I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Ro
bo
t
ics a
nd
Aut
o
m
a
t
io
n
(
I
J
RA)
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
,
p
p
.
6
7
8
~
6
8
9
I
SS
N:
2
7
2
2
-
2
5
8
6
,
DOI
: 1
0
.
1
1
5
9
1
/ijra
.
v
1
5
i
3
.
p
p
6
7
8
-
689
678
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
r
a
.
ia
esco
r
e.
co
m
PIK
ER
-
NET
:
Mu
lti
-
cla
ss
re
tinal di
sea
se cla
ss
ificatio
n using
Pied
K
ing
fishe
r o
ptimiza
tion
-
ba
sed
impro
v
ed residu
a
l net
wo
rk
L
is
s
y
Dev
a
s
a
ha
y
a
m
1
,
Ra
my
a
Dev
i Mu
rug
a
da
s
a
n
2
,
Ana
n
dh
i Sa
m
uel V
ij
a
y
a
la
k
s
hm
i
3
,
Cha
nthiy
a
P
uh
a
lent
hi
4
,
Ra
m
na
t
h M
uthus
a
m
y
3
,
Ahila
n Appa
t
hu
ra
i
5
,
Na
ta
r
a
ja
n
Mo
h
a
n
a
S
u
g
a
n
th
i
1
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
V
e
l
T
e
c
h
R
a
n
g
a
r
a
j
a
n
D
r
.
S
a
g
u
n
t
h
a
l
a
R
&D
I
n
st
i
t
u
t
e
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
A
v
a
d
i
,
Ta
mi
l
N
a
d
u
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
H
i
n
d
u
s
t
h
a
n
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
a
n
d
T
e
c
h
n
o
l
o
g
y
,
C
o
i
m
b
a
t
o
r
e
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
D
a
t
a
S
c
i
e
n
c
e
,
R
a
mc
o
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
R
a
j
a
p
a
l
a
y
a
m
,
T
a
mi
l
N
a
d
u
,
I
n
d
i
a
4
D
e
p
a
r
t
me
n
t
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
D
a
t
a
S
c
i
e
n
c
e
,
D
r
.
S
i
v
a
n
t
h
i
A
d
i
t
a
n
a
r
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
,
T
h
i
r
u
c
h
e
n
d
u
r
,
Ta
mi
l
N
a
d
u
,
I
n
d
i
a
5
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
P
S
N
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
Te
c
h
n
o
l
o
g
y
,
T
i
r
u
n
e
l
v
e
l
i
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
May
1
,
2
0
2
5
R
ev
is
ed
J
u
n
e
29
,
2
0
2
6
Acc
ep
ted
J
u
ly
20
,
2
0
2
6
Re
ti
n
a
l
d
ise
a
se
s
a
re
v
isio
n
-
t
h
re
a
ten
in
g
c
o
n
d
it
i
o
n
s,
i
n
c
lu
d
in
g
a
g
e
-
re
late
d
m
a
c
u
lar
d
e
g
e
n
e
ra
ti
o
n
(ARMD)
,
d
iab
e
ti
c
re
ti
n
o
p
a
t
h
y
(DR),
a
n
d
g
lau
c
o
m
a
,
th
a
t
re
q
u
ire
e
a
rly
a
n
d
a
c
c
u
ra
te
d
e
tec
ti
o
n
to
p
re
v
e
n
t
b
li
n
d
n
e
ss
.
Ho
we
v
e
r,
e
x
isti
n
g
m
e
th
o
d
s
o
fte
n
str
u
g
g
le
with
li
m
it
e
d
fe
a
tu
re
re
p
re
se
n
ta
ti
o
n
,
h
ig
h
in
ter
-
c
las
s
sim
il
a
rit
y
,
in
tra
-
c
las
s
v
a
riab
il
it
y
,
a
n
d
re
d
u
c
e
d
p
e
rfo
r
m
a
n
c
e
in
h
a
n
d
l
in
g
n
o
isy
a
n
d
lo
w
-
q
u
a
li
t
y
r
e
ti
n
a
l
ima
g
e
s.
To
a
d
d
re
ss
th
e
se
c
h
a
ll
e
n
g
e
s,
a
n
o
v
e
l
P
IKER
-
NET
fra
m
e
wo
rk
is
p
ro
p
o
se
d
fo
r
a
c
c
u
ra
te
m
u
lt
i
-
c
l
a
ss
re
ti
n
a
l
d
ise
a
se
c
las
sifica
ti
o
n
.
Th
e
i
n
p
u
t
f
u
n
d
u
s
ima
g
e
s
fro
m
th
e
RF
M
iD
d
a
tas
e
t
a
re
p
re
-
p
ro
c
e
ss
e
d
u
si
n
g
a
sc
a
lab
le
r
a
n
g
e
a
d
a
p
ti
v
e
b
il
a
tera
l
(S
CRAB
)
fil
ter
to
e
n
h
a
n
c
e
ima
g
e
c
larity
b
y
p
re
s
e
rv
in
g
e
d
g
e
s
w
h
il
e
re
d
u
c
in
g
n
o
ise
.
Th
e
imp
ro
v
e
d
re
sid
u
a
l
n
e
two
r
k
-
re
sc
a
led
(Im
Re
s
Ne
t
-
RS
)
in
teg
ra
ted
with
tem
p
o
ra
l
a
tt
e
n
ti
o
n
is
th
e
n
e
m
p
l
o
y
e
d
t
o
e
x
trac
t
d
e
e
p
h
iera
rc
h
ica
l
fe
a
tu
re
s
with
e
n
h
a
n
c
e
d
d
isc
rimin
a
ti
v
e
p
o
we
r.
P
ied
Kin
g
f
ish
e
r
o
p
ti
m
iza
ti
o
n
(P
KO
)
a
lg
o
rit
h
m
is
u
ti
li
z
e
d
f
o
r
fe
a
tu
re
se
lec
ti
o
n
,
e
ffe
c
ti
v
e
l
y
re
d
u
c
in
g
re
d
u
n
d
a
n
t
in
fo
rm
a
ti
o
n
wh
i
le
re
tain
in
g
t
h
e
m
o
st
re
lev
a
n
t
fe
a
tu
re
s.
Re
sid
u
a
l
m
u
lt
il
a
y
e
r
p
e
rc
e
p
tro
n
(Re
sML
P
)
is
u
se
d
t
o
c
las
sify
re
ti
n
a
l
d
ise
a
se
s
in
to
ARMD,
b
ra
n
c
h
re
ti
n
a
l
v
e
in
o
c
c
l
u
sio
n
(BRVO
),
d
iab
e
ti
c
n
e
u
ro
p
a
th
y
(DN
),
DR,
h
e
a
lt
h
y
,
a
n
d
m
y
o
p
ia
(M
YA
).
Th
e
P
IKER
-
NE
T
a
c
h
iev
e
s
a
n
o
v
e
ra
ll
a
c
c
u
ra
c
y
o
f
9
8
.
1
4
%
a
n
d
F
1
-
sc
o
re
o
f
9
7
.
0
6
%
.
Th
e
P
IKER
-
NET
a
p
p
ro
a
c
h
imp
r
o
v
e
s
o
v
e
ra
ll
a
c
c
u
ra
c
y
b
y
3
.
2
4
%
,
4
.
2
4
%
,
6
.
2
3
%
,
a
n
d
2
.
0
0
%
c
o
m
p
a
re
d
t
o
E
y
e
De
e
p
-
Ne
t,
IDL
-
M
RDD
,
De
e
p
Dia
b
e
ti
c
,
a
n
d
Visio
n
De
e
p
-
AI,
re
sp
e
c
ti
v
e
ly
.
Th
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
h
a
s
stro
n
g
c
li
n
ica
l
re
lev
a
n
c
e
b
y
s
u
p
p
o
rti
n
g
e
a
rli
e
r
d
ise
a
se
sc
re
e
n
in
g
,
re
d
u
c
in
g
m
is
d
iag
n
o
si
s,
a
n
d
e
n
a
b
li
n
g
fa
ste
r
d
iag
n
o
sis
to
a
ss
ist
o
p
h
th
a
lmo
l
o
g
ists
in
imp
ro
v
in
g
p
a
ti
e
n
t
o
u
tco
m
e
s.
K
ey
w
o
r
d
s
:
Feat
u
r
e
s
el
ec
t
io
n
Pied
Kin
g
f
is
h
er
o
p
tim
izatio
n
R
esid
u
al
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
R
etin
al
d
is
ea
s
e
Scal
ab
le
r
an
g
e
ad
a
p
ti
v
e
b
il
ate
r
al
f
ilt
e
r
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
:
Nata
r
ajan
Mo
h
an
a
Su
g
an
th
i
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
,
Vel
T
ec
h
R
an
g
ar
ajan
Dr
.
Sag
u
n
th
ala
R
&
D
I
n
s
titu
te
o
f
Scien
ce
an
d
T
ec
h
n
o
lo
g
y
Av
ad
i,
T
am
il Na
d
u
,
I
n
d
ia
E
m
ail: m
o
h
an
asu
g
a
n
th
i1
5
4
n
@
o
u
tlo
o
k
.
co
m
1.
I
NT
RO
D
UCT
I
O
N
R
etin
a
is
o
n
e
o
f
th
e
p
r
im
ar
y
elem
en
ts
o
f
t
h
e
e
y
e
th
at
s
u
s
tain
s
v
is
io
n
[
1
]
.
I
ts
p
r
im
a
r
y
f
u
n
ctio
n
is
to
co
n
v
er
t
lig
h
t
e
n
ter
in
g
th
e
e
y
e
in
to
elec
tr
ical
s
ig
n
als
th
at
ar
e
tr
an
s
m
itted
to
th
e
b
r
ai
n
v
ia
th
e
o
p
tic
n
e
r
v
e.
I
t
is
s
itu
ated
in
th
e
r
ea
r
o
f
th
e
ey
e
[
2
]
.
B
ec
au
s
e
o
f
its
n
atu
r
e,
th
e
r
etin
a
ca
n
ex
h
i
b
it
b
o
th
e
y
e
-
s
p
ec
if
ic
d
is
ea
s
es
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S I
n
t J Ro
b
&
Au
to
m
I
SS
N:
2722
-
2
5
8
6
P
I
K
E
R
-
N
E
T:
Mu
lti
-
cla
s
s
r
etin
a
l d
is
ea
s
e
cla
s
s
ifica
tio
n
u
s
in
g
P
ied
K
in
g
fis
h
er
…
(
Lis
s
y
Dev
a
s
a
h
a
ya
m
)
679
m
o
r
e
g
en
er
al
p
h
y
s
io
lo
g
ical
ailm
en
ts
,
s
u
ch
as
cir
c
u
lato
r
y
an
d
b
r
ai
n
d
is
o
r
d
er
s
[
3
]
.
Ag
e
-
r
elate
d
m
ac
u
la
r
d
eg
en
er
atio
n
(
AR
MD
)
,
d
iab
et
ic
r
etin
o
p
ath
y
(
DR
)
,
an
d
g
lau
co
m
a
ar
e
am
o
n
g
t
h
e
d
is
ea
s
es
th
at
ca
u
s
e
b
lin
d
n
ess
to
o
v
er
1
0
m
illi
o
n
in
d
iv
id
u
als wo
r
ld
wid
e
ea
ch
y
ea
r
.
