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
3
9
0
~
1
4
0
1
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
3
9
0
-
1
4
0
1
1390
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Accurate
stroke a
rea cla
ss
ificatio
n
using
extreme gra
dient
bo
o
sting
wit
h mu
l
ti
-
fea
ture
ex
tracti
o
n
K
a
v
iko
nd
a
la
P
ra
v
ee
n K
u
m
a
r
Ra
o
1
,
M
a
ha
L
a
k
s
hm
i Bo
nd
la
2
,
B
o
m
m
a
r
a
j
u Sriniv
a
s
a
Ra
o
3
,
Am
bid
i N
a
v
ee
na
4
,
K
.
V
.
B
a
l
a
ra
ma
k
rish
na
5
,
Srini
v
a
s
a
ra
o
G
o
da
6
1
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
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
K
a
mal
a
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
a
n
d
S
c
i
e
n
c
e
,
S
i
n
g
a
p
u
r
,
I
n
d
i
a
2
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
,
V
i
g
n
a
n
's I
n
s
t
i
t
u
t
e
o
f
I
n
f
o
r
ma
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
V
i
s
a
k
h
a
p
a
t
n
a
m,
I
n
d
i
a
3
D
e
p
a
r
t
m
e
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
G
e
e
t
h
a
n
j
a
l
i
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
T
e
c
h
n
o
l
o
g
y
,
H
y
d
e
r
a
b
a
d
,
I
n
d
i
a
4
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
T
e
l
e
mat
i
c
s
En
g
i
n
e
e
r
i
n
g
,
G
.
N
a
r
a
y
a
n
a
m
ma
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
S
c
i
e
n
c
e
,
H
y
d
e
r
a
b
a
d
,
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
,
A
d
i
t
y
a
U
n
i
v
e
r
si
t
y
,
S
u
r
a
m
p
a
l
e
m
,
I
n
d
i
a
6
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
,
R
.
V
.
R
.
&
J.C
.
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
G
u
n
t
u
r
,
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
J
u
l
12
,
2
0
2
5
R
ev
is
ed
J
an
20
,
2
0
2
6
Acc
ep
ted
Feb
6
,
2
0
2
6
S
tro
k
e
,
o
n
e
o
f
th
e
m
o
st
c
o
m
m
o
n
n
e
u
ro
l
o
g
ica
l
d
iso
r
d
e
rs
lea
d
i
n
g
to
lo
n
g
-
term
d
isa
b
il
it
y
a
n
d
m
o
r
talit
y
,
re
q
u
ires
a
c
c
u
ra
te
d
e
tec
ti
o
n
o
f
a
ffe
c
ted
b
ra
in
re
g
io
n
s
fo
r
ti
m
e
ly
trea
tme
n
t
p
lan
n
in
g
.
Ho
we
v
e
r,
c
o
n
v
e
n
ti
o
n
a
l
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l
s
fa
c
e
c
h
a
ll
e
n
g
e
s
in
a
c
h
ie
v
in
g
p
re
c
ise
se
g
m
e
n
tatio
n
a
n
d
r
o
b
u
st
c
la
ss
ifi
c
a
ti
o
n
d
u
e
to
n
o
isy
i
n
p
u
ts,
we
a
k
fe
a
tu
re
re
p
re
se
n
tatio
n
,
a
n
d
p
o
o
r
g
e
n
e
ra
li
z
a
ti
o
n
.
To
a
d
d
re
ss
th
e
se
g
a
p
s,
t
h
is
st
u
d
y
i
n
t
ro
d
u
c
e
s
a
h
y
b
ri
d
fra
m
e
wo
rk
th
a
t
in
teg
ra
tes
th
e
Co
n
v
Ne
Xt
a
rc
h
it
e
c
tu
re
fo
r
stro
k
e
re
g
i
o
n
se
g
m
e
n
ta
ti
o
n
with
XG
Bo
o
st
-
b
a
se
d
c
las
sifica
ti
o
n
,
stre
n
g
th
e
n
e
d
t
h
ro
u
g
h
t
h
re
e
c
o
m
p
lem
e
n
ta
ry
fe
a
tu
re
e
x
trac
ti
o
n
m
e
th
o
d
s:
l
o
c
a
l
b
in
a
ry
p
a
tt
e
rn
s
(LBP
),
a
d
a
p
ti
v
e
th
re
sh
o
l
d
d
irec
ti
o
n
a
l
b
in
a
r
y
g
ra
d
ie
n
t
m
a
tri
x
(AT
-
DBG
M
),
a
n
d
wa
v
e
let
p
a
c
k
e
t
tran
sfo
rm
(W
P
T).
T
h
e
se
m
e
th
o
d
s
c
a
p
tu
re
tex
tu
ra
l,
d
irec
ti
o
n
a
l,
a
n
d
m
u
lt
i
-
re
so
lu
ti
o
n
fe
a
tu
re
s,
wh
ich
a
re
c
o
n
c
a
ten
a
ted
in
t
o
a
sta
c
k
e
d
v
e
c
to
r
a
n
d
c
las
sified
u
sin
g
XG
Bo
o
st.
P
re
p
r
o
c
e
ss
in
g
ste
p
s,
in
c
l
u
d
i
n
g
n
o
rm
a
li
z
a
ti
o
n
a
n
d
re
siz
in
g
,
e
n
su
re
imp
ro
v
e
d
in
p
u
t
c
o
n
siste
n
c
y
.
Ex
p
e
rime
n
tal
e
v
a
l
u
a
ti
o
n
s
o
n
b
e
n
c
h
m
a
rk
str
o
k
e
ima
g
i
n
g
d
a
tas
e
ts
sh
o
w
th
a
t
t
h
e
p
r
o
p
o
se
d
fr
a
m
e
wo
rk
a
c
h
iev
e
s
9
8
.
5
6
%
Dic
e
sim
il
a
rit
y
c
o
e
fficie
n
t
(DSC)
,
1
2
.
9
6
m
m
Ha
u
sd
o
rff
d
istan
c
e
(HD
)
,
9
9
.
1
2
%
a
c
c
u
ra
c
y
,
9
8
.
6
9
%
se
n
sit
iv
it
y
,
9
9
.
0
6
%
sp
e
c
ifi
c
it
y
,
9
8
.
9
8
%
p
re
c
isio
n
,
a
n
d
9
8
.
8
5
%
F
1
-
sc
o
re
.
K
ey
w
o
r
d
s
:
C
o
n
v
NeX
t
Dee
p
lear
n
in
g
Me
d
ical
im
ag
in
g
Stro
k
e
s
eg
m
en
tatio
n
W
av
elet
p
ac
k
et
tr
an
s
f
o
r
m
XGBo
o
s
t
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
:
Kav
ik
o
n
d
ala
Pra
v
ee
n
Ku
m
a
r
R
ao
Dep
ar
tm
en
t o
f
Ar
tific
ial
I
n
tellig
en
ce
an
d
Ma
ch
in
e
L
ea
r
n
in
g
,
Kam
ala
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
an
d
Scien
ce
Sin
g
ap
u
r
,
H
u
zu
r
ab
a
d
,
T
elan
g
a
n
a
,
I
n
d
ia
E
m
ail:
p
r
av
ee
n
k
u
m
ar
r
ao
.
k
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
A
s
tr
o
k
e
f
u
n
ctio
n
s
as
a
m
ed
ic
al
em
er
g
en
c
y
b
ec
au
s
e
it
o
b
s
tr
u
cts
b
lo
o
d
cir
cu
latio
n
to
t
h
e
b
r
ain
wh
ic
h
d
en
ies
o
x
y
g
e
n
an
d
n
u
tr
ien
ts
to
b
r
ain
ce
lls
.
Peo
p
le
ac
r
o
s
s
th
e
wo
r
ld
ex
p
er
ie
n
ce
s
tr
o
k
e
as
a
m
ajo
r
h
ea
lth
th
r
ea
t
b
ec
au
s
e
it
ca
u
s
es
b
o
th
p
r
em
atu
r
e
d
ea
th
an
d
p
e
r
s
is
ten
t
d
is
ab
i
liti
es
s
o
d
iag
n
o
s
is
an
d
tr
ea
tm
en
t
ap
p
r
o
ac
h
es
n
ee
d
to
b
e
im
p
r
o
v
e
d
.
T
h
e
m
ain
ty
p
es
o
f
s
tr
o
k
es
ex
is
t
a
s
is
ch
e
m
ic
s
tr
o
k
e
an
d
h
em
o
r
r
h
ag
ic
s
tr
o
k
e
wh
ich
p
r
esen
t
d
if
f
er
en
t
ca
u
s
es
an
d
f
ea
tu
r
es
t
h
r
o
u
g
h
o
u
t
ea
ch
co
n
d
itio
n
.
T
h
e
b
r
ain
ce
lls
lo
ca
ted
in
t
h
e
af
f
ec
ted
ar
ea
d
ie
d
u
e
to
o
x
y
g
e
n
an
d
n
u
tr
itio
n
al
s
u
p
p
ly
d
ep
r
iv
atio
n
t
h
at
o
cc
u
r
s
f
r
o
m
a
lack
o
f
b
lo
o
d
.
T
h
e
b
leed
in
g
o
f
b
lo
o
d
v
ess
els
cr
ea
tes
h
em
o
r
r
h
ag
ic
s
tr
o
k
e
b
y
ca
u
s
in
g
in
ter
n
al
b
r
ain
b
leed
i
n
g
[
1
]
.
Diag
n
o
s
tic
p
r
o
ce
d
u
r
es
h
ea
v
ily
d
ep
en
d
o
n
s
tr
o
k
e
s
eg
m
en
tatio
n
s
in
ce
i
t
allo
ws
p
h
y
s
ician
s
to
id
e
n
tify
an
d
m
a
r
k
s
tr
o
k
e
-
af
f
ec
ted
ar
ea
s
.
Stro
k
e
s
eg
m
en
tatio
n
r
elies
h
ea
v
ily
o
n
m
ag
n
etic
r
eso
n
an
ce
im
a
g
in
g
(
MRI)
as
well
a
s
co
m
p
u
ted
to
m
o
g
r
a
p
h
y
(
C
T
)
as
its
m
ain
im
ag
in
g
to
o
ls
[
2
]
.
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
A
cc
u
r
a
te
s
tr
o
ke
a
r
ea
cla
s
s
ific
a
tio
n
u
s
in
g
ex
tr
eme
g
r
a
d
ie
n
t b
o
o
s
tin
g
…
(
K
a
viko
n
d
a
la
P
r
a
v
ee
n
K
u
ma
r
R
a
o
)
1391
C
T
im
ag
in
g
s
er
v
es
as
t
h
e
ess
en
tial
d
iag
n
o
s
tic
to
o
l
f
o
r
s
tr
o
k
e
p
atien
t
e
v
alu
atio
n
b
ec
au
s
e
it
p
r
o
v
i
d
es
wid
e
av
ailab
ilit
y
an
d
q
u
ick
r
esu
lts
.
Hea
lth
ca
r
e
p
r
o
v
id
er
s
co
m
m
o
n
l
y
p
e
r
f
o
r
m
n
o
n
-
c
o
n
tr
ast
C
T
to
ev
alu
ate
b
r
ain
r
e
g
io
n
s
f
o
r
h
em
o
r
r
h
a
g
ic
s
tr
o
k
e
a
n
d
s
ea
r
c
h
f
o
r
i
n
f
ar
cti
o
n
in
d
icato
r
s
th
at
p
r
esen
t
as
h
y
p
o
atten
u
atio
n
[
3
]
d
esp
ite
th
eir
d
if
f
er
e
n
t
ca
p
a
b
ilit
ies.
Me
d
ical
im
ag
in
g
tech
n
o
lo
g
y
h
as
g
ai
n
ed
ex
te
n
s
iv
e
in
ter
est
f
r
o
m
r
esear
ch
er
s
af
ter
th
e
r
is
e
o
f
ar
tific
ial
in
tellig
en
ce
(
AI
)
.
Dee
p
lear
n
in
g
ap
p
licatio
n
s
h
av
e
p
r
o
v
en
h
ig
h
l
y
s
u
cc
ess
f
u
l
f
o
r
m
u
ltip
le
m
ed
ical
im
ag
in
g
p
u
r
p
o
s
es
wh
ich
in
clu
d
e
lesi
o
n
s
eg
m
en
tatio
n
im
a
g
e
s
y
n
th
esizin
g
an
d
d
is
ea
s
e
ea
r
ly
-
s
tag
e
d
ete
ctio
n
[
4
]
.
U
-
Net
as
well
as
V
-
Net
an
d
3
D
v
ar
ian
ts
s
er
v
e
d
if
f
er
en
t
p
u
r
p
o
s
es
in
m
ed
ica
l
im
ag
in
g
d
ata
a
n
aly
s
is
b
y
ef
f
ec
tiv
ely
p
r
o
ce
s
s
in
g
co
m
p
lex
d
ata
an
d
ex
tr
ac
tin
g
co
n
tex
tu
al
in
f
o
r
m
atio
n
[
5
]
.
