y
s
o
u
r
ce
s
an
d
em
p
lo
y
in
g
ad
v
an
ce
d
ML
m
eth
o
d
s
to
en
h
an
ce
th
e
p
r
ec
is
io
n
an
d
ef
f
ec
tiv
en
ess
o
f
im
ag
e
an
aly
s
is
.
T
h
e
s
tu
d
y
s
ee
k
s
to
co
n
n
ec
t c
u
r
r
en
t
r
esear
ch
with
ad
v
an
ce
d
tec
h
n
o
lo
g
y
b
y
o
u
tlin
in
g
th
e
m
er
its
an
d
d
r
awb
ac
k
s
o
f
v
ar
io
u
s
DL
m
eth
o
d
s
,
s
u
ch
as
h
y
b
r
id
m
o
d
els,
r
ec
u
r
r
e
n
t
n
eu
r
al
n
et
wo
r
k
s
(
R
NNs),
an
d
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs).
T
h
e
s
u
r
v
ey
a
d
d
it
io
n
ally
tack
les
th
e
d
if
f
icu
ltie
s
o
f
d
ata
q
u
ality
,
an
n
o
tatio
n
,
a
n
d
in
te
g
r
atio
n
,
h
ig
h
lig
h
tin
g
th
e
im
p
o
r
ta
n
ce
o
f
well
-
o
r
g
an
ized
d
atasets
to
d
ev
elo
p
r
eliab
le
an
d
b
r
o
ad
ly
a
p
p
licab
le
m
o
d
els.
T
h
e
s
u
r
v
ey
aim
s
to
p
in
p
o
in
t
cr
itical
ar
ea
s
f
o
r
f
u
r
th
er
in
v
esti
g
atio
n
an
d
cr
ea
te
a
f
r
am
ewo
r
k
f
o
r
ad
v
a
n
cin
g
m
o
r
e
ef
f
icien
t a
n
d
s
ca
lab
le
r
etin
al
OC
T
im
ag
e
-
to
-
tex
t tr
an
s
latio
n
s
y
s
tem
s
.
2.
O
VE
RVI
E
W
O
F
R
E
T
I
NAL
O
CT
I
M
AG
I
NG
2
.
1
.
B
a
s
ics o
f
O
CT
:
t
ec
hn
ica
l o
v
er
v
iew
o
f
O
CT
t
ec
hn
o
lo
g
y
Ph
y
s
ician
s
ca
n
o
b
tain
cr
o
s
s
-
s
ec
tio
n
al
im
ag
es
o
f
th
e
r
etin
a
with
m
icr
o
m
eter
r
eso
lu
tio
n
u
s
in
g
lig
h
t
wav
es a
n
d
OC
T
,
g
iv
in
g
th
em
a
th
o
r
o
u
g
h
u
n
d
er
s
tan
d
in
g
o
f
t
h
e
in
ter
n
al
s
tr
u
ctu
r
e
o
f
t
h
e
ey
e
[
1
4
]
.
OC
T
is
b
ased
o
n
th
e
lo
w
-
c
o
h
er
e
n
ce
in
ter
f
e
r
o
m
etr
y
p
r
in
cip
le,
wh
ich
s
p
lits
lig
h
t f
r
o
m
a
b
r
o
a
d
b
an
d
s
o
u
r
ce
in
to
two
p
ath
s
: o
n
e
is
d
ir
ec
ted
to
war
d
s
th
e
r
etin
a,
an
d
th
e
o
th
e
r
is
r
e
f
lecte
d
b
y
a
r
ef
e
r
en
ce
m
ir
r
o
r
.
T
h
e
i
n
ter
f
er
en
ce
p
atter
n
p
r
o
d
u
ce
d
b
y
th
e
r
ef
lecte
d
lig
h
t
is
th
en
p
r
o
ce
s
s
ed
to
p
r
o
v
id
e
h
ig
h
-
r
eso
lu
tio
n
im
ag
es
o
f
th
e
r
etin
al
lay
er
s
.
T
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
10
50
-
10
61
1052
s
tr
u
ctu
r
e
an
d
th
ick
n
ess
o
f
th
e
v
ar
io
u
s
r
etin
al
lay
er
s
,
s
u
ch
as
th
e
m
ac
u
la
an
d
r
etin
al
n
er
v
e
f
ib
er
lay
er
(
R
NFL)
,
ca
n
b
e
m
ea
s
u
r
ed
u
s
in
g
OC
T
,
a
n
o
n
-
in
v
asiv
e,
in
v
iv
o
tech
n
i
q
u
e
f
o
r
v
is
u
alizin
g
r
etin
al
ar
c
h
itectu
r
e
[
1
5
]
,
[
1
6
]
.
I
n
o
p
h
t
h
alm
o
lo
g
y
,
th
e
tech
n
o
lo
g
y
’
s
ab
ilit
y
to
ac
cu
r
ately
d
etec
t
an
d
m
o
n
ito
r
ch
a
n
g
es
in
th
e
r
etin
a
is
h
ig
h
ly
v
alu
ed
s
in
ce
it is
cr
u
cial
f
o
r
id
en
tify
in
g
a
n
d
tr
ea
tin
g
ey
e
c
o
n
d
itio
n
s
.
2
.
2
.
Clini
ca
l
re
lev
a
nce:
a
pp
lica
t
io
ns
o
f
O
CT
in
dia
g
no
s
in
g
re
t
ina
l dis
ea
s
es
OC
T
p
r
o
v
id
es
ac
cu
r
ate,
h
ig
h
-
r
eso
lu
tio
n
,
n
o
n
-
in
v
asiv
e
im
ag
i
n
g
o
f
r
etin
al
s
tr
u
ctu
r
es,
m
ak
in
g
it
a
v
ital
to
o
l
in
clin
ical
o
p
h
th
alm
o
lo
g
y
.
I
t
is
o
f
ten
u
s
ed
in
d
iag
n
o
s
in
g
an
d
tr
ea
tin
g
r
etin
al
d
is
o
r
d
er
s
,
s
u
ch
as
AM
D
,
as
it
f
ac
ilit
ates
th
e
ass
ess
m
en
t
o
f
t
r
ea
tm
en
t
e
f
f
icac
y
a
n
d
th
e
id
en
tific
atio
n
o
f
d
r
u
s
en
,
p
ig
m
en
t
ep
ith
elial
d
etac
h
m
en
ts
,
an
d
m
a
cu
lar
at
r
o
p
h
y
.
OC
T
aid
s
in
th
e
ea
r
ly
m
an
ag
e
m
en
t
o
f
DR
to
p
r
e
v
en
t
v
is
io
n
l
o
s
s
b
y
f
ac
ilit
atin
g
a
d
etailed
ass
ess
m
en
t
o
f
r
etin
al
th
ic
k
n
ess
,
m
ac
u
lar
ed
em
a,
a
n
d
n
e
o
v
ascu
l
ar
ch
an
g
es.
OC
T
is
cr
u
cial
f
o
r
ea
r
l
y
d
etec
tio
n
an
d
m
o
n
ito
r
in
g
o
f
g
lau
c
o
m
a
p
r
o
g
r
ess
io
n
b
y
ass
ess
in
g
th
e
th
ick
n
ess
o
f
th
e
R
NFL
.
Ass
es
s
in
g
is
s
u
es
s
u
ch
as
m
a
cu
lar
h
o
les,
v
itre
o
m
ac
u
lar
tr
a
ctio
n
,
an
d
r
etin
al
d
etac
h
m
en
t
is
also
v
ital,
a
s
i
t
ass
is
ts
in
d
ec
id
in
g
th
e
m
o
s
t a
p
p
r
o
p
r
iate
tr
ea
tm
en
t o
r
s
u
r
g
ical
ap
p
r
o
ac
h
.
2
.
3
.
C
h
a
l
l
en
g
e
s
i
n
O
CT
d
a
t
a
i
n
t
e
rp
r
e
t
a
t
i
o
n
:
h
u
ma
n
e
rr
o
r
s
,
t
i
me
c
o
n
s
u
mp
t
i
o
n
,
a
n
d
n
e
ed
f
o
r
a
u
t
o
ma
t
i
o
n
Alth
o
u
g
h
OC
T
g
e
n
er
ates
p
r
ec
is
e
im
ag
es
ess
en
tial
f
o
r
m
an
ag
in
g
r
etin
al
co
n
d
itio
n
s
,
in
ter
p
r
e
tin
g
th
ese
im
ag
es
ca
n
o
cc
asio
n
ally
b
e
d
if
f
icu
lt.
I
f
q
u
alif
ied
m
e
d
ical
p
r
o
f
ess
io
n
als
r
ely
s
o
lely
o
n
m
an
u
al
ass
ess
m
en
t,
h
u
m
an
er
r
o
r
m
ay
o
cc
u
r
,
r
esu
ltin
g
in
v
ar
ie
d
in
te
r
p
r
etatio
n
s
.
Dif
f
er
en
ce
s
in
th
e
clin
ician
’
s
ex
p
er
tis
e,
th
e
q
u
ality
o
f
th
e
im
ag
e,
a
n
d
th
e
e
x
is
ten
ce
o
f
ar
tifa
cts
ca
n
in
f
lu
e
n
ce
th
e
ac
cu
r
ac
y
o
f
a
d
iag
n
o
s
is
.
Fu
r
th
er
m
o
r
e,
t
h
e
lar
g
e
q
u
an
tity
o
f
OC
T
im
ag
es
p
r
o
d
u
ce
d
in
clin
ical
en
v
ir
o
n
m
e
n
ts
ca
n
b
e
o
v
e
r
wh
elm
in
g
,
r
eq
u
ir
in
g
ex
tr
a
tim
e
f
o
r
co
m
p
r
eh
e
n
s
iv
e
ev
alu
atio
n
.
T
h
is
d
em
an
d
m
a
y
wo
r
s
en
d
iag
n
o
s
is
d
elay
s
i
n
h
i
g
h
-
tr
a
f
f
i
c
s
ettin
g
s
s
u
ch
as
telem
ed
icin
e
s
y
s
tem
s
o
r
lar
g
e
ey
e
clin
ics.
W
h
ile
OC
T
p
r
o
v
id
es
ac
cu
r
ate
im
ag
es
cr
u
cial
f
o
r
h
an
d
lin
g
r
e
tin
al
is
s
u
es,
u
n
d
er
s
tan
d
in
g
th
e
s
e
im
ag
es
ca
n
s
o
m
etim
es
b
e
ch
allen
g
in
g
.
W
h
en
q
u
alif
ied
m
ed
ical
p
r
o
f
ess
io
n
als
d
ep
en
d
o
n
l
y
o
n
m
an
u
al
ev
alu
atio
n
,
h
u
m
an
m
is
tak
es
ca
n
h
ap
p
en
,
lead
in
g
t
o
d
i
f
f
er
en
t
in
ter
p
r
etatio
n
s
.
Var
iatio
n
s
i
n
th
e
clin
icia
n
’
s
k
n
o
wled
g
e,
th
e
q
u
ality
o
f
th
e
im
ag
e,
an
d
th
e
p
r
esen
ce
o
f
ar
tifa
cts
ca
n
im
p
ac
t
th
e
p
r
ec
is
io
n
o
f
a
d
iag
n
o
s
is
.
Ad
d
itio
n
ally
,
th
e
s
ig
n
if
ican
t
n
u
m
b
er
o
f
OC
T
i
m
ag
es
g
en
er
ate
d
in
cli
n
ical
s
ettin
g
s
ca
n
b
e
d
au
n
tin
g
,
n
ec
e
s
s
itatin
g
ad
d
itio
n
al
tim
e
f
o
r
t
h
o
r
o
u
g
h
ass
ess
m
en
t.
T
h
is
r
eq
u
est
c
o
u
ld
e
x
ac
er
b
at
e
d
elay
s
in
d
iag
n
o
s
is
in
b
u
s
y
en
v
ir
o
n
m
en
ts
lik
e
telem
ed
icin
e
p
latf
o
r
m
s
o
r
lar
g
e
o
p
h
th
alm
o
lo
g
y
clin
ics.
3.
DATA CO
L
L
E
C
T
I
O
N
AND
O
RG
ANIZ
A
T
I
O
N
3
.
1
.
I
m
po
r
t
a
nce
o
f
re
lia
ble da
t
a
s
o
urce
s
:
ens
uring
da
t
a
qu
a
lity
a
nd
a
cc
ura
cy
I
n
m
ed
ical
im
ag
in
g
,
wh
er
e
d
ata
q
u
ality
i
n
f
lu
en
ce
s
t
h
e
d
ep
en
d
a
b
ilit
y
an
d
p
r
ec
is
io
n
o
f
m
o
d
el
p
r
ed
ictio
n
s
,
tr
u
s
two
r
th
y
d
ata
s
o
u
r
ce
s
ar
e
cr
u
cial
f
o
r
th
e
s
u
cc
ess
o
f
an
y
ML
p
r
o
jec
t.
I
n
r
etin
al
OC
T
im
ag
e
-
to
-
tex
t
tr
an
s
latio
n
,
u
tili
zin
g
h
ig
h
-
q
u
ality
an
d
p
r
ec
is
e
d
ata
en
s
u
r
es
th
at
th
e
m
o
d
els
tr
ain
ef
f
ec
tiv
ely
an
d
g
en
er
alize
s
u
cc
ess
f
u
lly
to
u
n
f
am
iliar
in
p
u
ts
.