R
etin
al
d
is
ea
s
es
ar
e
ca
u
s
ed
b
y
f
ac
to
r
s
s
u
ch
as
g
en
etics,
ag
in
g
,
d
iab
etes,
h
y
p
er
ten
s
io
n
,
an
d
life
s
ty
le
ch
o
ices
[
4
]
.
I
t
is
cr
itical
to
d
e
tect
r
etin
al
d
is
ea
s
es
ea
r
ly
an
d
in
itiate
tr
ea
tm
en
t
as
s
o
o
n
as
p
o
s
s
ib
le
to
p
r
eser
v
e
v
is
io
n
an
d
m
an
a
g
e
th
em
[
5
]
.
An
ea
r
ly
d
iag
n
o
s
is
is
ac
h
iev
ed
with
th
e
aid
o
f
Flu
o
r
escein
an
g
io
g
r
a
p
h
y
,
f
u
n
d
u
s
im
ag
in
g
,
r
etin
al
i
m
ag
in
g
,
an
d
o
p
tical
co
h
er
e
n
ce
to
m
o
g
r
ap
h
y
(
OC
T
)
[
6
]
.
I
n
d
ee
p
lear
n
in
g
(
DL
)
[
7
]
te
ch
n
iq
u
es,
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN)
[
8
]
an
d
tr
an
s
f
o
r
m
er
ar
ch
itectu
r
es
m
o
r
e
r
ec
en
tly
.
A
s
ig
n
if
ican
t
am
o
u
n
t
o
f
ef
f
o
r
t
h
as
b
ee
n
m
ad
e
to
id
en
ti
f
y
p
r
ev
alen
t
r
etin
al
illn
ess
es
s
u
ch
g
lau
co
m
a,
d
iab
etic
r
etin
o
p
ath
y
,
an
d
ag
e
-
r
elat
ed
m
ac
u
lar
d
eg
e
n
er
atio
n
[
9
]
.
c
o
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
s
(
p
r
e
-
tr
ain
e
d
C
NN
a
n
d
VGG1
6
)
[
1
0
]
to
d
iag
n
o
s
e
DR
b
ased
o
n
lesi
o
n
p
atch
p
r
o
b
ab
ilit
y
[
1
1
]
.
I
t
cr
ea
ted
a
s
y
s
tem
to
ca
teg
o
r
ies
f
u
n
d
u
s
im
ag
es
in
to
r
ef
er
ab
l
e
DR
an
d
n
o
n
-
r
ef
er
a
b
le
DR
u
s
in
g
th
r
ee
C
NNs:
I
n
ce
p
tio
n
-
v
3
,
R
esNet1
5
2
,
an
d
I
n
ce
p
tio
n
-
R
esNet
-
v2
[
1
2
]
,
[
1
3
]
.
Sen
g
ar
et
a
l.
[
1
4
]
in
tr
o
d
u
ce
d
a
n
o
n
-
in
v
asiv
e
au
to
m
ated
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
to
id
en
tify
n
u
m
er
o
u
s
ey
e
illn
ess
es
u
s
in
g
co
lo
r
f
u
n
d
u
s
im
ag
es.
T
h
e
s
u
g
g
ested
m
eth
o
d
o
lo
g
y
'
s
ef
f
ec
tiv
en
ess
in
class
if
y
in
g
an
d
id
en
tify
in
g
d
is
ea
s
es
u
s
in
g
d
i
g
ital
f
u
n
d
u
s
im
ag
es
was
d
em
o
n
s
tr
ated
b
y
a
co
m
p
ar
is
o
n
w
ith
th
e
m
o
s
t
r
ec
en
t
ap
p
r
o
ac
h
es.
Go
u
r
an
d
Kh
an
n
a
[
1
5
]
p
r
o
p
o
s
ed
a
tr
an
s
f
er
lear
n
in
g
-
b
ased
C
NN
u
tili
zin
g
two
d
is
tin
ct
o
p
tim
izer
s
an
d
f
o
u
r
p
r
e
-
tr
ain
e
d
C
NN
ar
ch
itectu
r
es.
T
h
ey
d
eter
m
in
ed
th
at
th
e
s
to
ch
asti
c
g
r
ad
i
en
t
d
escen
t
(
SGD)
o
p
tim
izer
p
r
e
-
tr
ain
ed
o
n
VGG
1
6
o
f
o
cu
lar
d
is
ea
s
es f
o
r
m
u
lti
-
class
,
f
u
n
d
u
s
im
ag
in
e
class
if
icatio
n
.
Vaiy
ap
u
r
i
et
a
l.
[
1
6
]
p
r
esen
ted
an
in
tellig
en
t
d
ee
p
lear
n
in
g
-
b
ased
m
u
lti
-
r
etin
al
d
is
ea
s
e
d
iag
n
o
s
tic
(
I
DL
-
MRDD)
f
r
am
ewo
r
k
u
s
i
n
g
f
u
n
d
u
s
im
ag
e.
W
ith
a
m
ax
im
u
m
ac
cu
r
ac
y
o
f
0
.
9
6
3
,
th
e
ex
p
er
im
en
tal
v
alu
es
p
r
o
d
u
ce
d
b
etter
r
esu
lts
th
an
th
e
cu
r
r
en
t
m
eth
o
d
s
.
R
o
d
r
íg
u
ez
et
a
l.
[
1
7
]
d
ev
elo
p
ed
a
m
u
lti
-
lab
el
r
etin
al
d
is
ea
s
e
class
if
icatio
n
u
s
in
g
tr
an
s
f
o
r
m
er
s
.
C
NNs
(
p
r
e
-
tr
ain
ed
C
NN
an
d
VGG1
6
)
to
d
iag
n
o
s
e
DR
b
ased
o
n
lesi
o
n
p
atch
p
r
o
b
a
b
ilit
y
.
I
t
c
r
ea
ted
a
s
y
s
tem
to
ca
teg
o
r
ies
f
u
n
d
u
s
im
ag
es
in
to
r
e
f
er
ab
le
DR
an
d
n
o
n
-
r
ef
er
a
b
le
DR
u
s
in
g
th
r
ee
C
NNs: I
n
ce
p
tio
n
-
v
3
,
R
esNet1
5
2
,
an
d
I
n
ce
p
tio
n
-
R
esNet
-
v2.
Alb
elaih
i
an
d
I
b
r
ah
im
[
1
8
]
i
n
tr
o
d
u
ce
d
a
Dee
p
Diab
etic
a
p
p
r
o
ac
h
f
o
r
i
d
en
tific
atio
n
o
f
d
iab
etic
ey
e
d
is
ea
s
es
u
s
in
g
o
f
d
ee
p
n
eu
r
a
l
n
etwo
r
k
s
.
T
h
e
ac
cu
r
ac
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
9
8
.
7
6
%,
wh
er
ea
s
th
e
ac
cu
r
ac
y
o
f
th
e
ea
r
lier
r
esear
ch
was
o
n
ly
8
8
.
3
3
%,
8
9
.
5
4
%
,
9
7
.
2
3
%,
an
d
8
0
.
3
3
%,
r
esp
ec
tiv
ely
.
J
o
s
h
i
et
a
l.
[
1
9
]
ex
am
in
e
d
a
Vis
io
n
Dee
p
-
AI
ap
p
r
o
ac
h
f
o
r
r
etin
al
b
lo
o
d
v
ess
els
s
eg
m
en
tatio
n
an
d
m
u
lti
-
class
class
if
icatio
n
f
r
am
ewo
r
k
f
o
r
ey
e
d
iag
n
o
s
is
.
Mu
lti
-
m
o
d
a
l
d
ee
p
f
ea
tu
r
e
-
f
u
s
ed
class
if
icatio
n
ar
ch
itectu
r
e
ac
h
iev
ed
a
test
ac
cu
r
ac
y
o
f
8
1
.
5
0
%,
with
a
s
p
ec
if
icity
o
f
9
3
.
8
3
%.
Yan
g
et
a
l.
[
2
0
]
d
ev
elo
p
e
d
a
C
R
AT
ap
p
r
o
ac
h
u
s
in
g
tr
an
s
f
o
r
m
er
-
b
ased
DL
alg
o
r
ith
m
in
OC
T
i
m
ag
e
class
if
icatio
n
.
T
h
ey
ar
e
9
7
.
3
3
%,
9
6
.
3
3
%
,
9
7
.
0
8
%,
9
9
.
1
7
%,
an
d
9
8
.
7
4
%
o
n
th
e
e
x
ter
n
al
test
d
ataset,
r
esp
ec
tiv
ely
.
E
x
is
tin
g
m
eth
o
d
s
in
th
e
liter
atu
r
e
r
ev
iew
h
av
e
m
an
y
s
h
o
r
tc
o
m
in
g
s
th
at
lim
it
th
eir
ef
f
ec
ti
v
en
ess
in
r
etin
al
d
is
ea
s
e
class
if
icatio
n
.
Ho
wev
er
,
ex
is
tin
g
m
et
h
o
d
s
o
f
ten
s
tr
u
g
g
le
with
ac
c
u
r
ate
cla
s
s
if
icatio
n
o
f
m
u
lti
-
class
r
etin
al
d
is
ea
s
e
s
d
u
e
to
lim
ited
f
ea
tu
r
e
r
e
p
r
esen
tatio
n
,
p
o
o
r
g
en
er
aliza
tio
n
ac
r
o
s
s
d
iv
er
s
e
d
atasets
,
h
ig
h
in
ter
-
class
s
im
ilar
ity
,
an
d
in
tr
a
-
class
v
ar
iab
ilit
y
.
Ad
d
itio
n
all
y
,
m
an
y
ap
p
r
o
ac
h
es
a
r
e
n
o
t
p
er
f
o
r
m
e
d
to
r
ed
u
ce
n
o
is
e
an
d
lo
w
-
q
u
ality
im
ag
es
,
an
d
r
ed
u
ce
d
ac
c
u
r
ac
y
in
ea
r
ly
-
s
tag
e
d
is
ea
s
e
d
etec
tio
n
.
T
o
o
v
er
c
o
m
e
th
ese
ch
allen
g
es,
a
n
o
v
el
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
was
ac
cu
r
ately
class
if
y
in
g
r
etin
al
d
is
ea
s
es
en
h
an
cin
g
d
iag
n
o
s
tic
ef
f
icien
cy
a
n
d
im
p
r
o
v
in
g
ea
r
ly
d
is
ea
s
e
d
etec
tio
n
.
