I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
,
p
p
.
1188
~
1
1
9
8
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
1
6
i
3
.
pp
1
1
8
8
-
1
1
9
8
1188
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Pros
tate
ma
g
neti
c re
so
na
nce ima
g
i
ng
/trans
rect
a
l ul
t
ra
so
und
regi
stra
tion usin
g
v
isio
n
trans
form
er
a
nd
co
nv
o
lutio
na
l neural
network
H
a
na
e
M
a
hm
o
ud
i,
H
iba
Ra
m
a
da
n,
J
a
m
a
l R
if
f
i,
H
a
m
id
T
a
iri
LI
S
A
C
La
b
o
r
a
t
o
r
y
,
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
s
D
h
a
r
-
M
a
h
r
a
z
,
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
,
F
e
z
,
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
Sep
2
9
,
2
0
2
4
R
ev
is
ed
J
an
2
4
,
2
0
2
6
Acc
ep
ted
Ma
r
1
6
,
2
0
2
6
M
u
lt
imo
d
a
l
re
g
istratio
n
o
f
3
D
m
e
d
ica
l
ima
g
e
s
(3
D
-
M
Re
g
)
p
la
y
s
a
k
e
y
r
o
le
in
se
v
e
ra
l
m
e
d
ica
l
a
p
p
li
c
a
ti
o
n
s
a
n
d
re
m
a
in
s
a
v
e
ry
c
h
a
ll
e
n
g
in
g
tas
k
a
s
it
d
e
a
ls
with
m
u
lt
imo
d
a
l
ima
g
e
s
a
n
d
v
o
lu
m
e
tri
c
o
b
jec
ts
a
t
t
h
e
s
a
m
e
ti
m
e
.
Re
c
e
n
tl
y
,
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
s)
b
a
se
d
a
p
p
r
o
a
c
h
e
s
h
a
v
e
b
e
e
n
p
ro
p
o
se
d
to
s
o
lv
e
3
D
-
M
Re
g
.
Ho
we
v
e
r,
t
h
e
se
tec
h
n
iq
u
e
s
c
a
n
n
o
t
p
re
se
rv
e
th
e
g
lo
b
a
l
sp
a
ti
a
l
c
o
n
te
x
t
re
q
u
ire
d
fo
r
a
c
c
u
ra
te
a
ffin
e
re
g
istratio
n
sin
c
e
th
e
y
re
ly
o
n
c
o
n
v
o
lu
ti
o
n
a
n
d
re
g
i
o
n
a
l
c
l
u
ste
rin
g
o
p
e
ra
ti
o
n
s
.
To
so
l
v
e
th
e
se
p
r
o
b
lem
s,
we
p
ro
p
o
se
a
su
p
e
rv
ise
d
a
p
p
ro
a
c
h
th
a
t
c
o
m
b
in
e
s
b
o
t
h
CNN
a
n
d
t
h
e
v
isi
o
n
tra
n
sfo
rm
e
r
(Vi
T)
to
p
re
d
ict
a
d
e
n
se
d
isp
lac
e
m
e
n
t
field
(DD
F
).
In
a
first
ste
p
,
o
u
r
m
e
th
o
d
in
v
e
stig
a
tes
t
h
e
p
o
we
r
o
f
ViT
t
o
c
a
p
tu
re
g
lo
b
a
l
v
o
x
e
ls
d
e
p
e
n
d
e
n
c
ies
f
o
r
i
n
it
ial
r
ig
i
d
a
li
g
n
m
e
n
t
.
T
h
e
n
we
e
x
p
l
o
it
th
e
fo
rc
e
o
f
CNN
s
to
f
o
c
u
s
o
n
l
o
c
a
l
d
e
tails
with
i
n
p
re
-
a
li
g
n
e
d
c
o
n
c
a
ten
a
ted
in
p
u
t
3
D
m
o
v
i
n
g
a
n
d
f
ix
e
d
ima
g
e
s
a
n
d
e
stim
a
te
DD
F
,
w
h
ich
is
t
h
e
n
a
p
p
li
e
d
to
t
h
e
m
o
v
in
g
lab
e
ls.