R
eliab
le
d
ata
s
o
u
r
ce
s
in
clu
d
e
clin
ical
d
atab
ases
,
ac
ad
em
ic
r
esear
ch
d
atasets
,
an
d
au
th
e
n
tic
p
u
b
lic
r
ep
o
s
ito
r
ies
th
at
o
f
f
er
s
tan
d
ar
d
ized
,
h
ig
h
-
r
eso
l
u
tio
n
OC
T
im
ag
es
ac
co
m
p
an
ied
b
y
ap
p
r
o
p
r
iate
lab
elin
g
.
E
n
s
u
r
in
g
d
ata
q
u
a
lity
r
eq
u
ir
es
v
er
if
y
in
g
th
e
p
r
ec
is
io
n
o
f
r
elate
d
m
etad
ata,
in
clu
d
i
n
g
p
atien
t
d
em
o
g
r
ap
h
ics
an
d
d
iag
n
o
s
tic
lab
els,
as
well
as
m
ain
tain
in
g
u
n
if
o
r
m
ity
in
d
ata
f
o
r
m
ats
an
d
im
ag
e
in
teg
r
ity
.
Hig
h
-
q
u
ality
d
ata
m
in
im
iz
es
th
e
ch
an
ce
s
o
f
b
iases
o
r
m
is
tak
es
b
ein
g
in
co
r
p
o
r
ated
in
t
o
th
e
m
o
d
el,
r
esu
ltin
g
in
m
o
r
e
r
o
b
u
s
t
an
d
th
er
ap
e
u
tically
r
elev
a
n
t
o
u
tco
m
es.
Up
h
o
ld
i
n
g
eth
ical
s
tan
d
ar
d
s
is
f
u
r
th
er
s
u
p
p
o
r
ted
b
y
u
tili
zin
g
in
f
o
r
m
ati
o
n
f
r
o
m
r
eliab
le
s
o
u
r
ce
s
,
p
ar
ti
cu
lar
ly
in
h
a
n
d
lin
g
co
n
f
id
en
tial m
e
d
ical
in
f
o
r
m
atio
n
.
3
.
2
.
M
et
ho
ds
o
f
da
t
a
a
cquis
it
io
n:
t
ec
hn
iqu
e
s
f
o
r
o
bta
ini
ng
O
CT
im
a
g
es
f
ro
m
clinica
l
da
t
a
ba
s
es,
p
ub
li
c
da
t
a
s
et
s
,
a
nd
o
t
her
s
o
urce
s
A
cr
itical
s
tag
e
in
cr
ea
tin
g
r
eliab
le
DL
m
o
d
els
is
d
ata
ac
q
u
is
itio
n
,
wh
ich
en
tails
g
ath
e
r
in
g
OC
T
im
ag
es
f
r
o
m
m
an
y
s
o
u
r
ce
s
to
g
u
ar
an
tee
t
h
o
r
o
u
g
h
tr
ain
in
g
a
n
d
v
alid
atio
n
.
L
a
r
g
e
am
o
u
n
ts
o
f
r
ea
l
-
wo
r
ld
OC
T
s
ca
n
s
ar
e
av
ailab
le
in
clin
ical
d
atab
ases
k
ep
t
u
p
to
d
ate
b
y
h
o
s
p
itals
an
d
ey
e
clin
ics.
T
h
ese
d
atab
ases
m
u
s
t
b
e
ac
ce
s
s
ed
with
th
e
p
r
o
p
er
eth
ic
al
ap
p
r
o
v
als
an
d
p
atien
t
co
n
s
en
t
in
o
r
d
e
r
to
ad
h
er
e
to
p
r
iv
ac
y
laws
lik
e
g
en
er
al
d
ata
p
r
o
tectio
n
r
eg
u
latio
n
(
G
DPR
)
an
d
h
ea
lth
in
s
u
r
a
n
ce
p
o
r
tab
ilit
y
an
d
ac
co
u
n
tab
ilit
y
ac
t
(
HI
PAA
)
.
T
h
e
Du
k
e
OC
T
d
ataset
an
d
r
etin
al
OC
T
to
o
l
f
o
r
h
ig
h
-
q
u
ality
ch
allen
g
e
(
R
E
T
OUCH
)
ar
e
two
p
u
b
licly
ac
ce
s
s
ib
le
d
atasets
th
at
p
r
o
v
id
e
s
tan
d
ar
d
ized
,
well
-
an
n
o
tated
im
a
g
es
th
at
ar
e
f
r
eq
u
en
tly
u
ti
lized
in
r
esear
ch
.
Fu
r
th
er
m
o
r
e
,
ac
ce
s
s
to
h
ig
h
-
q
u
ality
OC
T
d
ata
with
ex
p
er
t
a
n
n
o
tatio
n
s
is
m
ad
e
p
o
s
s
ib
le
th
r
o
u
g
h
p
a
r
tn
er
s
h
ip
s
with
ac
ad
em
ic
an
d
r
esear
ch
o
r
g
an
izatio
n
s
,
an
d
th
e
cr
ea
tio
n
o
f
s
y
n
th
etic
d
ata
u
s
in
g
m
eth
o
d
s
lik
e
g
en
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
s
(
GANs
)
ca
n
s
u
p
p
lem
en
t sm
all
d
atasets
an
d
en
h
an
ce
m
o
d
el
g
e
n
er
aliz
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
S
tr
u
ctu
r
ed
d
a
ta
c
o
llectio
n
a
n
d
d
ee
p
lea
r
n
in
g
fo
r
r
etin
a
l O
C
T
ima
g
e
-
to
-
text
tr
a
n
s
la
tio
n
…
(
Ud
a
y
Ma
n
d
e
)
1053
3
.
3
.
D
a
t
a
a
n
n
o
t
a
t
i
o
n
a
n
d
l
a
b
e
l
i
n
g
:
s
t
r
a
t
e
g
i
e
s
f
o
r
a
n
n
o
t
a
t
i
n
g
O
C
T
i
m
a
g
e
s
,
i
n
c
l
u
d
i
n
g
m
a
n
u
a
l
a
n
d
a
u
t
o
m
a
t
e
d
a
p
p
r
o
a
c
h
e
s
Acc
u
r
ate
d
ata
a
n
n
o
tatio
n
an
d
lab
elin
g
ar
e
c
r
u
cial
f
o
r
s
u
cc
ess
f
u
l
m
o
d
el
tr
ain
i
n
g
,
v
ali
d
atio
n
,
a
n
d
test
in
g
in
s
u
p
er
v
is
ed
lear
n
i
n
g
t
ask
s
th
at
tr
an
s
f
er
OC
T
im
ag
es to
tex
tu
al
d
escr
ip
tio
n
s
.
Alth
o
u
g
h
it tak
es a
lo
t
o
f
tim
e
an
d
r
eso
u
r
ce
s
,
m
an
u
al
a
n
n
o
tatio
n
b
y
s
u
b
ject
m
atter
s
p
ec
ialis
ts
lik
e
o
p
h
th
alm
o
lo
g
is
ts
y
ield
s
ex
tr
em
ely
ac
cu
r
ate
lab
elin
g
a
n
d
th
o
r
o
u
g
h
d
escr
ip
tio
n
s
.
Au
t
o
m
ated
an
n
o
tatio
n
tec
h
n
iq
u
es
u
s
e
ML
m
o
d
els
o
r
im
a
g
e
p
r
o
ce
s
s
in
g
to
p
r
o
d
u
ce
p
r
elim
in
ar
y
lab
els
th
at
ca
n
b
e
im
p
r
o
v
ed
th
r
o
u
g
h
ac
tiv
e
lear
n
i
n
g
an
d
ex
p
e
r
t
ass
ess
m
en
t.
Du
e
to
th
e
s
p
ec
ialized
n
atu
r
e
o
f
OC
T
in
ter
p
r
etatio
n
,
cr
o
wd
s
o
u
r
cin
g
p
latf
o
r
m
s
allo
w
f
o
r
s
ca
lab
le
lab
elin
g
b
u
t
n
ec
ess
itate
s
tr
in
g
en
t
q
u
ality
co
n
tr
o
l.
I
n
co
n
tr
ast,
h
y
b
r
i
d
tech
n
iq
u
es
th
at
co
m
b
in
e
au
t
o
m
at
ed
to
o
ls
with
ex
p
er
t
o
v
er
s
ig
h
t p
r
o
v
id
e
a
wo
r
k
a
b
le
b
alan
ce
b
etwe
en
ef
f
icien
cy
a
n
d
ac
cu
r
ac
y
.
3
.
4
.
D
a
t
a
p
r
ep
r
o
c
es
s
in
g
:
s
t
e
ps
f
o
r
p
r
e
p
a
ri
n
g
d
a
t
a
f
o
r
DL
mo
d
e
l
s
,
i
n
cl
ud
i
n
g
n
o
r
ma
l
i
z
a
t
i
o
n
,
a
u
g
me
n
t
a
t
i
o
n
,
a
n
d
s
e
g
me
n
t
a
t
i
o
n
A
cr
u
cial
s
tag
e
in
tr
an
s
f
o
r
m
in
g
u
n
p
r
o
ce
s
s
ed
OC
T
p
ictu
r
es
i
n
to
a
f
o
r
m
at
ap
p
r
o
p
r
iate
f
o
r
ef
f
icien
t
DL
m
o
d
el
tr
ain
i
n
g
is
d
ata
p
r
ep
r
o
ce
s
s
in
g
.
I
n
o
r
d
er
to
s
tab
ilize
lear
n
in
g
an
d
less
en
v
ar
ia
b
ilit
y
b
r
o
u
g
h
t
o
n
b
y
v
ar
io
u
s
im
a
g
in
g
s
ettin
g
s
,
it
i
n
v
o
lv
es
n
o
r
m
alizin
g
p
ix
el
in
ten
s
ity
v
alu
es
to
a
c
o
m
m
o
n
s
ca
le.
T
o
in
c
r
ea
s
e
d
ataset
d
iv
er
s
ity
,
b
o
o
s
t
g
en
er
aliza
tio
n
,
an
d
r
ed
u
ce
o
v
er
f
itti
n
g
,
d
ata
au
g
m
en
tatio
n
tech
n
i
q
u
es
lik
e
r
o
tatio
n
,
f
lip
p
in
g
,
s
ca
lin
g
,
a
n
d
n
o
is
e
in
jectio
n
ar
e
u
s
ed
.
W
h
ile
d
i
m
en
s
io
n
ality
r
ed
u
ctio
n
m
eth
o
d
s
lik
e
p
r
in
cip
al
co
m
p
o
n
en
t
an
aly
s
is
(
PC
A
)
ca
n
f
u
r
th
er
s
im
p
lify
f
e
atu
r
e
r
e
p
r
esen
tatio
n
an
d
b
o
o
s
t
co
m
p
u
tatio
n
al
ef
f
icien
cy
,
s
eg
m
en
tatio
n
o
f
p
er
tin
en
t
an
at
o
m
ical
s
tr
u
ctu
r
es,
s
u
ch
as
r
eti
n
al
lay
er
s
,
alo
n
g
with
ar
tifa
ct
an
d
n
o
is
e
r
em
o
v
al,
h
elp
s
th
e
m
o
d
el
c
o
n
ce
n
t
r
ate
o
n
clin
ically
s
ig
n
if
ican
t r
e
g
io
n
s
.
4.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
4
.
1
.
Rev
iew
o
f
ex
is
t
ing
a
pp
r
o
a
ches
4
.
1
.
1
.
O
v
e
r
v
i
e
w
o
f
p
a
s
t
a
n
d
c
u
r
r
e
n
t
m
e
t
h
o
d
o
l
o
g
i
e
s
u
s
e
d
f
o
r
r
e
t
i
n
a
l
O
C
T
i
m
a
g
e
a
n
a
l
y
s
i
s
a
n
d
t
e
x
t
c
o
n
v
e
r
s
i
o
n
A
co
m
m
o
n
n
o
n
-
in
v
asiv
e
i
m
ag
in
g
tec
h
n
iq
u
e
in
o
p
h
th
alm
o
lo
g
y
f
o
r
o
b
tain
i
n
g
h
ig
h
-
r
eso
lu
tio
n
cr
o
s
s
-
s
ec
tio
n
al
im
ag
es
o
f
th
e
r
etin
a
is
OC
T
.
Diag
n
o
s
in
g
an
d
tr
ac
k
in
g
r
etin
al
illn
ess
es
s
u
ch
g
lau
co
m
a
,
DR
,
an
d
AM
D
d
ep
en
d
h
ea
v
ily
o
n
th
e
an
aly
s
is
o
f
th
ese
im
ag
es.
T
h
e
f
o
llo
win
g
ca
teg
o
r
ies
ap
p
ly
to
ex
is
tin
g
m
eth
o
d
s
f
o
r
tex
t c
o
n
v
er
s
io
n
a
n
d
r
etin
al
OC
T
p
ictu
r
e
a
n
aly
s
is
:
i)
T
r
ad
itio
n
al
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es:
ea
r
ly
a
p
p
r
o
ac
h
es
m
o
s
tly
d
ep
en
d
ed
o
n
s
em
i
-
au
t
o
m
ated
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
a
n
d
m
a
n
u
al
in
ter
p
r
etatio
n
.
T
h
ese
m
et
h
o
d
s
in
clu
d
e
s
eg
m
en
tatio
n
tech
n
iq
u
es
lik
e
lev
el
s
et
an
d
ac
tiv
e
co
n
t
o
u
r
m
o
d
els,
th
r
esh
o
ld
in
g
,
an
d
ed
g
e
d
etec
tio
n
.