T
h
e
k
ey
r
esear
ch
co
n
tr
i
b
u
tio
n
s
o
f
th
e
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
ar
e
h
ig
h
lig
h
te
d
:
i)
Scalab
le
r
an
g
e
ad
a
p
tiv
e
b
ilat
er
al
(
SC
R
A
B
)
f
ilter
is
u
tili
ze
d
to
p
r
eser
v
e
ed
g
es
wh
ile
ef
f
ec
tiv
ely
r
ed
u
ci
n
g
n
o
is
e
in
r
etin
al
im
ag
es,
im
p
r
o
v
in
g
im
ag
e
q
u
ality
,
ii)
T
h
e
p
r
o
p
o
s
ed
I
m
R
esNet
-
R
S
in
teg
r
ated
with
tem
p
o
r
al
atten
tio
n
ap
p
r
o
ac
h
to
ca
p
t
u
r
e
h
ie
r
a
r
ch
ical
f
ea
tu
r
es
f
o
r
im
p
r
o
v
e
g
en
er
aliza
tio
n
,
an
d
s
tr
en
g
th
en
d
is
cr
im
in
ativ
e
lear
n
in
g
f
o
r
m
u
lti
-
class
d
is
ea
s
e
d
e
tectio
n
,
iii)
Pied
k
in
g
f
is
h
er
o
p
t
im
izatio
n
(
PKO)
is
u
tili
ze
d
to
s
elec
t
th
e
m
o
s
t
d
is
cr
im
in
ativ
e
f
ea
tu
r
es
b
y
b
alan
cin
g
ex
p
lo
r
atio
n
an
d
ex
p
lo
itatio
n
,
th
er
eb
y
en
h
an
cin
g
r
etin
al
d
is
ea
s
e
class
if
icatio
n
ac
cu
r
ac
y
,
an
d
iv
)
R
esML
P
f
o
r
p
r
ec
is
e
clas
s
if
icatio
n
o
f
m
u
ltip
le
r
etin
al
d
is
ea
s
es,
s
tab
ilizin
g
tr
ain
in
g
with
r
esid
u
al
c
o
n
n
ec
tio
n
s
.
T
h
e
r
em
ain
d
e
r
o
f
t
h
is
wo
r
k
i
s
o
r
g
an
ized
in
to
t
h
e
f
o
llo
win
g
s
ec
tio
n
s
.
Sectio
n
2
p
r
esen
ts
th
e
r
ec
en
t
wo
r
k
s
o
f
r
etin
al
d
is
ea
s
e
class
if
icatio
n
,
s
ec
tio
n
3
in
tr
o
d
u
ce
s
th
e
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
in
p
r
ac
tical
asp
ec
t,
s
ec
tio
n
4
d
em
o
n
s
tr
ates
th
e
ex
p
er
im
en
tal
ar
r
a
n
g
em
e
n
t
an
d
t
h
e
f
in
d
i
n
g
s
o
f
th
e
PIK
E
R
-
NE
T
,
f
in
ally
c
o
n
clu
s
io
n
an
d
f
u
t
u
r
e
wo
r
k
ar
e
i
n
clu
d
ed
in
s
ec
tio
n
5
.
T
h
e
p
r
o
p
o
s
ed
a
p
p
r
o
a
ch
r
aises
th
e
f
o
llo
win
g
s
cien
tific
q
u
esti
o
n
s
:
SQ1
:
Ho
w
ca
n
n
o
is
e
b
e
r
e
d
u
ce
d
an
d
im
ag
e
clar
ity
im
p
r
o
v
ed
in
r
etin
al
f
u
n
d
u
s
im
ag
es
f
o
r
b
etter
d
is
ea
s
e
d
etec
tio
n
?
SQ2
: Can
d
ee
p
lear
n
in
g
with
f
ea
tu
r
e
o
p
tim
izatio
n
ac
cu
r
ately
class
if
y
m
u
ltip
le
r
etin
al
d
is
ea
s
es?
SQ3
:
Ho
w
d
o
es
th
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
f
r
am
ewo
r
k
p
er
f
o
r
m
co
m
p
a
r
ed
to
ex
is
tin
g
r
etin
al
d
is
ea
s
e
class
if
icatio
n
m
eth
o
d
s
?
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S I
n
t J Ro
b
&
Au
to
m
,
Vo
l
.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
6
7
8
-
6
8
9
680
2.
M
E
T
H
O
D
I
n
th
is
r
esear
ch
,
a
n
o
v
el
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
is
p
r
o
p
o
s
ed
f
o
r
r
etin
al
d
is
ea
s
e
class
if
icatio
n
f
r
o
m
t
h
e
g
ath
er
ed
r
etin
al
f
u
n
d
u
s
m
u
lti
-
d
is
ea
s
e
im
ag
e
d
ataset
(
R
FM
i
D)
.
Fig
u
r
e
1
illu
s
tr
ates
th
e
o
v
er
all
p
r
o
ce
s
s
o
f
th
e
PIK
E
R
-
NE
T
m
eth
o
d
.
Fig
u
r
e
1
.
Pro
p
o
s
ed
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
2
.
1
.
Da
t
a
s
et
d
escript
io
n
T
h
e
R
FMiD
is
an
o
p
e
n
ly
a
cc
e
s
s
ib
le
d
ataset.
I
t
is
s
p
ec
if
ically
cr
ea
ted
f
o
r
th
e
ca
te
g
o
r
izatio
n
o
f
r
etin
al
d
is
o
r
d
er
s
in
to
m
u
ltip
le
class
es.
I
t
co
n
s
is
ts
o
f
3
,
2
0
0
h
i
g
h
-
r
e
s
o
lu
tio
n
f
u
n
d
u
s
im
ag
es
c
o
llected
f
r
o
m
r
ea
l
-
wo
r
ld
clin
ical
s
ettin
g
s
co
v
er
in
g
4
6
d
if
f
er
en
t
r
etin
al
c
o
n
d
itio
n
s
s
u
ch
h
y
p
er
ten
s
iv
e
r
etin
o
p
ath
y
,
AR
MD
,
g
lau
co
m
a,
an
d
DR
.
T
h
e
d
ataset
in
clu
d
es
ex
p
er
t
-
lab
elled
a
n
n
o
tatio
n
s
,
an
d
it
is
h
ig
h
ly
d
iv
e
r
s
e
f
ea
tu
r
i
n
g
a
wid
e
r
an
g
e
o
f
d
is
ea
s
e
s
ev
er
itie
s
,
im
ag
e
q
u
alities
,
an
d
v
ar
iatio
n
s
in
p
atien
t
d
em
o
g
r
a
p
h
ics.
T
h
e
d
is
tr
ib
u
tio
n
o
f
r
etin
al
f
u
n
d
u
s
im
ag
es
ac
r
o
s
s
d
if
f
er
en
t
d
is
ea
s
e
class
e
s
in
th
e
R
FMiD
d
atas
et
co
m
p
r
is
es
a
to
tal
o
f
3
,
3
7
2
im
ag
es.
T
h
e
d
ataset
co
v
er
s
1
0
ca
teg
o
r
ies
in
clu
d
in
g
p
ath
o
lo
g
ical
co
n
d
itio
n
s
s
u
c
h
as
ag
e
-
r
elate
d
m
ac
u
lar
d
eg
en
er
atio
n
(
AR
MD
)
,
b
r
an
ch
r
eti
n
al
v
ein
o
cc
lu
s
io
n
(
B
R
VO,
d
iab
etic
r
etin
o
p
ath
y
(
DR
)
,
h
ea
lth
y
,
an
d
m
y
o
p
ia
(
MY
A)
,
alo
n
g
with
h
ea
lth
y
ca
s
es.
T
h
e
Hea
lth
y
cl
ass
co
n
tain
s
th
e
h
ig
h
est
n
u
m
b
er
o
f
im
a
g
es
6
6
9
,
wh
ile
th
e
ODP
clas
s
h
as
th
e
f
ewe
s
t 1
1
5
.
T
h
is
d
ataset
p
r
o
v
i
d
es a
b
alan
ce
d
y
et
d
iv
er
s
e
r
e
p
r
esen
tatio
n
o
f
c
o
m
m
o
n
an
d
r
ar
e
r
etin
al
d
is
ea
s
es.
.
2
.
2
.
P
re
-
pro
ce
s
s
ing
us
ing
s
c
a
la
ble r
a
ng
e
a
da
ptiv
e
bil
a
t
er
a
l (
SCRAB
)
SC
R
AB
f
il
ter
is
u
tili
ze
d
to
e
n
h
an
ce
r
etin
al
im
ag
es
b
y
p
r
es
er
v
in
g
ed
g
es
wh
ile
r
ed
u
cin
g
n
o
is
e.
T
h
e
b
ilater
al
f
ilter
is
u
s
ed
f
o
r
n
o
i
s
e
elim
in
atio
n
,
lev
ellin
g
,
n
o
n
-
lin
ea
r
,
an
d
ed
g
e
-
p
r
eser
v
i
n
g
i
n
r
etin
al
im
ag
es.
A
m
ajo
r
s
h
o
r
tco
m
i
n
g
o
f
t
h
e
b
ilater
al
f
ilter
is
ac
q
u
ir
in
g
s
ig
n
if
ic
an
t
d
ata
ab
o
u
t
ed
g
e
f
lu
ctu
atio
n
s
in
n
o
is
y
im
ag
es.
T
h
is
tr
ick
y
is
s
o
lv
ed
b
y
th
e
S
C
R
A
B
f
ilter
,
wh
ich
is
d
eter
m
in
ed
as f
o
llo
ws:
(
,
,
)
=
∝
(
−
1
2
(
|
|
(
)
−
(
)
−
(
)
|
|
)
2
)
+
(
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S I
n
t J Ro
b
&
Au
to
m
I
SS
N:
2722
-
2
5
8
6
P
I
K
E
R
-
N
E
T:
Mu
lti
-
cla
s
s
r
etin
a
l d
is
ea
s
e
cla
s
s
ifica
tio
n
u
s
in
g
P
ied
K
in
g
fis
h
er
…
(
Lis
s
y
Dev
a
s
a
h
a
ya
m
)
681
(
)
=
{
|
(
)
−
(
Ω
)
|
|
−
|
≤
0
ℎ
(
2
)
w
h
er
e
Ω
d
en
o
tes
th
e
p
ix
el
-
s
et
o
f
(
2
+
1
)
∗
(
2
+
1
)
p
ix
el
win
d
o
w
wh
en
=
2
.
T
h
e
o
p
tim
is
tic
p
ar
am
eter
s
ar
e
∝
an
d
,
Ω
is
th
e
av
er
ag
e
v
alu
e
,
is
a
s
tab
le
v
ar
iab
le
an
d
(
)
is
a
r
an
g
e
-
b
ased
f
u
n
ctio
n
s
.
T
h
e
th
r
ee
v
ar
iab
les
ar
e
u
tili
ze
d
f
o
r
co
n
tr
o
llin
g
ar
e
th
e
s
ca
lin
g
f
ac
to
r
,
th
e
lin
ea
r
co
n
s
tan
t
co
ef
f
icien
ts
=
2
an
d
=
1
.
Sin
ce
th
ese
v
ar
iab
les
g
u
ar
an
tees
a
h
ig
h
d
eg
r
ee
o
f
v
is
u
al
s
im
ilar
ity
b
etwe
en
p
ix
el
in
th
e
ce
n
ter
an
d
p
ix
el
in
th
e
n
eig
h
b
o
r
in
g
ar
ea
.
T
h
is
SC
R
A
B
f
ilter
ef
f
ec
tiv
ely
en
h
an
ce
s
ed
g
e
p
r
eser
v
atio
n
an
d
n
o
is
e
r
ed
u
ctio
n
in
r
etin
al
im
a
g
es b
y
r
ef
in
in
g
in
ten
s
ity
d
if
f
er
e
n
ce
s
wh
ile
m
ain
tain
in
g
v
is
u
al
s
im
ilar
ity
.
2
.
3
.
