I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
2
,
A
p
r
il 2
0
2
6
,
p
p
.
1
6
1
3
~
1
6
2
2
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
1
6
1
3
-
1
6
2
2
1613
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Enha
ncing
imba
la
nced data
set
dia
g
no
sis
using
cla
ss
-
ba
sed inpu
t
ima
g
e comp
o
sitio
n
Azz
eddin
e
H
la
li,
M
a
j
id
B
en
Ya
k
hlef
,
So
ula
im
a
n E
l H
a
zz
a
t
En
g
i
n
e
e
r
i
n
g
S
c
i
e
n
c
e
s
P
o
l
y
d
i
s
c
i
p
l
i
n
a
r
y
L
a
b
o
r
a
t
o
r
y
,
F
a
c
u
l
t
y
o
f
Ta
z
a
,
S
i
d
i
M
o
h
a
me
d
B
e
n
A
b
d
e
l
l
a
h
U
n
i
v
e
r
s
i
t
y
(
U
S
M
B
A
)
,
F
e
s,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
8
,
2
0
2
5
R
ev
is
ed
J
an
2
7
,
2
0
2
6
Acc
ep
ted
Feb
6
,
2
0
2
6
De
e
p
lea
rn
in
g
m
o
d
e
ls
o
ften
fa
lt
e
r
wh
e
n
fa
c
e
d
wi
th
sm
a
ll
,
i
m
b
a
lan
c
e
d
d
a
tas
e
ts
o
r
d
e
g
ra
d
e
d
ima
g
e
q
u
a
li
ty
,
lea
d
in
g
to
u
n
a
c
c
e
p
tab
ly
h
ig
h
fa
lse
p
re
d
ictio
n
ra
tes
.
To
b
ri
d
g
e
th
i
s
g
a
p
,
we
in
tr
o
d
u
c
e
c
las
s
-
b
a
se
d
ima
g
e
c
o
m
p
o
siti
o
n
.
T
h
is
tec
h
n
i
q
u
e
re
fo
rm
u
late
s
train
in
g
i
n
p
u
ts
b
y
f
u
sin
g
m
u
lt
ip
le
in
tra
-
c
las
s
ima
g
e
s
in
t
o
u
n
iq
u
e
c
o
m
p
o
si
te
i
n
p
u
t
ima
g
e
s
(C
o
Im
g
).
By
c
o
n
c
e
n
tratin
g
in
f
o
rm
a
ti
o
n
d
e
n
s
it
y
a
n
d
a
m
p
li
f
y
i
n
g
i
n
tra
-
c
las
s
v
a
rian
c
e
,
Co
Im
g
f
o
rc
e
s
th
e
m
o
d
e
l
to
d
is
c
e
rn
su
b
tl
e
,
n
u
a
n
c
e
d
d
ise
a
se
p
a
tt
e
rn
s
th
a
t
m
ig
h
t
o
t
h
e
rwise
b
e
lo
st.
We
v
a
li
d
a
ted
th
is
a
p
p
ro
a
c
h
u
si
n
g
t
h
e
o
p
ti
c
a
l
c
o
h
e
re
n
c
e
to
m
o
g
ra
p
h
y
d
a
tas
e
t
fo
r
ima
g
e
-
b
a
se
d
d
e
e
p
lea
rn
i
n
g
m
e
th
o
d
s
(OCTDL),
a
c
o
ll
e
c
ti
o
n
o
f
se
v
e
n
imb
a
lan
c
e
d
re
ti
n
a
l
d
ise
a
se
sc
a
n
c
a
teg
o
ries
.
F
ro
m
th
is,
we
e
n
g
i
n
e
e
re
d
Co
-
O
CTDL:
a
p
e
rfe
c
tl
y
b
a
lan
c
e
d
v
a
ri
a
n
t
wh
e
re
e
a
c
h
train
i
n
g
sa
m
p
le
e
x
ists
a
s
a
3
×
1
lay
o
u
t
c
o
m
p
o
site.
To
m
e
a
su
re
th
e
imp
a
c
t
o
f
t
h
is
n
e
w
re
p
re
se
n
tatio
n
,
we
b
e
n
c
h
m
a
rk
e
d
th
e
o
ri
g
in
a
l
d
a
tas
e
t
a
g
a
in
st
it
s
c
o
m
p
o
site
c
o
u
n
terp
a
rt
u
sin
g
a
VG
G
1
6
a
rc
h
it
e
c
tu
re
.
P
re
c
isio
n
wa
s
p
a
ra
m
o
u
n
t.
We
m
a
in
tain
e
d
i
d
e
n
t
ic
a
l
h
y
p
e
r
p
a
ra
m
e
ters
a
n
d
m
o
d
e
l
stru
c
tu
re
s
a
c
ro
ss
a
ll
e
x
p
e
rime
n
ts
t
o
e
n
s
u
re
a
rig
o
ro
u
s,
fa
ir
c
o
m
p
a
riso
n
.
T
h
e
re
su
lt
s
we
re
tran
sfo
rm
a
ti
v
e
.
Wh
il
e
b
a
se
li
n
e
d
a
tas
e
ts
stru
g
g
led
,
th
e
e
n
h
a
n
c
e
d
Co
-
OCTDL
a
c
h
iev
e
d
a
n
e
a
r
-
p
e
rfe
c
t
F
1
-
sc
o
re
o
f
0
.
9
9
5
a
n
d
a
n
AU
C
o
f
0
.
9
9
9
6
.
Th
e
m
e
th
o
d
e
ffe
c
ti
v
e
l
y
n
e
u
tralize
d
th
e
risk
s
o
f
c
las
s
imb
a
lan
c
e
.
It
d
id
n
'
t
ju
s
t
imp
ro
v
e
t
h
e
n
u
m
b
e
rs;
it
re
fin
e
d
t
h
e
d
iag
n
o
stic relia
b
il
it
y
o
f
th
e
m
o
d
e
l.
K
ey
w
o
r
d
s
:
C
las
s
im
b
alan
ce
Fals
e
d
iag
n
o
s
tic
m
etr
ics
L
o
w
-
q
u
ality
f
ea
t
u
r
es
Me
d
ical
im
ag
e
an
aly
s
is
R
etin
al
d
is
ea
s
e
cla
s
s
if
icatio
n
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
:
Azz
ed
d
in
e
Hlali
E
n
g
in
ee
r
in
g
Scien
ce
s
Po
ly
d
is
cip
lin
ar
y
L
ab
o
r
ato
r
y
,
Facu
lty
o
f
T
az
a
Sid
i M
o
h
am
ed
B
en
Ab
d
ella
h
Un
iv
er
s
ity
(
USMBA)
Fes
,
Mo
r
o
cc
o
E
m
ail: a
zz
ed
d
in
e.
h
lali@
u
s
m
b
a.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
Dee
p
lear
n
in
g
m
o
d
els
h
av
e
u
n
d
o
u
b
ted
ly
b
ee
n
a
b
r
ea
k
th
r
o
u
g
h
in
m
ed
ical
im
ag
e
an
aly
s
is
,
b
u
t
th
eir
d
iag
n
o
s
tic
p
r
ec
is
io
n
lar
g
ely
d
ep
en
d
s
o
n
th
e
q
u
an
tity
an
d
q
u
ality
o
f
tr
ain
in
g
d
ata
[
1
]
.
T
h
e
d
o
m
ain
is
d
ea
lin
g
with
two
ch
allen
g
in
g
is
s
u
es,
wh
ich
ar
e
th
e
u
n
b
alan
ce
d
n
u
m
b
er
o
f
class
es
,
wh
er
e
th
e
n
u
m
b
er
o
f
p
ath
o
lo
g
ical
ca
s
es
is
co
n
s
id
er
ab
ly
less
th
an
th
at
o
f
h
ea
lth
y
co
n
tr
o
ls
[
2
]
an
d
th
e
f
ac
t
th
at
d
if
f
er
e
n
t
d
is
o
r
d
er
s
v
is
u
all
y
r
esem
b
le
ea
ch
o
th
er
v
er
y
clo
s
ely
[
3
]
.
Su
ch
p
r
o
b
lem
s
o
f
t
en
r
esu
lt
in
e
r
r
o
r
s
i
n
d
iag
n
o
s
is
[
4
]
.
T
h
e
r
etin
al
im
ag
in
g
b
r
an
ch
is
a
ty
p
ical
ex
am
p
le
wh
er
e
th
ese
d
i
f
f
icu
l
ties
ar
e
m
an
if
ested
,
an
d
f
o
r
wh
ich
ag
e
-
r
elate
d
m
ac
u
lar
d
e
g
en
er
atio
n
(
AM
D)
an
d
d
iab
etic
r
etin
o
p
ath
y
(
DR
)
s
h
ar
e
a
co
m
m
o
n
h
is
to
g
r
am
o
f
th
e
m
o
r
p
h
o
lo
g
ical
f
ea
tu
r
es
[
5
]
,
[
6
]
.
T
o
m
itig
ate
th
ese
ch
allen
g
es,
th
e
r
esear
ch
co
m
m
u
n
ity
h
as
ex
p
lo
r
ed
s
tr
ateg
ies
ac
r
o
s
s
th
r
e
e
p
r
im
ar
y
d
im
en
s
io
n
s
:
lo
s
s
m
o
d
u
latio
n
,
d
ata
au
g
m
en
tatio
n
,
an
d
a
r
ch
it
ec
tu
r
al
in
n
o
v
atio
n
.
T
o
a
d
d
r
ess
th
e
im
b
alan
ce
,
lo
s
s
re
-
weig
h
tin
g
tec
h
n
iq
u
es
h
a
v
e
b
ee
n
wid
ely
ad
o
p
ted
.
Ap
p
r
o
a
ch
es
s
u
ch
as
u
n
if
ied
f
o
ca
l
lo
s
s
[
7
]
,
class
-
b
alan
ce
d
lo
s
s
[
8
]
,
an
d
asy
m
m
etr
ic
lo
s
s
[
9
]
attem
p
t
to
m
o
d
u
late
th
e
in
f
lu
en
ce
o
f
ea
s
y
v
er
s
u
s
h
ar
d
ex
am
p
les.
Mo
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
6
1
3
-
1
6
2
2
1614
r
ec
en
tly
,
b
atch
-
b
alan
ce
d
f
o
ca
l
lo
s
s
m
itig
ates
th
e
in
h
er
en
t
cla
s
s
im
b
alan
ce
s
o
f
m
ed
ical
d
atasets
b
y
p
r
io
r
itizin
g
r
ar
e,
clin
ically
s
ig
n
if
ican
t
p
ath
o
lo
g
ical
f
ea
tu
r
es to
s
h
a
r
p
en
d
i
ag
n
o
s
tic
p
r
ec
is
io
n
[
1
0
]
.
I
n
ad
d
itio
n
,
ac
ce
s
s
to
d
atasets
h
as
b
ee
n
e
n
h
an
ce
d
n
o
t
o
n
l
y
th
r
o
u
g
h
d
ata
a
u
g
m
e
n
tatio
n
b
u
t
also
th
r
o
u
g
h
d
ata
s
y
n
th
esizin
g
.
I
t
h
as
b
ee
n
p
r
o
v
e
n
th
at
g
en
er
ati
v
e
ad
v
er
s
ar
ial
n
etwo
r
k
s
(
GANs)
ca
n
b
e
tr
ain
ed
v
er
y
ef
f
ec
tiv
ely
to
g
en
e
r
ate
m
ed
ical
lesi
o
n
s
[
1
1
]
,
[
1
2
]
.
At
t
h
e
s
am
e
tim
e,
tr
a
d
itio
n
al
au
g
m
en
tatio
n
s
in
s
p
ir
ed
b
y
th
e
d
o
m
ai
n
,
lik
e
Mix
Up
an
d
C
u
tMix
d
id
b
r
in
g
im
p
r
o
v
em
en
t
t
o
th
e
m
o
d
el
ca
lib
r
atio
n
[
1
3
]
.
L
ately
,
th
e
f
ield
o
f
m
e
d
ica
l
im
ag
e
g
en
er
atio
n
h
as
s
ee
n
th
e
in
t
r
o
d
u
ctio
n
o
f
d
if
f
u
s
io
n
m
o
d
els
ca
p
ab
le
o
f
g
en
er
atin
g
r
ea
lis
tic
-
lo
o
k
in
g
s
ca
n
s
[
1
4
]
.