O
u
r
m
e
th
o
d
h
a
s
b
e
e
n
v
a
li
d
a
ted
i
n
a
p
r
o
sta
t
e
m
a
g
n
e
ti
c
re
so
n
a
n
c
e
ima
g
in
g
/t
ra
n
sre
c
tal
u
lt
ra
so
u
n
d
(M
RI
/
TRUS
)
d
a
tas
e
t
a
n
d
a
c
h
iev
e
d
p
ro
m
isin
g
re
su
lt
s c
o
m
p
a
re
d
t
o
p
r
e
v
io
u
s w
o
rk
b
a
se
d
o
n
o
n
l
y
CNN
s
.
K
ey
w
o
r
d
s
:
3
D
m
ed
ical
im
ag
es
Af
f
in
e
r
eg
is
tr
atio
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Den
s
e
d
is
p
lace
m
en
t f
ield
Mu
ltimo
d
al
r
eg
is
tr
atio
n
Su
p
er
v
is
ed
r
eg
is
tr
atio
n
Vis
io
n
tr
an
s
f
o
r
m
er
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
:
Han
ae
Ma
h
m
o
u
d
i
L
I
SAC
L
ab
o
r
ato
r
y
,
De
p
ar
tm
e
n
t o
f
C
o
m
p
u
ter
Scien
ce
,
Facu
l
ty
o
f
Scien
ce
s
Dh
ar
-
Ma
h
r
az
,
Sid
i M
o
h
am
ed
B
en
Ab
d
ellah
Un
iv
er
s
ity
Fez
-
3
0
0
0
,
Mo
r
o
cc
o
E
m
ail:
h
an
ae
.
m
ah
m
o
u
d
i@
u
s
m
b
a.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
I
m
ag
e
r
eg
is
tr
atio
n
aim
s
to
ali
g
n
two
im
ag
es
o
f
th
e
s
am
e
s
ce
n
e
—
n
am
ely
a
r
ef
e
r
en
ce
im
ag
e
an
d
a
m
o
v
in
g
im
a
g
e
ac
q
u
ir
ed
at
d
if
f
er
en
t
tim
es,
f
r
o
m
d
is
tin
ct
v
iewp
o
in
ts
,
an
d
/o
r
u
s
in
g
d
if
f
er
en
t
im
ag
in
g
d
e
v
ices.
I
n
m
ed
ical
im
ag
in
g
,
th
e
o
b
je
ctiv
e
o
f
r
eg
is
tr
atio
n
is
to
esti
m
ate
an
ap
p
r
o
p
r
iate
s
p
atial
tr
an
s
f
o
r
m
atio
n
th
at
ac
h
iev
es
ac
cu
r
ate
alig
n
m
e
n
t
o
f
th
e
co
r
r
esp
o
n
d
in
g
an
ato
m
ic
al
s
tr
u
ctu
r
es.
Me
d
ical
im
ag
e
r
eg
is
tr
atio
n
(
MReg
)
is
th
er
ef
o
r
e
a
f
u
n
d
am
en
tal
c
o
m
p
o
n
en
t
i
n
m
a
n
y
clin
ical
an
d
r
esear
ch
a
p
p
licatio
n
s
[
1
]
,
in
clu
d
in
g
o
r
g
an
s
eg
m
en
tatio
n
[
2
]
,
atlas
co
n
s
tr
u
ctio
n
,
im
a
g
e
-
g
u
id
ed
in
ter
v
en
tio
n
s
f
o
r
d
iag
n
o
s
is
,
m
o
n
ito
r
i
n
g
,
an
d
th
er
a
p
y
,
as
well
as tele
-
s
u
r
g
er
y
an
d
p
o
s
t
-
o
p
er
ativ
e
ev
alu
atio
n
.
MReg
tech
n
iq
u
es
ca
n
b
e
cl
ass
if
ied
ac
co
r
d
in
g
to
s
ev
er
al
cr
iter
ia.