F
e
a
t
ure
e
x
t
r
a
ct
io
n
I
m
p
r
o
v
ed
r
esid
u
al
n
etwo
r
k
(
I
m
R
esNet
-
R
S)
in
teg
r
ated
with
tem
p
o
r
al
atten
tio
n
ap
p
r
o
ac
h
is
u
s
ed
f
o
r
ex
tr
ac
tin
g
h
ier
ar
ch
ical
f
ea
tu
r
e
s
f
r
o
m
r
etin
al
im
ag
es.
R
esNet
-
R
S
is
a
s
ca
led
R
esNet
ar
ch
ite
ctu
r
e
th
at
im
p
r
o
v
es
ac
cu
r
ac
y
an
d
e
f
f
icien
cy
b
y
s
y
s
tem
atica
lly
ad
ju
s
tin
g
d
ep
th
,
wid
th
,
an
d
in
p
u
t
r
eso
lu
tio
n
.
I
m
R
esNet
-
R
S
f
u
r
th
er
en
h
an
ce
s
th
is
d
esig
n
f
o
r
b
e
tter
p
er
f
o
r
m
an
ce
with
lo
w
co
m
p
u
tatio
n
al
co
s
t.
Fig
u
r
e
2
d
em
o
n
s
tr
ates
th
e
ar
ch
itectu
r
e
o
f
p
r
o
p
o
s
ed
R
esNet
-
R
S.
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
I
m
R
esNet
-
RS
T
h
e
p
r
im
a
r
y
c
o
n
tr
ib
u
ti
o
n
o
f
R
esNet
-
R
S
is
it
s
s
ca
lin
g
r
u
le,
wh
ich
ad
ju
s
ts
d
ep
th
(
)
,
wid
th
(
)
,
an
d
r
eso
lu
tio
n
(
)
ac
co
r
d
in
g
to
a
co
m
p
o
u
n
d
co
ef
f
icien
t
∅
.
T
h
is
s
ca
lin
g
f
o
llo
ws
(
3
)
:
=
∅
0
,
=
∅
0
,
=
∅
0
(
3
)
w
h
er
e
0
,
0
,
0
ar
e
th
e
in
itial
d
e
p
th
,
wid
th
,
an
d
r
eso
lu
ti
o
n
o
f
th
e
b
aselin
e
R
esNet
m
o
d
el,
wh
ile
,
,
ar
e
co
n
s
tan
ts
d
eter
m
in
ed
t
h
r
o
u
g
h
g
r
id
s
ea
r
ch
.
I
n
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
,
a
tem
p
o
r
al
a
tten
tio
n
m
o
d
u
le
is
em
b
ed
d
e
d
af
ter
co
n
v
o
lu
tio
n
a
l
lay
er
s
to
r
ef
in
e
th
e
ex
tr
ac
ted
f
ea
tu
r
es
b
y
weig
h
tin
g
th
e
m
o
s
t
in
f
o
r
m
ativ
e
p
atter
n
s
.
Ma
th
em
atica
lly
b
o
ttl
en
ec
k
b
lo
c
k
ex
p
r
ess
ed
as:
=
+
(
3
.
(
2
.
(
1
.
)
)
)
(
4
)
w
h
er
e
is
th
e
in
p
u
t
f
ea
tu
r
e
m
ap
1
,
2
,
3
ar
e
weig
h
t
m
atr
ices
f
o
r
th
e
1
×1
,
3
×
3
,
a
n
d
1
×1
co
n
v
o
lu
tio
n
,
r
ep
r
esen
ts
a
n
o
n
-
lin
ea
r
ity
an
d
is
th
e
f
u
n
ctio
n
t
h
at
ap
p
lies
b
atch
n
o
r
m
aliza
tio
n
an
d
ac
tiv
a
tio
n
an
d
is
th
e
f
in
al
o
u
tp
u
t
o
f
th
e
o
p
er
atio
n
.
I
m
R
esNet
-
R
S
en
h
an
ce
s
r
etin
al
im
ag
e
an
aly
s
is
b
y
ef
f
icien
tly
ex
tr
ac
tin
g
h
ier
ar
ch
ical
f
ea
tu
r
es wh
ile
m
a
in
tain
in
g
co
m
p
u
tatio
n
al
e
f
f
icien
cy
.
2
.
4
.
P
ied
k
ing
f
is
her
o
ptim
iz
a
t
io
n (
P
K
O
)
T
h
e
PKO
is
to
s
elec
t
th
e
m
o
s
t
d
is
cr
im
in
ativ
e
f
ea
tu
r
es
b
y
b
alan
cin
g
e
x
p
lo
r
atio
n
an
d
ex
p
lo
itatio
n
,
th
er
eb
y
en
h
an
cin
g
r
etin
al
d
is
e
ase
clas
s
if
icatio
n
ac
cu
r
ac
y
.
PKO
is
a
b
io
-
in
s
p
ir
ed
o
p
tim
iza
tio
n
alg
o
r
ith
m
th
at
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S I
n
t J Ro
b
&
Au
to
m
,
Vo
l
.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
6
7
8
-
6
8
9
682
m
im
ics
th
e
p
er
ch
in
g
,
h
o
v
er
in
g
,
an
d
d
i
v
in
g
b
eh
av
io
r
s
o
f
Pied
Kin
g
f
is
h
er
s
.
I
n
t
h
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
,
PKO
is
em
p
lo
y
ed
f
o
r
f
ea
tu
r
e
s
elec
tio
n
,
wh
er
e
p
er
c
h
in
g
en
a
b
les
g
lo
b
al
ex
p
lo
r
atio
n
o
f
th
e
f
ea
t
u
r
e
s
p
ac
e,
h
o
v
e
r
in
g
r
ef
in
es
p
r
o
m
is
in
g
f
ea
tu
r
e
s
u
b
s
ets,
an
d
d
iv
i
n
g
p
er
f
o
r
m
s
in
te
n
s
iv
e
lo
ca
l
e
x
p
lo
itatio
n
n
ea
r
t
h
e
b
est
ca
n
d
id
ates,
lead
in
g
to
o
p
tim
al
an
d
h
ig
h
l
y
d
is
cr
im
in
ativ
e
f
ea
t
u
r
e
s
elec
tio
n
.
T
h
is
o
p
tim
izatio
n
im
p
r
o
v
es
th
e
m
o
d
el’
s
p
er
f
o
r
m
an
ce
b
y
r
ed
u
cin
g
i
r
r
elev
an
t
f
ea
tu
r
es
f
o
r
en
h
a
n
cin
g
class
if
icatio
n
ac
cu
r
ac
y
wh
ile
lo
wer
in
g
co
m
p
u
tatio
n
al
c
o
s
t.
Step
1
: I
n
itializatio
n
T
h
e
in
itial in
p
u
t
p
ar
am
eter
s
s
u
ch
as p
o
p
u
latio
n
s
ize
an
d
f
ea
tu
r
e
s
u
b
s
et
len
g
th
a
r
e
s
et.
Step
2
: Ran
d
o
m
g
e
n
er
atio
n
C
an
d
id
ate
f
ea
tu
r
e
s
u
b
s
ets ar
e
g
en
er
ated
r
an
d
o
m
ly
f
r
o
m
th
e
ex
tr
ac
ted
f
ea
tu
r
es.
Step
3
: Fitn
ess
f
u
n
ctio
n
Fit
n
ess
i
s
d
eter
m
in
ed
b
y
th
e
cl
ass
if
icatio
n
p
er
f
o
r
m
a
n
ce
,
an
d
it is
d
ef
in
ed
as:
=
(
1
−
)
(
5
)
w
h
er
e
ac
cu
r
ac
y
r
ep
r
esen
ts
th
e
class
if
icatio
n
ac
cu
r
ac
y
o
f
th
e
I
m
R
esNet
-
R
S
ap
p
r
o
ac
h
w
h
e
n
ev
alu
ate
d
with
a
p
ar
ticu
lar
s
u
b
s
et
o
f
f
ea
tu
r
es.
Step
4
: E
x
p
lo
r
atio
n
PK’
s
h
o
v
er
in
g
an
d
p
e
r
ch
in
g
b
eh
av
io
r
s
u
p
p
o
r
ts
to
id
en
tify
s
u
b
s
ets
o
f
f
ea
tu
r
es
with
h
ig
h
p
o
ten
tial.
T
h
e
ex
p
lo
r
at
o
r
y
m
ec
h
an
is
m
allo
ws PKO
to
s
ca
n
th
e
f
ea
tu
r
e
s
p
ac
e
wid
ely
an
d
s
elec
t d
iv
er
s
e
ca
n
d
id
ate
s
u
b
s
ets
f
o
r
e
v
alu
atio
n
.
T
h
e
PK’
s
f
o
r
a
g
in
g
ac
ti
v
ity
is
u
s
ed
in
PKO
to
d
eter
m
i
n
e
th
e
lo
ca
tio
n
s
o
f
s
ea
r
ch
ag
en
ts
.
T
h
e
u
p
d
ated
p
o
s
itio
n
s
ar
e
ca
lcu
lat
ed
as:
(
+
1
)
=
(
)
+
∗
×
(
(
)
−
(
)
)
(
6
)
w
h
er
e
(
+
1
)
d
en
o
tes
s
o
lu
tio
n
f
o
r
n
ex
t
iter
atio
n
,
(
)
r
ep
r
esen
ts
th
e
p
o
s
itio
n
o
f
th
e
cu
r
r
en
t
iter
atio
n
,
an
d
d
en
o
tes
p
ar
am
eter
.
T
h
is
p
r
o
ce
s
s
en
s
u
r
es
wid
e
ex
p
lo
r
atio
n
o
f
f
ea
tu
r
e
s
u
b
s
ets,
th
er
eb
y
in
cr
ea
s
in
g
th
e
p
o
s
s
ib
ilit
y
o
f
s
elec
tin
g
th
e
m
o
s
t d
is
cr
im
in
ativ
e
f
ea
tu
r
es.
Step
5
: E
x
p
lo
itatio
n
T
h
e
d
iv
in
g
b
e
h
av
io
r
o
f
PK
en
s
u
r
es
r
ef
in
ed
ex
p
l
o
itatio
n
o
f
p
r
o
m
is
in
g
f
ea
tu
r
e
s
u
b
s
ets.
T
h
e
s
ea
r
c
h
p
r
o
ce
s
s
em
p
h
asizes
s
u
b
s
ets
t
h
at
m
ax
im
ize
ac
cu
r
ac
y
wh
il
e
m
in
im
izin
g
r
ed
u
n
d
an
cy
.
Ma
th
em
atica
lly
,
th
e
u
p
d
ated
ca
n
d
id
ate
f
e
atu
r
e
s
u
b
s
et
p
o
s
itio
n
s
ar
e
ex
p
r
ess
ed
as:
(
+
1
)
=
(
)
+
.
.
(
−
(
)
)
(
7
)
w
h
er
e
(
+
1
)
is
r
ep
r
esen
ts
th
e
u
p
d
a
ted
f
ea
tu
r
e
s
u
b
s
et
p
o
s
itio
n
at
iter
atio
n
+
1
,
an
d
(
)
d
en
o
tes
th
e
cu
r
r
en
t
f
e
atu
r
e
s
u
b
s
et
p
o
s
itio
n
at
iter
atio
n
,
th
e
is
th
e
lear
n
in
g
r
ate
th
at
c
o
n
tr
o
ls
th
e
s
tep
s
ize
to
war
d
s
th
e
b
est
f
ea
tu
r
e
s
u
b
s
et.
d
en
o
tes
th
e
co
n
tr
o
l
p
ar
am
eter
th
at
b
a
lan
ce
s
ex
p
lo
r
atio
n
an
d
e
x
p
lo
i
tatio
n
d
u
r
in
g
th
e
o
p
tim
izatio
n
.