On
th
e
ar
ch
itectu
r
al
f
r
o
n
t,
o
n
e
o
f
th
e
m
eth
o
d
s
is
th
e
in
teg
r
atio
n
o
f
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
with
tr
an
s
f
o
r
m
er
s
,
as
s
ee
n
f
r
o
m
th
e
ex
am
p
les
o
f
T
r
an
s
UNe
t
[
1
5
]
an
d
Swin
-
UNe
t
[
1
6
]
;
o
r
tr
ai
n
in
g
a
m
o
d
el
o
n
o
n
e
d
o
m
ain
an
d
th
en
f
in
e
-
tu
n
in
g
it
o
n
an
o
th
er
d
o
m
ain
b
y
u
s
in
g
p
r
e
-
tr
ain
e
d
b
ac
k
b
o
n
es
s
u
ch
as
R
esNet
-
50
[
1
7
]
an
d
E
f
f
icien
t
Net
-
B
3
[
1
8
]
.
Ho
wev
er
,
s
o
m
e
au
th
o
r
s
ar
g
u
e
th
at
tr
an
s
f
er
lear
n
in
g
f
r
o
m
th
e
ty
p
ical
im
ag
e
d
atasets
(
e.
g
.
,
I
m
ag
eNe
t)
m
a
y
n
o
t
b
e
v
er
y
b
en
ef
icial
in
th
e
ca
s
e
o
f
th
e
m
ed
ical
d
o
m
ain
[
1
9
]
.
Ho
wev
er
,
th
ese
m
eth
o
d
s
s
till
h
av
e
lar
g
e
lim
its
.
R
e
-
weig
h
tin
g
th
e
lo
s
s
m
ay
ca
u
s
e
th
e
f
ea
tu
r
es'
d
i
s
tr
ib
u
tio
n
to
b
e
in
co
n
s
is
ten
t,
an
d
th
e
s
y
n
th
esis
o
f
th
e
im
ag
es
m
ig
h
t
in
tr
o
d
u
ce
n
o
n
-
r
ea
lis
tic
ar
tifa
cts.
B
esid
es,
tech
n
iq
u
es
f
o
r
d
ea
lin
g
with
in
ter
-
class
s
i
m
ilar
ity
,
li
k
e
th
e
u
s
e
o
f
s
eg
m
en
tatio
n
to
g
u
i
d
e
class
if
icatio
n
[
2
0
]
o
r
atten
tio
n
-
b
ased
m
o
d
els
[
2
1
]
r
e
q
u
ir
e
a
la
r
g
e
am
o
u
n
t o
f
co
m
p
u
tatio
n
an
d
ar
e
ex
tr
em
ely
d
e
p
en
d
e
n
t o
n
th
e
av
ailab
ilit
y
an
d
q
u
ality
o
f
p
ix
el
-
le
v
el
an
n
o
tatio
n
s
[
2
2
]
.
Mo
r
e
im
p
o
r
tan
tly
,
th
e
m
eth
o
d
s
at
h
an
d
s
tr
u
g
g
le
to
tell
o
n
e
ty
p
e
o
f
d
is
ea
s
e
ap
ar
t f
r
o
m
th
e
o
th
e
r
wh
en
th
e
r
aw
im
a
g
es e
x
h
ib
ited
d
o
n
o
t
co
n
tain
a
n
y
s
u
b
s
tan
tial
o
r
clea
r
f
ea
tu
r
es.
W
e
,
t
h
e
r
e
f
o
r
e
,
p
r
o
p
o
s
e
a
s
i
m
p
le
y
e
t
e
f
f
e
c
t
i
v
e
m
e
t
h
o
d
t
o
b
o
t
h
e
n
r
i
c
h
a
n
d
i
m
p
r
o
v
e
t
h
e
s
o
l
u
t
i
o
n
s
.
C
la
s
s
-
b
a
s
e
d
i
n
p
u
t
i
m
a
g
e
c
o
m
p
o
s
i
t
i
o
n
(
C
B
-
I
m
g
C
o
m
p
)
is
a
t
e
c
h
n
i
q
u
e
t
h
a
t
ess
e
n
t
i
al
l
y
al
t
e
r
s
t
h
e
n
a
t
u
r
e
o
f
t
h
e
i
n
p
u
ts
d
u
r
i
n
g
t
r
a
i
n
i
n
g
.
C
B
-
I
m
g
C
o
m
p
d
e
p
a
r
t
s
f
r
o
m
t
h
e
t
r
a
d
i
t
i
o
n
al
a
u
g
m
e
n
t
a
t
i
o
n
s
o
r
o
v
e
r
s
a
m
p
li
n
g
m
e
t
h
o
d
s
.
I
t
g
o
es
a
b
o
u
t
g
e
n
e
r
a
t
i
n
g
s
t
r
u
c
t
u
r
e
d
l
ay
o
u
t
s
(
e
.
g
.
,
3
×
1
g
r
i
d
s
)
c
al
l
e
d
c
o
m
p
o
s
i
t
e
i
n
p
u
t
i
m
a
g
e
s
(
C
o
-
I
m
g
)
b
y
c
o
m
b
i
n
i
n
g
s
e
v
e
r
a
l
i
m
a
g
e
s
f
r
o
m
t
h
e
s
a
m
e
c
l
a
s
s
.
T
h
is
a
p
p
r
o
a
c
h
h
as
tw
o
m
a
i
n
f
u
n
c
t
i
o
n
s
.
O
n
t
h
e
o
n
e
h
a
n
d
,
i
t
w
o
r
k
s
a
s
a
t
a
r
g
e
t
e
d
a
u
g
m
e
n
t
at
i
o
n
m
et
h
o
d
t
h
a
t
i
s
c
a
p
a
b
l
e
o
f
r
e
-
b
a
la
n
c
i
n
g
t
h
e
n
u
m
b
e
r
o
f
s
a
m
p
l
es
i
n
t
h
e
u
n
d
e
r
-
r
e
p
r
e
s
e
n
t
e
d
c
l
a
s
s
es
.
O
n
t
h
e
o
t
h
e
r
h
a
n
d
,
i
t
i
s
a
v
e
r
y
e
f
f
e
ct
i
v
e
w
a
y
o
f
m
a
k
i
n
g
d
i
s
c
r
i
m
i
n
a
t
i
v
e
s
i
g
n
a
l
s
m
o
r
e
p
r
o
m
i
n
e
n
t
b
y
m
a
x
i
m
i
z
i
n
g
b
o
t
h
c
o
n
c
e
n
t
r
at
i
o
n
a
n
d
p
h
e
n
o
t
y
p
i
c
v
a
r
i
a
b
i
l
it
y
w
i
t
h
i
n
i
n
d
i
v
i
d
u
a
l
t
r
ai
n
i
n
g
s
a
m
p
l
e
s
.
F
e
a
t
u
r
es
f
r
o
m
v
a
r
i
o
u
s
i
m
a
g
es
w
it
h
i
n
a
c
o
m
p
o
s
i
t
e
a
r
e
m
i
x
e
d
,
l
e
a
d
i
n
g
t
o
a
h
i
g
h
d
e
g
r
e
e
o
f
f
e
a
t
u
r
e
d
i
v
e
r
s
i
t
y
,
w
h
i
c
h
is
i
n
s
t
r
u
m
e
n
t
al
i
n
p
r
e
v
e
n
t
i
n
g
t
h
e
o
v
e
r
-
f
i
t
t
i
n
g
o
n
m
i
n
o
r
i
t
y
c
l
a
s
s
e
s
w
it
h
o
u
t
a
n
y
l
o
s
s
o
f
s
e
m
a
n
ti
c
i
n
t
e
g
r
i
t
y
.
O
p
ti
c
a
l
c
o
h
e
r
e
n
c
e
t
o
m
o
g
r
a
p
h
y
(
O
C
T
)
i
m
a
g
e
s
w
il
l
b
e
u
s
e
d
t
o
s
u
b
s
ta
n
t
i
a
t
e
t
h
e
e
x
p
e
r
i
m
e
n
t
s
[
2
3
]
.
A
s
a
g
o
o
d
p
r
o
o
f
o
f
c
o
n
c
e
p
t
,
i
t
is
s
h
o
w
n
t
h
a
t
u
s
i
n
g
C
B
-
I
m
g
C
o
m
p
,
c
l
as
s
i
f
i
c
a
ti
o
n
e
r
r
o
r
s
c
a
n
b
e
s
e
r
i
o
u
s
l
y
l
o
w
e
r
ed
,
t
h
u
s
p
r
o
v
i
d
i
n
g
a
s
t
r
o
n
g
,
y
e
t
l
o
w
-
c
o
m
p
l
e
x
i
t
y
c
h
ec
k
b
o
a
r
d
f
o
r
t
h
e
p
e
r
f
o
r
m
a
n
c
e
g
r
o
w
t
h
o
n
s
m
a
l
l
a
n
d
i
m
b
a
l
a
n
c
e
d
d
a
t
a
s
et
s
.
2.
M
E
T
H
O
D
2
.
1
.
Resea
rc
h desi
g
n a
nd
da
t
a
a
cquis
it
io
n
T
h
e
r
esear
ch
was
m
ain
ly
f
o
c
u
s
ed
o
n
im
p
r
o
v
i
n
g
d
ee
p
lear
n
in
g
m
o
d
els'
d
iag
n
o
s
tic
p
er
f
o
r
m
an
ce
b
y
ad
d
r
ess
in
g
th
e
d
r
awb
ac
k
s
o
f
s
m
all,
im
b
alan
ce
d
d
atasets
an
d
lo
w
-
q
u
ality
in
p
u
ts
th
r
o
u
g
h
a
n
o
v
el
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
e.
W
e
ca
m
e
u
p
with
C
B
-
I
m
g
C
o
m
p
wh
ich
is
a
n
in
p
u
t
-
lev
el
m
eth
o
d
th
at
ch
a
n
g
es
t
h
e
tr
ain
in
g
d
ata
b
y
co
m
b
in
in
g
m
u
ltip
le
im
ag
es
f
r
o
m
th
e
s
am
e
class
in
to
u
n
if
i
ed
v
is
u
al
s
tr
u
ctu
r
es
—
Co
-
I
m
g
.
Delib
er
ate
v
is
u
al
co
m
p
o
s
itio
n
s
d
esig
n
ed
t
o
p
r
e
s
er
v
e
class
s
em
an
tics
wh
ile
a
m
p
lify
in
g
s
tr
u
ct
u
r
al
f
ea
tu
r
es.
W
e
s
o
u
r
ce
d
all
r
aw
s
am
p
les
d
ir
ec
tly
f
r
o
m
th
e
p
u
b
lic
o
p
tical
co
h
er
en
ce
to
m
o
g
r
ap
h
y
d
ataset
f
o
r
im
ag
e
-
b
ased
d
ee
p
lear
n
in
g
m
eth
o
d
s
(
OC
T
DL
)
[
2
3
]
.
T
h
is
d
ataset
in
clu
d
es
2
,
0
6
4
h
ig
h
-
r
eso
lu
tio
n
OC
T
s
ca
n
s
o
f
th
e
h
u
m
a
n
r
etin
a,
ca
teg
o
r
ized
in
to
s
ev
en
d
if
f
er
e
n
t
d
is
ea
s
e
clas
s
es.
Du
e
to
th
e
s
ig
n
if
ican
t
class
im
b
alan
ce
o
f
th
e
d
ataset,
it
was
u
s
ed
as th
e
b
aselin
e
d
ataset
f
o
r
o
u
r
e
x
p
er
im
e
n
t.
2
.
2
.
Resea
rc
h pro
ce
du
re
s
:
cla
s
s
-
ba
s
ed
in
pu
t
im
a
g
e
co
m
p
o
s
it
io
n
T
h
e
C
B
-
I
m
g
C
o
m
p
p
r
o
ce
d
u
r
e
was
ap
p
lied
to
in
cr
ea
s
e
th
e
d
iv
er
s
ity
o
f
d
ata
an
d
t
h
e
in
f
o
r
m
atio
n
d
en
s
ity
p
er
tr
ai
n
in
g
s
am
p
le,
as
illu
s
tr
ated
in
Fig
u
r
e
1
.
T
h
e
p
r
o
ce
s
s
in
v
o
lv
es
g
en
er
atin
g
C
o
-
I
m
g
s
.
T
h
is
m
eth
o
d
aim
s
to
o
b
tain
th
e
m
o
s
t
r
elev
an
t
in
f
o
r
m
atio
n
co
n
ten
t
f
r
o
m
e
ac
h
tr
ain
i
n
g
s
am
p
le
[
2
4
]
.
As
a
r
esu
lt,
th
e
o
r
ig
in
al
d
ataset
is
tr
an
s
f
o
r
m
e
d
in
t
o
a
n
ew
o
n
e
n
a
m
ed
C
o
-
OC
T
DL
,
wh
ich
is
m
a
d
e
u
p
o
f
C
o
-
I
m
g
.
T
h
e
s
ize
o
f
th
ese
Co
-
I
m
g
ca
n
b
e
c
h
an
g
e
d
,
s
o
th
at
d
if
f
er
en
t
m
atr
ix
co
n
f
i
g
u
r
ati
o
n
s
(
e.
g
.