Alth
o
u
g
h
h
elp
f
u
l,
th
ese
tech
n
iq
u
es
f
r
e
q
u
en
tly
ca
ll
f
o
r
a
h
ig
h
lev
el
o
f
s
k
ill
an
d
h
av
e
lim
itatio
n
s
wh
en
it
co
m
es
to
m
an
ag
in
g
th
e
v
ar
iab
ilit
y
in
OC
T
p
ictu
r
es b
r
o
u
g
h
t
o
n
b
y
n
o
is
e,
v
ar
io
u
s
im
ag
in
g
cir
cu
m
s
tan
ce
s
,
an
d
v
a
r
i
o
u
s
d
is
ea
s
es.
ii)
ML
-
b
ased
m
eth
o
d
s
:
with
th
e
r
is
e
o
f
ML
,
ad
v
an
ce
d
tech
n
iq
u
es
h
av
e
b
ee
n
u
tili
ze
d
f
o
r
OC
T
im
ag
e
an
aly
s
is
.
T
r
ad
itio
n
al
ML
tec
h
n
iq
u
es
s
u
ch
as
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
i
n
es
(
SVM)
,
r
an
d
o
m
f
o
r
ests
,
an
d
k
-
n
ea
r
est
n
eig
h
b
o
r
s
(
KNN)
h
av
e
b
ee
n
ap
p
lie
d
in
ar
ea
s
s
u
ch
as
f
ea
tu
r
e
ex
tr
ac
tio
n
,
clas
s
if
icatio
n
,
an
d
s
eg
m
en
tatio
n
.
No
n
eth
eless
,
th
ese
tech
n
iq
u
es
o
cc
asio
n
ally
r
eq
u
i
r
e
co
n
s
id
er
ab
le
h
u
m
an
f
ea
tu
r
e
en
g
in
ee
r
in
g
,
wh
ich
ca
n
b
e
la
b
o
r
-
in
ten
s
iv
e
an
d
r
elian
t
o
n
ex
p
er
t k
n
o
wled
g
e.
iii)
DL
-
b
ased
ap
p
r
o
ac
h
es:
r
ec
en
t
p
r
o
g
r
ess
in
DL
,
esp
ec
ially
in
C
NNs,
h
as
g
r
ea
tly
in
f
lu
en
ce
d
OC
T
im
ag
e
an
aly
s
is
.
DL
m
o
d
els
ca
n
au
to
m
atica
lly
d
er
iv
e
f
ea
tu
r
es
f
r
o
m
u
n
p
r
o
ce
s
s
ed
im
ag
e
d
ata,
r
em
o
v
in
g
th
e
n
ec
ess
ity
f
o
r
m
an
u
al
f
ea
tu
r
e
e
x
tr
ac
tio
n
.
Me
th
o
d
s
s
u
ch
as
U
-
Net,
f
u
lly
co
n
v
o
lu
tio
n
al
n
etwo
r
k
s
(
FC
Ns),
alo
n
g
with
m
o
r
e
in
tr
icate
a
r
ch
itectu
r
es
lik
e
R
esNet
an
d
Den
s
eNe
t,
h
av
e
d
em
o
n
s
tr
ate
d
en
c
o
u
r
ag
i
n
g
o
u
tco
m
es
f
o
r
task
s
in
clu
d
in
g
s
eg
m
en
tatio
n
,
class
if
icatio
n
,
an
d
an
o
m
al
y
d
etec
tio
n
in
r
etin
al
OC
T
im
ag
es.
Dee
p
g
en
er
ativ
e
m
o
d
els
lik
e
v
ar
iatio
n
al
au
to
e
n
co
d
er
s
(
VAE
s
)
an
d
GANs
h
av
e
b
ee
n
ex
p
lo
r
ed
f
o
r
th
e
p
u
r
p
o
s
es o
f
im
ag
e
s
y
n
t
h
esis
an
d
au
g
m
e
n
tatio
n
.
iv
)
T
ex
t
co
n
v
er
s
io
n
tech
n
iq
u
es
:
OC
T
im
ag
e
an
aly
s
is
in
f
o
r
m
a
tio
n
is
co
n
v
er
ted
in
to
tex
t
t
h
r
o
u
g
h
n
at
u
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P)
a
n
d
im
a
g
e
-
to
-
tex
t
tr
an
s
latio
n
t
ec
h
n
iq
u
es.
C
lin
ical
d
escr
ip
tio
n
s
h
av
e
b
ee
n
p
r
o
d
u
ce
d
u
s
in
g
c
o
n
v
e
n
tio
n
al
r
u
le
-
b
ased
alg
o
r
ith
m
s
d
er
iv
ed
f
r
o
m
s
tr
u
ctu
r
e
d
d
ata.
R
ec
en
tly
,
DL
m
o
d
els
lik
e
R
NNs,
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
n
etwo
r
k
s
(
L
STM
s
)
,
an
d
tr
an
s
f
o
r
m
er
-
b
ased
ar
ch
i
tectu
r
es
(
e.
g
.
,
b
id
ir
ec
tio
n
al
en
c
o
d
er
r
ep
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
e
r
s
(
B
E
R
T
)
an
d
g
e
n
er
ativ
e
p
r
e
-
tr
ain
e
d
tr
an
s
f
o
r
m
e
r
(
GPT
)
h
av
e
b
ee
n
em
p
lo
y
ed
t
o
p
r
o
d
u
ce
c
o
h
er
e
n
t
an
d
co
n
te
x
tu
ally
ap
p
r
o
p
r
iate
clin
ical
n
a
r
r
ativ
es
b
ased
o
n
im
ag
e
f
ea
tu
r
es.
Ma
lg
h
ee
t
et
a
l.
[
1
7
]
p
r
o
v
id
es
a
co
m
p
r
eh
en
s
iv
e
r
ev
iew
o
f
i
r
is
r
ec
o
g
n
itio
n
d
ev
el
o
p
m
en
t
tech
n
iq
u
es,
h
ig
h
lig
h
tin
g
th
e
ev
o
lu
tio
n
an
d
ef
f
ec
tiv
en
ess
o
f
th
ese
s
y
s
tem
s
in
v
ar
io
u
s
id
en
tific
atio
n
co
n
tex
ts
.
I
r
is
r
ec
o
g
n
itio
n
is
h
ig
h
ly
r
eg
ar
d
e
d
an
d
a
r
eliab
le
b
io
m
et
r
ic
f
o
r
s
ec
u
r
ity
ap
p
licatio
n
s
b
ec
au
s
e
th
e
h
u
m
a
n
ir
is
is
co
n
s
is
ten
t
an
d
u
n
i
q
u
e.
T
h
e
d
o
cu
m
en
t
d
escr
ib
es
th
e
s
ev
e
n
k
ey
s
tag
es
o
f
ir
is
r
ec
o
g
n
itio
n
s
y
s
tem
s
:
ac
q
u
is
itio
n
,
p
r
ep
r
o
ce
s
s
in
g
,
s
eg
m
en
tatio
n
,
n
o
r
m
aliza
tio
n
,
f
ea
t
u
r
e
e
x
tr
ac
tio
n
,
f
ea
tu
r
e
s
elec
tio
n
,
an
d
class
if
icatio
n
.
I
t
ex
p
lo
r
es
th
e
b
en
ef
its
an
d
d
r
awb
ac
k
s
o
f
b
o
th
DL
an
d
c
o
n
v
en
tio
n
al
m
eth
o
d
s
.
DL
al
g
o
r
ith
m
s
ex
ce
l
in
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
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
10
50
-
10
61
1054
p
er
f
o
r
m
an
ce
b
u
t
s
till
f
ac
e
ch
allen
g
es
in
u
n
co
n
tr
o
lled
s
itu
atio
n
s
,
wh
ile
class
ical
m
eth
o
d
s
,
th
o
u
g
h
estab
lis
h
ed
,
ca
n
s
tr
u
g
g
le
in
s
u
b
o
p
tim
al
co
n
d
itio
n
s
s
u
ch
as o
cc
lu
s
io
n
s
an
d
r
ef
lectio
n
s
.
T
h
e
a
r
ticle
also
ex
am
in
es th
e
im
p
ac
t
o
f
n
o
is
e
elem
en
ts
an
d
ey
ew
ea
r
o
n
s
y
s
tem
p
r
ec
is
io
n
,
h
ig
h
lig
h
tin
g
t
h
e
n
ee
d
f
o
r
im
p
r
o
v
ed
s
eg
m
e
n
tatio
n
alg
o
r
ith
m
s
to
e
n
h
an
ce
o
v
e
r
all
r
ec
o
g
n
itio
n
ef
f
ec
tiv
e
n
ess
.
T
o
e
n
h
an
ce
an
d
ad
v
a
n
ce
ir
is
r
ec
o
g
n
itio
n
tech
n
o
lo
g
ies,
u
p
co
m
in
g
r
esear
ch
will c
o
n
ce
n
tr
ate
o
n
th
ese
m
atter
s
.
Oh
et
a
l.
[
1
8
]
p
r
esen
ts
a
DL
-
b
ased
s
y
s
tem
f
o
r
ea
r
ly
d
etec
tio
n
o
f
DR
u
s
in
g
u
ltra
-
wid
e
-
f
ie
ld
(
UW
F)
f
u
n
d
u
s
im
ag
es.
A
m
ajo
r
ca
u
s
e
o
f
v
is
io
n
lo
s
s
is
DR
,
an
d
p
r
o
p
er
t
r
ea
tm
en
t
r
elies
o
n
tim
ely
d
etec
tio
n
.
T
h
is
wo
r
k
em
p
lo
y
s
UW
F
f
u
n
d
u
s
p
h
o
to
g
r
a
p
h
y
,
wh
ich
r
ec
o
r
d
s
as
m
u
ch
as
8
2
%
o
f
t
h
e
r
etin
al
s
u
r
f
ac
e
—
s
ig
n
if
ican
tly
s
u
r
p
ass
in
g
co
n
v
e
n
tio
n
al
tech
n
iq
u
es.
T
h
e
s
y
s
tem
co
n
ce
n
tr
ates
o
n
th
e
ea
r
ly
tr
ea
tm
en
t
d
iab
etic
r
etin
o
p
at
h
y
s
tu
d
y
(
E
T
DR
S)
7
-
s
tan
d
ar
d
f
ield
im
ag
es
f
r
o
m
UW
F
p
h
o
to
g
r
ap
h
y
,
em
p
lo
y
in
g
a
R
esNet
-
3
4
m
o
d
el
f
o
r
class
if
icatio
n
.
T
h
e
ar
ticle
s
ta
tes
th
at
em
p
lo
y
in
g
th
e
E
T
D
R
S
7
-
s
tan
d
ar
d
f
ield
p
h
o
to
s
en
h
an
ce
s
d
etec
tio
n
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
an
d
ar
ea
u
n
d
er
th
e
cu
r
v
e
(
AUC)
in
co
m
p
ar
is
o
n
to
o
th
er
tech
n
iq
u
es.
T
h
e
s
u
g
g
ested
m
eth
o
d
h
ig
h
lig
h
ted
th
e
s
ig
n
if
ican
ce
o
f
in
co
r
p
o
r
atin
g
p
er
ip
h
er
al
r
etin
al
in
f
o
r
m
atio
n
f
o
r
a
m
o
r
e
th
o
r
o
u
g
h
i
d
en
tific
atio
n
o
f
D
R
,
s
u
r
p
ass
in
g
m
o
d
els
th
at
r
e
lied
s
o
lely
o
n
co
n
v
en
tio
n
al
f
u
n
d
u
s
im
a
g
es.
T
h
e
r
esear
ch
h
ig
h
lig
h
ts
th
e
im
p
o
r
tan
ce
o
f
d
ep
e
n
d
ab
le
an
d
u
n
if
o
r
m
d
ata
g
at
h
er
in
g
m
et
h
o
d
s
to
en
h
an
ce
th
e
s
y
s
tem
'
s
ef
f
icien
cy
an
d
s
ca
la
b
ilit
y
.
Fu
r
th
er
m
o
r
e,
it
r
ec
o
m
m
en
d
s
au
to
m
ate
d
s
eg
m
en
tati
o
n
tech
n
i
q
u
es
,
an
d
lar
g
er
,
v
a
r
ied
d
atasets
.
Mo
r
ar
u
et
a
l.
[
1
9
]
d
is
cu
s
s
es
a
s
tu
d
y
o
n
t
h
e
d
e
v
elo
p
m
e
n
t
o
f
a
DL
-
b
ased
ap
p
r
o
ac
h
f
o
r
d
et
ec
tin
g
DR
u
s
in
g
r
etin
al
f
u
n
d
u
s
im
ag
es.
T
h
e
s
u
g
g
ested
ap
p
r
o
ac
h
e
m
p
lo
y
s
C
NNs
to
ca
teg
o
r
ize
r
etin
al
im
ag
es
in
to
d
if
f
er
en
t
s
tag
es
o
f
DR
,
r
an
g
in
g
f
r
o
m
n
o
DR
to
p
r
o
life
r
ativ
e
DR
,
to
p
r
o
v
id
e
a
p
r
ec
is
e
an
d
au
to
m
ated
d
iag
n
o
s
tic
s
o
lu
tio
n
.