W
ith
r
esp
ec
t
to
th
e
im
ag
in
g
m
o
d
alities
in
v
o
lv
e
d
,
s
u
c
h
as
m
ag
n
etic
r
eso
n
an
ce
im
a
g
in
g
(
MRI)
,
co
m
p
u
ted
to
m
o
g
r
ap
h
y
(
C
T
)
,
u
ltra
s
o
u
n
d
(
US)
,
an
d
o
t
h
er
s
,
r
eg
is
tr
atio
n
m
eth
o
d
s
ar
e
co
m
m
o
n
l
y
d
iv
id
ed
in
to
u
n
im
o
d
al
a
p
p
r
o
ac
h
es,
wh
er
e
b
o
th
im
ag
es
ar
e
ac
q
u
ir
ed
with
th
e
s
am
e
m
o
d
ality
(
e.
g
.
,
MRI
–
MRI,
C
T
–
C
T
,
US
–
US)
,
an
d
m
u
ltimo
d
al
ap
p
r
o
ac
h
es,
wh
er
e
th
e
im
ag
es
o
r
ig
in
ate
f
r
o
m
d
if
f
er
en
t
m
o
d
alities
(
e.
g
.
,
MRI
–
C
T
,
MRI
–
US)
.
Fro
m
a
d
im
e
n
s
io
n
al
v
iewp
o
i
n
t,
d
ep
en
d
i
n
g
o
n
w
h
eth
er
th
e
d
at
a
ar
e
p
lan
ar
o
r
v
o
lu
m
et
r
ic,
r
e
g
is
tr
atio
n
p
r
o
b
lem
s
ca
n
b
e
f
o
r
m
u
lated
as
2
D
–
2
D,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
r
o
s
ta
te
ma
g
n
etic
r
eso
n
a
n
ce
ima
g
in
g
/tr
a
n
s
r
ec
ta
l u
ltr
a
s
o
u
n
d
r
eg
is
tr
a
tio
n
…
(
Ha
n
a
e
Ma
h
mo
u
d
i
)
1189
2D
–
3
D,
o
r
3
D
–
3
D
alig
n
m
en
t.
Acc
o
r
d
in
g
to
th
e
tr
an
s
f
o
r
m
ati
o
n
m
o
d
el,
th
r
ee
p
r
in
cip
al
ca
te
g
o
r
ies ar
e
g
en
er
ally
co
n
s
id
er
ed
:
r
ig
i
d
,
af
f
in
e,
an
d
d
ef
o
r
m
a
b
le
r
eg
is
tr
atio
n
.
R
ig
id
r
eg
is
tr
atio
n
is
p
ar
a
m
eter
ize
d
b
y
s
ix
d
eg
r
ee
s
o
f
f
r
ee
d
o
m
co
r
r
esp
o
n
d
in
g
to
t
r
an
s
latio
n
s
an
d
r
o
tatio
n
s
.
Af
f
in
e
r
e
g
is
tr
atio
n
g
en
er
aliz
es
th
is
m
o
d
el
b
y
ad
d
itio
n
ally
allo
win
g
s
ca
lin
g
an
d
s
h
ea
r
i
n
g
,
lead
in
g
to
twelv
e
d
eg
r
ee
s
o
f
f
r
ee
d
o
m
.
B
y
co
n
t
r
ast,
d
ef
o
r
m
ab
le
o
r
non
-
r
i
g
id
r
e
g
is
tr
atio
n
esti
m
ates
s
p
atially
v
ar
y
in
g
tr
an
s
f
o
r
m
atio
n
s
at
th
e
v
o
x
el
le
v
el
an
d
t
y
p
ically
r
e
p
r
esen
ts
th
e
d
ef
o
r
m
atio
n
u
s
in
g
a
d
e
n
s
e
d
is
p
lace
m
en
t f
ield
(
DDF)
to
c
ap
tu
r
e
co
m
p
lex
n
o
n
-
lin
ea
r
an
a
to
m
ical
v
ar
iatio
n
s
.
Mu
ltimo
d
al
r
eg
is
tr
atio
n
o
f
3
D
m
ed
ical
im
ag
es
(
3
D
–
MRe
g
)
is
a
s
p
ec
if
ic
ty
p
e
o
f
r
eg
is
tr
atio
n
th
at
in
v
o
lv
es
th
e
alig
n
m
en
t
o
f
d
ata
co
n
tain
in
g
th
e
s
am
e
3
D
o
b
ject
b
u
t
ca
p
tu
r
ed
u
s
in
g
d
if
f
er
en
t
m
o
d
alities
.