T
h
e
n
in
d
icate
th
e
m
o
s
t
d
is
cr
im
in
ativ
e
f
ea
tu
r
e
s
u
b
s
et
id
en
tifie
d
an
d
b
ased
o
n
class
if
icatio
n
ac
cu
r
ac
y
.
−
(
)
s
y
m
b
io
s
is
th
e
d
ir
ec
tio
n
o
f
m
o
v
em
e
n
t
to
war
d
th
e
o
p
tim
al
s
u
b
s
et
in
th
e
f
ea
tu
r
e
s
p
ac
e.
Step
6
: U
p
d
atin
g
T
h
e
b
est
s
u
b
s
et
is
u
p
d
ated
iter
ativ
ely
u
s
in
g
class
if
icat
io
n
ac
cu
r
ac
y
as
th
e
o
b
ject
iv
e
u
n
til
co
n
v
er
g
en
ce
.
Step
7
: T
er
m
in
atio
n
T
h
e
p
r
o
ce
s
s
ter
m
in
ates
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
r
e
p
etitio
n
s
is
r
ea
ch
ed
o
r
wh
e
n
n
o
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t
in
f
ea
tu
r
e
s
elec
tio
n
is
o
b
s
er
v
ed
.
T
h
e
f
in
al
s
elec
ted
f
ea
tu
r
es
ar
e
th
en
p
ass
ed
to
th
e
class
if
ier
f
o
r
f
in
al
r
etin
al
d
is
ea
s
e
p
r
ed
ictio
n
.
2
.
5
.
ResM
L
P
R
esid
u
al
m
u
ltil
ay
er
p
er
c
ep
tr
o
n
(
R
esML
P)
is
u
s
ed
to
class
if
y
th
e
r
etin
al
d
is
ea
s
e
AR
MD
,
B
R
V
O,
DN,
DR
,
h
ea
lth
y
,
MY
A.
R
esML
P
is
an
ad
v
an
ce
d
v
ar
ian
t
o
f
ML
P
th
at
in
teg
r
ates
r
esid
u
al
co
n
n
ec
tio
n
s
to
s
tab
ilize
tr
ain
in
g
a
n
d
e
n
h
an
ce
f
ea
tu
r
e
lear
n
in
g
.
A
R
esML
P
u
s
u
ally
co
m
p
r
is
es
m
u
ltip
le
h
i
d
d
en
la
y
er
s
in
ter
co
n
n
ec
ted
with
r
esid
u
al
s
k
ip
co
n
n
ec
tio
n
s
alo
n
g
with
th
e
o
u
tp
u
t
lay
er
.
W
eig
h
ted
co
n
n
ec
tio
n
s
ar
e
u
s
ed
to
lin
k
ea
ch
n
eu
r
o
n
to
th
e
n
eu
r
o
n
s
in
th
e
lay
er
b
elo
w.
Fig
u
r
e
3
illu
s
tr
ates
th
e
R
esML
P
s
tr
u
ctu
r
e
f
o
r
R
etin
al
d
is
ea
s
e
class
if
icatio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S I
n
t J Ro
b
&
Au
to
m
I
SS
N:
2722
-
2
5
8
6
P
I
K
E
R
-
N
E
T:
Mu
lti
-
cla
s
s
r
etin
a
l d
is
ea
s
e
cla
s
s
ifica
tio
n
u
s
in
g
P
ied
K
in
g
fis
h
er
…
(
Lis
s
y
Dev
a
s
a
h
a
ya
m
)
683
Fig
u
r
e
3
.
Stru
ctu
r
e
o
f
R
esML
P
T
h
e
f
u
n
d
am
e
n
tal
u
n
it
o
f
a
R
esML
P
is
th
e
p
e
r
ce
p
tr
o
n
,
w
h
ich
in
tr
o
d
u
ce
s
n
o
n
-
lin
ea
r
ity
b
y
ap
p
ly
in
g
an
ac
tiv
atio
n
f
u
n
ctio
n
an
d
ca
lcu
l
atin
g
th
e
i
n
p
u
ts
weig
h
ted
q
u
a
n
tity
.
Ma
th
em
atica
lly
a
g
iv
e
n
n
eu
r
o
n
j,
th
e
o
u
tp
u
t
is
g
iv
en
b
y
(
8
)
.
=
∑
=
1
+
(
8
)
w
h
er
e
r
ep
r
esen
ts
th
e
in
p
u
t
r
ep
r
esen
ts
th
e
weig
h
t
ass
ig
n
ed
to
ev
er
y
i
n
p
u
t
co
n
n
ec
tio
n
an
d
is
th
e
b
ias
ter
m
.
T
h
is
weig
h
ted
s
u
m
g
o
e
s
th
r
o
u
g
h
a
n
ac
tiv
atio
n
f
u
n
cti
o
n
af
te
r
th
at
lik
e
th
e
tan
h
f
u
n
ctio
n
R
eL
U
o
r
s
ig
m
o
id
.
=
(
)
(
9
)
A
R
esML
P
with
m
u
ltip
le
lay
er
s
an
d
r
esid
u
al
lin
k
s
is
m
ath
em
atica
lly
r
ep
r
esen
ted
u
s
i
n
g
m
atr
i
x
n
o
tatio
n
.
Giv
en
an
in
p
u
t v
ec
to
r
an
d
we
ig
h
t m
atr
ices
,
th
e
o
u
tp
u
t
o
f
a
l
ay
er
wr
itten
as
(
1
0
)
.
(
)
=
(
(
)
(
−
1
)
+
(
)
)
+
(
−
1
)
(
1
0
)
w
h
er
e
(
)
s
y
m
b
o
lizes th
e
lay
er
ac
tiv
atio
n
s
at
lay
er
,
(
)
is
th
e
weig
h
t m
atr
ix
f
o
r
th
e
lay
er
,
an
d
(
)
is
th
e
b
ias
v
ec
to
r
.
T
h
e
r
esid
u
al
co
n
n
ec
tio
n
(
−
1
)
s
u
p
p
o
r
ts
g
r
ad
ien
ts
an
d
ac
ce
ler
ates
co
n
v
er
g
e
n
ce
.
T
h
is
R
es
ML
P
-
b
ased
ap
p
r
o
ac
h
en
h
an
ce
s
class
if
icatio
n
o
f
r
etin
al
d
is
ea
s
e
ty
p
es
b
y
r
esid
u
al
lear
n
i
n
g
f
r
o
m
ex
tr
ac
ted
f
ea
tu
r
es
t
o
ac
h
iev
e
p
r
ec
is
e
class
if
icatio
n
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
d
etails
th
e
ex
p
e
r
im
en
tal
r
esu
lts
o
f
th
e
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
was
im
p
lem
e
n
ted
u
s
in
g
MA
T
L
AB
2
0
2
2
b
.
T
h
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
w
as
ass
es
s
ed
u
s
in
g
a
n
u
m
b
er
o
f
m
etr
ics
in
th
is
s
ec
tio
n
in
clu
d
in
g
s
p
ec
if
icity
,
ac
cu
r
ac
y
,
F1
s
co
r
e,
r
ec
all,
an
d
p
r
ec
is
io
n
b
ased
o
n
im
a
g
es
tak
en
f
r
o
m
th
e
R
FMiD
d
ataset.
T
h
e
ex
p
er
im
en
tal
r
esu
lts
o
f
th
e
PIK
E
R
-
N
E
T
ap
p
r
o
ac
h
ar
e
illu
s
tr
ated
in
Fig
u
r
e
4
f
o
r
r
etin
a
l
d
is
ea
s
e
class
if
icatio
n
.
T
h
e
ex
p
er
im
e
n
tal
r
esu
lt
o
f
th
e
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
is
s
h
o
wn
in
Fig
u
r
e
4
.
I
n
c
o
lu
m
n
1
,
th
e
in
p
u
t
r
etin
al
im
ag
es
ar
e
co
llected
f
r
o
m
th
e
R
FMiD
d
ataset.
I
n
c
o
lu
m
n
2
,
SC
R
A
B
to
en
h
an
ce
r
etin
al
im
ag
es
b
y
p
r
eser
v
in
g
ed
g
es
wh
ile
r
ed
u
ci
n
g
n
o
is
e
f
o
llo
wed
b
y
d
ata
au
g
m
en
tatio
n
in
co
lu
m
n
3
.
Feat
u
r
e
e
x
tr
ac
tio
n
u
s
in
g
I
m
R
esNet
-
R
S
to
ex
tr
ac
t
th
e
f
ea
tu
r
e
im
ag
es
in
co
lu
m
n
4
.
Fi
n
ally
,
th
e
R
esML
P
is
u
s
ed
to
class
if
y
th
e
r
etin
al
d
is
ea
s
e
n
am
ely
AR
MD
,
B
R
V
O,
DN,
DR
,
Hea
lth
y
,
an
d
MY
A
in
co
lu
m
n
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S I
n
t J Ro
b
&
Au
to
m
,
Vo
l
.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
6
7
8
-
6
8
9
684
Fig
u
r
e
4
.
E
x
p
er
im
e
n
tal
r
esu
lt o
f
th
e
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
3
.
1
.
P
er
f
o
r
m
a
nce
a
na
ly
s
is
T
h
e
PIK
E
R
-
NE
T
was
ass
e
s
s
e
d
in
th
is
s
ec
tio
n
u
s
in
g
a
n
u
m
b
er
o
f
m
etr
ics,
in
clu
d
in
g
s
p
ec
if
icity
(
Sp
)
,
F1
s
co
r
e,
ac
cu
r
ac
y
(
Ac)
,
p
r
ec
i
s
io
n
(
Pr)
,
an
d
r
ec
all
(
R
c)
,
o
n
t
h
e
g
ath
er
e
d
R
FMiD
d
ataset.
=
+
(
1
1
)
=
+
+
+
+
(
1
2
)
=
+
(
1
3
)
=
+
(
1
4
)
1
=
2
(
∗
+
)
(
1
5
)
w
h
er
e
T
P
a
n
d
T
N
in
d
icate
tr
u
e
p
o
s
itiv
es
an
d
n
eg
ativ
es
o
f
th
e
s
am
p
le
im
ag
es,
FP
an
d
FN
in
d
icate
s
f
alse
p
o
s
itiv
es a
n
d
n
eg
ativ
es o
f
th
e
s
am
p
le
im
ag
es.
T
ab
le
1
p
r
esen
ts
th
e
s
p
ec
if
icity
,
F1
s
co
r
e,
r
ec
all,
ac
cu
r
ac
y
,
a
n
d
p
r
ec
is
io
n
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
f
o
r
class
es
with
m
u
ltip
le
lev
els.
T
ab
le
1
d
is
p
lay
s
th
e
ef
f
icien
c
y
m
etr
ics
o
f
th
e
p
r
o
p
o
s
ed
m
e
th
o
d
f
o
r
th
e
class
es
lis
ted
b
elo
w:
AR
MD
,
B
R
V
O,
DN,
DR
,
Hea
lth
y
,
an
d
MY
A
.
T
h
e
ac
cu
r
ac
y
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
is
9
6
.