T
h
e
p
r
o
b
lem
o
f
ac
cu
r
ate
s
eg
m
en
tatio
n
o
f
s
tr
o
k
e
ar
ea
in
m
ed
ical
im
ag
es
is
a
co
m
p
lic
ated
is
s
u
e
b
ec
au
s
e
o
f
t
h
e
ir
r
e
g
u
lar
s
h
a
p
e
o
f
lesi
o
n
an
d
f
u
zz
y
b
o
u
n
d
ar
ies
as
well
as
h
ig
h
in
te
r
-
p
atien
t
v
a
r
iatio
n
.
T
h
e
tr
ad
itio
n
al
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
b
ased
m
eth
o
d
s
s
u
ch
as
U
-
Net
an
d
D
ee
p
L
ab
V3
ca
n
n
o
t
ty
p
ically
d
eliv
er
f
in
e
-
g
r
ain
e
d
s
tr
u
ctu
r
al
a
n
d
h
ig
h
-
le
v
el
s
em
an
tic
f
ea
tu
r
es
at
th
e
s
am
e
tim
e,
r
esu
ltin
g
in
s
u
b
o
p
tim
al
b
o
u
n
d
a
r
y
d
etec
tio
n
o
f
t
h
e
s
tr
o
k
e
-
a
f
f
ec
ted
a
r
ea
s
.
Han
d
cr
af
ted
f
ea
tu
r
es,
th
o
u
g
h
u
s
ef
u
l
in
m
o
d
elin
g
lo
ca
l
in
ten
s
ity
v
ar
iatio
n
s
an
d
ca
p
tu
r
in
g
tex
tu
r
e
h
as
b
ee
n
s
h
o
wn
n
o
t
to
g
en
er
alize
ac
r
o
s
s
d
if
f
er
en
t
d
atasets
.
T
h
is
s
tu
d
y
h
y
p
o
th
esizes
th
at
a
h
y
b
r
id
d
ee
p
lear
n
in
g
f
r
am
ewo
r
k
,
wh
ich
co
m
b
in
es
C
o
n
v
NeX
t
-
b
ased
m
u
lti
-
s
ca
le
f
ea
tu
r
e
ex
tr
ac
tio
n
with
h
an
d
cr
a
f
ted
d
escr
ip
to
r
s
—
lo
ca
l
b
in
ar
y
p
atter
n
(
L
B
P)
,
ad
ap
tiv
e
th
r
esh
o
ld
d
ir
ec
tio
n
al
b
i
n
ar
y
g
r
ad
ie
n
t
m
atr
ix
(
AT
-
DB
GM
)
,
an
d
w
av
elet
p
ac
k
et
tr
an
s
f
o
r
m
(
W
PT)
—
an
d
XGBo
o
s
t
class
if
icatio
n
,
ca
n
o
v
er
co
m
e
t
h
ese
lim
itatio
n
s
.
Sp
ec
if
ically
,
we
p
r
o
p
o
s
e
th
at
th
e
i
n
teg
r
atio
n
o
f
d
ee
p
s
em
an
tic
f
ea
tu
r
es
with
h
an
d
cr
a
f
te
d
tex
t
u
r
e
d
escr
ip
to
r
s
will
en
h
an
ce
b
o
th
b
o
u
n
d
ar
y
p
r
ec
is
io
n
an
d
lesi
o
n
lo
ca
lizatio
n
,
u
ltima
tely
lead
in
g
t
o
s
u
p
e
r
io
r
s
eg
m
en
tatio
n
ac
c
u
r
ac
y
an
d
g
en
er
aliza
b
ilit
y
co
m
p
ar
ed
to
ex
is
tin
g
s
tan
d
alo
n
e
m
eth
o
d
s
.
T
h
e
p
r
o
p
o
s
ed
r
esear
ch
u
tili
ze
s
th
e
C
o
n
v
NeX
t
f
r
a
m
ewo
r
k
f
o
r
s
tr
o
k
e
ar
ea
s
eg
m
en
tatio
n
alo
n
g
with
XGBo
o
s
t
clas
s
if
icatio
n
th
r
o
u
g
h
ad
v
an
ce
d
f
ea
tu
r
e
ex
tr
ac
tio
n
tech
n
iq
u
es
to
ac
h
ie
v
e
b
etter
a
cc
u
r
ac
y
lev
els.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
co
m
b
in
es
C
o
n
v
NeX
t
with
h
an
d
cr
af
ted
f
e
atu
r
e
ex
tr
ac
tio
n
m
eth
o
d
s
(
L
B
P,
AT
-
DB
GM
,
an
d
W
PT)
to
im
p
r
o
v
e
s
tr
o
k
e
r
eg
io
n
r
ep
r
esen
tatio
n
an
d
u
s
es
a
ca
s
ca
d
ed
f
ea
tu
r
e
f
u
s
io
n
ap
p
r
o
ac
h
with
tex
tu
r
e,
g
r
ad
ien
t,
a
n
d
f
r
e
q
u
en
c
y
-
b
ase
d
f
ea
tu
r
es
to
ac
h
ie
v
e
r
o
b
u
s
t
class
if
icatio
n
alo
n
g
with
XGBo
o
s
t
f
o
r
ef
f
icien
t
s
tr
o
k
e
-
af
f
ec
ted
r
eg
io
n
class
if
icatio
n
with
h
ig
h
p
r
ec
is
io
n
.
Th
e
m
ain
co
n
tr
ib
u
tio
n
s
o
f
p
r
o
p
o
s
ed
wo
r
k
ar
e:
i)
P
r
o
p
o
s
e
a
n
o
v
el
h
y
b
r
id
ap
p
r
o
ac
h
th
at
i
n
teg
r
ates
th
e
C
o
n
v
NeX
t
ar
ch
itectu
r
e
f
o
r
s
eg
m
en
tatio
n
wit
h
XGBo
o
s
t
class
if
icatio
n
,
co
m
b
in
in
g
d
ee
p
s
em
an
tic
lear
n
in
g
with
g
r
ad
ien
t
-
b
o
o
s
ted
d
ec
i
s
io
n
tr
ee
s
f
o
r
en
h
an
ce
d
s
tr
o
k
e
ar
ea
d
etec
tio
n
.
ii)
D
esig
n
a
s
tack
ed
f
ea
tu
r
e
ex
tr
ac
tio
n
th
at
f
u
s
es
d
ee
p
f
ea
tu
r
es
f
r
o
m
C
o
n
v
NeX
t
with
h
an
d
cr
a
f
ted
d
escr
ip
to
r
s
—
L
B
P,
AT
-
D
B
G
M
an
d
W
PT
to
ca
p
tu
r
e
b
o
th
g
lo
b
al
s
em
an
tic
an
d
f
in
e
-
g
r
ain
ed
tex
tu
r
a
l
ch
ar
ac
ter
is
tics
o
f
s
tr
o
k
e
r
eg
i
o
n
s
.
iii)
P
er
f
o
r
m
s
tatis
tical
s
ig
n
if
ican
ce
test
in
g
(
p
air
ed
t
-
test
s
,
W
ilco
x
o
n
s
ig
n
e
d
-
r
an
k
test
s
,
ef
f
ec
t
s
ize
ca
lcu
latio
n
s
)
,
co
n
f
i
r
m
in
g
th
a
t
th
e
im
p
r
o
v
em
en
ts
o
v
er
b
a
s
elin
e
m
o
d
els
ar
e
s
tatis
ticall
y
s
ig
n
if
ican
t
(
p
<
0
.
0
1
)
a
n
d
n
o
t d
u
e
to
ch
an
ce
.
T
h
e
p
ap
e
r
h
as
a
s
tr
u
ct
u
r
ed
f
o
r
m
at
th
at
in
clu
d
es
s
ec
tio
n
2
f
o
r
r
ev
iewin
g
c
u
r
r
e
n
t
r
esear
ch
i
n
th
e
f
ield
.
F
o
llo
wed
b
y
s
ec
tio
n
3
wh
ich
e
x
p
lain
s
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
in
d
etail.
T
h
e
e
x
p
er
im
en
tal
f
in
d
i
n
g
s
an
d
m
eth
o
d
co
m
p
ar
is
o
n
a
p
p
ea
r
i
n
s
ec
tio
n
4
.
T
h
e
r
esear
ch
f
in
is
h
es with
t
h
e
co
n
clu
s
io
n
in
s
ec
tio
n
5
.
2.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
T
h
e
s
tu
d
y
o
f
s
tr
o
k
es
in
m
ed
ical
im
ag
es
attr
ac
ts
s
ig
n
if
ic
an
t
r
esear
ch
in
ter
est
d
u
e
to
im
p
r
o
v
in
g
p
er
f
o
r
m
an
ce
th
r
o
u
g
h
d
ee
p
lear
n
in
g
an
d
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es.
T
h
e
f
o
r
m
er
m
eth
o
d
s
ap
p
lied
h
u
m
a
n
ex
p
er
ts
f
o
r
s
eg
m
en
tatio
n
b
u
t
t
h
ese
m
eth
o
d
s
r
elied
o
n
m
a
n
u
al
t
ec
h
n
iq
u
es
an
d
r
u
le
-
b
ased
p
r
o
to
co
ls
wh
ich
le
d
to
in
cr
ea
s
ed
p
r
o
ce
s
s
in
g
tim
e
a
n
d
in
c
o
n
s
is
ten
t
r
esu
lts
ac
r
o
s
s
ex
p
er
ts
.
Stro
k
e
r
eg
i
o
n
s
eg
m
e
n
tatio
n
s
h
o
wed
h
u
g
e
p
r
o
g
r
ess
th
r
o
u
g
h
d
ee
p
lear
n
i
n
g
m
o
d
els
o
p
er
atin
g
C
NNs
s
p
ec
if
ically
in
U
-
Net,
Seg
Net
,
a
n
d
Dee
p
L
a
b
V3
.
T
h
e
m
o
d
els
p
er
f
o
r
m
p
o
o
r
ly
b
ec
a
u
s
e
th
ey
d
o
n
o
t
co
n
tain
p
r
o
p
er
f
ea
t
u
r
e
e
x
tr
ac
tio
n
p
r
o
ce
s
s
es
wh
ich
lead
s
t
o
m
is
tak
es
wh
en
id
en
tify
in
g
s
tr
o
k
e
-
af
f
ec
te
d
r
eg
i
o
n
s
.
T
h
e
f
ie
ld
o
f
s
tr
o
k
e
in
v
esti
g
atio
n
a
p
p
lies
d
ee
p
lear
n
in
g
r
esear
ch
th
at
c
o
m
b
in
es
d
ee
p
lear
n
in
g
tech
n
iq
u
es
with
m
an
u
ally
d
esig
n
ed
p
atter
n
r
ec
o
g
n
itio
n
s
y
s
tem
s
u
tili
zin
g
L
B
P a
n
d
W
av
elet
t
r
an
s
f
o
r
m
s
alo
n
g
with
g
r
a
d
ien
t
-
b
ased
m
eth
o
d
s
to
e
n
h
an
ce
s
eg
m
en
tatio
n
r
esu
lts
.
W
ei
et
a
l.
[
6
]
an
al
y
ze
d
b
r
ain
MRI
im
ag
es
o
f
ac
u
te
is
ch
em
i
c
s
tr
o
k
e
(
AI
S)
p
atien
ts
b
etwe
en
2
0
1
7
an
d
2
0
2
0
at
a
te
r
tiar
y
teac
h
i
n
g
h
o
s
p
ital
wh
ile
d
ev
el
o
p
in
g
s
em
an
tic
s
eg
m
en
tatio
n
g
u
i
d
e
d
d
etec
to
r
n
etwo
r
k
(
SGD
-
Net)
as
a
m
u
lti
-
s
tag
e
n
etwo
r
k
.
T
h
e
n
etwo
r
k
in
clu
d
e
d
f
ir
s
t
a
U
-
s
h
ap
ed
m
o
d
el
f
o
r
d
if
f
u
s
io
n
-
weig
h
ted
im
ag
in
g
(
DW
I
)
s
eg
m
e
n
tatio
n
an
d
s
ec
o
n
d
a
b
in
a
r
y
class
if
icatio
n
m
o
d
el
ass
ess
in
g
lesi
o
n
s
i
ze
s
ag
ain
s
t
lacu
n
e
an
d
n
o
n
-
lacu
n
e
a
n
d
cir
cu
lato
r
y
ter
r
ito
r
ies
f
o
r
an
ter
i
o
r
o
r
p
o
s
ter
io
r
lo
ca
tio
n
s
.
W
e
tr
an
s
f
o
r
m
ed
th
e
two
-
s
tag
e
d
ee
p
lear
n
i
n
g
m
o
d
el
to
SGD
-
Net
p
lu
s
th
r
o
u
g
h
a
n
au
to
m
atic
p
r
o
ce
s
s
wh
ich
s
eg
m
e
n
ted
A
I
S
lesi
o
n
s
in
DW
I
im
ag
es a
n
d
th
en
r
eg
is
ter
ed
th
e
ir
p
o
s
itio
n
in
T
1
-
weig
h
ted
im
a
g
es a
n
d
b
r
ai
n
atlases
[
6
]
.
Z
h
an
g
et
a
l.
[
7
]
p
r
esen
t
an
a
u
to
m
atic
ap
p
r
o
ac
h
to
d
etec
t
ac
u
te
is
ch
em
ic
s
tr
o
k
e
in
DW
I
s
th
r
o
u
g
h
d
ee
p
3
-
D
C
NNs.