Def
o
r
m
ab
le
r
e
g
is
tr
atio
n
p
lay
s
an
ess
en
tial
r
o
le
in
m
u
ltimo
d
al
r
eg
is
tr
atio
n
b
y
o
f
f
er
i
n
g
a
m
o
r
e
s
o
p
h
is
ticated
an
d
f
lex
ib
le
ap
p
r
o
ac
h
to
alig
n
im
ag
es
f
r
o
m
d
if
f
er
en
t
m
o
d
alities
.
Def
o
r
m
ab
le
r
eg
is
tr
atio
n
m
ak
es
it
p
o
s
s
ib
le
to
m
o
d
el
lo
ca
l
an
d
n
o
n
-
lin
ea
r
d
ef
o
r
m
atio
n
s
b
etwe
en
im
a
g
es,
o
f
f
er
in
g
a
m
o
r
e
ac
c
u
r
ate
co
r
r
esp
o
n
d
e
n
ce
.
T
h
is
ab
ilit
y
to
tak
e
ac
co
u
n
t
o
f
c
o
m
p
lex
an
ato
m
ical
v
ar
iatio
n
s
,
lo
ca
l
d
is
to
r
tio
n
s
,
an
d
d
if
f
er
en
ce
s
in
co
n
tr
ast
b
etwe
en
m
o
d
alities
m
ak
es
d
ef
o
r
m
ab
le
r
e
g
is
tr
atio
n
p
ar
tic
u
lar
ly
well
s
u
ited
to
th
e
ch
allen
g
es
p
o
s
ed
b
y
m
u
ltimo
d
al
im
a
g
es.
I
n
th
e
o
t
h
er
h
a
n
d
,
ap
p
ly
in
g
th
e
a
f
f
in
e
tr
an
s
f
o
r
m
atio
n
is
n
ec
ess
ar
y
f
o
r
t
h
e
m
u
ltimo
d
al
r
eg
is
tr
atio
n
p
r
o
ce
s
s
,
p
ar
ticu
lar
ly
as
an
in
itial
s
tep
b
ef
o
r
e
d
ef
o
r
m
ab
le
r
eg
is
tr
atio
n
.
I
t
en
s
u
r
e
s
g
lo
b
al
alig
n
m
en
t
an
d
u
n
if
o
r
m
in
itializatio
n
f
o
r
im
ag
es
f
r
o
m
d
if
f
er
e
n
t
m
o
d
a
liti
es.
T
h
e
af
f
in
e
tr
an
s
f
o
r
m
ati
o
n
cr
ea
tes
a
s
o
lid
f
o
u
n
d
atio
n
th
at
ac
cu
r
ately
h
an
d
les
s
p
atial
d
is
p
ar
ities
,
ac
c
o
u
n
tin
g
f
o
r
v
ar
iatio
n
s
in
s
ca
le
an
d
o
r
ien
tatio
n
,
ef
f
ec
tiv
ely
lim
itin
g
th
e
s
ea
r
ch
s
p
ac
e
f
o
r
s
u
b
s
eq
u
en
t
d
ef
o
r
m
ab
le
r
e
g
is
tr
atio
n
.
B
y
co
m
b
in
in
g
a
f
f
in
e
a
n
d
d
ef
o
r
m
a
b
le
r
eg
is
tr
atio
n
,
th
e
m
u
ltimo
d
al
r
eg
is
tr
atio
n
p
r
o
ce
s
s
b
ec
o
m
es
m
o
r
e
r
esil
ien
t
an
d
ca
p
ab
le
o
f
p
r
o
v
id
in
g
an
ac
cu
r
ate
o
v
er
lay
,
tak
in
g
in
to
ac
c
o
u
n
t
g
lo
b
al
an
d
l
o
ca
l
v
ar
iatio
n
s
b
etwe
en
i
m
a
g
es
f
r
o
m
d
if
f
er
en
t
m
o
d
alities
.