T
h
e
s
tu
d
y
em
p
h
asizes
th
e
s
ig
n
if
ican
ce
o
f
tim
ely
id
en
tific
atio
n
a
n
d
t
r
ea
tm
en
t
o
f
DR
to
av
er
t
v
is
io
n
im
p
air
m
en
t,
al
o
n
g
with
th
e
a
d
v
an
ta
g
es
o
f
DL
alg
o
r
ith
m
s
co
m
p
ar
ed
to
co
n
v
en
tio
n
al
im
ag
e
p
r
o
ce
s
s
in
g
m
eth
o
d
s
f
o
r
m
an
a
g
in
g
e
x
ten
s
iv
e
d
atasets
an
d
c
o
m
p
lex
p
atter
n
s
in
m
e
d
ical
i
m
ag
in
g
.
T
h
e
r
esear
ch
ass
es
s
ed
v
ar
io
u
s
C
NN
ar
ch
it
ec
tu
r
es
with
an
ex
ten
s
iv
e
d
a
taset
o
f
lab
eled
r
etin
al
f
u
n
d
u
s
im
ag
es
p
r
io
r
to
ch
o
o
s
in
g
a
m
o
d
el
th
at
d
em
o
n
s
tr
ated
h
ig
h
ac
c
u
r
ac
y
in
r
e
co
g
n
izin
g
DR
s
tag
es
alo
n
g
with
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
.
T
h
e
f
in
d
in
g
s
in
d
i
ca
te
th
at
th
e
DL
m
eth
o
d
g
r
ea
tly
im
p
r
o
v
es
d
iag
n
o
s
tic
ac
cu
r
ac
y
,
d
ec
r
ea
s
in
g
m
is
tak
es
an
d
in
co
n
s
is
ten
cies,
in
co
n
tr
ast
to
th
e
m
an
u
al
g
r
a
d
in
g
b
y
o
p
h
th
alm
o
l
o
g
is
ts
.
T
h
e
co
n
clu
s
io
n
o
f
th
e
p
ap
er
ex
p
lo
r
es
h
o
w
th
e
p
r
o
p
o
s
ed
s
y
s
tem
co
u
ld
b
e
i
n
teg
r
ate
d
in
to
clin
ical
wo
r
k
f
lo
ws.
T
o
en
h
an
ce
th
e
m
o
d
el
’
s
d
iag
n
o
s
tic
u
tili
ty
,
it
em
p
h
asizes
th
e
im
p
o
r
tan
ce
o
f
ad
d
itio
n
al
v
alid
atio
n
ac
r
o
s
s
v
ar
io
u
s
p
o
p
u
latio
n
s
an
d
th
e
o
p
p
o
r
tu
n
ity
to
im
p
r
o
v
e
th
e
m
o
d
el
with
d
if
f
er
e
n
t d
ata
ty
p
es,
lik
e
OC
T
im
ag
es.
T
o
n
g
et
a
l.
[
2
0
]
d
is
cu
s
s
es
th
e
ap
p
licatio
n
o
f
ML
in
o
p
h
th
al
m
ic
im
ag
in
g
,
h
ig
h
lig
h
tin
g
its
p
o
ten
tial
to
en
h
an
ce
th
e
d
iag
n
o
s
is
an
d
tr
ea
tm
en
t
o
f
e
y
e
d
is
ea
s
es.
I
t
s
h
o
ws
h
o
w
in
tr
icate
m
ed
ic
al
im
ag
es
ca
n
b
e
in
ter
p
r
eted
co
r
r
ec
tly
an
d
q
u
i
ck
ly
th
r
o
u
g
h
ML
an
d
DL
.
T
h
e
p
iece
em
p
h
asizes
th
e
ap
p
licatio
n
o
f
ML
in
v
ar
io
u
s
o
c
u
lar
im
a
g
in
g
te
ch
n
iq
u
es,
in
cl
u
d
in
g
OC
T
,
f
u
n
d
u
s
p
h
o
to
g
r
a
p
h
y
,
an
d
s
lit
-
lam
p
im
ag
in
g
.
Op
h
th
alm
o
lo
g
is
ts
ca
n
id
en
tif
y
co
n
d
itio
n
s
s
u
ch
as
DR
,
g
l
au
co
m
a,
an
d
AM
D
b
y
em
p
lo
y
in
g
ML
m
et
h
o
d
s
,
in
clu
d
in
g
s
u
p
er
v
is
ed
a
n
d
u
n
s
u
p
er
v
is
ed
lear
n
in
g
,
to
d
etec
t
an
d
ca
teg
o
r
ize
p
ath
o
lo
g
ical
s
ig
n
s
.
T
h
e
r
ep
o
r
t
an
aly
ze
s
th
e
o
b
s
tacle
s
an
d
f
u
tu
r
e
p
ath
way
s
o
f
ar
tific
ial
in
tellig
en
ce
(
AI
)
in
o
p
h
th
alm
o
lo
g
y
,
alo
n
g
with
th
e
m
eth
o
d
o
l
o
g
y
f
o
r
c
r
ea
tin
g
AI
m
o
d
els.
Aly
o
u
b
i
et
a
l.
[
2
1
]
r
e
v
iews
r
ec
en
t
ad
v
an
ce
m
en
ts
in
au
to
m
ated
d
etec
tio
n
an
d
class
if
icati
o
n
o
f
DR
u
s
in
g
DL
tech
n
iq
u
es,
p
a
r
ticu
lar
ly
C
NNs.
I
f
n
o
t a
d
d
r
ess
ed
,
DR
—
a
f
r
eq
u
en
t o
u
tco
m
e
o
f
d
i
ab
etes
—
m
ay
lead
to
v
is
io
n
lo
s
s
.
Diag
n
o
s
in
g
r
etin
a
l
f
u
n
d
u
s
im
ag
es
m
an
u
ally
is
tim
e
-
co
n
s
u
m
in
g
,
p
r
o
n
e
t
o
m
i
s
tak
es,
an
d
co
s
tly
.
T
h
e
r
esear
ch
h
ig
h
lig
h
ts
th
at
DL
tech
n
iq
u
es
—
s
p
ec
if
icall
y
,
C
NNs
—
ar
e
s
u
p
er
io
r
f
o
r
DR
d
etec
tio
n
an
d
class
if
icatio
n
in
th
e
an
al
y
s
is
o
f
m
ed
ical
im
a
g
es.
T
h
e
p
u
b
licatio
n
in
cl
u
d
es
v
ar
i
o
u
s
DL
m
o
d
els,
th
ei
r
ar
ch
itectu
r
es,
an
d
t
r
ain
in
g
a
n
d
v
alid
atio
n
d
atasets
lik
e
Me
s
s
i
d
o
r
,
Kag
g
le,
an
d
DI
AR
E
T
DB
1
.
An
d
r
ab
an
d
Gu
p
ta
[
2
2
]
r
ev
iews
r
ec
en
t
d
ev
elo
p
m
en
ts
in
u
s
in
g
DL
tech
n
iq
u
es
f
o
r
au
to
m
ated
d
etec
tio
n
an
d
class
if
icatio
n
o
f
DR
u
s
in
g
r
etin
a
im
ag
es.
I
f
n
o
t
ad
d
r
ess
ed
,
DR
,
a
f
r
eq
u
e
n
t
r
esu
lt
o
f
d
iab
etes,
m
ay
lead
to
v
is
io
n
im
p
ai
r
m
en
t.
Fu
n
d
u
s
im
ag
es
ar
e
u
tili
ze
d
i
n
th
e
tim
e
-
co
n
s
u
m
i
n
g
an
d
m
is
tak
e
-
p
r
o
n
e
m
an
u
al
d
etec
tio
n
o
f
DR
.
A
m
o
r
e
ef
f
icien
t
alter
n
ativ
e
is
o
f
f
er
ed
b
y
au
to
m
ated
s
y
s
tem
s
th
at
em
p
l
o
y
DL
p
ar
ticu
lar
l
y
C
NNs.
T
h
e
r
esear
ch
in
cl
u
d
es
v
ar
io
u
s
s
tag
es
o
f
DL
,
tec
h
n
i
q
u
es
f
o
r
im
a
g
e
p
r
ep
r
o
ce
s
s
in
g
,
p
u
b
licly
av
ailab
le
r
etin
al
d
atasets
,
an
d
th
e
p
er
f
o
r
m
an
ce
m
etr
ics
u
s
ed
in
th
e
s
e
m
o
d
els.
I
t
co
m
p
ar
es
b
in
a
r
y
an
d
m
u
lti
-
class
class
if
icatio
n
m
eth
o
d
s
,
s
h
o
wca
s
es
ef
f
ec
tiv
en
ess
o
f
C
N
Ns
i
n
d
etec
tin
g
an
d
class
if
y
in
g
D
R
,
an
d
em
p
h
asizes
th
e
im
p
o
r
tan
ce
o
f
m
o
r
e
e
x
ten
s
iv
e,
h
ig
h
e
r
-
q
u
ality
d
atasets
to
en
h
an
ce
m
o
d
el
ac
cu
r
ac
y
a
n
d
d
ep
en
d
ab
ilit
y
.
Nag
asato
et
a
l.
[
2
3
]
e
x
p
l
o
r
e
s
th
e
u
s
e
o
f
D
L
a
n
d
SV
M
t
e
c
h
n
iq
u
e
s
t
o
d
e
t
e
c
t
n
o
n
p
e
r
f
u
s
i
o
n
a
r
e
a
s
(
N
PA
)
c
a
u
s
e
d
b
y
r
et
i
n
a
l
v
e
i
n
o
cc
l
u
s
i
o
n
(
R
VO
)
i
n
o
p
t
i
c
al
c
o
h
e
r
e
n
c
e
to
m
o
g
r
a
p
h
y
a
n
g
i
o
g
r
a
p
h
y
(
O
C
T
A
)
i
m
a
g
es
.
A
d
e
ep
C
N
N
a
n
d
a
S
V
M
m
o
d
e
l
w
e
r
e
d
e
v
e
l
o
p
e
d
a
n
d
e
v
a
l
u
a
t
e
d
u
s
i
n
g
a
d
a
t
a
s
e
t
c
o
n
t
a
i
n
i
n
g
3
2
2
O
C
T
A
i
m
a
g
e
s
,
o
f
w
h
i
c
h
1
7
4
s
h
o
w
e
d
N
P
A
r
e
s
u
l
ti
n
g
f
r
o
m
R
V
O
.
I
n
c
o
m
p
a
r
i
s
o
n
t
o
t
h
e
S
V
M
,
w
h
i
c
h
s
h
o
w
e
d
a
n
A
U
C
o
f
0
.
8
8
0
,
a
s
e
n
s
i
t
i
v
it
y
o
f
7
9
.
3
%
,
a
n
d
a
s
p
e
c
i
f
i
c
it
y
o
f
8
1
.
1
%
,
t
h
e
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
(
DNN
)
o
u
t
p
e
r
f
o
r
m
e
d
i
t
,
a
c
h
i
e
v
i
n
g
a
n
A
U
C
o
f
0
.
9
8
6
,
a
s
e
n
s
it
i
v
it
y
o
f
9
3
.
7
%
,
a
n
d
a
s
p
e
c
i
f
ic
i
t
y
o
f
9
7
.
3
%
.
T
h
e
DN
N
’
s
AU
C
an
d
s
p
e
c
i
f
i
ci
t
y
al
s
o
s
u
r
p
a
s
s
e
d
t
h
o
s
e
o
f
s
e
v
e
n
o
p
h
t
h
a
l
m
o
l
o
g
i
s
t
s
,
a
n
d
it
r
e
q
u
i
r
e
d
s
i
g
n
i
f
i
c
a
n
t
l
y
l
es
s
t
i
m
e
t
o
d
i
ag
n
o
s
e
p
a
t
i
e
n
t
s
.
T
h
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
S
tr
u
ctu
r
ed
d
a
ta
c
o
llectio
n
a
n
d
d
ee
p
lea
r
n
in
g
fo
r
r
etin
a
l O
C
T
ima
g
e
-
to
-
text
tr
a
n
s
la
tio
n
…
(
Ud
a
y
Ma
n
d
e
)
1055
s
t
u
d
y
’
s
r
es
u
l
ts
i
n
d
i
c
at
e
t
h
a
t
D
L
a
n
d
O
C
T
A
c
o
ll
e
c
ti
v
e
l
y
o
f
f
e
r
s
i
g
n
i
f
i
c
a
n
t
a
c
c
u
r
a
c
y
i
n
d
e
t
e
c
ti
n
g
N
P
As
,
p
o
t
e
n
t
i
al
l
y
i
m
p
r
o
v
i
n
g
c
l
i
n
i
c
al
p
r
a
c
ti
c
es
an
d
r
e
t
i
n
a
l
s
c
r
e
e
n
i
n
g
t
e
c
h
n
i
q
u
es
.
A
c
o
m
p
a
r
is
o
n
o
f
s
e
v
e
r
a
l
ML
a
n
d
DL
m
e
t
h
o
d
s
u
s
e
d
f
o
r
O
C
T
i
m
a
g
e
a
n
a
l
y
s
is
i
s
s
h
o
w
n
i
n
T
a
b
l
e
1
.
T
ab
le
1
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
o
n
r
etin
al
OC
T
im
ag
e
a
n
aly
s
is
P
a
p
e
r
M
e
t
h
o
d
s
u
se
d
D
a
t
a
s
e
t
s
P
e
r
f
o
r
ma
n
c
e
me
t
r
i
c
s
M
a
i
n
f
i
n
d
i
n
g
s
[
2
4
]
Tr
a
d
i
t
i
o
n
a
l
i
ma
g
e
p
r
o
c
e
ss
i
n
g
(
t
h
r
e
s
h
o
l
d
i
n
g
,
a
n
d
e
d
g
e
d
e
t
e
c
t
i
o
n
)
P
r
i
v
a
t
e
O
C
T
d
a
t
a
s
e
t
(
2
0
0
i
ma
g
e
s)
D
i
c
e
c
o
e
f
f
i
c
i
e
n
t
:
0
.