T
h
is
m
eth
o
d
ef
f
ec
tiv
ely
p
r
o
ce
s
s
es
m
u
ltid
im
en
s
io
n
al
co
n
te
x
tu
al
d
ata
wh
ile
d
ev
elo
p
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
3
9
0
-
1
4
0
1
1392
h
ig
h
ly
d
is
cr
im
in
ativ
e
f
ea
tu
r
es a
s
p
ar
t
o
f
an
e
n
d
-
to
-
en
d
d
ata
-
d
r
iv
en
o
p
er
atio
n
.
Ou
r
n
etwo
r
k
in
co
r
p
o
r
ates
d
e
n
s
e
co
n
n
ec
tiv
ity
to
o
v
e
r
co
m
e
th
e
tr
ain
in
g
c
h
allen
g
es
o
f
d
ee
p
3
-
D
C
NN
b
y
allo
win
g
e
f
f
o
r
tl
ess
p
r
o
p
ag
atio
n
o
f
i
n
f
o
r
m
atio
n
as
well
as
g
r
ad
i
en
t
d
ata
th
r
o
u
g
h
o
u
t
th
e
n
et
wo
r
k
s
tr
u
ctu
r
e
[
7
]
.
T
h
e
d
ee
p
r
esid
u
al
atten
tio
n
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
DR
ANe
t)
th
at
L
iu
et
a
l.
[
8
]
d
ev
elo
p
ed
e
n
ab
les
ac
cu
r
ate
s
im
u
ltan
eo
u
s
lesi
o
n
se
g
m
en
tatio
n
a
n
d
q
u
an
tific
ati
o
n
o
f
is
ch
em
ic
s
tr
o
k
e
to
g
eth
e
r
with
wh
ite
m
atter
h
y
p
er
i
n
ten
s
ity
(
W
MH
)
lesi
o
n
s
in
MRI
im
ag
es.
T
h
e
U
-
n
et
d
esig
n
f
ea
tu
r
es
o
f
DR
ANe
t
en
ab
le
th
e
n
etwo
r
k
to
e
x
tr
ac
t
h
ig
h
-
q
u
ality
f
ea
tu
r
es
f
r
o
m
in
p
u
t
im
a
g
es
wh
ile
in
teg
r
atin
g
a
n
o
v
el
atte
n
tio
n
m
o
d
u
l
e.
T
h
e
tr
ain
i
n
g
o
f
DR
ANe
t
d
ep
en
d
s
o
n
Dice
lo
s
s
f
u
n
ctio
n
b
ec
au
s
e
th
is
lo
s
s
r
ed
u
ce
s
d
ata
im
b
alan
ce
is
s
u
es
in
th
e
tr
ain
in
g
d
ataset.
T
h
e
t
r
ain
in
g
an
d
ev
al
u
atio
n
p
r
o
ce
s
s
o
f
DR
ANe
t
tak
es
p
la
ce
o
n
7
4
2
2
D
MRI
im
a
g
es
o
b
tain
ed
f
r
o
m
th
e
s
u
b
-
ac
u
te
is
ch
em
ic
s
tr
o
k
e
ar
ea
s
eg
m
en
tatio
n
(
SISS)
ch
allen
g
e.
E
v
alu
atio
n
test
s
d
em
o
n
s
tr
ate
th
at
DR
A
Net
ac
h
iev
es
s
u
p
er
io
r
r
esu
lts
co
m
p
ar
ed
to
m
u
ltip
le
lea
d
in
g
s
eg
m
en
tatio
n
ap
p
r
o
ac
h
es o
f
it
s
tim
e
[
8
]
.
Yalç
ın
an
d
V
u
r
al
[
9
]
d
ev
elo
p
ed
U
-
Net
as
a
n
en
c
o
d
er
-
d
e
co
d
er
d
ee
p
lea
r
n
in
g
-
b
ased
C
NN
wh
ich
s
er
v
es
as
a
s
o
lu
tio
n
f
o
r
b
r
ain
s
tr
o
k
e
class
if
icatio
n
an
d
s
eg
m
en
tatio
n
.
A
co
n
v
o
lu
ti
o
n
al
d
ee
p
n
etwo
r
k
ar
ch
itectu
r
e
in
clu
d
es
an
o
p
tim
ized
d
im
en
s
io
n
al
U
-
Net
(
D
-
UNe
t)
th
r
o
u
g
h
b
lo
ck
in
g
an
d
ad
ap
tiv
e
co
n
v
o
lu
tio
n
lay
er
s
eq
u
en
ci
n
g
c
o
m
b
in
e
d
wi
th
ac
tiv
atio
n
f
u
n
ctio
n
an
d
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
.
T
h
e
p
r
o
p
o
s
ed
a
n
aly
s
is
m
eth
o
d
ap
p
lies
C
T
im
ag
e
p
r
o
ce
s
s
in
g
to
ev
alu
ate
b
r
ain
s
tr
o
k
e
p
r
esen
ce
in
t
h
e
d
ataset.
T
h
e
m
eth
o
d
h
elp
s
id
en
tify
s
tr
o
k
e
ca
u
s
es
b
etwe
en
is
ch
em
ic
an
d
h
em
o
r
r
h
a
g
ic
co
n
d
itio
n
s
af
ter
a
s
tr
o
k
e
tak
es
p
lace
.
T
h
e
p
r
o
p
o
s
e
d
m
eth
o
d
p
r
o
v
id
es
ex
ac
t
lo
ca
tio
n
id
en
tific
atio
n
o
f
r
eg
io
n
s
s
elec
ted
b
y
r
ad
io
lo
g
is
ts
an
d
p
er
f
o
r
m
s
s
eg
m
en
tatio
n
o
n
ac
tiv
e
s
tr
o
k
e
ar
ea
s
.
A
c
o
m
p
ar
is
o
n
o
f
t
h
e
p
r
o
p
o
s
ed
m
e
th
o
d
with
ex
is
tin
g
C
NN
-
ty
p
e
ar
ch
itectu
r
es
o
cc
u
r
s
th
r
o
u
g
h
v
ar
io
u
s
ex
p
er
im
e
n
ts
r
u
n
o
n
t
h
e
s
am
e
r
ea
l
d
ataset
u
s
in
g
Py
th
o
n
s
cr
ip
ts
[
9
]
.
Wu
et
a
l.
[
1
0
]
p
r
esen
t
W
-
Net
as
th
eir
n
ew
two
-
s
tag
e
ap
p
r
o
ac
h
f
o
r
b
r
ain
MRI
lesi
o
n
s
eg
m
en
tatio
n
.
W
-
Net
im
p
lem
en
ts
C
NN
an
d
tr
an
s
f
o
r
m
er
-
b
ased
m
eth
o
d
as
its
co
r
e
s
tr
u
ctu
r
e
wh
ile
in
teg
r
atin
g
b
o
u
n
d
ar
y
d
e
f
o
r
m
atio
n
m
o
d
u
le
(
B
DM
)
an
d
b
o
u
n
d
ar
y
co
n
s
tr
ain
t
m
o
d
u
le
(
B
C
M)
to
p
r
o
ce
s
s
u
n
clea
r
b
o
u
n
d
ar
ies.
T
h
e
B
DM
em
p
lo
y
s
cir
cu
lar
co
n
v
o
lu
tio
n
m
eth
o
d
s
to
f
i
x
th
e
in
itial
b
o
u
n
d
ar
y
w
h
i
le
B
C
M
im
p
lem
en
ts
d
ilated
co
n
v
o
lu
tio
n
to
p
r
o
v
i
d
e
d
y
n
am
ic
o
b
ject
b
o
u
n
d
ar
y
co
n
tr
o
l.
T
h
e
W
-
Net
r
ec
eiv
es
o
p
tim
izatio
n
th
r
o
u
g
h
o
u
r
d
esig
n
ed
m
u
lti
-
task
lea
r
n
in
g
lo
s
s
f
u
n
ctio
n
wh
ich
p
er
f
o
r
m
s
o
p
tim
izatio
n
f
r
o
m
b
o
th
r
e
g
io
n
a
n
d
b
o
u
n
d
ar
y
o
r
ien
tatio
n
s
[
1
0
]
.
Alth
o
u
g
h
s
ig
n
if
ica
n
t
ad
v
an
ce
m
en
ts
h
av
e
b
ee
n
m
ad
e
to
d
ate
in
th
e
u
s
e
o
f
C
NN
-
b
ased
ar
ch
itectu
r
es,
in
clu
d
in
g
U
-
Net,
W
-
Net,
an
d
DR
ANe
t,
th
er
e
ar
e
v
ar
io
u
s
c
h
allen
g
es
th
at
r
ed
u
ce
t
h
eir
p
o
ten
tial
clin
ical
u
s
e.
Mo
s
t
cu
r
r
en
t
m
o
d
els
ar
e
b
as
ed
o
n
h
u
g
e
an
n
o
tated
d
atasets
an
d
th
ey
ar
e
n
o
t
g
o
o
d
at
g
en
er
alizin
g
ac
r
o
s
s
d
if
f
er
en
t
p
atien
t
g
r
o
u
p
s
d
u
e
t
o
is
s
u
es
o
f
class
im
b
alan
ce
.
Alg
o
r
ith
m
s
s
u
ch
as
SGD
-
Net
an
d
3
D
C
NNs
ar
e
b
etter
at
lesi
o
n
d
etec
t
io
n
,
b
u
t
ca
n
b
e
c
o
m
p
u
tatio
n
ally
i
n
ten
s
iv
e
an
d
f
ail
to
r
eso
lv
e
f
in
e
-
s
ca
le
lesi
o
n
b
o
u
n
d
ar
ies,
esp
ec
ially
in
s
m
a
ll
o
r
s
m
all
s
tr
o
k
e
lo
ca
tio
n
s
.
T
r
an
s
f
o
r
m
er
-
b
ased
an
d
h
y
b
r
id
m
o
d
els
h
av
e
m
o
r
e
r
ich
f
ea
t
u
r
es,
b
u
t
a
r
e
co
s
tly
a
n
d
p
r
o
n
e
to
m
is
class
if
y
in
g
th
e
b
o
u
n
d
a
r
ie
s
.
T
h
ese
d
r
awb
ac
k
s
u
n
d
er
s
co
r
e
a
n
ee
d
to
h
av
e
a
f
r
a
m
ewo
r
k
t
h
at
ca
n
tr
ad
e
o
f
f
b
etwe
en
r
ich
s
em
an
ti
c
r
ep
r
esen
tatio
n
an
d
tex
tu
r
e
c
u
es
th
at
ar
e
cr
ea
ted
b
y
h
an
d
,
r
e
f
in
e
b
o
u
n
d
ar
y
p
r
ec
is
io
n
,
an
d
cr
ea
te
s
tr
o
n
g
s
eg
m
e
n
tatio
n
with
o
u
t r
el
y
in
g
o
n
lar
g
e
tr
ain
in
g
s
ets.
Ou
r
r
esear
ch
,
in
s
p
ir
ed
b
y
th
ese
g
a
p
s
,
p
r
o
p
o
s
es
a
C
o
n
v
NeX
t
-
XGBo
o
s
t
h
y
b
r
id
m
o
d
el
th
at
ca
n
b
e
u
s
ed
to
en
h
an
ce
th
e
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
ac
cu
r
ac
y
o
f
s
tr
o
k
e
d
e
tectio
n
b
y
u
s
in
g
m
u
lti
-
s
ca
le
d
ee
p
f
ea
tu
r
es
a
n
d
h
an
d
cr
a
f
ted
d
escr
ip
to
r
s
.
3.
M
E
T
H
O
D
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
em
p
lo
y
s
C
o
n
v
NeX
t
as
th
e
b
ac
k
b
o
n
e
f
o
r
f
ea
t
u
r
e
ex
t
r
ac
tio
n
,
wh
er
e
in
p
u
t
m
ed
ical
im
ag
es
ar
e
p
r
o
ce
s
s
ed
to
g
en
e
r
ate
m
u
lti
-
s
ca
le
f
e
atu
r
e
m
ap
s
f
r
o
m
d
if
f
er
en
t
n
etwo
r
k
s
tag
es.
T
h
e
s
h
allo
w
lay
er
s
ca
p
tu
r
e
f
in
e
-
g
r
ain
ed
s
tr
u
ctu
r
al
d
etails
s
u
ch
as
ed
g
es
an
d
tex
tu
r
es,
wh
il
e
th
e
d
ee
p
er
lay
er
s
en
co
d
e
h
i
g
h
-
lev
el
s
em
an
tic
c
o
n
tex
t
o
f
lesi
o
n
r
e
g
io
n
s
,
p
r
o
v
id
in
g
a
h
ier
ar
c
h
ical
r
ep
r
esen
t
atio
n
o
f
th
e
im
a
g
e.
T
h
ese
m
u
lti
-
s
ca
le
f
ea
tu
r
es
ar
e
th
en
p
r
o
g
r
ess
iv
ely
u
p
s
am
p
led
an
d
f
u
s
ed
with
in
a
U
-
Net
–
lik
e
d
ec
o
d
e
r
t
o
r
ec
o
n
s
tr
u
ct
s
p
atial
r
eso
lu
tio
n
a
n
d
ac
cu
r
ately
d
elin
ea
te
lesi
o
n
b
o
u
n
d
ar
ies.