,
3
×
1
,
2
×
5
,
4
×2
)
ca
n
b
e
u
s
ed
d
ep
en
d
in
g
o
n
th
e
m
o
d
el
an
d
th
e
in
itial d
a
taset.
Her
e,
we
h
av
e
ch
o
s
en
a
3
×
1
lay
o
u
t p
atter
n
f
o
r
C
o
-
I
m
g
cr
ea
tio
n
.
2
.
2
.
1
.
CB
-
I
m
g
Co
m
p
f
un
ct
io
n def
ini
t
io
n
T
h
e
C
B
-
I
m
g
C
o
m
p
f
u
n
ctio
n
o
p
er
ates
o
n
th
r
ee
co
r
e
v
a
r
iab
le
s
:
th
e
d
im
en
s
io
n
al
lay
o
u
t
o
f
t
h
e
C
o
-
I
m
g
,
th
e
s
o
u
r
ce
d
ataset,
an
d
a
b
o
o
lean
to
g
g
le
f
o
r
i
n
tr
a
-
co
m
p
o
s
ite
im
ag
e
r
ep
etitio
n
.
In
(
1
)
f
o
r
m
alize
s
th
is
r
elatio
n
s
h
ip
.
B
y
ex
p
licitly
d
ef
in
in
g
th
ese
p
ar
am
eter
s
,
we
r
eg
u
late
th
e
ar
ch
itectu
r
al
c
o
m
p
o
s
itio
n
o
f
ea
ch
tr
ain
in
g
s
am
p
le
to
e
n
s
u
r
e
m
a
x
im
u
m
d
ata
d
i
v
er
s
ity
with
o
u
t r
ed
u
n
d
a
n
t o
v
e
r
lap
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
imb
a
l
a
n
ce
d
d
a
ta
s
e
t d
ia
g
n
o
s
is
u
s
in
g
cla
s
s
-
b
a
s
ed
in
p
u
t ima
g
e
co
mp
o
s
itio
n
(
A
z
z
ed
d
in
e
Hla
li
)
1615
(
,
,
)
=
{
|
=
(
1
,
…
,
;
)
,
=
1
,
…
,
,
∈
,
=
(
)
,
∈
,
≠
=
0
,
=
1
,
…
,
,
=
1
,
…
,
}
(
1
)
W
h
er
e
is
th
e
s
ize
o
f
th
e
C
o
-
I
m
g
o
f
class
c
in
th
e
n
ew
d
atas
et,
L
is
th
e
lay
o
u
t
(
e.
g
.
,
3
×1
)
,
D
is
th
e
o
r
ig
in
al
d
ataset,
∈
{
0
,
1
}
in
d
icate
s
th
e
r
ep
etitio
n
f
lag
,
a
n
d
⊆
is
th
e
s
et
o
f
s
am
p
les f
r
o
m
class
c
.
Fig
u
r
e
1
.
Sy
n
th
esizin
g
in
tr
a
-
cl
ass
s
am
p
les in
to
h
ig
h
-
d
en
s
ity
co
m
p
o
s
ite
tr
ain
in
g
in
p
u
ts
2
.
2
.
2
.
CB
-
I
m
g
Co
m
p
f
un
ct
io
n a
lg
o
rit
hm
W
e
d
esig
n
a
p
r
ec
is
e
alg
o
r
ith
m
ic
p
ip
elin
e
to
ass
em
b
le
th
e
co
m
p
o
s
ite
im
ag
e
d
ata
s
et
—
clea
n
,
r
ep
ea
tab
le,
an
d
tig
h
tly
co
n
tr
o
l
led
f
r
o
m
s
tar
t
to
f
i
n
is
h
.
I
n
s
tea
d
o
f
r
an
d
o
m
a
g
g
r
e
g
atio
n
,
t
h
e
m
eth
o
d
b
u
ild
s
ea
ch
co
m
p
o
s
ite
im
ag
e
f
r
o
m
class
-
s
p
ec
if
ic
s
am
p
les
u
s
in
g
f
ix
ed
s
p
atial
d
im
en
s
io
n
s
,
wh
ile
e
x
p
licitly
m
an
ag
in
g
wh
eth
er
im
a
g
es
m
ay
r
ea
p
p
ea
r
ac
r
o
s
s
co
m
b
in
atio
n
s
o
r
r
em
a
in
u
n
i
q
u
e.
Alg
o
r
ith
m
1
e
n
co
d
es
th
is
lo
g
ic
i
n
th
e
CB
-
I
m
g
C
o
m
p
f
u
n
ctio
n
,
wh
er
e
class
-
wi
s
e
im
ag
e
s
elec
tio
n
a
n
d
s
tr
u
ctu
r
ed
co
m
p
o
s
itio
n
co
n
v
er
g
e
in
to
a
s
in
g
le
g
en
er
atio
n
f
lo
w
th
at
p
r
o
d
u
ce
s
th
e
en
r
ich
ed
d
ataset
D
final
.
Alg
o
r
ith
m
1
.
CB
-
I
m
g
C
o
m
p
f
u
n
ctio
n
Input:
Dataset D, dimensions m × n, repetition flag r
∈
{0,1}
Output:
Enriched dataset D
final
1: Initialize D
final
←
∅
{Empty list for CoImg}
2: k ← m × n {Number of images per CoImg}
3:
for
each class c
∈
D
do
4: N
c
← |D
c
| {Number of images in class c}
5:
if
r = 1
then
6:
Combinations
c
← generate_multicombinations(D
c
, k) {with repetition}
7:
else
8: Combinations
c
← generate_combinations(D
c
, k) {without repetition}
9:
end if
10:
for
each combination W
∈
Combinations
c
do
11: CoImg ←
create_CoImg(W, m, n)
12: D
final
← D
final
∪
{CoImg}
13:
end for
14:
end for
15: return D
final
Help
er
f
u
n
ctio
n
s
:
i)
Gen
er
ate
co
m
b
in
atio
n
s
(
with
o
u
t r
ep
etitio
n
)
function
generate_combinations(
S, k
)
return
all size
-
k
subsets of
S
(no
repetition)
ii)
Gen
er
ate
m
u
lti
-
co
m
b
in
atio
n
s
(
with
r
ep
etitio
n
)
function
generate_multicombinations(S, k)
return
all size
-
k multisets from S (with repetition)
iii)
C
r
ea
te
co
m
p
o
s
ite
im
ag
e
C
o
I
m
g
function
create_CoImg(W, m, n)
CoImg ← initialize_grid(m, n)
for
i
= 1 to m
do
for
j = 1 to n
do
CoImg[i, j] ← W [(i
-
1) · n + j]
end forend for
return
CoImg
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
6
1
3
-
1
6
2
2
1616
T
h
e
s
ch
em
a
as
s
h
o
wn
i
n
Fig
u
r
e
2
,
o
u
tlin
es
th
e
p
r
o
ce
s
s
o
f
cr
ea
tin
g
a
n
ew
d
ataset
r
e
p
r
esen
tatio
n
th
r
o
u
g
h
th
e
C
B
-
I
m
g
C
o
m
p
f
u
n
ctio
n
.
I
t
h
as
led
to
th
e
g
e
n
er
ati
o
n
o
f
a
f
r
esh
d
ata
c
o
llectio
n
w
h
er
e
ev
e
r
y
p
ictu
r
e
is
a
b
len
d
o
f
th
r
ee
p
ictu
r
es
f
r
o
m
th
e
r
aw
d
ataset.
E
ac
h
co
m
p
o
s
ite
im
ag
e
is
th
er
ef
o
r
e
a
f
u
s
io
n
o
f
m
u
ltip
le
elem
en
ts
an
d
tr
an
s
f
er
s
m
o
r
e
s
o
u
r
ce
s
ig
n
als
at
a
tim
e,
r
esu
ltin
g
in
a
m
u
c
h
s
tr
o
n
g
er
f
ea
tu
r
e
s
ig
n
al
th
a
n
a
n
y
o
f
th
e
o
r
ig
in
al
s
in
g
le
im
a
g
es.
Fig
u
r
e
3
p
r
esen
ts
a
f
ew
in
s
tan
ce
s
o
f
im
ag
es f
r
o
m
th
is
n
ew
d
ata
co
llectio
n
.
Fig
u
r
e
2
.
R
ed
ef
in
i
n
g
tr
ain
i
n
g
i
n
p
u
ts
th
r
o
u
g
h
class
-
b
ased
im
a
g
e
s
y
n
th
esis
Fig
u
r
e
3
.
Co
-
OC
DT
L
AM
D
an
d
R
VO
class
es
ex
am
p
les
2
.
3
.
P
er
f
ec
t
ba
la
ncing
cla
s
s
re
presenta
t
io
n
T
h
e
d
ataset
u
n
d
er
g
o
es
co
m
b
i
n
ato
r
ial
ex
p
an
s
io
n
a
f
ter
co
m
p
o
s
ite
-
im
ag
e
g
en
e
r
atio
n
is
ap
p
lied
.
Mo
s
t
im
p
o
r
tan
tly
,
th
e
n
u
m
b
e
r
o
f
i
n
f
o
r
m
ativ
e
s
am
p
les
in
th
e
m
in
o
r
ity
class
in
cr
ea
s
es
s
ig
n
if
i
ca
n
tly
,
th
u
s
it
g
ets
b
etter
r
ep
r
es
en
ted
.
T
h
e
r
ea
s
o
n
b
eh
in
d
th
is
is
th
at
we
co
n
s
id
er
it
as
th
e
b
aselin
e
an
d
all
o
th
er
class
es
m
u
s
t
b
e
alig
n
ed
with
it.
2
.
3
.
1
.
B
a
la
ncing
ba
s
ed
o
n t
he
m
ino
rit
y
cl
a
s
s
Fo
r
th
e
m
ajo
r
ity
o
f
class
es,
we
en
f
o
r
ce
s
tr
ict
n
o
n
-
r
e
u
s
e
d
u
r
in
g
co
m
p
o
s
itio
n
.
E
ac
h
im
a
g
e
ap
p
ea
r
s
o
n
ce
.
No
r
ep
etitio
n
;
th
is
way
,
th
e
n
u
m
b
er
o
f
co
m
p
o
s
ites
o
f
th
ese
class
e
s
i
s
n
atu
r
ally
ca
p
p
ed
,
wh
ile
th
e
m
o
s
t
im
p
o
r
tan
t in
f
o
r
m
atio
n
in
ea
c
h
co
m
p
o
s
ite
is
p
r
eser
v
ed
.
T
o
b
alan
ce
class
es,
we
s
et
a
t
ar
g
et
n
u
m
b
er
o
f
c
o
m
p
o
s
ites
(
T
)
,
wh
ich
c
o
r
r
esp
o
n
d
s
to
t
h
e
am
o
u
n
t
o
f
ex
p
an
s
io
n
ac
h
iev
ed
f
o
r
th
e
m
in
o
r
ity
class
.
C
o
m
p
o
s
ites
f
o
r
th
e
r
est
o
f
th
e
class
es
ar
e
p
r
o
d
u
ce
d
u
n
d
e
r
th
e
f
o
llo
win
g
p
r
in
cip
le:
i)
W
e
en
f
o
r
ce
a
s
tr
ict
"n
o
-
r
e
p
etitio
n
"
p
o
licy
with
in
ea
ch
C
o
I
m
g
co
m
p
o
s
ite
to
en
s
u
r
e
ev
er
y
tr
ain
in
g
s
am
p
le
p
r
esen
ts
a
u
n
iq
u
e
v
is
u
al
co
n
te
x
t.
Ho
wev
er
,
d
ata
s
ca
r
city
o
cc
asio
n
ally
d
is
r
u
p
ts
th
is
.
ii)
Sh
o
u
ld
a
m
ajo
r
ity
class
f
ail
t
o
r
ea
ch
th
e
tar
g
et
T
u
n
d
er
th
ese
r
ig
id
co
n
s
tr
ain
ts
,
we
b
r
id
g
e
th
e
g
a
p
b
y
in
teg
r
atin
g
th
e
r
e
m
ain
in
g
a
v
ai
lab
le
s
ca
n
s
,
s
u
b
jecte
d
to
lig
h
t
g
eo
m
etr
ic
an
d
p
h
o
to
m
etr
ic
a
u
g
m
en
tatio
n
s
lik
e
s
u
b
tle
r
o
tatio
n
s
,
tr
a
n
s
latio
n
s
,
o
r
co
n
tr
ast
s
h
if
ts
,
to
c
o
m
p
lete
th
e
lay
o
u
t.
T
h
is
b
ala
n
ce
b
etwe
en
r
aw
d
ata
in
teg
r
ity
an
d
s
y
n
th
etic
v
a
r
iety
.