3
7
%
f
o
r
AR
MD
,
9
8
.
5
6
%
f
o
r
B
R
VO,
9
7
.
8
1
%
f
o
r
DN,
9
9
.
1
1
%
f
o
r
DR
,
9
9
.
8
7
f
o
r
Hea
lt
h
y
,
a
n
d
9
7
.
1
6
%
f
o
r
MY
A
r
esp
ec
tiv
ely
.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
ac
h
iev
es a
to
tal
ac
cu
r
ac
y
r
ate
o
f
9
8
.
1
4
%.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S I
n
t J Ro
b
&
Au
to
m
I
SS
N:
2722
-
2
5
8
6
P
I
K
E
R
-
N
E
T:
Mu
lti
-
cla
s
s
r
etin
a
l d
is
ea
s
e
cla
s
s
ifica
tio
n
u
s
in
g
P
ied
K
in
g
fis
h
er
…
(
Lis
s
y
Dev
a
s
a
h
a
ya
m
)
685
T
h
e
p
er
f
o
r
m
a
n
ce
ev
alu
atio
n
o
f
th
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
a
p
p
r
o
ac
h
ac
r
o
s
s
1
0
0
ep
o
ch
s
is
illu
s
tr
ated
in
Fig
u
r
e
5
,
s
h
o
win
g
b
o
th
t
h
e
ac
cu
r
ac
y
a
n
d
lo
s
s
cu
r
v
es
d
u
r
in
g
th
e
tr
ain
in
g
an
d
test
in
g
s
tag
es.
Fig
u
r
e
5
(
a)
d
em
o
n
s
tr
ates
th
e
ac
c
u
r
ac
y
c
u
r
v
e
o
f
th
e
PIK
E
R
-
NE
T
a
p
p
r
o
ac
h
b
ased
o
n
th
e
g
ath
e
r
ed
R
FMiD
d
ataset.
T
h
e
PIK
E
R
-
NE
T
tr
ain
in
g
an
d
test
i
n
g
ac
cu
r
ac
y
c
u
r
v
e
d
is
p
lay
s
a
h
ig
h
ac
cu
r
ac
y
lev
el
o
f
9
8
.
1
4
%.
T
h
e
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
lo
s
s
cu
r
v
e,
wh
ich
d
is
p
lay
s
1
0
0
ep
o
ch
s
,
is
illu
s
tr
ate
d
in
Fig
u
r
e
5
(
b
)
.
I
t
ac
h
ie
v
es
a
lo
w
lo
s
s
o
f
1
.
8
6
%,
in
d
icatin
g
th
at
th
e
PIK
E
R
-
NE
T
m
o
d
el
p
e
r
f
o
r
m
s
ef
f
ec
tiv
ely
d
u
r
in
g
b
o
th
th
e
test
in
g
an
d
tr
a
in
in
g
s
tag
es.
T
ab
le
1
.
Per
f
o
r
m
an
ce
an
aly
s
is
o
f
PIK
E
R
-
NE
T
m
eth
o
d
C
l
a
s
ses
A
c
c
u
r
a
c
y
%
S
p
e
c
i
f
i
c
i
t
y
%
P
r
e
c
i
s
i
o
n
%
R
e
c
a
l
l
%
F
1
sc
o
r
e
%
A
R
M
D
9
6
.
3
7
9
7
.
4
0
9
5
.
3
2
9
6
.
3
8
9
8
.
0
2
B
R
V
O
9
8
.
5
6
9
8
.
1
1
9
4
.
6
5
9
8
.
3
0
9
7
.
2
1
DN
9
7
.
8
1
9
6
.
3
2
9
4
.
2
2
9
5
.
8
1
9
6
.
8
9
DR
9
9
.
1
1
9
8
.
9
7
9
6
.
3
7
9
8
.
4
2
9
9
.
0
4
H
e
a
l
t
h
y
9
9
.
8
7
9
9
.
2
1
9
7
.
9
4
9
5
.
3
9
9
8
.
2
7
M
Y
A
9
7
.
1
6
9
1
.
8
0
9
4
.
3
7
9
3
.
9
4
9
2
.
9
5
(
a)
(
b
)
Fig
u
r
e
5
.
Per
f
o
r
m
an
c
e
cu
r
v
es o
f
th
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
o
n
th
e
R
FMiD
d
atas
et,
s
h
o
win
g
(
a)
th
e
tr
ain
in
g
an
d
test
in
g
ac
cu
r
ac
y
c
u
r
v
e
an
d
(
b
)
th
e
tr
ai
n
in
g
a
n
d
t
esti
n
g
lo
s
s
cu
r
v
e
3
.
2
.
Co
m
pa
ra
t
iv
e
a
na
ly
s
is
T
h
is
s
ec
tio
n
in
clu
d
es
a
co
m
p
ar
ativ
e
ass
es
s
m
en
t
of
tr
ad
itio
n
al
n
eu
r
al
n
etwo
r
k
s
an
d
th
e
PI
KE
R
-
NE
T
ap
p
r
o
ac
h
.
I
t
h
as
b
ee
n
d
em
o
n
s
tr
ated
to
b
e
m
o
r
e
p
r
o
d
u
ctiv
e
th
an
e
x
is
tin
g
m
eth
o
d
s
as
d
em
o
n
s
tr
ated
b
y
th
e
co
m
p
ar
is
o
n
o
f
p
er
f
o
r
m
an
ce
m
etr
ics
in
clu
d
in
g
p
r
ec
is
io
n
,
s
p
ec
if
icity
,
F1
s
co
r
e,
r
ec
all,
an
d
ac
cu
r
ac
y
.
T
h
e
co
m
p
ar
ativ
e
e
v
alu
atio
n
i
n
v
o
lv
es th
e
ex
is
tin
g
DL
Netwo
r
k
an
d
p
r
o
p
o
s
ed
m
o
d
el.
T
ab
le
2
d
em
o
n
s
tr
ates
th
e
co
m
p
ar
is
o
n
o
f
ex
is
tin
g
ap
p
r
o
ac
h
e
s
an
d
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
.
T
h
e
ac
cu
r
ac
y
o
b
tain
ed
b
y
t
h
e
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
an
d
ex
is
tin
g
m
eth
o
d
s
s
u
ch
as
C
NN,
VGG1
6
,
D
B
N,
an
d
ML
P.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
p
er
f
o
r
m
s
7
.
7
0
%,
5
.
2
1
%,
3
.
3
9
%,
a
n
d
2
.
9
8
%
b
etter
th
an
th
e
co
n
v
en
tio
n
a
l
ap
p
r
o
ac
h
es,
s
u
ch
as
C
NN,
VGG1
6
,
DB
N
an
d
ML
P.
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
o
f
ex
is
tin
g
m
o
d
els with
p
r
o
p
o
s
ed
m
o
d
el
T
e
c
h
n
i
q
u
e
s
A
c
c
u
r
a
c
y
%
S
p
e
c
i
f
i
c
i
t
y
%
P
r
e
c
i
s
i
o
n
%
R
e
c
a
l
l
%
F
1
sc
o
r
e
%
C
N
N
9
1
.
1
2
8
9
.
6
2
8
8
.
0
7
9
1
.
6
3
8
9
.
2
3
V
G
G
1
6
9
3
.
2
8
9
1
.
7
4
9
0
.
9
6
9
2
.
9
5
9
2
.
6
7
D
B
N
9
4
.
9
2
9
3
.
3
1
9
2
.
3
5
9
4
.
0
2
9
3
.
9
1
M
LP
9
5
.
3
0
9
4
.
7
5
9
3
.
6
9
9
5
.
9
3
9
4
.
3
8
P
r
o
p
o
se
d
R
e
s
M
LP
9
8
.
1
4
9
5
.
3
2
9
6
.
2
3
9
7
.
1
4
9
8
.
9
7
T
ab
le
3
s
h
o
ws
th
e
co
m
p
ar
is
o
n
o
f
ex
is
tin
g
m
o
d
els
an
d
th
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
.
T
h
e
ev
id
en
ce
in
clu
d
es
a
co
m
p
r
e
h
en
s
iv
e
ev
alu
atio
n
o
f
th
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
ap
p
r
o
a
ch
u
s
in
g
m
u
ltip
le
p
er
f
o
r
m
an
ce
m
etr
ics
o
n
th
e
R
FMiD
d
ataset.
T
h
e
PIK
E
R
-
N
E
T
ap
p
r
o
ac
h
ac
h
iev
es
a
h
ig
h
er
ac
cu
r
ac
y
r
ate
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S I
n
t J Ro
b
&
Au
to
m
,
Vo
l
.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
6
7
8
-
6
8
9
686
9
8
.
1
4
%
in
class
if
y
in
g
r
etin
al
d
is
ea
s
es,
o
u
tp
er
f
o
r
m
in
g
ex
is
tin
g
m
et
h
o
d
s
b
y
3
.
2
4
%,
4
.
2
4
%,
6
.
2
3
%,
an
d
2
.
0
0
%
co
m
p
ar
ed
to
E
y
eDe
e
p
-
Net,
I
DL
-
MRDD,
Dee
p
Diab
etic,
an
d
Vis
io
n
Dee
p
-
AI
r
esp
ec
tiv
ely
.
T
h
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
m
o
d
el
s
h
o
ws th
at
th
e
co
m
p
ar
is
o
n
o
f
t
h
e
ac
cu
r
a
cy
is
s
u
p
er
io
r
to
t
h
e
ex
is
tin
g
a
p
p
r
o
ac
h
.
T
ab
le
3
.
Co
m
p
a
r
is
o
n
o
f
th
e
ex
is
tin
g
an
d
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
es
Au
th
o
rs
M
e
th
o
d
s
Ac
c
u
ra
c
y
S
e
n
g
a
r
e
t
a
l.
[
1
4
]
Ey
e
De
e
p
-
Ne
t
9
5
.
0
6
%
Va
iy
a
p
u
ri
e
t
a
l.
[1
6
]
IDL
-
M
RDD
9
4
.
1
4
%
Alb
e
laih
i
a
n
d
I
b
ra
h
im [1
8
]
De
e
p
Dia
b
e
ti
c
9
2
.
3
8
%
Jo
sh
i
e
t
a
l.
[
1
9
]
Visio
n
De
e
p
-
AI
9
6
.
2
1
%
P
ro
p
o
se
d
M
CRNET
-
RS
9
8
.
1
4
%
4.
CO
NCLU
SI
O
N
T
h
is
r
esear
ch
p
r
o
p
o
s
ed
a
n
o
v
el
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
f
o
r
ac
cu
r
ate
m
u
lti
-
class
r
etin
al
d
is
ea
s
e
class
if
icatio
n
.
T
h
e
in
p
u
t
f
u
n
d
u
s
im
ag
es
f
r
o
m
th
e
R
FMiD
d
ataset
ar
e
p
r
e
-
p
r
o
ce
s
s
ed
u
s
in
g
a
SC
R
AB
f
ilter
f
o
r
en
h
an
ce
s
im
ag
e
clar
ity
b
y
p
r
eser
v
in
g
e
d
g
es
wh
ile
r
ed
u
cin
g
n
o
is
e.
T
h
e
I
m
R
esNet
-
R
S
in
teg
r
ated
with
T
em
p
o
r
al
Atten
tio
n
is
th
e
n
em
p
lo
y
ed
to
ex
tr
ac
t
d
ee
p
h
i
er
ar
ch
ical
f
ea
t
u
r
es
with
en
h
a
n
ce
d
d
is
cr
im
in
ativ
e
p
o
wer
.