Sk
ip
co
n
n
ec
tio
n
s
ar
e
in
co
r
p
o
r
ated
to
in
teg
r
ate
s
h
allo
w
f
ea
tu
r
e
m
ap
s
with
co
r
r
esp
o
n
d
in
g
d
ec
o
d
er
lay
er
s
,
th
er
eb
y
p
r
eser
v
in
g
f
in
e
ed
g
e
d
etails
th
at
m
ig
h
t
o
th
er
wis
e
b
e
lo
s
t
in
d
ee
p
er
r
ep
r
e
s
en
tatio
n
s
.
Fin
ally
,
th
e
f
u
s
ed
f
ea
tu
r
e
m
ap
s
ar
e
p
a
s
s
ed
th
r
o
u
g
h
a
1
×1
co
n
v
o
l
u
tio
n
lay
e
r
f
o
llo
we
d
b
y
a
So
f
tMa
x
ac
ti
v
atio
n
f
u
n
cti
o
n
to
p
r
o
d
u
ce
th
e
f
in
al
s
eg
m
en
tatio
n
,
ef
f
ec
tiv
ely
h
ig
h
lig
h
tin
g
th
e
s
tr
o
k
e
r
eg
io
n
.
Fig
u
r
e
1
s
h
o
ws th
e
b
lo
ck
d
iag
r
am
o
f
t
h
e
p
r
o
p
o
s
ed
m
eth
o
d
.
T
h
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
f
o
r
s
t
r
o
k
e
ar
ea
s
eg
m
en
tatio
n
an
d
cl
ass
if
icatio
n
im
p
lem
en
ts
a
d
ee
p
lear
n
in
g
h
y
b
r
id
s
y
s
tem
wh
ic
h
c
o
m
b
in
es
C
o
n
v
NeX
t
with
a
d
v
an
ce
d
f
ea
tu
r
es
ex
tr
ac
tio
n
m
eth
o
d
s
a
n
d
XGBo
o
s
t
-
b
ased
class
if
icatio
n
f
o
r
b
etter
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
.
T
h
e
s
y
s
t
em
h
as
two
m
ai
n
o
p
er
atio
n
al
p
ar
ts
:
tr
ain
in
g
an
d
test
in
g
.
T
h
e
tr
ain
i
n
g
p
h
ase
b
eg
in
s
with
m
ed
ical
im
a
g
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
wh
ich
r
esize
th
e
im
ag
es
wh
ile
ap
p
ly
in
g
f
ilter
s
f
o
llo
wed
b
y
n
o
r
m
aliza
tio
n
tech
n
iq
u
es
to
e
n
h
an
ce
p
ictu
r
e
q
u
ality
th
r
o
u
g
h
n
o
is
e
r
em
o
v
al
an
d
p
ix
el
in
ten
s
ity
s
tan
d
ar
d
izatio
n
.
T
h
e
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
en
s
u
r
e
th
e
im
ag
es
m
ee
t
r
e
q
u
ir
em
en
ts
b
ef
o
r
e
f
ea
t
u
r
e
ex
tr
ac
tio
n
an
d
s
eg
m
en
tatio
n
p
r
o
ce
d
u
r
es.
T
h
e
p
r
o
ce
s
s
ed
i
m
ag
es
en
ter
C
o
n
v
NeX
t
f
o
r
a
n
aly
s
is
b
ec
au
s
e
th
is
u
p
d
ated
n
eu
r
al
n
etwo
r
k
ex
t
r
ac
ts
b
o
t
h
d
ee
p
s
p
atial
an
d
co
n
tex
tu
al
f
ea
tu
r
es
f
r
o
m
m
ed
ical
s
ca
n
s
.
T
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
cc
u
r
a
te
s
tr
o
ke
a
r
ea
cla
s
s
ific
a
tio
n
u
s
in
g
ex
tr
eme
g
r
a
d
ie
n
t b
o
o
s
tin
g
…
(
K
a
viko
n
d
a
la
P
r
a
v
ee
n
K
u
ma
r
R
a
o
)
1393
s
eg
m
en
tatio
n
f
u
n
ctio
n
o
f
C
o
n
v
NeX
t
g
en
e
r
ates
d
etailed
h
ig
h
-
r
eso
lu
tio
n
b
i
n
ar
y
s
eg
m
en
t
atio
n
m
ap
s
w
h
ich
s
h
o
w
s
tr
o
k
e
r
eg
io
n
s
clea
r
ly
.
Af
ter
s
eg
m
en
tatio
n
,
th
e
s
y
s
tem
ap
p
lies
a
n
a
d
v
an
ce
d
f
ea
tu
r
e
ex
tr
a
ctio
n
a
p
p
r
o
ac
h
t
o
e
n
h
an
ce
th
e
p
r
ec
is
io
n
o
f
s
tr
o
k
e
a
r
ea
cla
s
s
if
icatio
n
.
T
h
e
s
eg
m
en
ted
i
m
ag
es
r
ec
eiv
e
p
r
o
ce
s
s
in
g
w
ith
th
r
ee
a
d
v
an
ce
d
tech
n
iq
u
es
in
clu
d
i
n
g
L
B
P,
AT
-
DB
GM
,
an
d
W
PT.
L
B
P
id
en
tifie
s
r
eg
io
n
al
tex
tu
r
e
ch
ar
ac
ter
is
tics
f
o
r
m
icr
o
-
p
atter
n
an
aly
s
is
b
ec
au
s
e
th
ese
p
atter
n
s
d
is
tin
g
u
is
h
ar
ea
s
s
u
f
f
er
in
g
f
r
o
m
s
tr
o
k
e
f
r
o
m
u
n
af
f
ec
ted
tis
s
u
e
r
eg
io
n
s
.
AT
-
DB
GM
h
elp
s
ex
t
r
ac
t
g
r
a
d
ien
t
-
b
ased
d
ir
ec
tio
n
al
f
ea
tu
r
es
t
o
e
n
h
an
ce
th
e
d
etec
t
io
n
o
f
e
d
g
es
wh
ile
im
p
r
o
v
in
g
th
e
d
if
f
er
e
n
tiatio
n
o
f
s
tr
u
ctu
r
es
p
r
esen
t
in
s
tr
o
k
e
ar
ea
.
W
PT
d
ec
o
m
p
o
s
es
im
ag
es
in
to
s
ep
ar
ate
f
r
eq
u
e
n
cy
s
u
b
-
b
an
d
s
s
o
th
at
it
ca
n
ca
p
tu
r
e
b
o
th
co
ar
s
e
an
d
f
in
e
tex
tu
r
e
an
d
s
h
ap
e
d
etails.
T
h
e
th
r
ee
ex
tr
ac
ted
f
ea
tu
r
es
m
er
g
e
in
to
o
n
e
ca
s
ca
d
ed
v
ec
to
r
th
at
d
e
v
elo
p
s
a
m
u
lti
-
d
im
en
s
io
n
al
m
o
d
el
f
o
r
th
e
s
eg
m
en
ted
s
tr
o
k
e
ar
ea
s
.
T
h
e
ca
s
ca
d
ed
f
ea
tu
r
e
r
ep
r
esen
tatio
n
en
ter
s
an
XGBo
o
s
t
clas
s
if
ier
f
o
r
lear
n
in
g
s
tr
o
k
e
r
eg
i
o
n
class
if
icatio
n
p
atter
n
s
with
h
ig
h
p
r
ec
is
io
n
.
Fig
u
r
e
1
.
B
lo
ck
d
iag
r
am
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
3
.
1
.
P
re
pro
ce
s
s
ing
I
m
ag
e
r
esizin
g
is
a
p
r
ep
r
o
ce
s
s
in
g
s
tep
th
at
ad
ju
s
ts
th
e
d
im
en
s
io
n
s
o
f
an
im
ag
e
to
a
s
p
e
cif
ic
tar
g
et
s
ize,
en
s
u
r
in
g
u
n
if
o
r
m
in
p
u
t d
im
en
s
io
n
s
f
o
r
m
o
d
els in
co
m
p
u
ter
v
is
io
n
task
s
[
1
1
]
.
I
n
th
is
wo
r
k
we
r
esized
th
e
im
ag
es
in
to
2
5
6
×2
5
6
[
1
2
]
.
O
n
e
o
f
th
e
d
ata
tr
an
s
f
o
r
m
atio
n
p
r
o
ce
d
u
r
es
u
s
ed
in
d
ata
m
in
i
n
g
is
n
o
r
m
aliza
tio
n
,
wh
ich
s
ca
les
n
u
m
er
ical
ch
ar
ac
ter
is
tics
with
i
n
a
m
o
r
e
co
n
s
tr
ain
ed
r
an
g
e
.
I
n
f
in
ite
n
e
g
ativ
e
an
d
p
o
s
itiv
e
v
alu
es
m
ak
e
u
p
th
e
Z
-
s
co
r
e
v
al
u
e
[
1
3
]
–
[
1
5
]
.
Z
-
s
co
r
e
n
o
r
m
aliza
tio
n
is
a
d
ata
n
o
r
m
aliza
tio
n
tech
n
iq
u
e
th
at
u
s
es
th
e
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
t
o
p
r
o
v
id
e
d
ata
r
an
g
es
[
1
6
]
.
=
−
(
1)
W
h
er
e
μ
is
th
e
m
ea
n
p
ix
el
i
n
ten
s
ity
an
d
σ
is
th
e
s
tan
d
ar
d
d
ev
iatio
n
o
f
p
i
x
el
in
ten
s
ities
.
A
n
o
n
-
li
n
ea
r
d
ig
ital
f
ilter
in
g
tech
n
i
q
u
e
f
o
r
r
em
o
v
in
g
n
o
is
e
f
r
o
m
a
s
ig
n
al
o
r
im
ag
e
is
th
e
m
ed
ia
n
f
il
ter
.
On
e
co
m
m
o
n
p
r
e
-
p
r
o
ce
s
s
in
g
m
eth
o
d
to
im
p
r
o
v
e
ed
g
e
r
ec
o
g
n
itio
n
i
n
an
i
m
ag
e
is
n
o
is
e
r
ed
u
ctio
n
[
1
7
]
.
Fo
r
a
g
iv
en
win
d
o
w
s
ize,
th
e
m
ed
ian
f
ilter
o
u
t
p
e
r
f
o
r
m
s
Gau
s
s
ian
b
lu
r
in
r
em
o
v
in
g
n
o
is
e
wh
ile
k
ee
p
in
g
e
d
g
es
f
o
r
s
m
all
to
m
o
d
er
ate
am
o
u
n
ts
o
f
Gau
s
s
ian
n
o
is
e
[
1
8
]
.
3
.
2
.
Co
nv
Ne
x
t
C
o
n
v
NeX
t
is
a
m
o
d
er
n
C
NN
th
at
en
h
an
ce
s
tr
a
d
itio
n
al
C
NN
ar
ch
itectu
r
es
b
y
in
teg
r
atin
g
d
esig
n
p
r
in
cip
les
in
s
p
ir
ed
b
y
v
is
io
n
tr
an
s
f
o
r
m
er
s
(
ViT
s
)
.
A
k
ey
in
n
o
v
atio
n
in
C
o
n
v
NeX
t
is
la
y
er
s
ca
lin
g
,
wh
ic
h
s
tab
ilizes
tr
ain
in
g
b
y
in
tr
o
d
u
c
in
g
a
lea
r
n
ab
le
s
ca
lin
g
f
ac
to
r
i
n
r
esid
u
al
c
o
n
n
ec
tio
n
s
.
Un
lik
e
tr
ad
itio
n
al
C
NNs,
C
o
n
v
NeX
t
o
p
tim
izes
p
e
r
f
o
r
m
an
ce
th
r
o
u
g
h
s
tr
u
ct
u
r
ed
n
o
r
m
a
lizatio
n
,
b
atch
n
o
r
m
aliza
tio
n
a
f
ter
ac
tiv
atio
n
,
an
d
d
o
wn
s
am
p
lin
g
v
ia
p
atc
h
if
y
in
g
co
n
v
o
lu
tio
n
s
in
s
tead
o
f
p
o
o
lin
g
.
T
h
ese
im
p
r
o
v
em
en
ts
e
n
ab
le
C
o
n
v
NeX
t
t
o
ac
h
iev
e
h
ig
h
ac
c
u
r
ac
y
with
l
o
wer
co
m
p
u
tatio
n
al
co
s
t,
m
a
k
in
g
it
a
p
o
wer
f
u
l
b
ac
k
b
o
n
e
f
o
r
m
ed
ical
im
ag
e
s
eg
m
en
tatio
n
,
in
clu
d
in
g
s
tr
o
k
e
ar
ea
d
etec
tio
n
u
s
in
g
C
o
n
v
Ne
Xt
-
UNe
t.
′
=
(
4
,
4
)
(
)
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
3
9
0
-
1
4
0
1
1394
W
h
er
e
X
is
th
e
in
p
u
t
im
ag
e
.
C
on
v
(
4
,
4
)
is
a
co
n
v
o
lu
tio
n
lay
er
with
k
e
r
n
el
s
ize
4
×4
an
d
s
tr
id
e
4
,
p
r
o
jectin
g
to
ch
an
n
el
C
.