T
h
is
p
r
o
ce
s
s
g
u
ar
an
tees
th
at
all
class
es
ev
en
tu
ally
h
it
th
e
s
am
e
g
o
al,
h
en
ce
a
p
e
r
f
ec
tly
b
alan
ce
d
d
ata
s
et
is
o
b
tain
ed
wh
ile
th
e
co
n
ten
t
r
e
m
ain
s
in
f
o
r
m
ativ
e
a
n
d
d
iv
er
s
e,
as
s
h
o
wn
in
T
ab
le
1
,
wh
er
e
ea
ch
d
is
ea
s
e
class
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
imb
a
l
a
n
ce
d
d
a
ta
s
e
t d
ia
g
n
o
s
is
u
s
in
g
cla
s
s
-
b
a
s
ed
in
p
u
t ima
g
e
co
mp
o
s
itio
n
(
A
z
z
ed
d
in
e
Hla
li
)
1617
(
e.
g
.
,
AM
D
an
d
d
iab
etic
m
ac
u
lar
ed
em
a
(
DM
E
)
)
is
eith
er
ex
p
an
d
ed
o
r
r
ed
u
ce
d
u
p
to
n
o
n
-
r
ep
etitiv
e
co
m
p
o
s
ites
o
r
a
s
lig
h
t in
cr
ea
s
e
.
T
ab
le
1
.
Dis
tr
ib
u
tio
n
o
f
g
en
er
ated
s
am
p
les ac
r
o
s
s
r
etin
al
d
is
ea
s
e
class
e
s
D
i
sea
s
e
c
l
a
ss
R
a
w
s
c
a
n
s
C
o
I
mg
N
e
w
r
e
p
r
e
se
n
t
a
t
i
o
n
o
f
d
a
t
a
se
t
A
g
e
-
r
e
l
a
t
e
d
ma
c
u
l
a
r
d
e
g
e
n
e
r
a
t
i
o
n
1
,
2
3
1
3
1
0
,
1
4
4
,
295
M
a
j
o
r
i
t
y
c
l
a
ss
e
s.
Ex
p
a
n
d
/
r
e
d
u
c
e
u
p
t
o
T(
1
,
5
4
0
)
D
i
a
b
e
t
i
c
mac
u
l
a
r
e
d
e
ma
1
4
7
5
1
8
,
6
6
5
Ep
i
r
e
t
i
n
a
l
m
e
m
b
r
a
n
e
(
E
R
M
)
1
5
5
6
0
8
,
6
8
5
N
o
r
mal
3
3
2
6
,
0
4
4
,
0
6
0
R
e
t
i
n
a
l
v
e
i
n
o
c
c
l
u
si
o
n
(
R
V
O
)
1
0
1
1
6
6
,
6
5
0
V
i
t
r
e
o
ma
c
u
l
a
r
i
n
t
e
r
f
a
c
e
d
i
sea
se
(
V
I
D
)
76
70
,
300
R
e
t
i
n
a
l
a
r
t
e
r
y
o
c
c
l
u
s
i
o
n
(
R
A
O
)
22
1
,
5
4
0
M
i
n
o
r
i
t
y
c
l
a
ss.
M
i
n
i
ma
l
n
u
m
b
e
r
o
f
sa
mp
l
e
s (T
=1
,
5
4
0
)
.
To
t
a
l
2
,
0
6
4
3
1
7
,
5
5
4
,
195
10
,
7
8
0
(
T×
7
)
2
.
3
.
2
.
M
ino
rit
y
-
cla
s
s
-
driv
en
ba
la
ncing
a
lg
o
rit
hm
C
las
s
im
b
alan
ce
d
is
to
r
ts
lear
n
in
g
,
esp
ec
ially
wh
en
co
m
p
o
s
ite
d
ata
am
p
lifie
s
th
e
d
o
m
in
an
ce
o
f
alr
ea
d
y
o
v
e
r
r
ep
r
esen
te
d
class
e
s
.
Naiv
e
co
m
p
o
s
ite
g
en
er
atio
n
d
o
es e
x
ac
tly
th
at:
it m
u
ltip
lies
wh
at
is
ab
u
n
d
an
t,
s
tar
v
es
wh
at
is
r
ar
e,
an
d
tr
ai
n
s
th
e
m
o
d
el
o
n
a
s
k
ewe
d
r
e
ality
th
at
lo
o
k
s
s
tatis
tically
r
ich
b
u
t
s
em
a
n
tically
p
o
o
r
.
Alg
o
r
ith
m
2
b
r
ea
k
s
th
is
p
atter
n
b
y
en
f
o
r
cin
g
a
s
tr
ict
eq
u
ilib
r
iu
m
:
i
t
f
ix
es
th
e
n
u
m
b
e
r
o
f
co
m
p
o
s
ites
p
er
class
u
s
in
g
f
ea
s
ib
le
k
-
co
m
b
in
atio
n
s
,
n
o
t
co
n
v
en
ien
ce
o
r
av
ailab
ilit
y
,
f
o
r
cin
g
th
e
d
ataset
to
g
r
o
w
s
y
m
m
etr
ically
r
ath
er
th
a
n
o
r
g
an
ically
.
T
h
is
s
tr
ateg
y
p
r
eser
v
es
class
d
iv
er
s
ity
b
y
r
eta
in
in
g
ev
er
y
s
ca
n
,
av
o
id
in
g
th
e
d
ata
lo
s
s
ty
p
icall
y
r
eq
u
ir
e
d
f
o
r
p
a
r
ity
.
Fig
u
r
e
4
illu
s
tr
ates
th
e
r
esu
ltin
g
C
o
-
OC
T
DL
alig
n
m
en
t,
a
p
er
f
ec
tly
b
ala
n
ce
d
d
ataset.
Alg
o
r
ith
m
2
.
B
alan
ce
d
co
-
d
ataset
g
en
er
atio
n
Input
: A dataset D with N classes, where class i has N
i
images; CoImg size k
Output
: A
balanced dataset where each class has T images (CoImg)
1: Use the procedure defined in Algorithm 1 as a CBCFunction
2: T ← min
i=1,…,N
(
)
{Baseline: minimum possible k
-
combinations across classes}
3:
for
each class C with N
c
← |C|
do
4: M ←
(
)
{Total unique k
-
combinations in class C}
5:
if
M > T
then
6: CBCFunction(k, D, 0) {Generate without intra
-
composite repetition}
7:
else
8:
CB
CF
un
ct
io
n(
k
,
D,
1)
{G
en
er
at
e
al
l
un
i
qu
e
co
mp
os
it
es
;
if
T
is
n
ot
re
ac
he
d
,
complete with
slight augmentation}
9:
end if
10: Store the resulting T composite candidates (Co
-
Img) for class C
11:
end for
12: Merge all class
-
specific T composite sets into the final balanced dataset
13: return Final balanced dataset
Fig
u
r
e
4
.
B
en
ch
m
ar
k
in
g
b
asel
in
e
p
er
f
o
r
m
an
ce
a
g
ain
s
t th
e
b
a
lan
ce
d
Co
-
OC
T
DL
v
ar
ian
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
6
1
3
-
1
6
2
2
1618
2
.
4
.
E
x
perim
ent
a
l
s
et
up
a
nd
t
esting
T
o
v
alid
ate
C
B
-
I
m
g
C
o
m
p
,
we
co
n
s
tr
u
cted
a
r
ig
o
r
o
u
s
co
m
p
ar
ativ
e
f
r
am
ewo
r
k
.
W
e
s
tr
ip
p
ed
awa
y
co
n
f
o
u
n
d
in
g
v
ar
iab
les.
T
h
is
c
o
n
tr
o
l
g
u
ar
an
tees
th
at
o
b
s
er
v
ed
p
e
r
f
o
r
m
an
ce
g
ain
s
s
tem
s
o
lely
f
r
o
m
s
u
p
er
io
r
d
ata
r
ep
r
esen
tatio
n
,
n
o
t
f
r
o
m
a
r
ch
itectu
r
al
q
u
ir
k
s
o
r
tr
ai
n
in
g
an
o
m
alies.
2
.
4
.
1
.
I
m
plem
ent
a
t
i
o
n f
ra
m
e
wo
rk
T
o
g
u
ar
an
tee
co
m
p
u
tatio
n
al
r
ep
r
o
d
u
cib
ilit
y
,
we
im
p
lem
en
ted
th
e
p
ip
elin
e
u
s
in
g
Py
T
o
r
ch
with
in
th
e
Kag
g
le
en
v
ir
o
n
m
e
n
t.
W
e
o
p
tim
ized
th
e
c
o
n
f
ig
u
r
atio
n
to
p
r
o
ce
s
s
h
ig
h
-
r
eso
lu
tio
n
co
m
p
o
s
ite
im
ag
es
ef
f
icien
tly
.
C
r
u
cially
,
we
m
ain
tain
ed
s
tr
ict
alg
o
r
ith
m
ic
co
n
s
is
ten
cy
th
r
o
u
g
h
o
u
t th
e
v
alid
ati
o
n
p
r
o
ce
s
s
.
2
.
4
.
2
.
Da
t
a
s
et
c
o
nfig
ura
t
io
n
T
o
m
ea
s
u
r
e
th
e
s
p
ec
if
ic
i
m
p
ac
t o
f
o
u
r
ap
p
r
o
ac
h
,
we
co
m
p
ar
e
d
two
d
is
tin
ct
d
ataset
co
n
f
ig
u
r
atio
n
s
:
i)
B
aselin
e
co
n
f
ig
u
r
atio
n
:
we
u
t
ilized
th
e
r
aw
OC
T
DL
d
atase
t
[
2
3
]
.
C
o
m
p
r
is
in
g
2
,
0
6
4
OC
T
s
ca
n
s
ac
r
o
s
s
s
ev
en
im
b
alan
ce
d
class
es,
th
is
s
et
is
u
n
m
o
d
if
ied
.
W
e
ap
p
lied
n
o
b
alan
cin
g
tech
n
iq
u
e
s
.
I
t
f
u
n
ctio
n
s
s
tr
ictly
as a
co
n
tr
o
l b
en
c
h
m
ar
k
.
ii)
E
x
p
er
im
en
tal
co
n
f
ig
u
r
atio
n
(
C
o
-
OC
T
DL
)
:
th
e
p
r
o
p
o
s
ed
d
ataset
g
e
n
er
ated
v
ia
t
h
e
C
B
-
I
m
g
C
o
m
p
alg
o
r
ith
m
.
T
h
is
co
n
f
i
g
u
r
atio
n
co
m
p
r
is
es
1
0
,
7
8
0
co
m
p
o
s
i
te
im
ag
es,
s
tr
u
ctu
r
ed
i
n
a
3
×1
lay
o
u
t.
As
d
etailed
in
s
u
b
-
s
ec
tio
n
2
.
3
,
th
i
s
d
ataset
i
s
p
er
f
ec
tly
b
alan
ce
d
,
with
ea
ch
o
f
th
e
s
ev
en
class
es
n
o
r
m
alize
d
to
a
tar
g
et
co
u
n
t o
f
T
=1
,
5
4
0
s
am
p
les.
2
.
4
.
3
.
M
o
del a
rc
hite
ct
ure
a
nd
t
ra
ini
ng
T
o
en
s
u
r
e
a
v
a
l
i
d
c
o
m
p
ar
i
s
o
n
,
t
h
e
s
a
m
e
V
G
G1
6
n
e
t
w
o
r
k
wa
s
u
s
e
d
f
o
r
b
o
t
h
d
a
ta
s
e
t
c
o
n
f
i
g
u
r
a
t
io
n
s
.
T
h
i
s
a
r
ch
i
t
e
c
tu
r
e
w
a
s
ch
o
s
e
n
to
b
e
i
n
l
i
n
e
w
i
t
h
t
h
e
m
e
t
h
o
d
u
s
e
d
i
n
th
e
o
r
ig
i
n
a
l
O
C
T
D
L
d
a
t
a
s
e
t
b
e
n
c
h
m
a
r
k
[
2
3
]
,
th
u
s
m
a
k
in
g
i
t
a
c
o
n
s
i
s
t
e
n
t
b
a
s
e
l
i
n
e
f
o
r
ev
a
l
u
a
t
i
o
n
.
E
a
c
h
o
f
t
h
e
m
o
d
e
l
s
w
a
s
t
r
a
i
n
ed
w
i
t
h
th
e
s
a
m
e
s
e
t
o
f
h
y
p
e
r
p
a
r
am
e
t
e
r
s
to
r
e
m
a
i
n
t
h
e
o
n
l
y
v
a
r
i
ab
l
e
f
o
r
th
e
ch
a
n
g
in
g
in
p
u
t
d
a
t
a
c
o
n
f
i
g
u
r
a
t
io
n
(
2
4
4
a
s
i
n
p
u
t
s
i
z
e
,
1
0
0
ep
o
ch
s
,
6
4
b
a
t
ch
s
i
z
e
,
A
d
am
a
s
o
p
t
im
i
z
a
t
i
o
n
a
lg
o
r
i
th
m
,
0
.