6
5
,
A
c
c
u
r
a
c
y
:
7
5
%
Tr
a
d
i
t
i
o
n
a
l
m
e
t
h
o
d
s
sh
o
w
m
o
d
e
r
a
t
e
p
e
r
f
o
r
m
a
n
c
e
;
s
e
n
s
i
t
i
v
e
t
o
n
o
i
s
e
a
n
d
v
a
r
i
a
b
i
l
i
t
y
i
n
i
m
a
g
e
q
u
a
l
i
t
y
.
[
2
5
]
S
V
M
,
r
a
n
d
o
m
f
o
r
e
st
,
a
n
d
K
N
N
D
u
k
e
O
C
T
d
a
t
a
se
t
(
5
0
0
i
ma
g
e
s)
A
c
c
u
r
a
c
y
:
8
8
%
,
S
e
n
s
i
t
i
v
i
t
y
:
8
5
%
,
S
p
e
c
i
f
i
c
i
t
y
:
9
0
%
M
L
mo
d
e
l
s
o
u
t
p
e
r
f
o
r
m
t
r
a
d
i
t
i
o
n
a
l
met
h
o
d
s
w
i
t
h
b
e
t
t
e
r
a
c
c
u
r
a
c
y
;
r
e
q
u
i
r
e
f
e
a
t
u
r
e
e
n
g
i
n
e
e
r
i
n
g
.
[
2
6
]
C
N
N
(
V
G
G
N
e
t
a
n
d
R
e
sN
e
t
)
O
C
T2
0
1
7
d
a
t
a
se
t
(
1
,
0
0
0
i
ma
g
e
s)
A
c
c
u
r
a
c
y
:
9
4
%
,
S
e
n
s
i
t
i
v
i
t
y
:
9
2
%
,
S
p
e
c
i
f
i
c
i
t
y
:
9
5
%
C
N
N
m
o
d
e
l
s
a
c
h
i
e
v
e
h
i
g
h
a
c
c
u
r
a
c
y
;
t
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
e
n
h
a
n
c
e
s
p
e
r
f
o
r
m
a
n
c
e
o
n
l
i
m
i
t
e
d
d
a
t
a
.
[
2
7
]
C
N
N
+
mu
l
t
i
m
o
d
a
l
d
a
t
a
f
u
si
o
n
M
i
x
e
d
d
a
t
a
s
e
t
s
(
O
C
T+
F
u
n
d
u
s
,
2
,
0
0
0
i
ma
g
e
s)
A
U
C
:
0
.
9
8
,
A
c
c
u
r
a
c
y
:
9
5
%
C
o
m
b
i
n
i
n
g
O
C
T
w
i
t
h
f
u
n
d
u
s
i
ma
g
e
s
i
mp
r
o
v
e
s
d
i
a
g
n
o
st
i
c
a
c
c
u
r
a
c
y
;
mu
l
t
i
m
o
d
a
l
l
e
a
r
n
i
n
g
i
s
e
f
f
e
c
t
i
v
e
.
[
2
8
]
U
-
N
e
t
a
n
d
F
C
N
R
ETO
U
C
H
d
a
t
a
set
(
3
0
0
i
ma
g
e
s)
D
i
c
e
c
o
e
f
f
i
c
i
e
n
t
:
0
.
9
2
,
S
e
n
s
i
t
i
v
i
t
y
:
9
0
%
U
-
N
e
t
a
r
c
h
i
t
e
c
t
u
r
e
p
r
o
v
i
d
e
s
st
a
t
e
-
of
-
t
h
e
-
a
r
t
se
g
m
e
n
t
a
t
i
o
n
p
e
r
f
o
r
ma
n
c
e
;
r
o
b
u
s
t
t
o
n
o
i
se
.
[
2
9
]
G
A
N
s (P
i
x
2
P
i
x
a
n
d
C
y
c
l
e
G
A
N
)
P
r
i
v
a
t
e
O
C
T
d
a
t
a
s
e
t
(
4
0
0
i
ma
g
e
s)
P
S
N
R
:
3
0
d
B
,
S
S
I
M
:
0
.
8
5
G
A
N
s
e
f
f
e
c
t
i
v
e
l
y
e
n
h
a
n
c
e
i
m
a
g
e
q
u
a
l
i
t
y
;
u
s
e
f
u
l
f
o
r
r
e
d
u
c
i
n
g
n
o
i
se
a
n
d
i
mp
r
o
v
i
n
g
i
m
a
g
e
c
l
a
r
i
t
y
.
[
3
0
]
Pre
-
t
r
a
i
n
e
d
C
N
N
s
(
I
n
c
e
p
t
i
o
n
V
3
a
n
d
R
e
sN
e
t
5
0
)
O
C
T2
0
1
7
d
a
t
a
se
t
(
1
,
2
0
0
i
ma
g
e
s)
A
c
c
u
r
a
c
y
:
9
6
%
,
F1
-
s
c
o
r
e
:
0
.
9
4
Tr
a
n
sf
e
r
l
e
a
r
n
i
n
g
i
m
p
r
o
v
e
s
p
e
r
f
o
r
m
a
n
c
e
o
n
sm
a
l
l
d
a
t
a
s
e
t
s
;
f
a
st
e
r
c
o
n
v
e
r
g
e
n
c
e
w
i
t
h
f
e
w
e
r
e
p
o
c
h
s
.
[
3
1
]
D
e
e
p
a
u
t
o
e
n
c
o
d
e
r
s
P
r
i
v
a
t
e
O
C
T
d
a
t
a
s
e
t
(
3
5
0
i
ma
g
e
s)
A
U
C
:
0
.
9
3
,
P
r
e
c
i
s
i
o
n
:
8
9
%
,
R
e
c
a
l
l
:
9
1
%
A
u
t
o
e
n
c
o
d
e
r
s
a
r
e
e
f
f
e
c
t
i
v
e
f
o
r
u
n
s
u
p
e
r
v
i
s
e
d
a
n
o
mal
y
d
e
t
e
c
t
i
o
n
;
u
s
e
f
u
l
i
n
i
d
e
n
t
i
f
y
i
n
g
r
a
r
e
p
a
t
h
o
l
o
g
i
e
s.
[
3
2
]
C
N
N
+
e
x
p
l
a
i
n
a
b
l
e
A
I
(
G
r
a
d
-
C
A
M
a
n
d
LI
M
E)
A
C
R
I
M
A
d
a
t
a
se
t
(
5
0
0
i
ma
g
e
s)
A
c
c
u
r
a
c
y
:
9
2
%
,
I
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
sc
o
r
e
:
H
i
g
h
Ex
p
l
a
i
n
a
b
l
e
A
I
t
e
c
h
n
i
q
u
e
s
e
n
h
a
n
c
e
mo
d
e
l
t
r
a
n
sp
a
r
e
n
c
y
;
u
sef
u
l
f
o
r
c
l
i
n
i
c
a
l
a
c
c
e
p
t
a
n
c
e
a
n
d
t
r
u
s
t
.
[
3
3
]
Li
g
h
t
w
e
i
g
h
t
C
N
N
(
M
o
b
i
l
e
N
e
t
a
n
d
S
q
u
e
e
z
e
N
e
t
)
P
r
i
v
a
t
e
O
C
T
d
a
t
a
s
e
t
(
1
5
0
i
ma
g
e
s)
A
c
c
u
r
a
c
y
:
8
9
%
,
I
n
f
e
r
e
n
c
e
t
i
me
:
5
0
ms
p
e
r
i
ma
g
e
Li
g
h
t
w
e
i
g
h
t
m
o
d
e
l
s
a
l
l
o
w
f
o
r
i
mm
e
d
i
a
t
e
a
n
a
l
y
si
s
;
i
d
e
a
l
f
o
r
p
o
i
n
t
-
of
-
c
a
r
e
se
t
t
i
n
g
s
4
.
2
.
M
a
chine
lea
rning
a
pp
li
ca
t
io
ns
in m
edica
l im
a
g
ing
Key
s
tu
d
ies an
d
ad
v
a
n
ce
m
en
t
s
in
u
s
in
g
ML
f
o
r
d
iag
n
o
s
tic
im
ag
in
g
:
i)
C
las
s
if
icatio
n
:
v
ar
io
u
s
r
esear
c
h
h
as
s
h
o
wn
t
h
e
ef
f
ec
tiv
e
n
ess
o
f
C
NNs
in
class
if
y
in
g
r
etin
al
OC
T
im
ag
es
in
to
th
r
ee
ca
teg
o
r
ies:
n
o
r
m
al
,
AM
D,
an
d
DR
.
Fo
r
attain
in
g
h
i
g
h
class
if
icatio
n
p
r
ec
is
io
n
f
o
r
r
etin
a
l
d
is
ea
s
es,
tr
an
s
f
er
lear
n
in
g
u
tili
zin
g
p
r
e
-
tr
ain
ed
m
o
d
els
s
u
ch
as
VGGN
et,
I
n
ce
p
tio
n
,
an
d
R
esNet
h
as
b
ee
n
co
m
m
o
n
ly
em
p
lo
y
ed
.
ii)
Seg
m
en
tatio
n
:
ass
ess
in
g
th
e
ad
v
an
ce
m
e
n
t
o
f
th
e
d
is
ea
s
e
n
ec
ess
itates
th
e
ac
cu
r
ate
d
if
f
er
en
tiatio
n
o
f
r
etin
al
lay
er
s
an
d
ir
r
eg
u
lar
ities
.
DL
m
o
d
els,
in
clu
d
in
g
U
-
Net
an
d
its
alter
n
ativ
es,
h
av
e
wid
ely
b
ee
n
em
p
lo
y
ed
f
o
r
r
etin
al
lay
er
s
eg
m
en
tatio
n
,
d
em
o
n
s
tr
atin
g
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
co
m
p
ar
ed
t
o
co
n
v
e
n
tio
n
a
l
tech
n
iq
u
es.
Netwo
r
k
s
f
ea
tu
r
i
n
g
m
u
lti
-
s
ca
le
an
d
m
u
lti
-
lev
el
f
ea
tu
r
e
f
u
s
io
n
h
av
e
b
ee
n
in
tr
o
d
u
ce
d
to
en
h
an
ce
s
eg
m
e
n
tatio
n
ac
cu
r
ac
y
.
iii)
An
o
m
aly
d
etec
tio
n
:
r
esear
c
h
h
as
b
ee
n
d
o
n
e
o
n
u
s
in
g
au
to
en
co
d
er
s
an
d
GANs
to
d
etec
t
an
o
m
alies
in
r
etin
al
OC
T
im
ag
es.
T
h
ese
m
o
d
els
ca
n
id
en
tify
ch
a
n
g
es
in
t
h
e
n
o
r
m
al
ap
p
ea
r
a
n
ce
o
f
th
e
r
etin
a
th
at
m
a
y
in
d
icate
d
is
ea
s
e.
An
o
m
aly
d
et
ec
tio
n
is
p
ar
ticu
lar
ly
u
s
ef
u
l
i
n
u
n
s
u
p
er
v
is
ed
o
r
s
em
i
-
s
u
p
er
v
i
s
ed
co
n
tex
ts
wh
er
e
lab
eled
d
ata
is
s
ca
r
ce
.
iv
)
Mu
ltimo
d
al
lear
n
in
g
:
in
a
n
ef
f
o
r
t
to
in
c
r
ea
s
e
d
iag
n
o
s
tic
ac
cu
r
ac
y
,
r
ec
e
n
t
s
tu
d
ies
h
av
e
lo
o
k
e
d
in
to
m
er
g
in
g
OC
T
p
ictu
r
es
with
ad
d
itio
n
al
m
o
d
alities
s
u
ch
f
u
n
d
u
s
p
h
o
t
o
g
r
a
p
h
y
an
d
p
atien
t
d
em
o
g
r
ap
h
ic
in
f
o
r
m
atio
n
.
T
h
r
o
u
g
h
th
e
u
s
e
o
f
co
m
p
lem
e
n
tar
y
in
f
o
r
m
ati
o
n
f
r
o
m
m
an
y
d
ata
s
o
u
r
ce
s
,
m
u
ltimo
d
al
DL
m
o
d
els ca
n
im
p
r
o
v
e
p
er
f
o
r
m
a
n
ce
in
task
s
in
v
o
lv
i
n
g
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
.
4
.
3
.
Co
m
pa
ra
t
iv
e
a
na
ly
s
is
o
f
t
ec
hn
iqu
es
B
ased
o
n
im
p
o
r
ta
n
t
ev
alu
atio
n
cr
iter
ia,
T
ab
le
2
p
r
esen
ts
a
c
o
m
p
ar
is
o
n
o
f
DL
,
ML
,
an
d
co
n
v
en
tio
n
al
im
ag
e
p
r
o
ce
s
s
in
g
m
eth
o
d
s
.
I
t
h
ig
h
lig
h
ts
th
e
tr
ad
e
-
o
f
f
s
b
etwe
en
ac
cu
r
ac
y
,
c
o
m
p
u
tati
o
n
al
r
eq
u
ir
e
m
en
ts
,
g
en
er
aliza
b
ilit
y
,
an
d
in
ter
p
r
e
tab
ilit
y
.