'
X′ is th
e
d
o
wn
s
am
p
led
f
ea
tu
r
e
m
ap
.
=
(
(
(
,
)
,
1
(
′
)
)
)
(
3
)
W
h
er
e
C
on
v
(
k
,
k
)
,
1
is
a
d
ep
th
wis
e
co
n
v
o
lu
tio
n
(
k
er
n
el
s
ize
k
×k
)
.
B
N
is
b
atch
n
o
r
m
aliza
tio
n
.
σ
is
th
e
non
-
l
in
ea
r
ity
ac
tiv
atio
n
f
u
n
cti
o
n
(
Gau
s
s
ian
er
r
o
r
lin
ea
r
u
n
it
(
GE
L
U
)
)
.
X′ is th
e
in
p
u
t
f
r
o
m
th
e
p
r
ev
io
u
s
lay
er
.
=
(
1
,
1
)
,
(
(
(
)
)
)
(
4
)
′
=
(
1
,
1
)
,
(
(
(
)
)
)
(
5
)
W
h
er
e
r
is
th
e
ex
p
an
s
io
n
r
ati
o
.
C
on
v
(
1
,
1
)
,
rC
ex
p
an
d
s
th
e
n
u
m
b
er
o
f
ch
an
n
els.
C
on
v
(
1
,
1
)
,
C
p
r
o
jects
it
b
ac
k
to
th
e
o
r
ig
in
al
c
h
an
n
el
c
o
u
n
t.
=
+
⋅
′
(
6
)
W
h
er
e
Y
is
th
e
r
esid
u
al
in
p
u
t.
Y′ is th
e
tr
an
s
f
o
r
m
ed
o
u
tp
u
t,
λ
is
a
lear
n
ab
le
weig
h
t p
a
r
am
et
er
=
+
⋅
(
1
,
1
)
,
(
(
(
(
1
,
1
)
,
(
(
(
(
,
)
,
1
(
)
)
)
)
)
)
)
(
7
)
T
ab
le
1
in
clu
d
es
th
e
ess
en
tial
tr
ain
in
g
p
ar
a
m
eter
s
o
f
C
o
n
v
NeX
t
th
at
o
p
tim
ize
its
p
er
f
o
r
m
an
ce
in
s
tr
o
k
e
ar
ea
s
eg
m
en
tatio
n
.
T
h
e
m
o
d
el
co
n
tr
o
ls
its
in
f
o
r
m
atio
n
p
r
o
ce
s
s
in
g
s
co
p
e
th
r
o
u
g
h
a
k
er
n
el
s
ize
o
f
7
wh
ich
s
im
u
ltan
eo
u
s
ly
allo
ws
it
to
u
n
d
e
r
s
tan
d
lar
g
e
p
atter
n
s
alo
n
g
s
id
e
m
ain
tain
in
g
its
ab
ilit
y
to
d
etec
t
p
r
ec
is
e
d
etai
ls
.
T
h
e
p
r
o
p
o
s
ed
C
o
n
v
N
eXt
–
h
y
b
r
id
f
r
am
ewo
r
k
,
with
its
co
m
b
in
atio
n
o
f
d
ee
p
s
em
an
tic
f
ea
tu
r
es
an
d
h
an
d
cr
a
f
ted
d
escr
ip
to
r
s
,
ca
n
b
e
ad
ap
ted
to
o
th
er
m
e
d
ical
im
ag
e
s
eg
m
en
tatio
n
task
s
,
s
u
ch
as
tu
m
o
r
,
lesi
o
n
,
o
r
o
r
g
an
d
elin
ea
tio
n
,
b
y
r
etr
ain
i
n
g
o
n
t
h
e
r
ele
v
an
t
d
at
asets
wh
ile
p
r
eser
v
in
g
t
h
e
m
u
lti
-
s
ca
le
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
f
u
s
io
n
s
tr
ateg
y
.
T
ab
le
1
.
T
r
ai
n
in
g
p
ar
am
eter
s
o
f
C
o
n
v
Nex
t
P
a
r
a
me
t
e
r
Ty
p
i
c
a
l
v
a
l
u
e
s
K
e
r
n
e
l
s
i
z
e
k
7
Ex
p
a
n
si
o
n
r
a
t
i
o
r
4
La
y
e
r
s
c
a
l
i
n
g
λ
1e
-
6
P
a
t
c
h
i
f
y
c
onv
4
×
4
,
s
t
r
i
d
e
4
O
p
t
i
mi
z
e
r
A
d
a
mW
Le
a
r
n
i
n
g
r
a
t
e
1e
-
4
Ep
o
c
h
s
1
0
0
B
a
t
c
h
si
z
e
16
3
.
3
.
Ada
ptiv
e
t
hresh
o
ld dire
ct
io
na
l bina
ry
g
ra
dient
ma
t
r
ix
T
h
e
p
r
o
p
o
s
ed
DB
GM
in
tr
o
d
u
ce
s
a
n
o
v
el
ap
p
r
o
ac
h
to
f
ea
t
u
r
e
ex
tr
ac
tio
n
in
b
in
a
r
y
im
ag
es,
lev
er
ag
in
g
ad
ap
tiv
e
win
d
o
win
g
,
g
r
a
d
ie
n
t
-
b
ased
d
i
r
ec
tio
n
al
b
in
n
in
g
,
an
d
m
u
lti
-
s
ca
le
an
aly
s
is
t
o
ca
p
tu
r
e
co
m
p
lex
p
atter
n
s
.
L
et
g
(
i,
j)
b
e
th
e
g
r
ad
ien
t a
t p
ix
el
(
i,
j)
.
T
h
is
ca
n
b
e
co
m
p
u
ted
u
s
in
g
b
i
n
ar
y
-
a
d
ap
te
d
s
o
b
el
o
p
e
r
ato
r
s
.
=
[
−
1
0
1
−
2
0
2
−
1
0
1
]
(
8
)
=
[
−
1
−
2
−
1
0
0
0
1
2
1
]
(
9
)
(
,
)
=
∑
∑
(
,
)
.
(
+
,
+
)
1
=
−
1
1
=
−
1
(
1
0
)
(
,
)
=
∑
∑
(
,
)
.
(
+
,
+
)
1
=
−
1
1
=
−
1
(
1
1
)
(
,
)
=
√
(
,
)
2
+
(
,
)
2
(
1
2
)
(
,
)
=
(
(
,
)
(
,
)
)
(
1
3
)
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
A
cc
u
r
a
te
s
tr
o
ke
a
r
ea
cla
s
s
ific
a
tio
n
u
s
in
g
ex
tr
eme
g
r
a
d
ie
n
t b
o
o
s
tin
g
…
(
K
a
viko
n
d
a
la
P
r
a
v
ee
n
K
u
ma
r
R
a
o
)
1395
T
h
e
r
an
g
e
o
f
p
o
s
s
ib
le
g
r
ad
ie
n
t d
ir
ec
tio
n
s
θ(
i,
j)
is
d
i
v
id
ed
in
t
o
K
b
in
s
Θ
k
(
e.
g
.
,
K
=8
b
in
s
,
e
ac
h
co
v
er
i
n
g
45
°
):
=
[
,
+
)
,
=
1
,
2
,
…
.
(
1
4
)
(
,
)
=
1
∣
(
,
)
∣
∑
(
,
)
(
,
)
∈
(
,
)
(
1
5
)
(
,
)
∣
=
(
(
.
1
(
,
)
,
)
,
)
(
1
6
)
W
h
er
e
α
is
a
s
ca
lin
g
f
ac
to
r
,
a
n
d
an
d
ar
e
th
e
m
in
im
u
m
a
n
d
m
ax
im
u
m
win
d
o
w
s
izes,
r
esp
ec
tiv
ely
.
(
,
)
=
∑
1
{
(
,
)
=
1
}
.
1
{
(
,
)
=
1
}
(
,
)
∈
(
,
)
1
{
(
,
)
=
}
.
1
{
(
,
)
=
}
(
1
7
)
=
∑
=
1
(
1
8
)
=
−
∑
(
,
)
(
,
)
,
(
1
9
)
=
∑
=
1
(
2
0
)
(
)
(
,
=
)
(
2
1
)
T
h
e
m
u
lti
-
s
ca
le
co
n
s
is
ten
cy
C
(
s
1
,
s
2
)
b
etwe
en
s
ca
les
s
1
an
d
s
2
.
(
1
,
2
)
=
1
|
(
1
)
|
.
|
(
2
)
|
∑
(
1
)
(
,
)
⋅
(
2
)
(
,
)
,
(
2
2
)
=
[
,
(
1
,
2
)
,
,
,
,
,
…
]
(
2
3
)
3
.
4
.
L
o
ca
l bina
ry
pa
t
t
er
n
L
B
P
s
er
v
es
as
a
wid
ely
u
s
ed
im
ag
e
-
p
r
o
ce
s
s
in
g
f
ea
tu
r
e
e
x
tr
ac
tio
n
m
eth
o
d
[
1
9
]
.
T
h
e
L
B
P
o
p
er
ates
f
r
eq
u
e
n
tly
b
ec
au
s
e
it p
r
o
ce
s
s
es a
p
p
licatio
n
s
in
r
ea
l
-
tim
e
wh
ile
r
em
ain
in
g
s
im
p
le
an
d
lo
w
-
c
o
s
t f
o
r
ca
lcu
latio
n
s
an
d
ex
tr
ac
tin
g
f
ea
tu
r
es e
f
f
ec
tiv
ely
.
,
=
∑
(
−
)
2
−
1
=
0
(
2
4
)
=
{
1
,
≥
0
0
,
<
0
(
2
5
)
W
h
er
e
P is
th
e
n
u
m
b
er
o
f
n
eig
h
b
o
r
in
g
p
ix
els an
d
R
is
th
e
r
ad
iu
s
.
3
.
5
.
Wa
v
elet
pa
c
k
et
t
r
a
ns
f
o
rm
W
PT
is
an
ad
v
a
n
ce
d
ex
ten
s
i
o
n
o
f
t
h
e
d
is
cr
ete
wav
elet
tr
a
n
s
f
o
r
m
(
DW
T
)
,
ca
p
a
b
le
o
f
p
r
o
v
id
in
g
a
m
u
lti
-
r
eso
lu
tio
n
an
al
y
s
is
o
f
im
ag
es
b
y
d
ec
o
m
p
o
s
in
g
b
o
th
ap
p
r
o
x
im
atio
n
an
d
h
ig
h
-
f
r
e
q
u
en
cy
co
m
p
o
n
e
n
ts
at
ea
ch
lev
el.
T
h
e
W
PT
d
ec
o
m
p
o
s
itio
n
at
lev
el
+
1
is
d
ef
in
ed
as
:
+
1
(
)
(
)
=
∑
ℎ
(
)
(
)
(
2
−
)
(
2
6
)
+
1
(
)
(
)
=
∑
(
)
(
)
(
2
−
)
(
2
7
)
W
h
er
e
W
(
n
)
r
ep
r
esen
ts
th
e
wav
el
et
p
ac
k
et
co
ef
f
icien
ts
at
lev
el
j.
h
(
k
)
an
d
g
(
k
)
ar
e
t
h
e
lo
w
-
p
ass
an
d
h
ig
h
-
p
ass
f
ilter
s
,
r
esp
ec
tiv
ely
.
W
+
1
(
n
)
(
t
)
an
d
V
+
1
(
n
)
(
t
)
d
en
o
te
th
e
n
ew
d
ec
o
m
p
o
s
ed
s
u
b
-
b
an
d
s
at
th
e
n
ex
t le
v
el.
=
∑
|
(
)
|
2
(
2
8
)
W
h
er
e
r
ep
r
esen
ts
th
e
en
er
g
y
d
is
tr
ib
u
tio
n
in
d
if
f
er
en
t
f
r
e
q
u
en
cy
s
u
b
-
b
a
n
d
s
,
ca
p
tu
r
in
g
ess
en
tial
tex
tu
r
al
p
atter
n
s
f
o
r
s
tr
o
k
e
r
eg
i
o
n
an
al
y
s
is
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
3
9
0
-
1
4
0
1
1396
3
.
6
.
E
x
t
re
m
e
g
r
a
dient
bo
o
s
t
ing
T
h
e
tr
ee
-
b
ased
alg
o
r
ith
m
ca
ll
ed
XGBo
o
s
t
h
as
g
ain
ed
p
o
p
u
l
ar
ity
lately
f
o
r
d
ata
ca
teg
o
r
iz
a
tio
n
an
d
is
a
v
er
y
s
u
cc
ess
f
u
l
tech
n
i
q
u
e.
A
v
er
y
s
ca
lab
le
en
d
-
to
-
e
n
d
tr
ee
-
b
o
o
s
tin
g
m
eth
o
d
,
XGBo
o
s
t
is
u
s
ed
in
m
ac
h
in
e
lear
n
in
g
f
o
r
p
r
o
b
lem
s
in
clu
d
i
n
g
r
e
g
r
ess
io
n
an
d
class
if
icatio
n
[
2
0
]
.
Giv
en
a
d
ataset
with
n
s
am
p
les
an
d
m
f
ea
tu
r
es,
let
{
(
,
)
}
=
1
r
ep
r
esen
t
th
e
tr
ai
n
in
g
d
ata,
wh
er
e
∈
d
en
o
tes
th
e
in
p
u
t
f
ea
tu
r
es
an
d
is
th
e
co
r
r
esp
o
n
d
in
g
la
b
el.