0
0
0
3
a
s
l
e
a
r
n
in
g
r
a
t
e,
an
d
c
r
o
s
s
e
n
tr
o
p
y
a
s
l
o
s
s
f
u
n
c
t
io
n
)
.
2
.
4
.
4
.
E
v
a
lua
t
io
n m
e
t
rics
W
e
q
u
an
tifie
d
d
iag
n
o
s
tic
r
elia
b
ilit
y
u
s
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
,
F1
-
s
co
r
e
,
an
d
AUC.
On
to
p
o
f
th
at,
f
o
r
b
o
th
r
u
n
s
,
co
n
f
u
s
io
n
m
atr
ices
wer
e
cr
ea
t
ed
with
th
e
in
ten
tio
n
o
f
v
is
u
alizin
g
class
-
s
p
ec
if
ic
m
is
class
if
icatio
n
s
an
d
e
v
alu
at
in
g
th
e
d
e
g
r
ee
to
w
h
ich
i
n
ter
-
class
co
n
f
u
s
io
n
was
r
ed
u
ce
d
.
Fig
u
r
e
5
s
h
o
ws
t
h
e
co
m
p
lete
ex
p
er
im
en
tal
p
r
o
ce
d
u
r
e,
f
r
o
m
in
itial
d
ata
in
p
u
t
an
d
C
o
-
OC
T
DL
cr
ea
tio
n
,
t
h
r
o
u
g
h
t
o
VGG
-
1
6
tr
ain
in
g
an
d
co
m
p
ar
is
o
n
ev
al
u
atio
n
.
Fig
u
r
e
5
.
E
x
p
er
im
e
n
t
wo
r
k
f
lo
w
s
ch
em
a
f
o
r
C
B
-
I
m
g
C
o
m
p
v
alid
atio
n
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
Resul
t
s
As
in
d
icate
d
in
T
ab
le
2
,
t
h
e
m
o
d
el
tr
ain
e
d
o
n
C
o
-
OC
T
DL
d
ata
r
ea
ch
es
n
ea
r
-
p
e
r
f
ec
t
p
e
r
f
o
r
m
an
ce
,
with
ac
cu
r
ac
y
clim
b
in
g
to
9
9
.
7
0
%
an
d
th
e
F1
-
s
co
r
e
s
tab
ilizin
g
at
0
.
9
9
7
,
a
m
ar
g
in
t
h
at
d
e
cisi
v
ely
s
ep
ar
ates
it
f
r
o
m
th
e
b
aselin
e
m
o
d
el,
w
h
ich
r
em
ai
n
s
co
n
s
tr
ain
e
d
at
8
5
.
9
%
ac
cu
r
ac
y
an
d
0
.
8
6
9
F1
-
s
co
r
e
.
T
h
is
g
a
p
is
n
o
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
imb
a
l
a
n
ce
d
d
a
ta
s
e
t d
ia
g
n
o
s
is
u
s
in
g
cla
s
s
-
b
a
s
ed
in
p
u
t ima
g
e
co
mp
o
s
itio
n
(
A
z
z
ed
d
in
e
Hla
li
)
1619
co
s
m
etic,
it
r
ef
lects
a
s
tr
u
ctu
r
al
im
p
r
o
v
e
m
en
t
in
r
e
p
r
esen
t
atio
n
lear
n
in
g
,
wh
er
e
C
o
-
OC
T
DL
r
esh
ap
es
th
e
f
ea
tu
r
e
s
p
ac
e
r
ath
er
th
a
n
m
e
r
ely
in
cr
ea
s
in
g
s
am
p
le
v
o
lu
m
e.
T
h
e
m
o
s
t
tellin
g
s
h
if
t
ap
p
ea
r
s
in
th
e
AUC,
wh
ich
r
is
es
f
r
o
m
0
.
9
7
7
to
0
.
9
9
9
6
,
s
ig
n
alin
g
an
al
m
o
s
t
id
ea
l
s
ep
ar
atio
n
b
o
u
n
d
ar
y
b
etw
ee
n
r
etin
al
d
is
ea
s
e
ca
teg
o
r
ies.
I
n
p
r
ac
tical
ter
m
s
,
th
e
m
o
d
el
n
o
lo
n
g
er
ap
p
r
o
x
im
ates
class
d
is
tin
ctio
n
s
;
it
r
eso
lv
es
th
em
with
n
ea
r
-
d
eter
m
i
n
is
tic
p
r
ec
is
io
n
.
T
ab
le
2
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
o
f
th
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
a
g
ain
s
t th
e
b
aselin
e
M
e
t
h
o
d
A
c
c
u
r
a
c
y
F1
-
sc
o
r
e
AUC
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
B
a
se
l
i
n
e
(
V
G
G
1
6
-
O
C
TD
L)
0
.
8
5
9
0
.
8
6
9
0
.
9
7
7
0
.
8
8
8
0
.
8
5
9
CB
-
I
mg
C
o
mp
(
V
G
G
1
6
-
Co
-
O
C
TD
L)
0
.
9
9
7
0
3
2
0
.
9
9
7
0
3
6
0
.
9
9
9
6
4
3
0
.
9
9
7
0
7
9
0
.
9
9
7
0
3
2
3
.
1
.
1
.
Ana
ly
s
is
o
f
m
is
cla
s
s
if
ica
t
io
n e
rr
o
rs
Fig
u
r
e
6
s
h
o
ws
th
e
c
o
m
p
a
r
ativ
e
an
aly
s
is
o
f
co
n
f
u
s
io
n
m
at
r
ices
.
T
h
e
b
aselin
e
m
o
d
el
,
as
s
h
o
wn
i
n
Fig
u
r
e
6
(
a)
[
2
3
]
s
h
o
ws
th
e
s
p
r
ea
d
o
f
m
is
class
if
icatio
n
s
,
es
p
ec
ially
am
o
n
g
class
es
th
at
ar
e
s
u
p
p
o
s
ed
ly
m
o
s
t
s
im
ilar
;
f
o
r
ex
am
p
le,
th
e
co
n
f
u
s
io
n
b
etwe
en
R
AO
an
d
R
V
O
is
s
h
o
wn
,
as
wel
l
as
co
n
f
u
s
io
n
b
etwe
en
AM
D
an
d
n
o
r
m
al
(
NO)
class
es.
B
y
co
m
p
ar
is
o
n
,
th
e
m
o
d
el
with
o
u
r
C
B
-
I
m
g
C
o
m
p
m
eth
o
d
,
as
s
h
o
wn
in
Fig
u
r
e
6
(
b
)
h
as
a
n
ea
r
ly
p
e
r
f
ec
t
d
ia
g
o
n
al
p
atter
n
r
e
p
r
esen
tin
g
a
tr
e
m
e
n
d
o
u
s
d
ec
r
ea
s
e
in
m
is
tak
es.
Pra
ctica
lly
,
th
e
f
alse
p
r
ed
ictio
n
s
wer
e
k
ep
t
to
n
e
g
lig
ib
le
o
u
tlier
s
,
o
n
ly
a
f
ew
ca
s
es
l
ik
e
DM
E
wr
o
n
g
ly
id
en
tifie
d
as
R
AO
(
5
s
am
p
les)
an
d
E
R
M
as
R
V
O
(
2
s
am
p
les)
h
en
ce
,
th
e
m
eth
o
d
'
s
s
u
cc
es
s
in
s
o
lv
in
g
th
e
b
aselin
e's
lim
ita
tio
n
o
f
f
ea
tu
r
e
am
b
i
g
u
ity
is
th
er
eb
y
c
er
tifie
d
.
(
a)
(
b
)
Fig
u
r
e
6
.
C
o
m
p
a
r
ativ
e
an
aly
s
i
s
o
f
co
n
f
u
s
io
n
m
atr
ices f
o
r
(
a)
th
e
b
aselin
e
m
o
d
el
a
n
d
(
b
)
th
e
p
r
o
p
o
s
ed
CB
-
I
m
g
C
o
m
p
ap
p
r
o
a
ch
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
6
1
3
-
1
6
2
2
1620
3
.
1
.
2
.
Adv
a
nta
g
es o
f
t
he
new
da
t
a
s
et
W
e
co
m
b
in
e
s
ev
er
al
im
ag
es
o
f
th
e
s
am
e
class
th
at
b
elo
n
g
to
d
if
f
er
e
n
t
s
am
p
les
th
r
o
u
g
h
m
e
r
g
in
g
an
d
th
u
s
in
t
h
is
way
,
th
e
s
am
p
les
t
h
at
wer
e
in
itially
n
o
t
as
s
tr
o
n
g
b
ec
o
m
e
s
tr
en
g
th
en
ed
b
y
th
e
s
tr
o
n
g
o
n
es
an
d
th
e
n
ew
r
esu
ltin
g
co
m
p
o
s
ites
th
at
h
av
e
a
l
o
t
o
f
f
ea
tu
r
es
d
r
asti
ca
lly
im
p
r
o
v
e
th
e
tr
ai
n
in
g
p
r
o
ce
s
s
.
T
h
is
m
eth
o
d
ess
en
tially
o
v
er
co
m
es
th
e
p
r
o
b
lem
s
o
f
h
av
in
g
a
s
m
all
n
u
m
b
er
o
f
d
ata,
o
r
l
o
w
-
q
u
ality
im
a
g
es,
an
d
at
th
e
s
am
e
tim
e,
it g
iv
es th
e
m
o
d
el
a
p
o
w
er
f
u
l in
p
u
t th
at
r
esu
lts
in
b
ette
r
g
en
er
aliza
tio
n
.
i)
B
alan
ce
d
class
es
o
f
d
ataset:
by
s
ettin
g
ev
er
y
class
to
a
f
ix
ed
tar
g
et
co
u
n
t
(
T
)
,
th
e
d
ataset
h
as
b
ee
n
m
ad
e
p
er
f
ec
tly
b
alan
ce
d
,
th
is
way
m
ajo
r
ity
-
class
b
ias
is
co
m
p
letely
r
em
o
v
ed
,
an
d
d
iv
er
s
e
c
o
n
ten
t
is
s
til
l
p
r
eser
v
ed
as c
an
b
e
s
ee
n
clea
r
ly
in
(
T
ab
le
1
)
.
ii)
E
n
h
an
cin
g
v
ar
iab
ilit
y
: e
ac
h
Co
-
I
m
g
f
u
s
es m
u
ltip
le
im
ag
es t
o
co
n
s
o
lid
ate
r
elev
an
t v
is
u
al
cu
es a
n
d
en
r
ich
in
f
o
r
m
ativ
e
co
n
ten
t,
en
ab
lin
g
th
e
m
o
d
el
to
ex
tr
ac
t
s
tr
o
n
g
er
,
m
o
r
e
g
en
er
aliza
b
le
f
ea
tu
r
es
f
o
r
im
p
r
o
v
ed
lear
n
in
g
.
Fig
u
r
e
3
d
is
p
lay
s
ex
am
p
le
co
m
p
o
s
ite
im
ag
es
f
o
r
t
h
e
AM
D
an
d
R
VO
cla
s
s
e
s
w
ith
in
th
e
n
ew
d
ataset
r
ep
r
esen
tatio
n
.
3
.
2
.
Dis
cus
s
io
n
T
h
e
VGG1
6
p
er
f
o
r
m
an
c
e
o
n
C
o
-
OC
T
DL
u
p
en
d
s
th
e
cu
r
r
en
t
f
ix
atio
n
o
n
co
m
p
le
x
ar
c
h
itectu
r
al
m
o
d
if
icatio
n
s
.
I
n
p
u
t
-
lev
el
o
p
tim
izatio
n
p
r
o
v
es
its
elf
to
b
e
th
e
s
u
p
er
io
r
s
tr
ateg
y
.
Stan
d
ar
d
a
u
g
m
en
tatio
n
o
f
te
n
f
ails
to
b
r
id
g
e
th
e
s
em
an
tic
g
ap
b
etwe
en
m
in
o
r
ity
an
d
m
aj
o
r
ity
class
es.
C
B
-
Img
C
o
m
p
s
u
cc
ee
d
s
.
I
t
d
o
es
n
o
t
s
im
p
ly
r
ep
licate
d
ata;
it
c
o
n
s
tr
u
cts
s
em
an
tically
d
en
s
e
r
e
p
r
esen
tatio
n
s
.
T
h
ese
f
o
r
ce
t
h
e
n
etwo
r
k
to
lear
n
r
o
b
u
s
t
class
m
ar
k
er
s
r
ath
e
r
th
an
b
ac
k
g
r
o
u
n
d
n
o
is
e.
T
h
is
p
r
ec
is
io
n
er
a
d
icate
s
m
is
clas
s
if
icatio
n
b
etwe
en
clin
ically
m
ir
r
o
r
in
g
p
ath
o
lo
g
ie
s
,
s
p
ec
if
ically
R
AO
an
d
R
VO.