Ma
n
y
co
n
v
en
tio
n
al
alg
o
r
ith
m
s
h
av
e
b
ee
n
d
ev
el
o
p
ed
a
n
d
s
tu
d
ied
f
o
r
3
D
–
MReg
[
3
]
,
[
4
]
b
ased
o
n
o
p
tim
izin
g
a
s
im
ilar
ity
m
etr
ic,
s
u
ch
as
a
s
u
m
o
f
s
q
u
ar
ed
d
if
f
er
en
ce
s
(
SS
D)
o
r
m
u
tu
al
in
f
o
r
m
atio
n
(
MI
)
[
5
]
,
b
y
ch
an
g
in
g
th
e
tr
a
n
s
f
o
r
m
atio
n
'
s
s
ettin
g
s
.
T
h
e
o
p
tim
izatio
n
p
r
o
ce
s
s
f
o
r
tr
ad
itio
n
al
r
eg
is
tr
atio
n
tech
n
iq
u
es
o
f
3
D
m
u
ltimo
d
al
im
ag
es
f
r
eq
u
en
tly
b
eg
in
s
with
a
p
r
elim
in
ar
y
esti
m
atio
n
o
f
th
e
tr
an
s
f
o
r
m
atio
n
p
ar
am
eter
s
,
wh
ich
ca
n
lead
to
co
n
v
er
g
en
ce
to
in
co
r
r
ec
t
r
e
g
is
tr
atio
n
r
e
s
u
lts
.
I
n
th
e
s
am
e
way
,
th
eir
p
r
o
ce
s
s
in
g
is
tim
e
-
co
n
s
u
m
in
g
,
p
ar
ticu
lar
ly
f
o
r
h
i
g
h
-
r
eso
lu
tio
n
o
r
la
r
g
e
v
o
lu
m
e
s
o
f
im
ag
es.
Dee
p
lear
n
in
g
-
b
ased
r
eg
is
tr
at
io
n
m
eth
o
d
s
h
av
e
r
ec
en
tly
a
d
d
r
ess
ed
th
ese
lim
itatio
n
s
an
d
tr
ied
to
s
o
lv
e
th
e
p
r
o
b
lem
o
f
3
D
–
M
R
eg
[
6
]
,
[
7
]
u
s
in
g
t
h
e
lear
n
i
n
g
ca
p
ab
ilit
ies
o
f
n
e
u
r
al
n
et
wo
r
k
s
.
I
n
f
ac
t,
d
ee
p
lear
n
in
g
-
b
ased
ap
p
r
o
ac
h
es
ca
n
au
to
m
atica
lly
esti
m
ate
tr
an
s
f
o
r
m
atio
n
p
ar
am
eter
s
wh
ile
lear
n
in
g
co
m
p
le
x
co
r
r
esp
o
n
d
en
ce
s
b
etwe
en
im
a
g
es.
C
o
n
v
o
l
u
tio
n
n
eu
r
al
n
etw
o
r
k
s
(
C
NNs)
b
ased
ap
p
r
o
ac
h
es
ar
e
o
n
e
o
f
d
ee
p
lear
n
in
g
tec
h
n
iq
u
es
th
at
ca
n
e
x
tr
ac
t
s
p
atial
an
d
in
ten
s
ity
d
at
a
an
d
p
r
ed
ict
DDF
to
p
er
f
o
r
m
3
D
–
MReg
[
8
]
,
[
9
]
.
Alth
o
u
g
h
C
NNs
ca
n
d
etec
t
lo
ca
l
f
ea
tu
r
es
b
ec
au
s
e
o
f
th
eir
d
ep
en
d
e
n
ce
o
n
r
eg
io
n
al
co
n
v
o
lu
tio
n
an
d
p
o
o
lin
g
p
r
o
ce
d
u
r
es,
th
ey
o
f
ten
n
ee
d
t
o
r
etain
th
e
g
lo
b
al
s
p
atial
co
n
tex
t
r
eq
u
i
r
ed
f
o
r
ac
cu
r
ate
3
D
–
MReg
,
esp
ec
ially
wh
en
co
m
p
le
x
an
d
ex
ten
s
iv
e
i
n
ter
ac
tio
n
s
ex
is
t b
etwe
en
s
ev
e
r
al
co
m
p
o
n
en
ts
o
f
t
h
e
im
ag
e
v
o
lu
m
e.