T
h
e
m
o
d
el
p
r
ed
icts
a
̂
u
s
in
g
d
ec
is
io
n
tr
ee
s
as in
(
2
9
)
.
̂
=
∑
(
)
∈
=
1
(
2
9
)
W
h
er
e
F
is
th
e
s
p
ac
e
o
f
r
eg
r
ess
io
n
tr
ee
s
.
T
h
e
o
b
jectiv
e
f
u
n
cti
o
n
o
f
XGBo
o
s
t
co
n
s
is
t
s
o
f
tw
o
co
m
p
o
n
en
ts
:
th
e
lo
s
s
f
u
n
ctio
n
an
d
th
e
r
eg
u
lar
i
za
tio
n
ter
m
Ω
,
g
iv
en
b
y
(
3
0
)
.
=
∑
(
,
̂
)
+
∑
(
)
=
1
=
1
(
3
0
)
W
h
er
e
th
e
lo
s
s
f
u
n
ctio
n
(
,
̂
)
m
ea
s
u
r
es th
e
d
if
f
e
r
en
ce
b
etwe
en
th
e
ac
tu
al
an
d
p
r
ed
icted
v
alu
es.
(
)
=
+
1
2
∑
2
=
1
(
3
1
)
W
h
er
e
is
n
u
m
b
er
o
f
leav
es in
tr
ee
,
r
ep
r
esen
ts
leaf
weig
h
ts
an
d
γ
,
a
n
d
λ
ar
e
r
eg
u
lar
izatio
n
p
ar
am
eter
s
.
(
)
≈
∑
[
(
)
+
1
2
ℎ
2
(
)
]
=
1
+
(
)
(
3
2
)
W
h
er
e
=
∂
L
(
y
,
̂
)
∂
̂
an
d
ℎ
=
∂
2
L
(
y
,
̂
)
∂
̂
2
ar
e
th
e
f
ir
s
t
-
a
n
d
s
ec
o
n
d
-
o
r
d
er
g
r
ad
ien
ts
o
f
th
e
l
o
s
s
f
u
n
ctio
n
.
=
1
2
[
(
∑
)
2
,
∑
+
ℎ
,
+
(
∑
)
2
,
∑
+
ℎ
,
−
(
∑
)
2
∑
+
ℎ
]
−
(
3
3
)
W
h
er
e
,
an
d
ℎ
,
ℎ
d
e
n
o
te
th
e
g
r
ad
ien
t
an
d
Hess
ian
v
alu
es
f
o
r
th
e
lef
t
an
d
r
ig
h
t
ch
ild
n
o
d
es,
r
esp
ec
tiv
ely
.
T
ab
le
2
s
h
o
ws
th
e
tr
ain
in
g
p
a
r
am
eter
s
o
f
XGB
o
o
s
t.
T
ab
le
2
.
T
r
ai
n
in
g
p
ar
am
eter
s
o
f
XGBo
o
s
t
P
a
r
a
me
t
e
r
Ty
p
i
c
a
l
v
a
l
u
e
s
l
e
a
r
n
i
n
g
_
r
a
t
e
0
.
0
2
max
_
d
e
p
t
h
5
mi
n
_
c
h
i
l
d
_
w
e
i
g
h
t
6
l
a
m
b
d
a
8
g
a
mm
a
5
t
r
e
e
_
me
t
h
o
d
h
i
s
t
b
o
o
s
t
e
r
g
b
t
r
e
e
o
b
j
e
c
t
i
v
e
b
i
n
a
r
y
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
W
e
im
p
lem
en
ted
an
d
ev
alu
at
ed
th
e
p
r
o
p
o
s
ed
m
eth
o
d
s
u
s
in
g
th
e
Py
T
o
r
c
h
6
f
r
am
ewo
r
k
in
Py
th
o
n
3
.
6
,
e
n
s
u
r
in
g
ef
f
icien
t
d
ee
p
-
lear
n
in
g
m
o
d
el
ex
ec
u
tio
n
.
T
h
e
e
x
p
er
im
e
n
ts
wer
e
co
n
d
u
cte
d
o
n
a
h
ig
h
-
p
e
r
f
o
r
m
an
ce
co
m
p
u
tin
g
s
etu
p
co
n
s
is
tin
g
o
f
a
PC
eq
u
ip
p
ed
wi
th
1
6
GB
o
f
R
AM
,
an
I
n
tel
C
o
r
e
i7
-
4
7
9
0
p
r
o
ce
s
s
o
r
clo
c
k
ed
at
3
.
6
0
G
Hz,
an
d
an
NVI
DI
A
T
I
T
AN
X
GPU.
T
h
is
h
ar
d
war
e
c
o
n
f
ig
u
r
atio
n
en
ab
le
d
ac
ce
ler
ated
tr
ain
in
g
an
d
in
f
e
r
en
ce
,
f
ac
ilit
atin
g
th
e
ef
f
ec
tiv
e
p
r
o
ce
s
s
in
g
o
f
lar
g
e
-
s
ca
le
m
ed
ical
im
ag
in
g
d
ata.
T
h
e
d
ataset
was
s
p
lit
in
to
7
0
%
f
o
r
tr
ain
in
g
,
1
5
%
f
o
r
v
ali
d
atio
n
,
an
d
1
5
%
f
o
r
test
in
g
t
o
en
s
u
r
e
b
ala
n
ce
d
ev
alu
atio
n
an
d
p
r
e
v
en
t d
ata
leak
ag
e.
4
.
1
.
Da
t
a
s
et
T
h
er
e
ar
e
two
b
r
ain
h
em
o
r
r
h
a
g
e
d
atasets
u
tili
ze
d
in
o
u
r
s
tu
d
y
.
T
h
e
first
o
n
e,
as
th
e
s
o
u
r
ce
d
o
m
ain
o
f
tr
an
s
f
er
le
ar
n
i
n
g
,
is
th
e
R
ad
io
lo
g
ical
So
ciety
o
f
No
r
th
A
m
er
ica
(
R
SNA
)
in
tr
ac
r
an
ial
h
em
o
r
r
h
ag
e
(
I
C
H)
d
etec
tio
n
d
ataset
[
2
1
]
.
W
h
ich
is
a
p
u
b
lic
d
ataset,
co
n
s
is
tin
g
o
f
1
9
,
5
3
0
ca
s
es
o
f
b
r
ai
n
C
T
,
in
clu
d
in
g
8
,
0
0
3
ca
s
es
with
I
C
H
an
d
1
1
,
5
2
7
n
o
h
em
o
r
r
h
a
g
e
ca
s
es.
I
n
th
e
8
,
0
0
3
I
C
H
ca
s
e
s
,
co
v
er
in
g
fiv
e
k
in
d
s
o
f
s
u
b
ty
p
es
o
f
I
C
H
in
clu
d
in
g
s
u
b
ar
ac
h
n
o
id
h
em
o
r
r
h
ag
e
(
SAH)
,
s
u
b
d
u
r
al
h
em
o
r
r
h
ag
e
(
SDH)
,
ep
id
u
r
al
h
em
o
r
r
h
ag
e
(
E
DH)
,
in
tr
ap
ar
en
c
h
y
m
al
h
em
o
r
r
h
ag
e
(
I
PH)
,
a
n
d
in
tr
av
e
n
tr
icu
lar
h
e
m
o
r
r
h
a
g
e
(
I
VH)
[
2
2
]
.
Fig
u
r
e
2
s
h
o
ws
th
e
s
am
p
le
im
ag
es f
r
o
m
th
e
d
ataset.
T
ab
l
e
3
s
h
o
ws th
e
d
escr
ip
tio
n
o
f
t
h
e
R
SNA
I
C
H
d
etec
tio
n
d
ataset
.
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
A
cc
u
r
a
te
s
tr
o
ke
a
r
ea
cla
s
s
ific
a
tio
n
u
s
in
g
ex
tr
eme
g
r
a
d
ie
n
t b
o
o
s
tin
g
…
(
K
a
viko
n
d
a
la
P
r
a
v
ee
n
K
u
ma
r
R
a
o
)
1397
Fig
u
r
e
2
.
Sam
p
le
im
a
g
es f
r
o
m
th
e
d
ataset
T
ab
le
3
.
Descr
ip
tio
n
o
f
R
SNA
I
C
H
d
etec
tio
n
d
ataset
C
l
a
s
ses
C
a
ses
I
n
t
r
a
p
a
r
e
n
c
h
y
ma
l
4
,
7
9
6
I
n
t
r
a
v
e
n
t
r
i
c
u
l
a
r
3
,
3
1
3
S
u
b
a
r
a
c
h
n
o
i
d
3
,
5
4
9
S
u
b
d
u
r
a
l
3
,
4
4
2
Ep
i
d
u
r
a
l
3
1
3
A
n
y
8
,
0
0
3
4
.
2
.
P
er
f
o
r
m
a
nce
m
e
t
rics
T
h
e
ev
alu
atio
n
o
f
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
m
o
d
els
r
elies
o
n
s
ev
er
al
q
u
an
titativ
e
p
er
f
o
r
m
a
n
ce
m
etr
ics
to
ass
es
s
th
eir
ac
cu
r
ac
y
an
d
r
elia
b
ilit
y
[
2
3
]
.
Fo
r
s
e
g
m
en
tatio
n
task
s
,
m
etr
ics
s
u
ch
as
Dice
s
im
ilar
ity
co
ef
f
icien
t
(
DS
C
)
,
J
ac
ca
r
d
in
d
ex
o
r
in
ter
s
ec
tio
n
o
v
er
u
n
i
o
n
(
I
o
U)
,
v
o
l
u
m
etr
ic
o
v
e
r
lap
er
r
o
r
(
VOE
)
,
an
d
m
ea
n
s
u
r
f
ac
e
d
is
tan
ce
(
MSD)
ar
e
co
m
m
o
n
l
y
u
s
ed
to
m
ea
s
u
r
e
t
h
e
o
v
er
lap
b
etwe
en
p
r
ed
icte
d
an
d
g
r
o
u
n
d
tr
u
th
r
eg
io
n
s
[
2
4
]
.
T
h
e
Dice
co
ef
f
icien
t
an
d
J
ac
ca
r
d
i
n
d
ex
q
u
an
tif
y
th
e
s
p
atial
s
im
ilar
ity
o
f
s
eg
m
en
ted
ar
ea
s
,
wh
ile
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
ev
a
lu
ate
th
e
m
o
d
el
’
s
ab
ilit
y
to
c
o
r
r
ec
tly
id
e
n
tify
tr
u
e
p
o
s
itiv
e
an
d
tr
u
e
n
eg
ativ
e
p
ix
els,
r
esp
ec
tiv
ely
[
2
5
]
,
[
2
6
]
.
Fo
r
class
if
icatio
n
task
s
,
m
et
r
ics
s
u
ch
s
u
ch
as
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
,
an
d
F1
-
s
co
r
e
ar
e
em
p
lo
y
ed
to
ass
ess
th
e
m
o
d
el’
s
d
is
cr
im
in
ativ
e
ca
p
ab
ilit
y
am
o
n
g
class
es.
Fig
u
r
e
3
s
h
o
ws
th
e
s
a
m
p
le
s
e
g
m
e
n
t
ed
im
ag
es.
T
h
e
p
e
r
f
o
r
m
a
n
ce
ass
ess
m
e
n
t
o
f
t
h
e
p
r
o
p
o
s
e
d
m
et
h
o
d
o
lo
g
y
a
p
p
ea
r
s
i
n
T
ab
l
es
4
a
n
d
5
f
o
r
b
o
t
h
cl
ass
i
f
ic
ati
o
n
al
o
n
g
s
e
g
m
e
n
t
ati
o
n
.
T
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
d
is
p
la
y
s
h
i
g
h
a
cc
u
r
ac
y
r
at
es
b
etwe
e
n
9
8
.
1
7
%
an
d
9
9
.
4
2
%
w
h
e
n
i
d
e
n
t
if
y
i
n
g
i
n
t
r
a
p
a
r
e
n
ch
y
m
al
,
i
n
t
r
a
v
e
n
t
r
i
c
u
la
r
,
s
u
b
a
r
ac
h
n
o
i
d
,
s
u
b
d
u
r
a
l,
a
n
d
e
p
i
d
u
r
al
o
f
h
em
o
r
r
h
ag
es
f
o
r
c
la
s
s
if
i
ca
t
io
n
p
u
r
p
o
s
es
.
T
h
e
d
et
e
cti
o
n
ca
p
ab
ilit
ies
o
f
th
e
m
et
h
o
d
ar
e
s
t
r
o
n
g
b
e
ca
u
s
e
SDH
d
e
m
o
n
s
tr
ate
d
t
h
e
h
i
g
h
est
s
e
n
s
it
iv
it
y
v
a
lu
e
o
f
9
9
.
1
2
%.
Fig
u
r
e
3
.
Sam
p
le
s
eg
m
en
ted
i
m
ag
es
T
ab
le
4
.