Ou
r
co
m
p
o
s
ite
ap
p
r
o
ac
h
p
r
eser
v
es
f
in
e
-
g
r
ain
ed
v
ascu
la
r
d
etails
o
f
te
n
o
b
liter
ated
b
y
s
tan
d
ar
d
d
o
wn
-
s
am
p
lin
g
.
B
y
s
tab
ilizin
g
th
e
tr
ain
in
g
d
is
tr
ib
u
tio
n
,
we
p
r
o
v
id
e
a
c
o
n
cr
ete
b
lu
e
p
r
in
t
f
o
r
d
ata
-
s
ca
r
ce
en
v
ir
o
n
m
en
ts
.
T
h
e
leap
f
r
o
m
0
.
8
5
9
to
0
.
9
9
7
ac
c
u
r
ac
y
is
n
o
t
ju
s
t
a
s
tatis
tica
l
g
ain
.
I
t
m
ar
k
s
a
d
ef
in
itiv
e
tr
an
s
itio
n
f
r
o
m
ex
p
er
im
e
n
tal
u
n
ce
r
tain
ty
to
clin
ical
u
tili
ty
.
3
.
2
.
1
.
Clini
ca
l r
elev
a
nce
a
nd
im
pa
ct
On
e
o
f
th
e
way
s
C
B
-
I
m
g
C
o
m
p
in
cr
ea
s
es
th
e
tr
u
s
two
r
th
in
ess
o
f
d
iag
n
o
s
tics
(
wh
ich
m
ed
ic
al
AI
tr
u
s
t
d
ep
en
d
s
o
n
)
is
b
y
g
r
ea
tly
cu
ttin
g
d
o
wn
o
n
th
e
n
u
m
b
er
o
f
f
alse
p
o
s
itiv
e
an
d
f
alse
n
eg
ativ
e
r
esu
lts
.
T
h
e
en
h
an
ce
d
p
r
ec
is
io
n
an
d
r
ec
all
im
p
ly
th
at
th
e
m
o
d
el
is
q
u
ite
s
tu
r
d
y
.
Hen
ce
,
it
is
ca
p
ab
le
o
f
s
af
ely
m
in
im
izin
g
er
r
o
r
r
ates
,
wh
ich
is
o
f
u
tm
o
s
t im
p
o
r
tan
ce
f
o
r
p
atien
ts
'
s
af
ety
[
2
5
]
,
[
2
6
]
.
3
.
2
.
2
.
I
m
pli
ca
t
io
ns
o
f
f
ind
ing
s
Ad
d
r
ess
in
g
s
m
all
an
d
im
b
ala
n
ce
d
d
atasets
:
cr
u
cially
,
t
h
e
t
ec
h
n
iq
u
e
th
r
iv
es
o
n
s
m
all,
u
n
b
alan
ce
d
d
atasets
.
Fu
n
ctio
n
in
g
as
a
s
e
m
an
tic
au
g
m
en
tatio
n
[
2
7
]
m
e
th
o
d
,
C
B
-
I
m
g
C
o
m
p
b
alan
ce
s
class
e
s
d
ir
ec
tly
at
th
e
in
p
u
t
lev
el.
T
h
is
s
tr
ateg
y
o
u
tp
er
f
o
r
m
s
s
tan
d
ar
d
al
g
o
r
ith
m
ic
f
ix
es
lik
e
weig
h
ted
lo
s
s
f
u
n
ctio
n
s
.
B
y
s
u
p
p
ly
in
g
h
ig
h
-
f
id
elity
,
f
e
atu
r
e
-
r
ich
ex
am
p
les,
we
ac
h
ie
v
e
s
tab
le
tr
ain
in
g
an
d
s
u
p
e
r
io
r
g
en
e
r
aliza
tio
n
f
o
r
th
e
m
in
o
r
ity
class
.
Min
im
izin
g
f
alse
d
iag
n
o
s
tic
r
esu
lts
:
clin
ical
u
tili
ty
r
eq
u
ir
es
m
in
im
izin
g
d
iag
n
o
s
tic
er
r
o
r
s
[
2
6
]
.
B
y
s
h
ar
p
en
in
g
p
ath
o
lo
g
ical
r
ep
r
e
s
en
tatio
n
s
,
we
en
ab
led
th
e
m
o
d
el
to
ac
h
ie
v
e
s
u
p
er
io
r
d
is
cr
im
in
ativ
e
p
o
wer
.
T
h
is
r
ed
u
ctio
n
in
u
n
ce
r
tain
ty
is
p
iv
o
tal.
I
t
s
u
b
s
tan
tiates
r
ea
l
-
wo
r
ld
d
e
p
lo
y
m
e
n
t,
s
tr
ictly
lim
itin
g
th
e
r
is
k
s
o
f
m
is
s
ed
d
iag
n
o
s
es o
r
u
n
n
ec
ess
ar
y
tr
ea
tm
en
ts
.
4.
CO
NCLU
SI
O
N
W
e
b
y
p
a
s
s
ed
t
h
e
c
o
n
s
t
r
a
in
t
s
o
f
s
m
a
l
l,
i
m
b
a
l
a
n
c
e
d
m
e
d
i
c
a
l
d
a
t
a
s
e
t
s
b
y
i
n
tr
o
d
u
c
i
n
g
C
B
-
I
m
g
C
o
m
p
.
T
h
i
s
m
e
th
o
d
s
y
n
th
e
s
i
z
e
s
in
f
o
r
m
a
t
io
n
-
d
en
s
e
,
b
a
l
an
c
ed
t
r
a
in
i
n
g
s
a
m
p
l
e
s
.
T
h
e
i
m
p
a
c
t
w
a
s
d
i
s
t
i
n
c
t
.
U
s
i
n
g
s
t
an
d
ar
d
V
G
G
1
6
ar
c
h
i
t
e
c
t
u
r
e
o
n
th
e
O
C
T
D
L
d
a
t
as
e
t
,
w
e
a
c
h
i
ev
ed
9
9
.
7
%
ac
cu
r
a
cy
an
d
0
.
9
9
9
6
A
U
C
.
W
e
d
e
l
i
b
e
r
a
t
e
ly
p
r
i
o
r
i
ti
z
e
d
i
n
p
u
t
o
p
t
im
i
z
a
t
i
o
n
o
v
e
r
a
r
c
h
i
t
ec
t
u
r
a
l
c
o
m
p
l
ex
i
t
y
.
T
h
es
e
r
e
s
u
l
t
s
c
o
n
f
i
r
m
t
h
a
t
r
e
f
in
i
n
g
d
a
t
a
r
e
p
r
e
s
e
n
ta
ti
o
n
i
s
a
p
o
t
en
t
,
m
o
d
e
l
-
a
g
n
o
s
t
i
c
a
l
t
e
r
n
a
t
iv
e
to
en
g
in
e
er
i
n
g
h
e
av
y
n
e
t
wo
r
k
s
.
O
u
r
ap
p
r
o
a
c
h
e
s
t
a
b
l
i
s
h
e
s
a
v
i
ab
l
e
p
a
th
f
o
r
d
ep
l
o
y
in
g
r
o
b
u
s
t
AI
in
d
a
t
a
-
s
ca
r
c
e
e
n
v
i
r
o
n
m
e
n
t
s
.
F
u
t
u
r
e
in
v
e
s
t
i
g
a
t
io
n
s
w
i
l
l
ex
t
e
n
d
b
ey
o
n
d
O
C
T
i
m
a
g
er
y
.
W
e
i
n
t
en
d
to
v
a
l
id
a
t
e
g
e
n
er
a
l
i
z
ab
i
l
i
t
y
o
n
m
a
g
n
e
t
i
c
r
e
s
o
n
an
c
e
i
m
a
g
in
g
(
M
R
I
)
a
n
d
h
i
s
to
p
a
th
o
lo
g
y
a
n
d
a
s
s
e
s
s
i
n
t
e
g
r
a
t
io
n
w
i
th
e
m
er
g
i
n
g
T
r
a
n
s
f
o
r
m
e
r
-
b
a
s
e
d
a
r
ch
i
t
e
c
tu
r
e
s
.
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
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
imb
a
l
a
n
ce
d
d
a
ta
s
e
t d
ia
g
n
o
s
is
u
s
in
g
cla
s
s
-
b
a
s
ed
in
p
u
t ima
g
e
co
mp
o
s
itio
n
(
A
z
z
ed
d
in
e
Hla
li
)
1621
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
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Azz
ed
d
in
e
Hlali
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ma
jid
B
en
Yak
h
lef
✓
✓
✓
✓
✓
✓
✓
So
u
laim
an
E
l H
az
za
t
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
ts
th
e
f
i
n
d
in
g
s
o
f
th
is
s
tu
d
y
a
r
e
o
p
e
n
ly
a
v
ailab
le
in
Kag
g
le.
–
T
h
e
ex
p
e
r
im
en
tal
d
ataset
is
ac
ce
s
s
ib
le
at
h
ttp
s
://w
ww.
k
ag
g
le.
co
m
/d
atasets
/azd
in
eh
/c
-
d
ataset
-
2025
–
T
h
e
b
aselin
e
d
ata
s
et
(
OC
T
DL
)
is
av
ailab
le
at
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/o
r
v
ile/o
ctd
l
-
o
p
tical
-
co
h
er
en
ce
-
to
m
o
g
r
ap
h
y
-
d
atase
t
RE
F
E
R
E
NC
E
S
[
1
]
M
.
J.
W
i
l
l
e
m
i
n
k
e
t
a
l
.
,
“
P
r
e
p
a
r
i
n
g
m
e
d
i
c
a
l
i
m
a
g
i
n
g
d
a
t
a
f
o
r
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
,
”
R
a
d
i
o
l
o
g
y
,
v
o
l
.
2
9
5
,
n
o
.
1
,
p
p
.
4
–
1
5
,
A
p
r
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
4
8
/
r
a
d
i
o
l
.
2
0
2
0
1
9
2
2
2
4
.
[
2
]
J.
M
.
Jo
h
n
so
n
a
n
d
T
.
M
.
K
h
o
s
h
g
o
f
t
a
a
r
,
“
S
u
r
v
e
y
o
n
d
e
e
p
l
e
a
r
n
i
n
g
w
i
t
h
c
l
a
ss
i
mb
a
l
a
n
c
e
,
”
J
o
u
r
n
a
l
o
f
Bi
g
D
a
t
a
,
v
o
l
.
6
,
n
o
.
1
,
D
e
c
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
0
5
3
7
-
0
1
9
-
0
1
9
2
-
5.
[
3
]
J.
Z
h
a
n
g
,
Y
.
X
i
e
,
Q
.
W
u
,
a
n
d
Y
.
X
i
a
,
“
M
e
d
i
c
a
l
i
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
s
y
n
e
r
g
i
c
d
e
e
p
l
e
a
r
n
i
n
g
,
”
Me
d
i
c
a
l
I
m
a
g
e
A
n
a
l
y
s
i
s
,
v
o
l
.
5
4
,
p
p
.
1
0
–
1
9
,
M
a
y
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
m
e
d
i
a
.
2
0
1
9
.
0
2
.
0
1
0
.
[
4
]
S
.
M
.
A
n
w
a
r
,
M
.
M
a
j
i
d
,
A
.
Q
a
y
y
u
m
,
M
.
A
w
a
i
s,
M
.
A
l
n
o
w
a
mi
,
a
n
d
M
.
K
.
K
h
a
n
,
“
M
e
d
i
c
a
l
i
ma
g
e
a
n
a
l
y
si
s
u
s
i
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
s
:
a
r
e
v
i
e
w
,
”
J
o
u
r
n
a
l
o
f
M
e
d
i
c
a
l
S
y
st
e
m
s
,
v
o
l
.
4
2
,
n
o
.
1
1
,
N
o
v
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
9
1
6
-
0
1
8
-
1
0
8
8
-
1.
[
5
]
J.
M
o
r
a
n
o
,
Á
.
S
.
H
e
r
v
e
l
l
a
,
J
.
R
o
u
c
o
,
J.
N
o
v
o
,
J.
I
.
F
.
-
V
i
g
o
,
a
n
d
M
.
O
r
t
e
g
a
,
“
I
mp
r
o
v
i
n
g
A
M
D
d
i
a
g
n
o
s
i
s
b
y
t
h
e
s
i
m
u
l
t
a
n
e
o
u
s
i
d
e
n
t
i
f
i
c
a
t
i
o
n
o
f
a
ss
o
c
i
a
t
e
d
r
e
t
i
n
a
l
l
e
s
i
o
n
s
,
”
i
n
I
m
a
g
e
A
n
a
l
y
si
s
a
n
d
Pr
o
c
e
ssi
n
g
–
I
C
I
A
P
2
0
2
2
,
S
p
r
i
n
g
e
r
,
C
h
a
m
,
2
0
2
2
,
p
p
.