T
h
an
k
s
to
th
e
g
r
ea
t
s
u
cc
ess
o
f
t
r
an
s
f
o
r
m
er
[
1
0
]
in
n
at
u
r
al
la
n
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P),
a
b
ig
f
o
c
u
s
is
g
iv
en
ac
tu
ally
to
s
elf
-
atten
ti
o
n
(
SA)
m
ec
h
an
is
m
-
b
ased
a
r
ch
itectu
r
es
in
m
an
y
c
o
m
p
u
t
er
v
is
io
n
task
s
an
d
m
ed
ical
im
ag
e
ap
p
licatio
n
s
[
1
1
]
to
im
p
r
o
v
e
th
e
n
o
n
lo
ca
l
m
o
d
elin
g
ca
p
ab
ilit
y
.
Sin
ce
p
r
io
r
itizin
g
g
lo
b
al
p
ictu
r
e
in
f
o
r
m
atio
n
is
n
ec
ess
ar
y
f
o
r
r
eg
is
tr
atio
n
task
s
,
v
is
io
n
tr
an
s
f
o
r
m
e
r
s
[
1
2
]
m
a
y
b
e
e
m
p
lo
y
ed
to
lea
r
n
th
e
o
v
er
all
im
ag
e
r
ep
r
esen
tatio
n
s
an
d
g
ath
er
s
em
an
tic
el
em
en
ts
cr
u
cial
f
o
r
3
D
-
MR
eg
u
s
in
g
atten
tio
n
m
ec
h
an
is
m
s
[
1
3
]
.
I
n
s
p
ir
ed
b
y
th
e
wo
r
k
i
n
[
1
4
]
,
wh
ich
p
r
o
p
o
s
es
m
o
n
o
-
m
o
d
al
3
D
im
ag
e
r
eg
is
tr
atio
n
b
ased
o
n
th
e
v
is
io
n
tr
an
s
f
o
r
m
er
(
ViT
)
,
an
d
th
e
w
o
r
k
s
[
8
]
,
[
9
]
,
wh
er
e
t
h
e
au
th
o
r
s
d
ev
elo
p
a
C
NN
-
b
ased
m
u
ltim
o
d
al
3
D
-
MReg
,
we
p
r
o
p
o
s
e
a
n
ew
ap
p
r
o
ac
h
o
f
s
u
p
er
v
is
ed
d
ef
o
r
m
a
b
le
m
u
ltim
o
d
al
im
ag
e
r
eg
is
tr
atio
n
b
ased
o
n
th
e
co
m
b
i
n
atio
n
o
f
ViT
a
n
d
C
NN
wh
ich
we
n
am
e
3D
-
ViT
.
Ou
r
m
et
h
o
d
lev
er
ag
es
an
ato
m
ical
lab
els
t
o
esti
m
ate
v
o
x
el
-
wis
e
tr
an
s
f
o
r
m
atio
n
s
.
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
is
ev
alu
ated
o
n
3
D
tr
an
s
r
ec
tal
u
ltra
s
o
u
n
d
(
T
R
US)
an
d
T2
-
weig
h
ted
MRI
d
ata
f
r
o
m
p
r
o
s
tate
ca
n
ce
r
p
atien
ts
a
n
d
d
em
o
n
s
tr
ates
en
co
u
r
ag
in
g
p
er
f
o
r
m
an
ce
f
o
r
th
e
3
D
m
ed
ical
im
ag
e
r
e
g
is
tr
atio
n
task
.
T
h
e
m
ai
n
co
n
t
r
ib
u
tio
n
s
o
f
t
h
is
wo
r
k
ar
e
s
u
m
m
a
r
ized
:
a.
I
n
a
f
ir
s
t
s
tep
,
we
ex
p
lo
it
th
e
p
o
wer
o
f
ViT
to
p
r
o
p
o
s
e
a
ViT
-
b
ased
n
etwo
r
k
th
at
we
n
am
e
3D
-
ViT
to
ac
h
iev
e
ac
cu
r
ate
a
f
f
in
e
im
ag
e
r
e
g
is
tr
atio
n
.