Ov
er
all,
wh
ile
DL
o
f
f
er
s
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
an
d
ad
ap
ta
b
ilit
y
,
it
d
em
an
d
s
h
ig
h
e
r
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
an
d
p
o
s
es
ch
allen
g
es
in
m
o
d
el
in
ter
p
r
eta
b
ilit
y
co
m
p
ar
ed
to
co
n
v
en
tio
n
al
tech
n
iq
u
es.
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
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
10
50
-
10
61
1056
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
o
f
th
e
d
i
f
f
er
en
t te
ch
n
iq
u
es u
s
ed
f
o
r
r
etin
al
OC
T
im
ag
e
an
aly
s
is
,
h
ig
h
l
ig
h
tin
g
th
eir
s
tr
en
g
th
s
an
d
lim
itatio
n
s
in
v
a
r
io
u
s
cr
iter
ia
C
r
i
t
e
r
i
a
Tr
a
d
i
t
i
o
n
a
l
i
ma
g
e
p
r
o
c
e
ss
i
n
g
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
D
e
e
p
l
e
a
r
n
i
n
g
A
c
c
u
r
a
c
y
a
n
d
p
e
r
f
o
r
m
a
n
c
e
M
o
d
e
r
a
t
e
a
c
c
u
r
a
c
y
;
p
e
r
f
o
r
m
a
n
c
e
v
a
r
i
e
s
w
i
t
h
i
ma
g
e
q
u
a
l
i
t
y
a
n
d
n
o
i
se.
H
i
g
h
e
r
a
c
c
u
r
a
c
y
t
h
a
n
t
r
a
d
i
t
i
o
n
a
l
met
h
o
d
s
;
d
e
p
e
n
d
s
o
n
f
e
a
t
u
r
e
e
n
g
i
n
e
e
r
i
n
g
a
n
d
q
u
a
l
i
t
y
o
f
f
e
a
t
u
r
e
s.
H
i
g
h
a
c
c
u
r
a
c
y
;
su
p
e
r
i
o
r
p
e
r
f
o
r
ma
n
c
e
d
u
e
t
o
a
u
t
o
ma
t
i
c
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
a
n
d
h
i
e
r
a
r
c
h
i
c
a
l
f
e
a
t
u
r
e
l
e
a
r
n
i
n
g
.
C
o
m
p
u
t
a
t
i
o
n
a
l
c
o
m
p
l
e
x
i
t
y
a
n
d
r
e
so
u
r
c
e
r
e
q
u
i
r
e
m
e
n
t
s
Lo
w
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
m
p
l
e
x
i
t
y
;
l
e
ss res
o
u
r
c
e
-
i
n
t
e
n
si
v
e
.
M
o
d
e
r
a
t
e
c
o
m
p
l
e
x
i
t
y
;
r
e
q
u
i
r
e
s
c
o
m
p
u
t
a
t
i
o
n
a
l
r
e
s
o
u
r
c
e
s
f
o
r
t
r
a
i
n
i
n
g
b
u
t
l
e
ss
t
h
a
n
DL
.
H
i
g
h
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
mp
l
e
x
i
t
y
;
r
e
q
u
i
r
e
s
si
g
n
i
f
i
c
a
n
t
c
o
mp
u
t
a
t
i
o
n
a
l
r
e
so
u
r
c
e
s
(
e
.
g
.
,
G
P
U
s)
a
n
d
l
a
r
g
e
d
a
t
a
se
t
s.
G
e
n
e
r
a
l
i
z
a
b
i
l
i
t
y
a
n
d
a
d
a
p
t
a
b
i
l
i
t
y
Li
mi
t
e
d
g
e
n
e
r
a
l
i
z
a
b
i
l
i
t
y
;
h
i
g
h
l
y
d
e
p
e
n
d
e
n
t
o
n
sp
e
c
i
f
i
c
i
ma
g
i
n
g
c
o
n
d
i
t
i
o
n
s
a
n
d
s
e
t
t
i
n
g
s
.
M
o
d
e
r
a
t
e
l
y
a
d
a
p
t
a
b
l
e
;
g
e
n
e
r
a
l
i
z
e
s
b
e
t
t
e
r
t
h
a
n
t
r
a
d
i
t
i
o
n
a
l
me
t
h
o
d
s
b
u
t
st
i
l
l
r
e
q
u
i
r
e
s fe
a
t
u
r
e
t
u
n
i
n
g
.
H
i
g
h
a
d
a
p
t
a
b
i
l
i
t
y
a
n
d
g
e
n
e
r
a
l
i
z
a
t
i
o
n
a
c
r
o
ss
d
i
f
f
e
r
e
n
t
d
a
t
a
se
t
s;
t
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
c
a
n
i
m
p
r
o
v
e
a
d
a
p
t
a
b
i
l
i
t
y
.
I
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
H
i
g
h
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
;
r
e
s
u
l
t
s
a
r
e
e
a
si
l
y
e
x
p
l
a
i
n
a
b
l
e
.
M
o
d
e
r
a
t
e
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
;
f
e
a
t
u
r
e
-
b
a
s
e
d
d
e
c
i
si
o
n
s
c
a
n
b
e
t
r
a
c
e
d
,
b
u
t
mo
d
e
l
s
a
r
e
l
e
ss
t
r
a
n
sp
a
r
e
n
t
t
h
a
n
t
r
a
d
i
t
i
o
n
a
l
m
e
t
h
o
d
s
.
Lo
w
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
;
c
o
n
s
i
d
e
r
e
d
"
b
l
a
c
k
-
b
o
x
"
m
o
d
e
l
s
,
t
h
o
u
g
h
e
x
p
l
a
i
n
a
b
i
l
i
t
y
t
e
c
h
n
i
q
u
e
s
(
e
.
g
.
,
a
t
t
e
n
t
i
o
n
me
c
h
a
n
i
s
ms)
a
r
e
e
m
e
r
g
i
n
g
.
4
.
4
.
G
a
ps
in
curr
ent
re
s
ea
rc
h
I
d
en
tify
in
g
th
e
lim
itatio
n
s
an
d
g
ap
s
in
th
e
e
x
is
tin
g
liter
atu
r
e
th
at
th
is
s
u
r
v
ey
aim
s
to
ad
d
r
es
s
:
i)
L
im
ited
g
en
er
aliza
tio
n
ac
r
o
s
s
d
iv
er
s
e
p
o
p
u
latio
n
s
:
m
an
y
p
r
io
r
s
tu
d
ies
h
av
e
b
ee
n
c
o
n
d
u
cted
o
n
v
er
y
h
o
m
o
g
en
eo
u
s
d
atasets
,
wh
ich
lim
its
th
eir
ap
p
licab
ilit
y
to
a
b
r
o
ad
s
p
ec
tr
u
m
o
f
p
atien
t
p
o
p
u
latio
n
s
.
T
o
cr
ea
te
m
o
d
els
th
at
ar
e
m
o
r
e
r
eliab
le
an
d
b
r
o
ad
ly
a
p
p
licab
le
,
f
u
r
th
er
s
tu
d
y
o
n
r
ep
r
esen
tati
v
e
an
d
d
iv
er
s
e
d
atasets
is
r
eq
u
ir
ed
.
ii)
Data
s
ca
r
city
an
d
im
b
alan
ce
:
lack
an
d
u
n
b
ala
n
ce
o
f
lab
el
ed
d
ata
is
a
m
ajo
r
o
b
s
tacle
in
OC
T
im
ag
e
an
aly
s
is
,
esp
ec
ially
f
o
r
u
n
co
m
m
o
n
d
is
o
r
d
e
r
s
.
R
esear
ch
o
n
m
eth
o
d
s
s
u
ch
as
s
em
i
-
s
u
p
er
v
is
ed
an
d
u
n
s
u
p
er
v
is
ed
lea
r
n
in
g
th
at
tac
k
le
d
ata
s
ca
r
city
is
s
till
cr
u
cial.
iii)
I
n
teg
r
atio
n
o
f
m
u
ltimo
d
al
d
a
ta:
m
o
r
e
r
esear
ch
is
r
eq
u
ir
ed
to
s
u
cc
ess
f
u
lly
in
teg
r
ate
d
at
a
f
r
o
m
m
an
y
s
o
u
r
ce
s
(
e.
g
.
,
f
u
n
d
u
s
p
h
o
t
o
g
r
ap
h
y
,
clin
ical
r
ec
o
r
d
s
,
an
d
OC
T
)
in
o
r
d
er
to
en
h
an
ce
d
is
e
ase
m
o
n
ito
r
in
g
an
d
d
iag
n
o
s
tic
ac
cu
r
ac
y
,
ev
en
if
m
u
ltimo
d
al
lear
n
in
g
h
as sh
o
wed
p
r
o
m
is
e.
iv
)
R
ea
l
-
tim
e
an
d
p
o
in
t
-
of
-
ca
r
e
ap
p
licatio
n
s
:
ex
is
tin
g
m
o
d
els
m
ay
n
o
t
b
e
ap
p
r
o
p
r
iate
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
o
r
u
s
e
in
en
v
ir
o
n
m
en
ts
with
r
estricte
d
r
eso
u
r
ce
s
s
in
ce
th
ey
f
r
e
q
u
en
tly
d
em
an
d
lar
g
e
am
o
u
n
ts
o
f
c
o
m
p
u
te
r
p
o
wer
.
I
t
is
n
ec
ess
ar
y
to
co
n
d
u
ct
r
esea
r
ch
o
n
e
f
f
ec
tiv
e
alg
o
r
ith
m
s
an
d
lig
h
tweig
h
t
m
o
d
els f
o
r
u
s
e
in
p
o
i
n
t
-
of
-
ca
r
e
an
d
r
ea
l
-
tim
e
a
p
p
licatio
n
s
.
5.
P
RO
P
O
SE
D
M
E
T
H
O
DO
L
O
G
Y
First,
we
g
ath
er
th
e
im
ag
es
tak
en
b
y
OC
T
d
ev
ices
to
m
ak
e
o
u
r
p
r
o
m
in
e
n
t
d
ata
s
et
f
o
r
f
u
r
th
e
r
p
r
o
ce
s
s
in
g
.
Af
ter
th
at
we
wi
ll
p
r
o
ce
s
s
th
e
im
ag
es
c
o
llected
f
o
r
n
o
is
e
r
em
o
v
al
b
y
ap
p
ly
in
g
eith
e
r
o
v
er
s
am
p
lin
g
o
r
u
n
d
er
s
am
p
lin
g
m
eth
o
d
s
an
d
will
ex
t
r
ac
t
s
o
m
e
f
ea
t
u
r
es.
W
e
ca
n
al
s
o
p
er
f
o
r
m
im
ag
e
s
eg
m
en
tatio
n
to
an
aly
ze
th
e
i
m
ag
e
f
o
r
s
elec
tin
g
it
f
o
r
clas
s
if
icatio
n
.
No
r
m
aliza
tio
n
is
t
h
e
n
e
x
t
s
tep
to
b
e
p
er
f
o
r
m
ed
to
c
h
ec
k
th
e
r
ea
d
in
ess
o
f
o
u
r
d
ataset.
Nex
t
s
tep
s
ar
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
an
d
f
ea
t
u
r
e
s
elec
tio
n
f
r
o
m
th
e
d
ataset
with
th
e
s
u
itab
le
al
g
o
r
ith
m
s
.
L
astl
y
,
we
ap
p
ly
a
n
ac
cu
r
ate
an
d
ef
f
icien
t
m
eth
o
d
o
f
class
if
icatio
n
to
p
r
ed
ict
th
e
d
ia
g
n
o
s
is
.
T
h
e
f
r
am
ewo
r
k
m
ak
es
u
s
e
o
f
OC
T
r
etin
al
im
ag
es
th
at
h
av
e
b
ee
n
en
h
an
ce
d
f
o
r
c
o
n
tr
ast,
n
o
is
e
r
em
o
v
al,
co
n
to
u
r
-
b
ased
ed
g
e
r
ec
o
g
n
itio
n
,
an
d
r
eti
n
al
lay
er
ex
tr
ac
tio
n
.
Dif
f
er
e
n
t
p
u
b
licl
y
ac
ce
s
s
ib
le
f
u
n
d
u
s
im
ag
e
co
llectio
n
s
ar
e
f
r
eq
u
e
n
tly
u
s
ed
.
Pictu
r
e
p
r
ep
r
o
ce
s
s
in
g
is
a
cr
u
cial
s
tep
to
elim
in
ate
im
ag
e
n
o
is
e,
im
p
r
o
v
e
im
ag
e
c
h
ar
ac
ter
is
tics
,
an
d
g
u
a
r
an
tee
im
a
g
e
co
n
s
is
t
en
cy
.
T
h
e
m
e
d
ian
f
ilter
,
Gau
s
s
ian
f
ilter
,
an
d
n
o
n
-
lo
ca
l
m
ea
n
s
d
e
n
o
is
in
g
m
eth
o
d
s
ar
e
th
e
m
o
s
t
o
f
ten
u
s
ed
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es
in
co
n
tem
p
o
r
ar
y
r
esear
ch
n
o
is
e
r
ed
u
ctio
n
s
tr
ateg
ies.
W
h
en
ce
r
tain
im
ag
e
class
es
wer
e
u
n
b
alan
ce
d
o
r
th
e
d
ataset
n
ee
d
ed
to
b
e
lar
g
e
r
,
d
ata
au
g
m
e
n
tatio
n
tech
n
i
q
u
es
wer
e
u
s
ed
.