PKO
alg
o
r
ith
m
is
u
ti
lized
f
o
r
f
ea
tu
r
e
s
elec
tio
n
,
ef
f
ec
tiv
ely
r
ed
u
ci
n
g
r
e
d
u
n
d
an
t
in
f
o
r
m
atio
n
wh
ile
r
etain
in
g
th
e
m
o
s
t
r
elev
a
n
t
f
e
atu
r
es.
R
esML
P
is
u
s
ed
to
class
if
y
r
etin
al
d
is
ea
s
es
in
to
AR
MD
,
B
R
VO,
DN,
DR
,
Hea
lth
y
,
an
d
MY
A.
T
h
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
is
ev
alu
ated
u
s
in
g
p
er
f
o
r
m
an
ce
m
etr
ics
s
u
ch
as
a
cc
u
r
ac
y
,
s
p
ec
if
icity
,
r
ec
all,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e.
T
h
e
PI
KE
R
-
NE
T
ap
p
r
o
ac
h
ac
h
iev
es
an
o
v
er
all
ac
cu
r
ac
y
o
f
9
8
.
1
4
%,
an
d
F1
s
co
r
e
o
f
9
7
.
0
6
%.
T
h
e
PIK
E
R
-
NE
T
ap
p
r
o
ac
h
im
p
r
o
v
es
an
o
v
er
all
ac
cu
r
ac
y
b
y
3
.
2
4
%,
4
.
2
4
%,
6
.
2
3
%,
an
d
2
.
0
0
%
co
m
p
ar
ed
t
o
E
y
eDe
ep
-
Net,
I
DL
-
MRDD,
Dee
p
Diab
etic,
an
d
Vis
io
n
Dee
p
-
AI
r
esp
ec
tiv
ely
.
I
n
th
e
f
u
tu
r
e,
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
ca
n
ex
ten
d
to
lar
g
er
m
u
ltimo
d
al
r
etin
al
d
atasets
to
im
p
r
o
v
e
s
tr
o
n
g
a
n
d
g
e
n
er
aliza
tio
n
.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
t
h
o
r
wo
u
ld
lik
e
to
ex
p
r
ess
h
is
h
ea
r
tf
elt
g
r
atitu
d
e
t
o
th
e
s
u
p
er
v
is
o
r
f
o
r
h
is
g
u
i
d
an
ce
a
n
d
u
n
wav
er
in
g
s
u
p
p
o
r
t d
u
r
in
g
t
h
is
r
esear
ch
f
o
r
h
is
g
u
id
an
ce
an
d
s
u
p
p
o
r
t.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT
)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
L
is
s
y
Dev
asah
ay
am
✓
✓
✓
✓
✓
✓
R
am
y
a
Dev
i
Mu
r
u
g
ad
asan
✓
✓
✓
✓
✓
✓
✓
An
an
d
h
i Sam
u
el
Vijay
alak
s
h
m
i
✓
✓
✓
✓
✓
✓
✓
C
h
an
th
iy
a
Pu
h
alen
th
i
✓
✓
✓
✓
✓
✓
✓
R
am
n
ath
Mu
th
u
s
am
y
✓
✓
✓
✓
✓
✓
✓
Ah
ilan
Ap
p
ath
u
r
ai
✓
✓
✓
✓
✓
✓
Nata
r
ajan
Mo
h
an
a
Su
g
an
th
i
✓
✓
✓
✓
✓
✓
✓
✓
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
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S I
n
t J Ro
b
&
Au
to
m
I
SS
N:
2722
-
2
5
8
6
P
I
K
E
R
-
N
E
T:
Mu
lti
-
cla
s
s
r
etin
a
l d
is
ea
s
e
cla
s
s
ifica
tio
n
u
s
in
g
P
ied
K
in
g
fis
h
er
…
(
Lis
s
y
Dev
a
s
a
h
a
ya
m
)
687
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
t
h
o
r
s
d
ec
lar
e
th
at
th
e
y
h
av
e
n
o
k
n
o
wn
c
o
m
p
etin
g
f
in
an
cial
in
ter
ests
o
r
p
er
s
o
n
al
r
el
atio
n
s
h
ip
s
th
at
co
u
ld
h
av
e
ap
p
ea
r
ed
t
o
in
f
lu
en
ce
th
e
wo
r
k
r
e
p
o
r
te
d
in
t
h
is
p
ap
er
.
I
NF
O
RM
E
D
CO
NS
E
N
T
Au
th
o
r
s
ce
r
tify
th
at
au
th
o
r
s
h
av
e
ex
p
lain
ed
th
e
n
atu
r
e
an
d
p
u
r
p
o
s
e
o
f
th
is
s
tu
d
y
to
th
e
ab
o
v
e
-
n
am
e
d
in
d
iv
id
u
al,
a
n
d
a
u
th
o
r
s
h
av
e
d
is
cu
s
s
ed
th
e
p
o
ten
tial
b
en
e
f
i
ts
o
f
th
is
s
tu
d
y
p
ar
ticip
atio
n
.
T
h
e
q
u
esti
o
n
s
th
e
in
d
iv
id
u
al
h
ad
a
b
o
u
t
th
is
s
tu
d
y
h
av
e
b
ee
n
an
s
wer
ed
,
an
d
we
will
a
lway
s
b
e
av
ailab
l
e
to
ad
d
r
ess
f
u
tu
r
e
q
u
esti
o
n
s
.
E
T
H
I
CAL AP
P
RO
V
AL
Ou
r
r
esear
ch
g
u
id
e
r
e
v
iewe
d
a
n
d
eth
ically
ap
p
r
o
v
ed
th
is
m
a
n
u
s
cr
ip
t f
o
r
p
u
b
lis
h
in
g
in
th
is
j
o
u
r
n
al.
DATA AV
AI
L
AB
I
L
I
T
Y
Data
s
h
ar
in
g
n
o
t
ap
p
licab
le
t
o
th
is
ar
ticle
as
n
o
d
atasets
we
r
eg
en
er
ated
o
r
an
al
y
s
ed
d
u
r
in
g
t
h
e
cu
r
r
en
t stu
d
y
.
RE
F
E
R
E
NC
E
S
[
1
]
M
.
E
l
sh
a
r
k
a
w
y
e
t
a
l
.
,
“
O
C
T
-
Tr
a
n
s
:
a
n
o
v
e
l
t
r
a
n
sf
o
r
m
e
r
b
a
c
k
b
o
n
e
w
i
t
h
m
u
l
t
i
m
o
d
a
l
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
i
n
O
C
T
-
b
a
se
d
r
e
t
i
n
a
l
d
i
s
e
a
se
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
2
0
2
5
I
EE
E
2
2
n
d
I
n
t
e
rn
a
t
i
o
n
a
l
S
y
m
p
o
si
u
m
o
n
Bi
o
m
e
d
i
c
a
l
I
m
a
g
i
n
g
(
I
S
BI
)
,
A
p
r
.
2
0
2
5
,
p
p
.
1
–
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
S
B
I
6
0
5
8
1
.
2
0
2
5
.
1
0
9
8
0
7
9
0
.
[
2
]
G
.
R
.
H
e
ma
l
a
k
sh
mi
,
M
.
M
u
r
u
g
a
p
p
a
n
,
M
.
Y
.
S
i
k
k
a
n
d
a
r
,
S
.
S
.
B
e
g
u
m,
a
n
d
N
.
B
.
P
r
a
k
a
sh
,
“
A
u
t
o
ma
t
e
d
r
e
t
i
n
a
l
d
i
s
e
a
se
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
h
y
b
r
i
d
t
r
a
n
sf
o
r
mer
mo
d
e
l
(
S
V
i
T)
u
si
n
g
o
p
t
i
c
a
l
c
o
h
e
r
e
n
c
e
t
o
m
o
g
r
a
p
h
y
i
m
a
g
e
s,
”
N
e
u
ra
l
C
o
m
p
u
t
i
n
g
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
6
,
n
o
.
1
6
,
p
p
.
9
1
7
1
–
9
1
8
8
,
J
u
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
5
2
1
-
0
2
4
-
0
9
5
6
4
-
7.
[
3
]
M
.
El
k
h
o
l
y
a
n
d
M
.
A
.
M
a
r
z
o
u
k
,
“
D
e
e
p
l
e
a
r
n
i
n
g
-
b
a
se
d
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
e
y
e
d
i
se
a
ses
u
si
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
O
C
T
i
ma
g
e
s,
”
F
ro
n
t
i
e
rs
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
5
,
J
a
n
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
8
9
/
f
c
o
mp
.
2
0
2
3
.
1
2
5
2
2
9
5
.
[
4
]
T.
D
.
N
g
u
y
e
n
,
D
.
-
T.
Le
,
J.
B
u
m,
S
.
K
i
m
,
S
.
J.
S
o
n
g
,
a
n
d
H
.
C
h
o
o
,
“
R
e
t
i
n
a
l
d
i
s
e
a
s
e
d
i
a
g
n
o
s
i
s
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
o
n
u
l
t
r
a
-
w
i
d
e
-
f
i
e
l
d
f
u
n
d
u
s i
ma
g
e
s,
”
D
i
a
g
n
o
s
t
i
c
s
,
v
o
l
.
1
4
,
n
o
.
1
,
p
.
1
0
5
,
J
a
n
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
st
i
c
s
1
4
0
1
0
1
0
5
.
[
5
]
S
.
D
.
K
h
a
n
,
S
.
B
a
s
a
l
a
ma
h
,
a
n
d
A
.
L
b
a
t
h
,
“
A
n
o
v
e
l
d
e
e
p
l
e
a
r
n
i
n
g
f
r
a
mew
o
r
k
f
o
r
r
e
t
i
n
a
l
d
i
s
e
a
s
e
d
e
t
e
c
t
i
o
n
l
e
v
e
r
a
g
i
n
g
c
o
n
t
e
x
t
u
a
l
a
n
d
l
o
c
a
l
f
e
a
t
u
r
e
s
c
u
e
s
f
r
o
m
r
e
t
i
n
a
l
i
mag
e
s,
”
Me
d
i
c
a
l
&
Bi
o
l
o
g
i
c
a
l
E
n
g
i
n
e
e
ri
n
g
& C
o
m
p
u
t
i
n
g
,
v
o
l
.
6
3
,
n
o
.
7
,
p
p
.
2
0
2
9
–
2
0
4
6
,
Ju
l
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
5
1
7
-
0
2
5
-
0
3
3
1
4
-
0.
[
6
]
C
.
V
e
n
k
a
t
a
i
a
h
e
t
a
l
.
,
“
A
n
o
v
e
l
e
y
e
d
i
se
a
se
se
g
m
e
n
t
a
t
i
o
n
a
n
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
m
o
d
e
l
u
si
n
g
a
d
v
a
n
c
e
d
d
e
e
p
l
e
a
r
n
i
n
g
n
e
t
w
o
r
k
,
”
Bi
o
m
e
d
i
c
a
l
S
i
g
n
a
l
Pr
o
c
e
s
si
n
g
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
1
0
5
,
p
.
1
0
7
5
6
5
,
J
u
l
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
b
s
p
c
.
2
0
2
5
.