T
h
e
o
u
tp
u
t
o
f
th
is
s
tag
e
will
b
e
f
ed
to
a
s
ec
o
n
d
n
etwo
r
k
to
p
er
f
o
r
m
d
e
f
o
r
m
a
b
le
r
eg
is
tr
atio
n
.
b.
I
n
th
e
s
ec
o
n
d
s
tep
,
we
h
ar
n
e
s
s
th
e
s
tr
en
g
th
o
f
C
NNs
to
f
o
cu
s
o
n
lo
ca
l
d
etails.
T
h
e
l
o
ca
l
-
n
et
tak
es
th
e
r
esu
lt
o
f
3D
-
ViT
to
p
r
ed
ict
th
e
lo
ca
l
DDF.
I
n
th
e
f
in
al
s
te
p
,
we
co
m
b
in
e
th
e
af
f
in
e
m
atr
ix
an
d
th
e
lo
ca
l D
DF to
p
r
ed
ict
th
e
f
i
n
al
o
u
tp
u
t D
DF w
h
ich
will b
e
ap
p
lied
to
th
e
m
o
v
i
n
g
lab
els.
c.
Valid
atio
n
o
n
a
p
u
b
lic
p
r
o
s
tate
ca
n
ce
r
MRI/T
R
US
d
ataset
d
em
o
n
s
tr
ates
a
m
e
d
ian
tar
g
et
r
eg
is
tr
atio
n
er
r
o
r
o
f
1
.
7
m
m
at
lan
d
m
ar
k
ce
n
tr
o
i
d
s
an
d
a
m
e
d
ian
Dice
co
ef
f
icien
t o
f
0
.
9
5
f
o
r
t
h
e
p
r
o
s
tate
g
lan
d
.
T
h
e
r
em
ain
d
er
o
f
th
is
ar
ticle
i
s
o
r
g
an
ized
as
f
o
llo
ws.
Sectio
n
2
r
e
v
iews
r
elate
d
wo
r
k
o
n
m
u
ltimo
d
al
m
ed
ical
im
ag
e
r
eg
is
tr
atio
n
.
Se
ctio
n
3
in
tr
o
d
u
ce
s
3D
-
ViT
,
t
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
f
o
r
m
u
ltimo
d
al
r
eg
is
tr
atio
n
o
f
v
o
l
u
m
etr
ic
m
ed
ical
im
ag
es
th
at
in
teg
r
ates
v
is
io
n
tr
an
s
f
o
r
m
er
s
an
d
co
n
v
o
lu
tio
n
al
n
eu
r
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
1
8
8
-
1
1
9
8
1190
n
etwo
r
k
s
.
Sectio
n
4
r
ep
o
r
ts
th
e
ex
p
er
im
en
tal
s
etu
p
an
d
e
v
al
u
atio
n
r
esu
lts
u
s
ed
to
ass
ess
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
.
Fin
ally
,
s
ec
tio
n
5
co
n
clu
d
es th
e
p
a
p
er
.
2.
RE
L
AT
E
D
WO
RK
Mu
ltimo
d
al
r
eg
is
tr
atio
n
p
lay
s
a
ce
n
tr
al
r
o
le
in
n
u
m
er
o
u
s
m
e
d
ical
im
ag
e
an
aly
s
is
p
ip
elin
es
.
I
n
r
ec
en
t
y
ea
r
s
,
d
ee
p
lear
n
in
g
–
b
ased
m
u
ltimo
d
al
im
ag
e
r
eg
is
tr
atio
n
h
as
attr
ac
ted
co
n
s
id
er
a
b
le
atten
tio
n
.
I
n
[
8
]
,
th
e
au
th
o
r
s
p
r
o
p
o
s
ed
a
C
NN
-
b
as
ed
f
r
am
ew
o
r
k
to
esti
m
ate
d
e
n
s
e
v
o
x
el
c
o
r
r
esp
o
n
d
en
ce
s
b
y
ex
p
lo
itin
g
m
u
ltip
le
id
en
tifia
b
le
an
ato
m
ical
lab
el
t
y
p
es.