1
4
8
–
1
5
9
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
031
-
0
6
4
2
7
-
2
_
1
3
.
[
6
]
H
.
G
h
e
b
r
e
c
h
r
i
s
t
o
s,
G
.
A
l
a
g
h
b
a
n
d
,
a
n
d
R
.
Y
.
H
w
a
n
g
,
“
R
e
t
i
N
e
t
—
f
e
a
t
u
r
e
e
x
t
r
a
c
t
o
r
f
o
r
l
e
a
r
n
i
n
g
p
a
t
t
e
r
n
s
o
f
d
i
a
b
e
t
i
c
r
e
t
i
n
o
p
a
t
h
y
a
n
d
a
g
e
-
r
e
l
a
t
e
d
ma
c
u
l
a
r
d
e
g
e
n
e
r
a
t
i
o
n
f
r
o
m
p
u
b
l
i
c
l
y
a
v
a
i
l
a
b
l
e
d
a
t
a
s
e
t
s,
”
i
n
2
0
1
7
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
S
c
i
e
n
c
e
a
n
d
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
(
C
S
C
I
)
,
D
e
c
.
2
0
1
7
,
p
p
.
1
6
4
3
–
1
6
4
8
,
d
o
i
:
1
0
.
1
1
0
9
/
C
S
C
I
.
2
0
1
7
.
2
8
6
.
[
7
]
M
.
Y
e
u
n
g
,
E.
S
a
l
a
,
C
.
-
B
.
S
c
h
ö
n
l
i
e
b
,
a
n
d
L.
R
u
n
d
o
,
“
U
n
i
f
i
e
d
f
o
c
a
l
l
o
ss:
g
e
n
e
r
a
l
i
si
n
g
d
i
c
e
a
n
d
c
r
o
ss e
n
t
r
o
p
y
-
b
a
s
e
d
l
o
ss
e
s t
o
h
a
n
d
l
e
c
l
a
ss
i
m
b
a
l
a
n
c
e
d
me
d
i
c
a
l
i
m
a
g
e
seg
m
e
n
t
a
t
i
o
n
,
”
C
o
m
p
u
t
e
r
i
ze
d
M
e
d
i
c
a
l
I
m
a
g
i
n
g
a
n
d
G
ra
p
h
i
c
s
,
v
o
l
.
9
5
,
Ja
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
m
p
m
e
d
i
ma
g
.
2
0
2
1
.
1
0
2
0
2
6
.
[
8
]
Y
.
C
u
i
,
M
.
J
i
a
,
T
.
-
Y
.
Li
n
,
Y
.
S
o
n
g
,
a
n
d
S
.
B
e
l
o
n
g
i
e
,
“
C
l
a
ss
-
b
a
l
a
n
c
e
d
l
o
ss
b
a
se
d
o
n
e
f
f
e
c
t
i
v
e
n
u
m
b
e
r
o
f
sa
mp
l
e
s,”
i
n
2
0
1
9
I
EEE/
C
VF
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
si
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
J
u
n
.
2
0
1
9
,
p
p
.
9
2
6
0
–
9
2
6
9
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
9
.
0
0
9
4
9
.
[
9
]
T.
R
i
d
n
i
k
e
t
a
l
.
,
“
A
sy
mm
e
t
r
i
c
l
o
ss
f
o
r
m
u
l
t
i
-
l
a
b
e
l
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
i
n
2
0
2
1
I
EE
E/
C
VF
I
n
t
e
r
n
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
Vi
si
o
n
(
I
C
C
V)
,
O
c
t
.
2
0
2
1
,
p
p
.
8
2
–
91
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
4
8
9
2
2
.
2
0
2
1
.
0
0
0
1
5
.
[
1
0
]
T.
H
u
a
n
g
e
t
a
l
.
,
“
M
a
s
k
e
d
d
i
st
i
l
l
a
t
i
o
n
w
i
t
h
r
e
c
e
p
t
i
v
e
t
o
k
e
n
s,
”
i
n
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
L
e
a
rn
i
n
g
Re
p
res
e
n
t
a
t
i
o
n
s
(
I
C
L
R
2
0
2
3
)
,
2
0
2
3
.
[
1
1
]
M
.
F
.
-
A
d
a
r
,
I
.
D
i
a
ma
n
t
,
E.
K
l
a
n
g
,
M
.
A
mi
t
a
i
,
J.
G
o
l
d
b
e
r
g
e
r
,
a
n
d
H
.
G
r
e
e
n
s
p
a
n
,
“
G
A
N
-
b
a
s
e
d
s
y
n
t
h
e
t
i
c
me
d
i
c
a
l
i
m
a
g
e
a
u
g
me
n
t
a
t
i
o
n
f
o
r
i
n
c
r
e
a
se
d
C
N
N
p
e
r
f
o
r
ma
n
c
e
i
n
l
i
v
e
r
l
e
si
o
n
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
N
e
u
ro
c
o
m
p
u
t
i
n
g
,
v
o
l
.
3
2
1
,
p
p
.
3
2
1
–
3
3
1
,
D
e
c
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m.
2
0
1
8
.
0
9
.
0
1
3
.
[
1
2
]
M
.
N
.
F
e
k
r
i
,
A
.
M
.
G
h
o
s
h
,
a
n
d
K
.
G
r
o
l
i
n
g
e
r
,
“
G
e
n
e
r
a
t
i
n
g
e
n
e
r
g
y
d
a
t
a
f
o
r
mac
h
i
n
e
l
e
a
r
n
i
n
g
w
i
t
h
r
e
c
u
r
r
e
n
t
g
e
n
e
r
a
t
i
v
e
a
d
v
e
r
sar
i
a
l
n
e
t
w
o
r
k
s,”
E
n
e
r
g
i
e
s
,
v
o
l
.
1
3
,
n
o
.
1
,
D
e
c
.
2
0
1
9
,
d
o
i
:
1
0
.
3
3
9
0
/
e
n
1
3
0
1
0
1
3
0
.
[
1
3
]
A
.
R
a
o
,
J
.
-
Y
.
Le
e
,
a
n
d
O
.
A
a
l
a
mi
,
“
S
t
u
d
y
i
n
g
t
h
e
i
m
p
a
c
t
o
f
a
u
g
me
n
t
a
t
i
o
n
s o
n
me
d
i
c
a
l
c
o
n
f
i
d
e
n
c
e
c
a
l
i
b
r
a
t
i
o
n
,
”
i
n
2
0
2
3
I
EEE
/
C
VF
I
n
t
e
r
n
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
Vi
s
i
o
n
W
o
rks
h
o
p
s
(
I
C
C
V
W)
,
O
c
t
.
2
0
2
3
,
p
p
.
2
4
5
4
–
2
4
6
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
W
6
0
7
9
3
.
2
0
2
3
.
0
0
2
6
0
.
[
1
4
]
S
.
P
a
n
e
t
a
l
.
,
“
S
y
n
t
h
e
t
i
c
C
T
g
e
n
e
r
a
t
i
o
n
f
r
o
m
M
R
I
u
s
i
n
g
3
D
t
r
a
n
sf
o
r
m
e
r
‐
b
a
se
d
d
e
n
o
i
si
n
g
d
i
f
f
u
s
i
o
n
m
o
d
e
l
,
”
Me
d
i
c
a
l
P
h
y
si
c
s
,
v
o
l
.
5
1
,
n
o
.
4
,
p
p
.
2
5
3
8
–
2
5
4
8
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
2
/
mp
.
1
6
8
4
7
.
[
1
5
]
J.
C
h
e
n
e
t
a
l
.
,
“
Tr
a
n
sU
N
e
t
:
t
r
a
n
sf
o
r
mers
ma
k
e
st
r
o
n
g
e
n
c
o
d
e
r
s
f
o
r
m
e
d
i
c
a
l
i
ma
g
e
s
e
g
me
n
t
a
t
i
o
n
,
”
2
0
2
1
,
a
r
X
i
v
:
2
1
0
2
.
0
4
3
0
6
.
[
1
6
]
H
.
C
a
o
e
t
a
l
.
,
“
S
w
i
n
-
U
n
e
t
:
U
n
e
t
-
l
i
k
e
p
u
r
e
t
r
a
n
sf
o
r
mer
f
o
r
me
d
i
c
a
l
i
m
a
g
e
s
e
g
m
e
n
t
a
t
i
o
n
,
”
i
n
C
o
m
p
u
t
e
r
V
i
si
o
n
–
EC
C
V
2
0
2
2
Wo
r
k
sh
o
p
s
,
2
0
2
3
,
p
p
.
2
0
5
–
2
1
8
.
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
1
-
2
5
0
6
6
-
8
_
9
.
[
1
7
]
E.
H
a
ss
a
n
e
t
a
l
.
,
“
E
n
h
a
n
c
e
d
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
f
o
r
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
r
e
t
i
n
a
l
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
mag
e
s,”
S
e
n
so
rs
,
v
o
l
.
2
3
,
n
o
.
1
2
,
J
u
n
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
s
2
3
1
2
5
3
9
3
.
[
1
8
]
N
.
J.
B
i
l
a
l
a
n
d
R
.
Z
.
Y
o
u
s
i
f
,
“
N
e
u
t
r
o
so
p
h
i
c
-
i
n
s
p
i
r
e
d
d
a
t
a
a
u
g
me
n
t
a
t
i
o
n
a
n
d
m
u
l
t
i
-
a
r
c
h
i
t
e
c
t
u
r
e
e
n
s
e
mb
l
e
m
o
d
e
l
f
o
r
r
o
b
u
st
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
,
”
E
n
g
i
n
e
e
ri
n
g
R
e
se
a
rc
h
E
x
p
r
e
ss
,
v
o
l
.
7
,
n
o
.
3
,
S
e
p
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
8
8
/
2
6
3
1
-
8
6
9
5
/
a
d
f
c
2
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
6
1
3
-
1
6
2
2
1622
[
1
9
]
M
.
R
a
g
h
u
,
C
.
Z
h
a
n
g
,
J.
K
l
e
i
n
b
e
r
g
,
a
n
d
S
.
B
e
n
g
i
o
,
“
Tr
a
n
sf
u
si
o
n
:
u
n
d
e
r
st
a
n
d
i
n
g
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
f
o
r
me
d
i
c
a
l
i
ma
g
i
n
g
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
3
rd
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
N
e
u
ra
l
I
n
f
o
rm
a
t
i
o
n
P
ro
c
e
ssi
n
g
S
y
st
e
m
s
,
2
0
1
9
,
p
p
.
3
3
4
7
–
3
3
5
7
.
[
2
0
]
W
.
Li
e
t
a
l
.
,
“
I
n
t
e
r
p
r
e
t
a
b
l
e
d
e
t
e
c
t
i
o
n
o
f
d
i
a
b
e
t
i
c
r
e
t
i
n
o
p
a
t
h
y
,
r
e
t
i
n
a
l
v
e
i
n
o
c
c
l
u
si
o
n
,
a
g
e
-
r
e
l
a
t
e
d
mac
u
l
a
r
d
e
g
e
n
e
r
a
t
i
o
n
,
a
n
d
o
t
h
e
r
f
u
n
d
u
s c
o
n
d
i
t
i
o
n
s,
”
D
i
a
g
n
o
s
t
i
c
s
,
v
o
l
.
1
4
,
n
o
.
2
,
J
a
n
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
s
t
i
c
s
1
4
0
2
0
1
2
1
.
[
2
1
]
Z.
Li
e
t
a
l
.
,
“
A
d
e
e
p
-
l
e
a
r
n
i
n
g
p
i
p
e
l
i
n
e
f
o
r
t
h
e
d
i
a
g
n
o
s
i
s
a
n
d
g
r
a
d
i
n
g
o
f
c
o
mm
o
n
b
l
i
n
d
i
n
g
o
p
h
t
h
a
l
mi
c
d
i
se
a
ses
b
a
s
e
d
o
n
l
e
si
o
n
-
f
o
c
u
se
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
mo
d
e
l
,
”
Fr
o
n
t
i
e
rs i
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
7
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
8
9
/
f
r
a
i
.
2
0
2
4
.
1
4
4
4
1
3
6
.
[
2
2
]
M
.
D
r
i
b
a
n
,
A
.
Y
a
n
,
A
.
S
e
l
v
a
m,
J.
O
n
g
,
K
.
K
.
V
u
p
p
a
r
a
b
o
i
n
a
,
a
n
d
J
.
C
h
h
a
b
l
a
n
i
,
“
A
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
i
n
c
h
o
r
i
o
r
e
t
i
n
a
l
p
a
t
h
o
l
o
g
y
t
h
r
o
u
g
h
f
u
n
d
o
s
c
o
p
y
:
a
c
o
mp
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Re
t
i
n
a
a
n
d
V
i
t
r
e
o
u
s
,
v
o
l
.