T
r
an
s
latio
n
,
r
o
tatio
n
,
s
h
ea
r
i
n
g
,
f
lip
p
in
g
,
co
n
tr
ast
s
ca
lin
g
,
an
d
r
esizin
g
ar
e
ex
am
p
l
es
o
f
d
ata
au
g
m
en
tatio
n
tech
n
i
q
u
es.
T
h
e
f
ea
tu
r
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
m
ad
e
ad
v
an
tag
e
o
f
th
e
clev
e
r
ed
g
e
m
eth
o
d
.
T
h
e
p
h
o
to
s
ar
e
p
r
ep
a
r
ed
to
b
e
u
s
ed
as
in
p
u
t
f
o
r
t
h
e
DL
af
ter
p
r
ep
r
o
ce
s
s
in
g
.
T
h
e
s
ca
le
o
f
th
e
d
atasets
r
eq
u
ir
ed
to
tr
ain
th
e
DL
s
y
s
tem
s
,
as
DL
d
em
an
d
s
a
v
ast
q
u
a
n
tity
o
f
d
at
a,
is
o
n
e
o
f
th
e
ch
allen
g
es
th
a
t
th
e
ap
p
licatio
n
o
f
DL
in
th
e
m
ed
ical
p
r
o
f
ess
io
n
f
ac
es.
T
h
e
am
o
u
n
t,
q
u
ality
,
an
d
class
b
alan
ce
o
f
th
e
tr
ain
in
g
d
ata
all
h
av
e
a
s
ig
n
if
ican
t
im
p
ac
t
o
n
h
o
w
we
ll
DL
s
y
s
tem
s
p
er
f
o
r
m
.
Fig
u
r
e
2
s
h
o
ws
th
e
ar
ch
itectu
r
e
f
o
r
r
etin
al
OC
T
im
a
g
e
to
tex
t c
o
n
v
er
s
io
n
u
s
in
g
DL
.
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
S
tr
u
ctu
r
ed
d
a
ta
c
o
llectio
n
a
n
d
d
ee
p
lea
r
n
in
g
fo
r
r
etin
a
l O
C
T
ima
g
e
-
to
-
text
tr
a
n
s
la
tio
n
…
(
Ud
a
y
Ma
n
d
e
)
1057
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
f
o
r
r
eti
n
al
OC
T
im
ag
e
to
tex
t c
o
n
v
er
s
io
n
u
s
in
g
DL
6.
DIS
CU
SS
I
O
N
S O
N
DAT
AS
E
T
OC
T
d
ataset:
OC
T
d
ataset
u
s
ed
f
o
r
an
ea
r
lier
s
tu
d
y
to
Ka
g
g
le
[
1
1
]
.
T
h
at
d
ataset,
wh
ic
h
in
clu
d
e
d
r
etin
al
OC
T
im
ag
es,
was
d
o
wn
lo
a
d
ed
f
r
o
m
th
e
Kag
g
le
web
s
ite
at
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/p
au
ltimo
th
y
m
o
o
n
ey
/k
er
m
a
n
y
2
0
1
8
(
ac
ce
s
s
ed
o
n
Oct.
5
,
2
0
2
3
)
.
T
h
is
p
u
b
lis
h
ed
d
ataset
in
clu
d
es
8
4
,
4
8
4
im
ag
es:
8
3
,
4
8
4
f
r
o
m
th
e
tr
ain
in
g
d
ataset
an
d
1
,
0
0
0
f
r
o
m
a
test
d
ataset.
T
h
e
d
ataset
in
clu
d
ed
f
ewe
r
OC
T
i
m
ag
es th
an
th
e
d
ataset
u
s
ed
f
o
r
th
e
ea
r
lier
s
tu
d
y
.
T
h
e
tr
ain
in
g
d
ataset
co
m
p
r
is
ed
3
7
,
2
0
5
im
ag
es
s
h
o
win
g
ch
o
r
o
id
al
n
eo
v
ascu
la
r
izatio
n
(
C
NV)
,
1
1
,
3
4
8
s
h
o
win
g
d
ia
b
etic
m
ac
u
lar
ed
em
a
(
DM
E
)
,
8
,
6
1
6
s
h
o
win
g
d
r
u
s
en
,
an
d
2
6
,
3
1
5
n
o
r
m
al
im
ag
es.
T
h
e
test
d
ataset
co
m
p
r
is
ed
2
5
0
im
ag
es
f
r
o
m
ea
c
h
class
.
W
e
d
iv
id
ed
th
e
tr
ain
in
g
d
ataset
in
to
a
s
u
b
-
tr
ain
in
g
d
ataset
an
d
a
v
alid
atio
n
d
ataset,
wh
ich
i
n
clu
d
ed
4
,
0
0
0
im
a
g
es
ex
tr
ac
ted
r
an
d
o
m
ly
f
r
o
m
1
,
0
0
0
im
ag
es
o
f
e
ac
h
class
.
T
h
e
s
u
b
-
tr
ain
in
g
d
ataset
in
clu
d
es
th
e
r
em
ain
in
g
tr
ain
i
n
g
d
ata.
T
h
e
im
ag
e
f
o
r
m
at
f
o
r
t
h
e
OC
T
d
ataset
is
jo
in
t
p
h
o
to
g
r
ap
h
ic
e
x
p
er
ts
’
g
r
o
u
p
8
-
b
it.
Fig
u
r
e
3
p
o
r
tr
ay
s
s
o
m
e
class
if
ied
OC
T
d
ataset
im
ag
es.
Fig
u
r
e
3
.
OC
T
im
ag
es in
th
e
OC
T
d
ataset
T
ab
le
3
.
Diab
etic
p
e
o
p
le
an
d
o
p
h
th
alm
o
lo
g
is
t r
atio
Y
e
a
r
N
o
o
f
p
a
t
i
e
n
t
w
i
t
h
d
i
a
b
e
t
e
s
N
o
o
f
o
p
h
t
h
a
l
mo
l
o
g
i
st
A
v
a
i
l
a
b
l
e
i
n
I
n
d
i
a
R
a
t
i
o
p
e
r
o
p
h
t
h
a
l
m
o
l
o
g
i
s
t
2
0
2
1
1
0
1
m
i
l
l
i
o
n
(
1
0
.
1
c
r
o
r
e
)
25
,
000
2
5
/
1
0
1
,
0
0
0
t
h
a
t
i
s
4
,
0
4
0
d
i
a
b
e
t
i
c
s
h
a
v
e
1
o
p
h
t
h
a
l
m
o
l
o
g
i
st
2
0
2
5
1
5
8
m
i
l
l
i
o
n
(
1
5
.
8
c
r
o
r
e
)
30
,
000
3
0
/
1
5
8
,
0
0
0
t
h
a
t
i
s
5
,
2
6
6
d
i
a
b
e
t
i
c
s
h
a
v
e
1
o
p
h
t
h
a
l
m
o
l
o
g
i
st
2
0
3
0
2
1
6
m
i
l
l
i
o
n
(
2
1
.
6
c
r
o
r
e
)
35
,
000
3
5
/
2
1
6
,
0
0
0
t
h
a
t
i
s
6
,
1
7
1
d
i
a
b
e
t
i
c
s
h
a
v
e
1
o
p
h
t
h
a
l
m
o
l
o
g
i
st
T
ab
le
3
illu
s
tr
ates
th
e
g
r
o
win
g
d
is
p
ar
ity
b
etwe
en
th
e
r
is
in
g
n
u
m
b
e
r
o
f
d
iab
etic
p
atien
ts
an
d
th
e
lim
ited
av
ailab
ilit
y
o
f
o
p
h
th
al
m
o
lo
g
is
ts
in
I
n
d
ia.
As d
iab
ete
s
p
r
ev
alen
ce
in
cr
ea
s
es sh
ar
p
ly
f
r
o
m
2
0
2
1
to
2
0
3
0
,
th
e
p
atien
t
-
to
-
o
p
h
th
alm
o
l
o
g
is
t
r
atio
wo
r
s
en
s
s
ig
n
if
ican
tl
y
,
h
ig
h
lig
h
tin
g
th
e
u
r
g
en
t
n
ee
d
f
o
r
s
ca
lab
le
,
AI
-
ass
is
ted
r
etin
al
s
cr
ee
n
in
g
an
d
d
iag
n
o
s
tic
s
o
lu
tio
n
s
.
C
o
n
s
id
er
in
g
t
h
e
ab
o
v
e
s
itu
atio
n
i.e
.
,
i
n
2
0
2
5
o
n
e
o
p
h
th
alm
o
lo
g
is
t
will
tr
ea
t
5
,
2
6
6
d
ia
b
etic
p
eo
p
le
is
a
p
o
in
t
o
f
wo
r
r
y
.
I
f
we
ig
n
o
r
e
t
h
e
f
a
cts,
it
will
f
u
r
th
er
d
eter
io
r
ate
o
f
6
,
1
7
1
.
So
,
we
co
m
e
to
co
n
cl
u
s
io
n
th
at
we
m
u
s
t
s
av
e
th
e
p
r
ec
io
u
s
tim
e
o
f
th
e
o
p
h
th
alm
o
lo
g
is
t.
No
r
m
al
r
etin
a:
n
o
r
m
al
r
etin
a
,
with
p
r
eser
v
ed
f
o
v
ea
l
co
n
t
o
u
r
an
d
a
b
s
en
ce
o
f
an
y
r
etin
al
f
lu
id
/
e
d
em
a.
T
h
e
f
o
llo
win
g
ar
e
th
e
p
ar
am
et
er
s
o
f
a
h
ea
lth
y
r
etin
a:
i)
No
r
m
al
r
etin
a:
in
d
ica
tes
th
at
th
e
r
etin
a
ap
p
ea
r
s
to
b
e
in
a
ty
p
ical,
h
ea
lth
y
s
tate
with
o
u
t
an
y
v
is
ib
le
ab
n
o
r
m
alities
in
im
ag
e
its
elf
.
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
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
10
50
-
10
61
1058
ii)
Pre
s
er
v
ed
f
o
v
ea
l
co
n
to
u
r
:
th
e
f
o
v
ea
,
th
e
p
a
r
t
o
f
th
e
r
etin
a
r
esp
o
n
s
ib
le
f
o
r
s
h
ar
p
ce
n
tr
al
v
is
io
n
,
an
d
m
ain
tain
s
its
n
o
r
m
al
s
h
ap
e
an
d
s
tr
u
ctu
r
e.
iii)
Ab
s
en
ce
o
f
r
etin
al
f
lu
id
/e
d
em
a:
th
er
e
is
n
o
f
lu
id
b
u
ild
u
p
o
r
s
wellin
g
in
th
e
r
etin
a,
wh
ich
is
a
p
o
s
itiv
e
s
ig
n
th
at
th
er
e
ar
e
n
o
u
n
d
e
r
ly
i
n
g
is
s
u
es su
ch
as r
etin
al
d
etac
h
m
en
t o
r
m
ac
u
lar
e
d
em
a.
iv
)
DM
E
:
DM
E
i
s
a
co
m
p
licatio
n
o
f
d
iab
etes
th
at
af
f
ec
ts
th
e
r
etin
a,
s
p
ec
if
ically
th
e
m
ac
u
la,
wh
ich
is
th
e
ce
n
tr
al
p
ar
t
o
f
th
e
r
etin
a
r
esp
o
n
s
ib
le
f
o
r
s
h
ar
p
,
d
etailed
v
is
io
n
.
I
n
DM
E
,
th
er
e
is
a
b
u
ild
u
p
o
f
f
lu
id
in
t
h
e
m
ac
u
la
d
u
e
to
lea
k
ag
e
f
r
o
m
d
am
ag
ed
b
lo
o
d
v
ess
els.
T
h
is
f
l
u
id
ac
cu
m
u
latio
n
ca
n
ca
u
s
e
a
d
v
er
s
e
r
etin
al
th
ick
en
in
g
a
n
d
ca
n
lead
t
o
v
is
io
n
im
p
air
m
e
n
t.
W
h
en
y
o
u
m
e
n
tio
n
“
r
etin
al
-
th
ick
en
in
g
-
ass
o
ciate
d
in
tr
ar
etin
al
f
lu
id
”
it
in
d
icate
s
th
at
in
th
e
im
ag
in
g
,
if
we
p
o
in
tin
g
to
ar
ea
s
wh
er
e
th
er
e
is
b
o
th
th
ick
en
in
g
o
f
th
e
r
etin
al
tis
s
u
e
an
d
th
e
p
r
esen
ce
o
f
f
lu
id
with
in
th
e
r
etin
a.
T
h
e
o
cc
u
r
r
en
ce
o
f
th
is
f
lu
id
ac
cu
m
u
latio
n
,
co
m
m
o
n
ly
k
n
o
wn
as
in
tr
ar
etin
al
f
lu
id
,
is
e
s
s
en
tial
f
o
r
d
iag
n
o
s
in
g
an
d
m
o
n
ito
r
in
g
DM
E
.
T
o
ef
f
ec
tiv
ely
h
a
n
d
le
DM
E
,
it
is
ess
en
tial
to
r
eg
u
late
b
lo
o
d
s
u
g
ar
lev
els.