C
lass
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
p
r
o
p
o
s
ed
m
eth
o
d
M
e
t
r
i
c
s
I
n
t
r
a
p
a
r
e
n
c
h
y
ma
l
I
n
t
r
a
v
e
n
t
r
i
c
u
l
a
r
S
u
b
a
r
a
c
h
n
o
i
d
S
u
b
d
u
r
a
l
Ep
i
d
u
r
a
l
A
c
c
u
r
a
c
y
9
9
.
0
1
9
8
.
7
2
9
9
.
4
2
9
8
.
7
4
9
8
.
1
7
S
e
n
s
i
t
i
v
i
t
y
9
5
.
2
5
9
8
.
2
7
9
8
.
4
5
9
9
.
1
2
9
8
.
7
2
S
p
e
c
i
f
i
c
i
t
y
9
8
.
8
6
9
9
.
3
0
9
7
.
4
9
9
7
.
6
9
9
9
.
4
3
P
r
e
c
i
s
i
o
n
9
7
.
4
5
9
5
.
7
4
9
8
.
6
9
9
9
.
1
3
9
9
.
5
8
T
ab
le
5
.
S
eg
m
e
n
tatio
n
p
e
r
f
o
r
m
an
ce
o
f
p
r
o
p
o
s
ed
m
eth
o
d
M
e
t
r
i
c
s
V
a
l
u
e
s
D
S
C
9
8
.
5
6
I
o
U
9
8
.
3
1
H
a
u
s
d
o
r
f
f
d
i
s
t
a
n
c
e
(
HD
)
1
2
.
9
6
VOE
9
7
.
9
6
M
S
D
9
4
.
9
0
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
u
n
d
er
g
o
es
ev
alu
atio
n
a
g
ain
s
t
ex
is
tin
g
ap
p
r
o
ac
h
es
u
s
in
g
C
NN
-
b
ased
m
o
d
els
an
d
W
-
Net
an
d
co
n
s
is
ten
t
p
e
r
ce
p
tio
n
g
en
e
r
ativ
e
ad
v
er
s
ar
i
al
n
etwo
r
k
(
C
PGAN)
a
s
s
h
o
wn
in
T
ab
le
6
.
T
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
3
9
0
-
1
4
0
1
1398
p
r
o
p
o
s
ed
m
eth
o
d
o
u
tp
er
f
o
r
m
s
all
b
aselin
es
in
k
ey
p
er
f
o
r
m
an
ce
m
et
r
ics.
T
h
e
m
o
d
el
d
em
o
n
s
tr
ates
th
e
b
est
ac
cu
r
ac
y
r
ate
o
f
9
9
.
1
2
%
alo
n
g
s
id
e
9
9
.
0
6
%
s
p
ec
if
icity
wh
ich
en
ab
les
d
ep
en
d
ab
le
class
if
icatio
n
with
f
ew
in
co
r
r
ec
t
r
esu
lts
.
T
h
e
d
etec
ti
o
n
ca
p
ab
ilit
ies
o
f
th
is
m
eth
o
d
ar
e
b
alan
ce
d
b
ec
au
s
e
it
d
e
m
o
n
s
tr
ates
9
8
.
6
9
%
s
en
s
itiv
ity
an
d
9
8
.
9
8
% p
r
ec
is
io
n
.
T
ab
le
7
p
r
esen
ts
an
ab
latio
n
s
tu
d
y
an
aly
zin
g
th
e
co
n
tr
ib
u
ti
o
n
o
f
ea
ch
f
ea
tu
r
e
ex
tr
ac
tio
n
m
eth
o
d
to
th
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
f
r
am
ew
o
r
k
.
Usi
n
g
C
o
n
v
NeX
t
alo
n
e
p
r
o
v
id
es
a
s
tr
o
n
g
b
aselin
e,
b
u
t
in
te
g
r
atin
g
h
an
d
cr
af
te
d
f
ea
tu
r
es
s
ig
n
if
ican
tl
y
en
h
a
n
ce
s
s
eg
m
en
tatio
n
ac
c
u
r
ac
y
.
Ad
d
in
g
L
B
P
im
p
r
o
v
es
b
o
u
n
d
ar
y
d
etec
tio
n
an
d
s
tr
u
ctu
r
al
d
etail
ca
p
tu
r
e
,
wh
ile
AT
-
DB
GM
co
n
tr
ib
u
t
es
to
m
o
r
e
p
r
ec
is
e
d
ir
ec
tio
n
al
p
atter
n
r
ec
o
g
n
itio
n
in
s
tr
o
k
e
a
r
ea
.
T
h
e
co
m
b
in
atio
n
o
f
all
th
r
ee
h
a
n
d
cr
a
f
te
d
f
ea
tu
r
es
with
C
o
n
v
NeX
t
y
ield
s
th
e
h
ig
h
es
t
p
er
f
o
r
m
an
ce
,
ac
h
iev
in
g
a
Dice
s
co
r
e
o
f
9
8
.
5
6
%,
HD
o
f
1
2
.
9
6
m
m
,
an
d
ac
cu
r
ac
y
o
f
9
9
.
1
2
%,
d
em
o
n
s
tr
atin
g
th
at
ea
ch
f
ea
tu
r
e
ex
tr
a
ctio
n
co
m
p
o
n
en
t
p
lay
s
a
co
m
p
lem
en
tar
y
r
o
le
in
im
p
r
o
v
in
g
lesi
o
n
d
elin
ea
tio
n
an
d
class
if
ic
atio
n
.
T
h
e
r
esu
lts
co
n
f
ir
m
e
d
th
at
o
u
r
h
y
b
r
id
f
r
am
ewo
r
k
ac
h
iev
ed
s
tatis
t
ically
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
,
with
DSC
an
d
F1
-
s
co
r
e
en
h
an
ce
m
e
n
ts
o
v
er
W
-
Net
an
d
C
PGAN
y
ield
in
g
p
<0
.
0
1
.
C
o
n
f
id
e
n
ce
in
ter
v
als
f
u
r
th
er
v
alid
ated
th
e
r
o
b
u
s
tn
ess
o
f
th
ese
f
in
d
in
g
s
,
wh
ile
ef
f
ec
t
s
iz
e
ca
lcu
latio
n
s
(
C
o
h
en
’
s
d
>
1
.
8
)
in
d
icate
d
a
s
tr
o
n
g
p
r
ac
tical
im
p
ac
t.
T
ab
le
6
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
P
e
r
f
o
r
ma
n
c
e
C
N
N
[
7
]
C
N
N
[
8
]
W
-
N
e
t
[
1
0
]
C
P
G
A
N
[
1
1
]
V
i
T
[
2
6
]
Th
i
s
w
o
r
k
D
S
C
7
9
.
1
3
9
7
.
4
6
8
5
.
6
0
6
1
.
7
9
6
.
9
9
98
.
5
6
H
D
(
mm
)
2
5
.
0
2
2
8
.
0
2
2
7
.
3
4
2
9
.
5
8
3
2
.
5
8
1
2
.
9
6
A
c
c
u
r
a
c
y
8
8
.
7
6
9
8
.
9
1
8
9
.
7
6
6
3
.
8
9
7
.
5
9
9
9
.
1
2
S
e
n
s
i
t
i
v
i
t
y
8
6
.
0
8
9
7
.
4
6
8
5
.
3
9
5
5
.
6
9
7
.
0
0
9
8
.
6
9
S
p
e
c
i
f
i
c
i
t
y
8
9
.
8
6
9
6
.
6
7
8
8
.
1
2
7
0
.
5
9
7
.
0
0
9
9
.
0
6
P
r
e
c
i
s
i
o
n
9
2
.
6
7
9
7
.
9
6
8
8
.
3
4
7
5
.
6
3
9
7
.
0
0
9
8
.
9
8
F1
-
sc
o
r
e
8
9
.
2
5
9
8
.
6
3
8
5
.
2
9
7
0
.
3
7
9
7
.
0
0
98
.
8
5
T
ab
le
7
.
Ab
latio
n
s
tu
d
y
o
f
p
r
o
p
o
s
ed
m
eth
o
d
C
o
n
f
i
g
u
r
a
t
i
o
n
D
S
C
HD
A
c
c
u
r
a
c
y
S
e
n
s
i
t
i
v
i
t
y
S
p
e
c
i
f
i
c
i
t
y
C
o
n
v
N
e
X
t
o
n
l
y
9
2
.
1
3
1
8
.
4
5
9
6
.
7
8
9
5
.
0
2
9
7
.
2
1
C
o
n
v
N
e
X
t
+
L
B
P
9
4
.
0
5
1
5
.
8
7
9
7
.
3
5
9
6
.
1
8
9
7
.
9
8
C
o
n
v
N
e
X
t
+
A
T
-
D
B
G
M
9
4
.
6
2
1
4
.
3
2
9
7
.
6
8
9
6
.
5
4
9
8
.
1
2
C
o
n
v
N
e
X
t
+
W
P
T
9
4
.
4
0
1
4
.
8
5
9
7
.
5
0
9
6
.
4
0
9
8
.
0
5
C
o
n
v
N
e
X
t
+
L
B
P
+
A
T
-
D
B
G
M
+
W
P
T
9
8
.
5
6
1
2
.
9
6
9
9
.
1
2
9
8
.
6
9
9
9
.
0
6
4
.
3
.
Dis
cus
s
io
ns
T
h
e
ex
p
e
r
im
en
tal
r
esu
lts
clea
r
ly
d
em
o
n
s
tr
ate
th
e
ef
f
ec
tiv
en
e
s
s
o
f
th
e
p
r
o
p
o
s
ed
C
o
n
v
NeX
t
-
XGBo
o
s
t
h
y
b
r
id
f
r
am
ew
o
r
k
i
n
s
tr
o
k
e
ar
ea
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
.
C
o
m
p
ar
ed
with
c
o
n
v
en
tio
n
al
C
NNs,
W
-
Net,
an
d
C
PGAN
-
b
ased
ap
p
r
o
ac
h
es,
o
u
r
m
eth
o
d
ac
h
ie
v
ed
th
e
h
ig
h
est
Dice
s
co
r
e
(
9
8
.
5
6
%),
p
r
ec
is
io
n
(
9
8
.
9
8
%),
an
d
F1
-
s
co
r
e
(
9
8
.
8
5
%),
wh
ile
also
r
e
p
o
r
tin
g
th
e
lo
west
HD
(
1
2
.
9
6
m
m
)
.
T
h
ese
im
p
r
o
v
em
en
ts
h
ig
h
lig
h
t
t
h
e
ab
ilit
y
o
f
o
u
r
f
r
a
m
ewo
r
k
t
o
g
e
n
er
ate
m
o
r
e
ac
c
u
r
ate
lesi
o
n
b
o
u
n
d
ar
ies,
m
i
n
im
ize
f
alse
p
o
s
itiv
es,
an
d
r
e
d
u
ce
o
v
e
r
-
s
eg
m
en
tatio
n
er
r
o
r
s
.
T
h
e
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
ca
n
b
e
attr
ib
u
ted
to
t
h
e
m
u
lti
-
lev
el
f
ea
tu
r
e
f
u
s
io
n
s
tr
ateg
y
.
C
o
n
v
NeX
t
ef
f
icien
tly
ex
tr
ac
ts
d
ee
p
s
em
an
tic
r
ep
r
esen
tatio
n
s
o
f
s
tr
o
k
e
ar
ea
,
wh
ile
h
an
d
cr
a
f
ted
d
escr
ip
to
r
s
s
u
c
h
as
L
B
P,
AT
-
DB
GM
,
an
d
W
PT
p
r
eser
v
e
lo
ca
l
tex
tu
r
e
an
d
s
tr
u
ctu
r
al
d
etails.
B
y
co
m
b
in
in
g
th
ese
co
m
p
lem
en
t
ar
y
f
ea
tu
r
e
s
ets,
th
e
m
o
d
el
ca
p
tu
r
es
b
o
th
g
lo
b
al
c
o
n
tex
t
a
n
d
f
in
e
-
g
r
ain
ed
lesi
o
n
p
atter
n
s
,
lead
in
g
to
b
etter
d
elin
ea
tio
n
o
f
ir
r
eg
u
la
r
an
d
s
m
all
s
tr
o
k
e
r
eg
io
n
s
.
Fu
r
th
er
m
o
r
e
,
th
e
in
teg
r
atio
n
o
f
XGBo
o
s
t
en
h
an
ce
s
class
if
icatio
n
r
o
b
u
s
tn
ess
b
y
ex
p
l
o
itin
g
g
r
a
d
ien
t
-
b
o
o
s
ted
d
ec
is
io
n
tr
ee
s
f
o
r
h
an
d
lin
g
co
m
p
lex
f
ea
t
u
r
e
in
ter
ac
tio
n
s
,
th
er
eb
y
o
u
tp
e
r
f
o
r
m
in
g
e
n
d
-
to
-
en
d
C
NN
-
o
n
ly
ap
p
r
o
ac
h
es.
S
t
a
ti
s
t
i
c
al
s
i
g
n
i
f
i
c
a
n
c
e
t
es
t
i
n
g
f
u
r
t
h
e
r
v
a
l
i
d
a
t
es
t
h
e
r
o
b
u
s
tn
e
s
s
o
f
o
u
r
r
es
u
l
ts
.