1
0
7
5
6
5
.
[
7
]
M
.
B
e
r
r
i
m
i
a
n
d
A
.
M
o
u
ss
a
o
u
i
,
“
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
i
d
e
n
t
i
f
y
i
n
g
a
n
d
c
l
a
s
si
f
y
i
n
g
r
e
t
i
n
a
l
d
i
s
e
a
s
e
s,
”
i
n
2
0
2
0
2
n
d
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s (I
C
C
I
S
)
,
O
c
t
.
2
0
2
0
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
I
S
4
9
2
4
0
.
2
0
2
0
.
9
2
5
7
6
7
4
.
[
8
]
B
.
R
.
M
.
,
K
.
B
.
V
a
r
d
h
a
n
,
M
.
N
i
d
h
i
s
h
,
S
.
K
i
r
a
n
C
.
,
D
.
N
a
h
i
d
S
h
a
m
e
e
m
,
a
n
d
V
.
S
a
i
C
h
a
r
a
n
,
“
E
y
e
d
i
s
e
a
s
e
d
e
t
e
c
t
i
o
n
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
s
w
i
t
h
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
,
”
I
C
S
T
T
r
a
n
sa
c
t
i
o
n
s
o
n
S
c
a
l
a
b
l
e
I
n
f
o
rm
a
t
i
o
n
S
y
s
t
e
m
s
,
v
o
l
.
1
1
,
Ju
l
.
2
0
2
4
,
d
o
i
:
1
0
.
4
1
0
8
/
e
e
t
s
i
s.
5
9
7
1
.
[
9
]
R
.
S
a
r
k
i
,
K
.
A
h
me
d
,
H
.
W
a
n
g
,
a
n
d
Y
.
Zh
a
n
g
,
“
A
u
t
o
ma
t
e
d
d
e
t
e
c
t
i
o
n
o
f
mi
l
d
a
n
d
m
u
l
t
i
-
c
l
a
ss
d
i
a
b
e
t
i
c
e
y
e
d
i
s
e
a
ses
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
a
n
d
S
y
st
e
m
s
,
v
o
l
.
8
,
n
o
.
1
,
p
.
3
2
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
3
7
5
5
-
0
2
0
-
0
0
1
2
5
-
5.
[
1
0
]
K
.
R
.
A
k
h
i
l
a
,
N
.
M
u
t
h
u
k
u
mara
n
,
a
n
d
A
.
A
h
i
l
a
n
,
“
C
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
c
e
r
v
i
c
a
l
c
a
n
c
e
r
u
s
i
n
g
a
n
a
u
t
o
e
n
c
o
d
e
r
a
n
d
c
a
s
c
a
d
e
d
m
u
l
t
i
l
a
y
e
r
p
e
r
c
e
p
t
r
o
n
,
”
I
ETE
J
o
u
r
n
a
l
o
f
R
e
se
a
rc
h
,
v
o
l
.
7
0
,
n
o
.
1
,
p
p
.
2
6
–
3
6
,
J
a
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
8
0
/
0
3
7
7
2
0
6
3
.
2
0
2
2
.
2
1
4
2
8
5
9
.
[
1
1
]
R
.
S
h
o
b
a
n
a
,
A
.
V
.
K
.
Li
j
a
,
a
n
d
S
.
S
.
M
a
r
i
c
e
,
“
G
D
2
-
EFF
I
N
ET:
G
l
a
u
c
o
m
a
d
i
se
a
se
d
e
t
e
c
t
i
o
n
a
n
d
se
g
me
n
t
a
t
i
o
n
o
f
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
s
u
s
i
n
g
Ef
f
i
c
i
e
n
t
N
e
t
B
0
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
u
rr
e
n
t
B
i
o
-
Me
d
i
c
a
l
En
g
i
n
e
e
r
i
n
g
(
I
J
C
BE)
,
v
o
l
.
1
,
n
o
.
2
,
p
p
.
4
5
–
5
1
,
2
0
2
3
.
[
1
2
]
J.
Je
n
c
y
a
n
d
S
.
S
h
u
n
mu
g
a
n
,
“
D
e
e
p
l
e
a
r
n
i
n
g
-
b
a
se
d
d
i
a
b
e
t
i
c
r
e
t
i
n
o
p
a
t
h
y
f
o
r
c
l
a
ss
i
f
y
i
n
g
r
e
t
i
n
a
l
i
m
a
g
e
s,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
C
u
rr
e
n
t
B
i
o
-
Me
d
i
c
a
l
En
g
i
n
e
e
ri
n
g
(
I
J
C
BE)
,
v
o
l
.
3
,
n
o
.
3
,
p
p
.
5
8
–
6
4
,
2
0
2
5
.
[
1
3
]
A
.
A
h
i
l
a
n
,
M
.
A
n
l
i
n
S
a
h
a
y
a
T
i
n
u
,
A
.
Jasmi
n
e
G
n
a
n
a
M
a
l
a
r
,
a
n
d
B
.
M
u
t
h
u
K
u
m
a
r
,
“
S
t
a
t
i
o
n
a
r
y
w
a
v
e
l
e
t
-
o
r
i
e
n
t
e
d
l
u
mi
n
a
n
c
e
e
n
h
a
n
c
e
me
n
t
a
p
p
r
o
a
c
h
f
o
r
b
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
w
i
t
h
m
u
l
t
i
-
m
o
d
a
l
i
t
y
i
ma
g
e
s,
”
i
n
E
v
o
l
u
t
i
o
n
i
n
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
.
FI
C
T
A
2
0
2
3
.
S
m
a
r
t
I
n
n
o
v
a
t
i
o
n
,
S
y
st
e
m
s
a
n
d
T
e
c
h
n
o
l
o
g
i
e
s
(
S
I
S
T
,
v
o
l
u
m
e
3
7
0
)
,
S
p
r
i
n
g
e
r
,
S
i
n
g
a
p
o
r
e
,
2
0
2
3
,
p
p
.
4
6
1
–
473
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
981
-
99
-
6
7
0
2
-
5
_
3
8
.
[
1
4
]
N
.
S
e
n
g
a
r
,
R
.
C
.
J
o
s
h
i
,
M
.
K
.
D
u
t
t
a
,
a
n
d
R
.
B
u
r
g
e
t
,
“
E
y
e
D
e
e
p
-
N
e
t
:
a
m
u
l
t
i
-
c
l
a
ss
d
i
a
g
n
o
si
s
o
f
r
e
t
i
n
a
l
d
i
s
e
a
s
e
s
u
si
n
g
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
N
e
u
r
a
l
C
o
m
p
u
t
i
n
g
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
5
,
n
o
.
1
4
,
p
p
.
1
0
5
5
1
–
1
0
5
7
1
,
M
a
y
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
5
2
1
-
0
2
3
-
0
8
2
4
9
-
x.
[
1
5
]
N
.
G
o
u
r
a
n
d
P
.
K
h
a
n
n
a
,
“
M
u
l
t
i
-
c
l
a
ss
m
u
l
t
i
-
l
a
b
e
l
o
p
h
t
h
a
l
m
o
l
o
g
i
c
a
l
d
i
s
e
a
se
d
e
t
e
c
t
i
o
n
u
si
n
g
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
b
a
s
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
Bi
o
m
e
d
i
c
a
l
S
i
g
n
a
l
P
ro
c
e
ssi
n
g
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
6
6
,
p
.
1
0
2
3
2
9
,
A
p
r
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
b
s
p
c
.
2
0
2
0
.
1
0
2
3
2
9
.
[
1
6
]
T.
V
a
i
y
a
p
u
r
i
e
t
a
l
.
,
“
I
n
t
e
l
l
i
g
e
n
t
d
e
e
p
l
e
a
r
n
i
n
g
b
a
s
e
d
m
u
l
t
i
-
r
e
t
i
n
a
l
d
i
sea
s
e
d
i
a
g
n
o
si
s
a
n
d
c
l
a
ssi
f
i
c
a
t
i
o
n
f
r
a
m
e
w
o
r
k
,
”
C
o
m
p
u
t
e
rs
,
Ma
t
e
r
i
a
l
s &
C
o
n
t
i
n
u
a
,
v
o
l
.
7
3
,
n
o
.
3
,
p
p
.
5
5
4
3
–
5
5
5
7
,
2
0
2
2
,
d
o
i
:
1
0
.
3
2
6
0
4
/
c
mc.
2
0
2
2
.
0
2
3
9
1
9
.
[
1
7
]
M
.
A
.
R
o
d
r
í
g
u
e
z
,
H
.
A
l
M
a
r
z
o
u
q
i
,
a
n
d
P
.
Li
a
t
s
i
s,
“
M
u
l
t
i
-
l
a
b
e
l
r
e
t
i
n
a
l
d
i
s
e
a
se
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
t
r
a
n
sf
o
r
mers
,
”
I
EEE
J
o
u
r
n
a
l
o
f
Bi
o
m
e
d
i
c
a
l
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
7
,
n
o
.
6
,
p
p
.
2
7
3
9
–
2
7
5
0
,
Ju
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
JB
H
I
.
2
0
2
2
.
3
2
1
4
0
8
6
.
[
1
8
]
A
.
A
l
b
e
l
a
i
h
i
a
n
d
D
.
M
.
I
b
r
a
h
i
m
,
“
D
e
e
p
D
i
a
b
e
t
i
c
:
a
n
i
d
e
n
t
i
f
i
c
a
t
i
o
n
sy
s
t
e
m
o
f
d
i
a
b
e
t
i
c
e
y
e
d
i
s
e
a
s
e
s
u
si
n
g
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
1
0
7
6
9
–
1
0
7
8
9
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
4
.
3
3
5
4
8
5
4
.
[
1
9
]
R
.
C
h
a
n
d
r
a
J
o
s
h
i
,
A
.
K
u
mar
S
h
a
r
ma,
a
n
d
M
.
K
i
sh
o
r
e
D
u
t
t
a
,
“
V
i
si
o
n
D
e
e
p
-
A
I
:
D
e
e
p
l
e
a
r
n
i
n
g
-
b
a
s
e
d
r
e
t
i
n
a
l
b
l
o
o
d
v
e
ss
e
l
s
seg
m
e
n
t
a
t
i
o
n
a
n
d
mu
l
t
i
-
c
l
a
ss
c
l
a
s
si
f
i
c
a
t
i
o
n
f
r
a
m
e
w
o
r
k
f
o
r
e
y
e
d
i
a
g
n
o
si
s,
”
Bi
o
m
e
d
i
c
a
l
S
i
g
n
a
l
Pro
c
e
ss
i
n
g
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
9
4
,
p
.
1
0
6
2
7
3
,
A
u
g
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
b
sp
c
.
2
0
2
4
.
1
0
6
2
7
3
.
[
2
0
]
M
.
Y
a
n
g
,
J
.
D
u
,
a
n
d
R
.
L
v
,
“
C
R
A
T:
A
d
v
a
n
c
e
d
t
r
a
n
sf
o
r
mer
-
b
a
se
d
d
e
e
p
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
ms
i
n
O
C
T
i
ma
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
Bi
o
m
e
d
i
c
a
l
S
i
g
n
a
l
Pr
o
c
e
s
si
n
g
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
1
0
4
,
p
.
1
0
7
5
4
4
,
J
u
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
b
s
p
c
.
2
0
2
5
.
1
0
7
5
4
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.