Ad
d
itio
n
al
tr
ea
tm
en
ts
th
at
co
u
ld
b
e
u
tili
ze
d
co
m
p
r
is
e
co
r
tico
s
ter
o
id
s
,
an
ti
-
v
ascu
lar
en
d
o
th
elial
g
r
o
wth
f
ac
to
r
(
an
ti
-
VE
GF
)
in
jectio
n
s
,
o
r
laser
th
er
a
p
y
to
en
h
an
ce
v
is
io
n
an
d
r
ed
u
ce
f
lu
i
d
leak
ag
e.
v)
Dr
u
s
en
:
ea
r
ly
AM
D
is
o
f
ten
a
s
s
o
ciate
d
with
d
r
u
s
en
,
wh
ic
h
ar
e
s
m
all,
y
ello
wis
h
s
p
o
ts
th
at
f
o
r
m
b
en
ea
t
h
th
e
r
etin
a.
Dr
u
s
en
,
v
is
ib
le
as
ar
r
o
wh
ea
d
s
o
n
s
ca
n
s
,
s
er
v
e
as
an
in
itial
s
ig
n
o
f
AM
D
an
d
m
ay
b
e
r
elate
d
to
ch
an
g
es
in
v
is
u
al
q
u
ality
.
Dr
u
s
en
ar
e
co
m
m
o
n
ly
r
e
g
ar
d
ed
as
th
e
in
itial
s
tag
e
o
f
A
MD
,
an
d
th
eir
p
r
esen
ce
ca
n
h
elp
in
d
iag
n
o
s
in
g
an
d
m
o
n
ito
r
in
g
th
e
p
r
o
g
r
e
s
s
io
n
o
f
th
e
d
is
ea
s
e.
Dr
u
s
en
m
ay
n
o
t
h
av
e
a
m
ajo
r
im
p
ac
t
o
n
v
is
io
n
,
b
u
t
p
eo
p
le
m
ay
e
x
p
er
ien
ce
litt
le
ch
an
g
es
in
th
eir
v
is
io
n
,
s
u
c
h
as
tr
o
u
b
le
r
ea
d
in
g
o
r
s
ee
in
g
tin
y
d
etails.
v
i)
C
NV:
in
th
e
co
n
tex
t
o
f
AM
D
,
C
NV
r
ef
er
s
to
ch
o
r
o
id
al
n
e
o
v
ascu
lar
izatio
n
.
T
h
is
co
n
d
itio
n
en
tails
th
e
f
o
r
m
atio
n
o
f
n
ew,
ir
r
e
g
u
lar
b
lo
o
d
v
ess
els
in
th
e
ch
o
r
o
id
l
ay
er
lo
ca
ted
b
e
n
ea
th
th
e
r
eti
n
a.
T
h
e
n
ewly
f
o
r
m
ed
b
lo
o
d
v
ess
els
ar
e
k
n
o
wn
as
th
e
n
eo
v
ascu
lar
m
em
b
r
an
e
(
in
d
icate
d
b
y
ar
r
o
wh
ea
d
s
)
,
wh
ich
ca
n
lead
to
leak
s
an
d
b
leed
in
g
.
T
h
e
f
lu
id
ac
cu
m
u
latin
g
b
etwe
en
th
e
r
etin
a
an
d
th
e
u
n
d
er
ly
i
n
g
c
h
o
r
o
id
d
u
e
to
leak
ag
e
f
r
o
m
t
h
ese
ab
n
o
r
m
al
ca
p
illar
ies
is
r
ef
er
r
ed
to
as
s
u
b
r
etin
al
f
lu
id
(
in
d
icate
d
b
y
ar
r
o
ws).
Su
d
d
e
n
v
is
io
n
d
eter
io
r
atio
n
an
d
d
is
to
r
tio
n
ca
n
r
esu
lt
f
r
o
m
th
e
s
wellin
g
o
f
th
is
f
lu
id
an
d
d
am
a
g
e
to
th
e
r
etin
a.
T
o
m
an
ag
e
s
y
m
p
to
m
s
an
d
p
r
e
v
e
n
t
f
u
r
th
er
v
is
io
n
lo
s
s
,
tr
ea
tm
en
t
f
o
r
C
NV,
ty
p
ically
as
s
o
ciate
d
with
wet
AM
D,
m
u
s
t
b
e
in
itiated
p
r
o
m
p
tly
.
An
ti
-
VE
GF
in
jectio
n
s
ar
e
co
m
m
o
n
ly
u
tili
ze
d
in
th
er
a
p
ies
to
tar
g
e
t
an
d
h
alt
th
e
p
r
o
g
r
ess
io
n
o
f
t
h
ese
ab
n
o
r
m
al
b
lo
o
d
v
ess
els.
R
o
u
tin
e
im
ag
in
g
m
o
n
ito
r
in
g
is
cr
u
cial
to
ass
es
s
tr
ea
tm
en
t e
f
f
ec
tiv
en
ess
an
d
im
p
lem
e
n
t n
ec
ess
ar
y
m
o
d
if
icatio
n
s
.
Ov
er
all,
th
ese
o
b
s
er
v
atio
n
s
s
u
g
g
est
th
at
th
e
r
etin
a
is
in
g
o
o
d
co
n
d
itio
n
.
I
f
th
is
is
p
ar
t
o
f
a
r
ep
o
r
t
o
r
ass
es
s
m
en
t,
it
g
en
er
all
y
im
p
lies
th
at
th
er
e
a
r
e
n
o
s
ig
n
s
o
f
r
etin
al
d
is
ea
s
e
o
r
d
am
a
g
e
at
th
e
tim
e
o
f
ex
am
in
atio
n
.
Fig
u
r
e
4
s
h
o
ws th
e
s
am
p
les o
f
n
o
r
m
al,
Fig
u
r
e
5
s
h
o
ws th
e
s
am
p
les o
f
DM
E
,
Fig
u
r
e
6
s
h
o
ws th
e
s
am
p
les o
f
d
r
u
s
en
,
an
d
last
ly
Fig
u
r
e
7
s
h
o
ws th
e
s
am
p
les o
f
C
NV.
Fig
u
r
e
4
.
Sam
p
le
o
f
n
o
r
m
al
Fig
u
r
e
5
.
Sam
p
le
o
f
DM
E
Fig
u
r
e
6
.
Sam
p
le
o
f
d
r
u
s
en
Fig
u
r
e
7
.
Sam
p
le
o
f
C
NV
7.
CO
NCLU
SI
O
N
I
n
r
esp
o
n
s
e
to
th
e
in
cr
ea
s
in
g
n
ee
d
f
o
r
a
u
to
m
ated
to
o
ls
in
o
p
h
th
alm
o
lo
g
y
,
th
is
r
esear
ch
p
r
esen
ts
a
s
o
lid
f
o
u
n
d
atio
n
f
o
r
tr
a
n
s
latin
g
r
etin
al
OC
T
im
ag
es
in
to
d
iag
n
o
s
tic
tex
t
b
y
c
o
m
b
in
i
n
g
s
tr
u
ctu
r
ed
d
ata
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
S
tr
u
ctu
r
ed
d
a
ta
c
o
llectio
n
a
n
d
d
ee
p
lea
r
n
in
g
fo
r
r
etin
a
l O
C
T
ima
g
e
-
to
-
text
tr
a
n
s
la
tio
n
…
(
Ud
a
y
Ma
n
d
e
)
1059
g
ath
er
in
g
with
DL
alg
o
r
ith
m
s
.
T
h
e
f
r
am
ewo
r
k
g
u
a
r
an
tees
th
at
h
ig
h
-
q
u
ality
OC
T
d
ata
is
r
ea
d
y
f
o
r
m
o
d
e
l
tr
ain
in
g
b
y
e
m
p
lo
y
i
n
g
s
o
p
h
is
ticated
im
ag
e
p
r
o
ce
s
s
in
g
m
et
h
o
d
s
lik
e
co
n
tr
ast
e
n
h
an
ce
m
e
n
t,
n
o
is
e
r
ed
u
ctio
n
,
an
d
ed
g
e
d
etec
tio
n
.
T
h
e
p
r
o
j
ec
t
ef
f
ec
tiv
ely
tack
les
class
im
b
alan
ce
s
with
in
th
e
d
ataset
o
f
8
4
,
4
8
4
r
etin
al
im
ag
es
th
r
o
u
g
h
th
e
ap
p
licati
o
n
o
f
p
r
ep
r
o
ce
s
s
in
g
an
d
d
at
a
au
g
m
en
tatio
n
tech
n
iq
u
es.
T
h
is
en
h
an
ce
s
th
e
m
o
d
el
’
s
ab
ilit
y
to
class
if
y
r
etin
al
d
is
o
r
d
er
s
lik
e
DM
E
,
d
r
u
s
en
,
an
d
C
NV
.
T
h
e
in
teg
r
atio
n
o
f
DL
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
an
d
s
elec
tio
n
alg
o
r
ith
m
s
au
to
m
ates
d
iag
n
o
s
is
wh
ile
en
h
an
cin
g
s
ca
lab
ilit
y
an
d
ef
f
icien
c
y
in
clin
ical
en
v
ir
o
n
m
en
ts
.
T
h
is
m
eth
o
d
m
ee
ts
th
e
ess
en
tial
n
ee
d
f
o
r
p
r
ec
is
e
a
n
d
p
r
o
m
p
t
i
d
en
tific
atio
n
o
f
r
etin
al
d
is
o
r
d
er
s
d
u
e
to
th
e
r
is
in
g
i
n
cid
en
ce
o
f
ailm
en
ts
lik
e
DR
.
T
h
e
f
r
am
ewo
r
k
ca
n
s
ig
n
if
ican
tly
allev
iate
th
e
wo
r
k
lo
ad
o
f
o
p
h
th
alm
o
l
o
g
is
ts
,
en
h
an
ce
clin
ical
d
ec
is
io
n
-
m
ak
in
g
,
an
d
b
o
o
s
t
p
atien
t
o
u
tc
o
m
es
b
y
f
ac
ilit
atin
g
ea
r
ly
tr
ea
tm
en
t
an
d
in
ter
v
e
n
tio
n
th
r
o
u
g
h
o
p
tim
izin
g
th
e
d
ia
g
n
o
s
tic
p
r
o
ce
s
s
.
Fu
tu
r
e
r
esear
ch
s
h
o
u
ld
f
o
c
u
s
o
n
ex
p
an
d
i
n
g
t
h
e
d
ataset
v
a
r
iety
an
d
ex
p
l
o
r
in
g
m
o
r
e
ad
v
an
ce
d
DL
ar
c
h
itectu
r
es
to
im
p
r
o
v
e
th
e
ac
c
u
r
ac
y
an
d
r
elev
an
ce
o
f
th
e
s
y
s
tem
ac
r
o
s
s
v
ar
io
u
s
clin
ical
s
itu
atio
n
s
.
ACK
NO
WL
E
DG
E
M
E
NT
W
e
wo
u
ld
lik
e
to
th
an
k
to
Or
ac
le
r
esear
ch
f
o
r
s
u
p
p
o
r
tin
g
th
is
p
r
o
ject
g
iv
in
g
f
in
a
n
cial
ass
is
tan
t
o
f
R
s
.
2
5
0
,
0
0
0
th
o
u
s
an
d
i
n
th
e
f
o
r
m
o
f
cl
o
u
d
s
er
v
ices.
B
8
9
0
0
9
an
d
B
8
8
2
0
6
C
lo
u
d
s
er
v
ices
o
f
f
e
r
ed
b
y
Or
ac
le
R
esear
ch
o
f
co
s
t 2
5
0
,
0
0
0
/
-
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
au
th
o
r
s
d
ec
lar
e
th
at
n
o
f
i
n
an
cial
s
u
p
p
o
r
t
o
r
f
u
n
d
i
n
g
w
as
r
ec
eiv
ed
f
o
r
t
h
e
r
esear
ch
,
au
th
o
r
s
h
ip
,
an
d
/o
r
p
u
b
licatio
n
o
f
th
is
ar
tic
le.
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
Ud
ay
Ma
n
d
e
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Sh
af
i Path
an
✓
✓
✓
✓
✓
✓
Pan
k
aj
C
h
an
d
r
e
✓
✓
✓
✓
✓
✓
Sh
ar
v
ar
i M
an
d
e
✓
✓
✓
✓
✓
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
T
h
e
au
th
o
r
s
d
ec
lar
e
t
h
at
th
e
y
h
av
e
n
o
k
n
o
wn
co
m
p
etin
g
f
in
an
cial
o
r
n
o
n
-
f
i
n
an
cial
in
ter
ests
th
a
t
co
u
ld
h
a
v
e
in
f
lu
e
n
ce
d
th
e
wo
r
k
r
ep
o
r
ted
in
t
h
is
p
ap
er
.
I
NF
O
RM
E
D
CO
NS
E
N
T
T
h
is
s
ec
tio
n
n
o
t
ap
p
licab
le,
as
th
is
s
tu
d
y
d
id
n
o
t
in
v
o
lv
e
h
u
m
an
p
ar
ticip
an
ts
o
r
t
h
e
u
s
e
o
f
id
en
tifia
b
le
p
er
s
o
n
al
d
ata.
E
T
H
I
CAL AP
P
RO
V
AL
T
h
is
s
ec
tio
n
n
o
t a
p
p
licab
le
,
as th
is
s
tu
d
y
d
id
n
o
t in
v
o
lv
e
h
u
m
an
p
ar
ticip
an
ts
o
r
an
im
al
s
u
b
jects.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
au
th
o
r
s
co
n
f
ir
m
th
at
th
e
d
ata
s
u
p
p
o
r
tin
g
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
av
ailab
le
with
in
th
e
ar
ticle
[
an
d
/o
r
its
s
u
p
p
lem
en
tar
y
m
ater
ials
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.