Pa
i
r
e
d
t
-
t
e
s
ts
a
n
d
W
i
lc
o
x
o
n
s
i
g
n
e
d
-
r
a
n
k
t
es
t
s
c
o
n
f
i
r
m
e
d
t
h
a
t
t
h
e
p
e
r
f
o
r
m
a
n
c
e
im
p
r
o
v
e
m
e
n
t
s
,
p
a
r
t
i
c
u
la
r
l
y
i
n
Di
c
e
a
n
d
I
o
U
m
e
t
r
i
c
s
,
w
e
r
e
s
t
a
t
is
ti
c
a
ll
y
s
i
g
n
i
f
i
c
a
n
t
(
p
<
0
.
0
1
)
w
h
e
n
c
o
m
p
a
r
e
d
w
i
t
h
W
-
N
et
a
n
d
C
N
N
b
as
e
l
i
n
es
.
T
h
e
s
e
f
i
n
d
i
n
g
s
p
r
o
v
i
d
e
s
t
r
o
n
g
e
v
i
d
e
n
c
e
t
h
a
t
t
h
e
o
b
s
e
r
v
e
d
i
m
p
r
o
v
e
m
e
n
t
s
a
r
e
n
o
t
d
u
e
t
o
r
a
n
d
o
m
c
h
a
n
c
e
b
u
t
s
t
e
m
f
r
o
m
t
h
e
a
r
c
h
it
e
c
t
u
r
al
i
n
n
o
v
a
t
i
o
n
s
i
n
t
r
o
d
u
c
e
d
i
n
t
h
i
s
w
o
r
k
.
D
e
s
p
i
t
e
t
h
e
p
r
o
p
o
s
e
d
C
o
n
v
N
e
X
t
-
h
a
n
d
c
r
a
f
t
e
d
f
e
a
t
u
r
e
h
y
b
r
i
d
w
h
i
c
h
i
s
o
p
e
r
a
t
e
d
u
s
i
n
g
XG
B
o
o
s
t
d
e
m
o
n
s
t
r
a
t
i
n
g
b
e
tt
e
r
p
e
r
f
o
r
m
a
n
c
e
i
n
t
e
r
m
s
o
f
s
t
r
o
k
e
a
r
e
a
s
e
g
m
en
t
a
t
i
o
n
,
a
n
u
m
b
e
r
o
f
l
i
m
i
t
at
i
o
n
s
c
a
n
b
e
n
o
t
e
d
.
F
i
r
s
t,
t
h
e
e
x
p
e
r
i
m
e
n
t
s
h
a
v
e
b
e
e
n
d
o
n
e
o
n
a
s
m
a
l
l
s
e
t
o
f
d
a
t
a
th
a
t
m
i
g
h
t
l
i
m
i
t
t
h
e
a
p
p
l
i
c
a
t
i
o
n
o
f
t
h
e
r
e
s
u
l
ts
t
o
t
h
e
w
i
d
e
r
a
n
d
m
o
r
e
h
et
e
r
o
g
e
n
e
o
u
s
c
l
i
n
i
c
a
l
s
e
tt
i
n
g
s
.
A
lt
h
o
u
g
h
t
h
e
p
r
o
p
o
s
e
d
f
r
a
m
e
w
o
r
k
w
a
s
e
v
a
l
u
a
t
e
d
o
n
t
h
e
R
S
N
A
d
a
t
as
e
t
,
it
s
r
o
b
u
s
t
p
e
r
f
o
r
m
a
n
c
e
i
n
d
i
c
a
t
e
s
s
t
r
o
n
g
p
o
t
e
n
t
i
a
l
f
o
r
g
e
n
e
r
a
l
i
z
a
ti
o
n
t
o
o
t
h
e
r
s
t
r
o
k
e
i
m
a
g
i
n
g
b
e
n
c
h
m
a
r
k
s
s
u
c
h
as
s
t
r
o
k
e
i
m
a
g
i
n
g
s
e
g
m
e
n
t
a
t
i
o
n
s
t
u
d
y
(
S
I
SS
)
a
n
d
i
s
c
h
e
m
ic
s
t
r
o
k
e
l
e
s
i
o
n
s
e
g
m
en
t
a
t
i
o
n
(
I
S
L
E
S
)
.
T
h
e
m
o
d
e
l
’
s
a
r
c
h
i
te
c
t
u
r
e
,
d
es
i
g
n
e
d
w
it
h
m
o
d
a
l
i
t
y
-
i
n
v
a
r
i
a
n
t
f
e
a
t
u
r
e
e
x
t
r
a
ct
i
o
n
a
n
d
s
p
at
i
a
l
a
t
t
e
n
t
i
o
n
m
e
c
h
a
n
is
m
s
,
c
a
n
a
d
ap
t
t
o
v
a
r
i
a
t
i
o
n
s
i
n
i
m
a
g
e
c
o
n
t
r
a
s
t
,
r
e
s
o
l
u
t
i
o
n
,
a
n
d
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
A
cc
u
r
a
te
s
tr
o
ke
a
r
ea
cla
s
s
ific
a
tio
n
u
s
in
g
ex
tr
eme
g
r
a
d
ie
n
t b
o
o
s
tin
g
…
(
K
a
viko
n
d
a
la
P
r
a
v
ee
n
K
u
ma
r
R
a
o
)
1399
a
c
q
u
i
s
it
i
o
n
s
e
tt
i
n
g
s
.
F
u
t
u
r
e
w
o
r
k
w
i
l
l
f
o
c
u
s
o
n
c
r
o
s
s
-
d
a
ta
s
et
v
a
l
i
d
a
t
i
o
n
a
n
d
m
u
l
t
i
-
c
e
n
t
e
r
s
t
u
d
ie
s
t
o
f
u
r
t
h
e
r
a
s
s
e
s
s
g
e
n
e
r
a
l
i
z
a
b
i
li
t
y
a
n
d
c
li
n
i
c
a
l
r
eli
a
b
i
l
it
y
a
c
r
o
s
s
d
i
v
e
r
s
e
p
a
t
i
e
n
t
p
o
p
u
l
a
t
i
o
n
s
a
n
d
i
m
a
g
i
n
g
p
r
o
t
o
co
l
s
.
T
h
e
en
h
a
n
ce
d
s
eg
m
en
tatio
n
a
n
d
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
h
o
ld
s
ig
n
if
ican
t
clin
ical
v
alu
e
i
n
s
tr
o
k
e
d
iag
n
o
s
is
an
d
m
a
n
ag
em
en
t.
Acc
u
r
ate
d
elin
ea
tio
n
o
f
th
e
s
tr
o
k
e
-
af
f
ec
ted
r
eg
io
n
s
ca
n
ass
is
t
r
a
d
io
lo
g
is
ts
in
r
ap
id
ly
id
en
tify
in
g
lesi
o
n
b
o
u
n
d
ar
ies,
r
ed
u
cin
g
in
te
r
p
r
etatio
n
tim
e
an
d
in
ter
-
o
b
s
er
v
er
v
ar
iab
ilit
y
.
T
h
e
p
r
ec
is
e
q
u
an
ti
f
icatio
n
o
f
in
f
a
r
cted
tis
s
u
e
s
u
p
p
o
r
ts
ea
r
l
y
d
ia
g
n
o
s
is
,
p
r
o
g
n
o
s
is
ass
e
s
s
m
en
t,
an
d
tr
ea
tm
en
t
p
lan
n
in
g
,
en
ab
lin
g
t
im
ely
th
er
ap
e
u
tic
d
ec
is
io
n
s
s
u
ch
as
th
r
o
m
b
o
ly
s
is
o
r
th
r
o
m
b
e
cto
m
y
.
I
n
teg
r
atio
n
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
i
n
to
clin
ical
wo
r
k
f
lo
ws
ca
n
t
h
u
s
s
er
v
e
as
a
d
ec
is
io
n
-
s
u
p
p
o
r
t
to
o
l,
im
p
r
o
v
in
g
d
iag
n
o
s
tic
co
n
s
is
ten
cy
an
d
f
ac
ilit
atin
g
p
er
s
o
n
alize
d
tr
ea
tm
en
t stra
teg
ies f
o
r
s
tr
o
k
e
p
at
ien
ts
.
5.
CO
NCLU
SI
O
N
T
h
e
r
esear
ch
d
ev
elo
p
e
d
a
d
ee
p
lear
n
in
g
f
u
s
io
n
s
y
s
tem
wh
i
ch
in
teg
r
ates
C
o
n
v
NeX
t
with
ad
v
an
ce
d
f
ea
tu
r
e
ex
tr
ac
tio
n
m
eth
o
d
s
an
d
XGBo
o
s
t
class
if
icatio
n
co
m
p
o
n
en
ts
to
e
x
ec
u
te
ac
cu
r
ate
s
tr
o
k
e
z
o
n
e
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
o
p
er
atio
n
s
.
T
h
e
s
eg
m
en
ta
tio
n
s
y
s
tem
d
eliv
er
s
h
ig
h
p
er
f
o
r
m
an
ce
ac
cu
r
ac
y
b
ec
au
s
e
it
im
p
lem
en
ts
d
ee
p
le
ar
n
in
g
al
g
o
r
ith
m
s
to
co
m
b
i
n
e
L
B
P
AT
-
DB
GM
an
d
W
PT
h
a
n
d
cr
af
ted
f
ea
t
u
r
es.
T
h
e
XGBo
o
s
t
clas
s
if
ier
p
er
f
o
r
m
s
ac
cu
r
ate
s
tr
o
k
e
id
en
tific
atio
n
th
r
o
u
g
h
its
ab
ilit
y
to
co
m
b
i
n
e
s
ev
er
al
f
ea
tu
r
es
d
u
r
in
g
ex
tr
ac
t
f
ea
tu
r
e
a
n
aly
s
is
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
p
r
o
c
ess
ed
d
ata
with
9
8
.
5
6
%
DSC
r
ate
an
d
1
2
.
9
6
m
m
HD
s
co
r
e
wh
ile
ac
h
iev
in
g
9
9
.
1
2
%
ac
cu
r
ac
y
alo
n
g
with
9
8
.
6
9
%
s
en
s
itiv
ity
,
9
9
.
0
6
%
s
p
ec
if
icity
,
9
8
.
9
8
%
p
r
ec
is
io
n
,
an
d
9
8
.
8
5
%
F1
-
s
co
r
e.
T
h
e
in
te
g
r
atio
n
o
f
m
a
n
u
all
y
d
er
iv
e
d
f
ea
tu
r
es
with
C
o
n
v
NeX
t
d
ee
p
lear
n
in
g
ca
p
ab
ilit
ies
cr
ea
tes
an
o
p
er
atio
n
-
r
ea
d
y
ac
cu
r
ate
m
e
d
ical
im
ag
e
s
tr
o
k
e
d
etec
tio
n
s
y
s
te
m
.
F
u
tu
r
e
wo
r
k
will
f
o
cu
s
o
n
r
ea
l
-
tim
e
d
ep
l
o
y
m
en
t,
in
teg
r
atio
n
with
p
ictu
r
e
ar
ch
iv
in
g
an
d
c
o
m
m
u
n
icatio
n
s
y
s
tem
(
PAC
S
)
s
y
s
tem
s
,
an
d
ex
ten
s
io
n
to
m
u
ltimo
d
al
im
ag
in
g
an
d
o
th
er
m
ed
ical
s
eg
m
en
tatio
n
task
s
s
u
ch
as
tu
m
o
r
an
d
lesi
o
n
d
etec
tio
n
.
T
h
e
f
r
am
ew
o
r
k
’
s
s
ca
lab
le
d
esig
n
d
em
o
n
s
tr
ates
p
o
ten
tial
f
o
r
b
r
o
ad
e
r
in
tellig
en
t
h
ea
lth
ca
r
e
ap
p
licatio
n
s
,
co
n
tr
i
b
u
tin
g
t
o
c
o
m
p
u
ter
v
is
io
n
,
clin
ical
d
ec
is
i
o
n
s
u
p
p
o
r
t,
an
d
AI
-
d
r
iv
en
d
ia
g
n
o
s
tic
s
y
s
tem
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
d
ec
la
r
es th
er
e
is
n
o
f
u
n
d
in
g
s
ar
e
p
r
o
v
id
ed
f
o
r
th
is
w
o
r
k
.
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
Kav
ik
o
n
d
ala
Pra
v
ee
n
Ku
m
ar
R
ao
✓
✓
✓
✓
✓
✓
✓
Ma
h
a
L
ak
s
h
m
i Bo
n
d
la
✓
✓
✓
✓
✓
✓
B
o
m
m
ar
aju
Srin
iv
asa
R
ao
✓
✓
✓
Am
b
id
i N
av
ee
n
a
✓
✓
✓
✓
✓
✓
K.
V.
B
alar
am
ak
r
is
h
n
a
✓
✓
✓
✓
✓
Srin
iv
asar
ao
Go
d
a
✓
✓
✓
✓
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
tates th
er
e
is
n
o
co
n
f
l
ict
o
f
in
ter
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
atasets
u
s
ed
an
d
/o
r
an
aly
ze
d
d
u
r
in
g
th
e
cu
r
r
e
n
t
s
tu
d
y
ar
e
av
ailab
le
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
au
th
o
r
,
[
KPKR
]
,
o
n
r
ea
s
o
n
a
b
le
r
eq
u
est.
Evaluation Warning : The document was created with Spire.PDF for Python.