1
0
,
n
o
.
1
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
0
9
4
2
-
024
-
0
0
5
54
-
4.
[
2
3
]
M
.
K
u
l
y
a
b
i
n
e
t
a
l
.
,
“
O
C
TD
L
:
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
d
a
t
a
s
e
t
f
o
r
i
mag
e
-
b
a
se
d
d
e
e
p
l
e
a
r
n
i
n
g
me
t
h
o
d
s,”
S
c
i
e
n
t
i
f
i
c
D
a
t
a
,
v
o
l
.
1
1
,
n
o
.
1
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
7
-
0
2
4
-
0
3
1
8
2
-
7.
[
2
4
]
Y
.
G
u
o
,
Y
.
G
a
o
,
a
n
d
D
.
S
h
e
n
,
“
D
e
f
o
r
ma
b
l
e
M
R
p
r
o
st
a
t
e
s
e
g
m
e
n
t
a
t
i
o
n
v
i
a
d
e
e
p
f
e
a
t
u
r
e
l
e
a
r
n
i
n
g
a
n
d
s
p
a
r
se
p
a
t
c
h
mat
c
h
i
n
g
,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
M
e
d
i
c
a
l
I
m
a
g
i
n
g
,
v
o
l
.
3
5
,
n
o
.
4
,
p
p
.
1
0
7
7
–
1
0
8
9
,
A
p
r
.
2
0
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
TM
I
.
2
0
1
5
.
2
5
0
8
2
8
0
.
[
2
5
]
S
.
C
h
a
t
i
o
,
N
.
A
.
A
n
sa
h
,
D
.
A
.
A
w
u
n
i
,
A
.
O
d
u
r
o
,
a
n
d
P
.
O
.
A
n
sah
,
“
C
o
mm
u
n
i
t
y
a
c
c
e
p
t
a
b
i
l
i
t
y
o
f
sea
s
o
n
a
l
mal
a
r
i
a
c
h
e
mo
p
r
e
v
e
n
t
i
o
n
o
f
m
o
r
b
i
d
i
t
y
a
n
d
m
o
r
t
a
l
i
t
y
i
n
y
o
u
n
g
c
h
i
l
d
r
e
n
:
a
q
u
a
l
i
t
a
t
i
v
e
st
u
d
y
i
n
t
h
e
u
p
p
e
r
W
e
s
t
R
e
g
i
o
n
o
f
G
h
a
n
a
,
”
PLO
S
O
N
E
,
v
o
l
.
1
4
,
n
o
.
5
,
M
a
y
2
0
1
9
,
d
o
i
:
1
0
.
1
3
7
1
/
j
o
u
r
n
a
l
.
p
o
n
e
.
0
2
1
6
4
8
6
.
[
2
6
]
E.
J.
T
o
p
o
l
,
“
H
i
g
h
-
p
e
r
f
o
r
ma
n
c
e
me
d
i
c
i
n
e
:
t
h
e
c
o
n
v
e
r
g
e
n
c
e
o
f
h
u
ma
n
a
n
d
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
,
”
N
a
t
u
re
Me
d
i
c
i
n
e
,
v
o
l
.
2
5
,
n
o
.
1
,
p
p
.
4
4
–
5
6
,
Ja
n
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
5
9
1
-
0
1
8
-
0
3
0
0
-
7.
[
2
7
]
Y
.
Z
h
u
,
X
.
C
a
i
,
X
.
W
a
n
g
,
X
.
C
h
e
n
,
Z
.
F
u
,
a
n
d
Y
.
Y
a
o
,
“
B
S
D
A
:
B
a
y
e
s
i
a
n
r
a
n
d
o
m
se
ma
n
t
i
c
d
a
t
a
a
u
g
me
n
t
a
t
i
o
n
f
o
r
med
i
c
a
l
i
m
a
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
S
e
n
s
o
rs
,
v
o
l
.
2
4
,
n
o
.
2
3
,
N
o
v
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
s
2
4
2
3
7
5
1
1
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Az
z
e
d
d
in
e
H
la
li
a
P
h
.
D.
stu
d
e
n
t
in
In
tell
ig
e
n
t
S
y
ste
m
s
a
n
d
Bio
m
e
d
ica
l
In
fo
rm
a
ti
c
s
,
h
e
is
a
M
a
ste
r’s
in
In
telli
g
e
n
t
a
n
d
M
o
b
il
e
S
y
ste
m
s
fro
m
S
id
i
M
o
h
a
m
e
d
Be
n
Ab
d
e
ll
a
h
Un
i
v
e
rsity
(2
0
2
1
).
Hi
s
re
se
a
rc
h
targ
e
ts
th
e
in
ters
e
c
ti
o
n
o
f
AI
a
n
d
h
e
a
lt
h
c
a
re
,
sp
e
c
ifi
c
a
ll
y
e
x
a
m
in
in
g
b
io
m
e
d
i
c
a
l
ima
g
in
g
a
n
d
e
lec
tro
n
ic
h
e
a
lt
h
re
c
o
rd
s.
P
re
v
io
u
sly
,
h
e
d
e
p
lo
y
e
d
d
e
e
p
lea
rn
in
g
m
o
d
e
ls
f
o
r
COV
ID
-
1
9
p
n
e
u
m
o
n
ia
sc
re
e
n
in
g
u
sin
g
ra
d
io
lo
g
ica
l
d
a
ta.
His d
o
c
to
ra
l
wo
rk
n
o
w aim
s to
e
n
g
in
e
e
r
in
telli
g
e
n
t
so
l
u
ti
o
n
s fo
r
m
a
ss
iv
e
h
e
a
lt
h
c
a
re
d
a
tas
e
ts t
o
sh
a
rp
e
n
d
ia
g
n
o
stic
a
c
c
u
ra
c
y
a
n
d
a
u
t
o
m
a
te
c
li
n
ica
l
d
e
c
isio
n
s
.
F
o
r
t
h
is
stu
d
y
,
h
e
led
c
o
n
c
e
p
tu
a
li
z
a
ti
o
n
a
n
d
m
e
th
o
d
o
lo
g
y
.
He
d
e
v
e
lo
p
e
d
t
h
e
so
ftwa
re
f
ra
m
e
wo
rk
,
v
a
li
d
a
ti
o
n
,
a
n
d
in
v
e
stig
a
t
io
n
,
m
a
n
a
g
e
d
d
a
ta
c
u
ra
ti
o
n
,
a
n
d
a
u
th
o
re
d
t
h
e
o
rig
i
n
a
l
d
ra
ft.
He
c
a
n
b
e
c
o
n
tac
ted
a
t:
a
z
z
e
d
d
in
e
.
h
lali@
u
sm
b
a
.
a
c
.
m
a
.
Ma
jid
Be
n
Ya
k
h
lef
h
o
l
d
in
g
a
Do
c
t
o
ra
te
fr
o
m
t
h
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
s
a
t
S
i
d
i
M
o
h
a
m
e
d
Be
n
Ab
d
e
ll
a
h
U
n
iv
e
rsity
(US
M
BA),
c
u
rre
n
tl
y
,
h
e
se
rv
e
s
a
s
a
p
ro
fe
ss
o
r
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
t
th
e
F
a
c
u
lt
y
o
f
M
u
lt
id
isc
ip
li
n
a
ry
S
t
u
d
ies
i
n
Taz
a
.
He
is
a
p
e
rm
a
n
e
n
t
m
e
m
b
e
r
o
f
th
e
E
n
g
i
n
e
e
rin
g
S
c
ien
c
e
s P
o
ly
d
i
sc
ip
li
n
a
ry
Lab
o
ra
to
r
y
.
His res
e
a
rc
h
c
re
a
t
e
s a
n
e
x
u
s
b
e
twe
e
n
c
o
m
p
u
ter
e
n
g
i
n
e
e
rin
g
,
a
u
to
m
a
ti
o
n
sy
ste
m
s,
a
n
d
a
rti
fici
a
l
in
telli
g
e
n
c
e
,
sp
e
c
ifi
c
a
ll
y
in
v
e
stig
a
t
in
g
i
n
telli
g
e
n
t
d
e
c
isio
n
-
m
a
k
in
g
tec
h
n
o
lo
g
ies
.
He
e
n
g
i
n
e
e
rs
AI
-
b
a
se
d
so
lu
ti
o
n
s
fo
r
c
o
m
p
lex
,
re
a
l
-
wo
rl
d
a
p
p
l
ica
ti
o
n
s.
Be
y
o
n
d
th
e
o
re
ti
c
a
l
in
q
u
ir
y
,
h
is
fo
c
u
se
s
o
n
a
d
v
a
n
c
e
d
c
o
m
p
u
tati
o
n
a
l
m
e
th
o
d
s
a
n
d
in
tel
li
g
e
n
t
c
o
n
tro
l
sy
ste
m
s.
He
b
a
lan
c
e
s
th
e
se
sc
h
o
larl
y
p
u
rs
u
it
s
with
th
e
ri
g
o
r
o
u
s
m
e
n
t
o
rsh
i
p
o
f
stu
d
e
n
ts,
a
c
ti
v
e
ly
d
riv
in
g
in
n
o
v
a
ti
o
n
wit
h
in
e
n
g
i
n
e
e
rin
g
sc
ien
c
e
s
wh
il
e
in
teg
ra
ti
n
g
e
m
e
rg
in
g
tec
h
n
o
l
o
g
ies
i
n
to
m
o
d
e
r
n
c
o
m
p
u
tati
o
n
a
l
fra
m
e
wo
rk
s
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
m
a
j
id
.
b
e
n
y
a
k
h
lef@u
sm
b
a
.
a
c
.
m
a
.
S
o
u
l
a
im
a
n
El
H
a
z
z
a
t
e
a
r
n
in
g
h
is
P
h
.
D.
i
n
C
o
m
p
u
ter
S
c
ien
c
e
fro
m
S
id
i
M
o
h
a
m
e
d
Be
n
A
b
d
e
ll
a
h
Un
iv
e
rsity
i
n
2
0
1
8
,
c
u
rre
n
tl
y
,
h
e
se
rv
e
s
a
s
a
p
ro
fe
ss
o
r
a
t
t
h
e
P
o
ly
d
isc
ip
l
in
a
ry
F
a
c
u
lt
y
o
f
Taz
a
.
A
c
o
re
m
e
m
b
e
r
o
f
th
e
E
n
g
i
n
e
e
rin
g
S
c
ien
c
e
s
P
o
ly
d
isc
ip
l
in
a
ry
Lab
o
ra
t
o
ry
,
h
e
d
riv
e
s
re
se
a
rc
h
in
a
rti
f
icia
l
i
n
telli
g
e
n
c
e
a
n
d
a
d
v
a
n
c
e
d
in
tera
c
ti
v
e
tec
h
n
o
l
o
g
ies
.
His
f
o
c
u
s
is
s
h
a
rp
.
It
sp
a
n
s
c
o
m
p
u
ter
v
isio
n
,
d
e
e
p
lea
rn
in
g
,
a
n
d
imm
e
rsiv
e
e
n
v
iro
n
m
e
n
ts
l
ik
e
V
R
a
n
d
AR.
Cru
c
iall
y
,
h
e
in
v
e
s
ti
g
a
tes
d
e
e
p
fa
k
e
d
e
tec
ti
o
n
,
e
n
g
in
e
e
rin
g
in
n
o
v
a
ti
v
e
m
e
th
o
d
s
t
o
i
d
e
n
ti
f
y
m
a
n
i
p
u
late
d
m
e
d
ia
a
n
d
b
o
lster
d
ig
it
a
l
se
c
u
rit
y
.
F
o
r
th
is
sp
e
c
ifi
c
stu
d
y
,
h
e
v
a
li
d
a
ted
a
n
d
in
terp
re
ted
th
e
e
x
p
e
rime
n
tal
re
su
lt
s.
He
e
n
su
re
d
sc
ien
ti
fic
rig
o
r
b
y
re
v
iew
i
n
g
a
n
d
re
fin
i
n
g
th
e
fin
a
l
m
a
n
u
sc
rip
t
.
Be
y
o
n
d
t
e
c
h
n
ica
l
a
n
a
ly
sis,
h
is
ro
le
e
x
ten
d
e
d
to
h
ig
h
-
lev
e
l
s
u
p
e
rv
isi
o
n
a
n
d
p
r
o
jec
t
a
d
m
in
istratio
n
,
i
n
c
lu
d
in
g
t
h
e
a
c
q
u
isi
ti
o
n
o
f
n
e
c
e
ss
a
ry
re
so
u
rc
e
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
so
u
laim
a
n
.
e
lh
a
z
z
a
t@u
sm
b
a
.
a
c
.
m
a
.
Evaluation Warning : The document was created with Spire.PDF for Python.