T
h
eir
m
eth
o
d
in
tr
o
d
u
ce
d
a
d
ed
icate
d
n
etwo
r
k
d
esig
n
t
o
ef
f
icien
tly
p
r
ed
ict
b
o
th
g
lo
b
al
a
n
d
lo
ca
l
d
ef
o
r
m
atio
n
s
ar
is
in
g
in
p
r
o
s
tate
ca
n
c
er
s
u
r
g
er
y
th
r
o
u
g
h
MRI/T
R
US
r
eg
is
tr
atio
n
.
I
n
a
s
u
b
s
eq
u
en
t
ex
te
n
s
io
n
[
9
]
,
a
u
n
if
ied
n
etwo
r
k
was
p
r
o
p
o
s
ed
to
d
ir
ec
tly
p
r
ed
ict
DDF
s
at
m
u
ltip
le
s
p
atial
r
eso
lu
tio
n
s
.
W
h
en
ex
p
licit
s
im
ilar
ity
m
ea
s
u
r
es
ar
e
u
n
av
ailab
le,
n
eu
r
al
n
etwo
r
k
–
d
r
iv
en
ap
p
r
o
ac
h
es
h
av
e
b
ee
n
em
p
lo
y
ed
to
in
f
er
d
e
n
s
e
v
o
x
e
l
co
r
r
esp
o
n
d
en
ce
s
f
r
o
m
a
v
ar
i
ety
o
f
a
n
ato
m
ical
lab
el
s
h
a
p
es,
th
er
eb
y
ca
p
tu
r
in
g
m
ea
n
in
g
f
u
l
an
at
o
m
ical
f
ea
tu
r
es
[
6
]
.
Hask
in
s
et
a
l.
[
1
5
]
i
n
tr
o
d
u
ce
d
a
d
ee
p
C
NN
f
r
a
m
ewo
r
k
t
o
lear
n
a
s
im
ilar
ity
m
etr
ic
f
o
r
MRI/T
R
US
r
eg
is
tr
atio
n
an
d
c
o
m
b
in
e
d
it
with
a
h
y
b
r
id
o
p
tim
izatio
n
s
tr
ateg
y
to
o
b
tain
s
u
itab
le
in
itializatio
n
s
f
o
r
s
ec
o
n
d
-
o
r
d
er
o
p
tim
izatio
n
.
Simo
n
o
v
s
k
y
et
a
l.
[
1
6
]
f
o
r
m
u
lated
3
D
m
ed
ical
im
ag
e
r
eg
is
tr
atio
n
as
a
cla
s
s
if
icatio
n
p
r
o
b
lem
to
d
is
tin
g
u
is
h
b
etwe
en
well
-
alig
n
ed
a
n
d
m
is
alig
n
ed
im
a
g
e
p
a
tch
es
ac
r
o
s
s
d
if
f
e
r
en
t
m
o
d
alit
ies
u
s
in
g
C
NNs.
I
n
[
1
7
]
,
th
e
c
o
h
er
e
n
t
p
o
in
t
d
r
if
t
a
lg
o
r
ith
m
was
a
d
o
p
ted
to
elastically
r
eg
is
ter
s
u
r
f
ac
es,
an
d
th
e
d
ef
o
r
m
atio
n
f
ield
o
b
tain
ed
f
r
o
m
th
in
-
p
late
s
p
li
n
es
was
p
r
o
p
a
g
ated
t
o
th
e
e
n
tire
g
lan
d
.
A
weig
h
ted
s
elf
-
s
im
ilar
ity
s
tr
u
ctu
r
e
v
ec
to
r
(
W
SS
V)
was
later
ap
p
lied
to
p
er
f
o
r
m
m
u
ltimo
d
al
r
eg
is
tr
atio
n
[
1
8
]
,
an
d
th
is
s
tr
at
eg
y
was
s
h
o
wn
t
o
im
p
r
o
v
e
th
e
in
t
r
ao
p
er
ati
v
e
lo
ca
lizatio
n
o
f
3
D
MRI
an
d
2
D
u
ltra
s
o
u
n
d
lesi
o
n
s
in
f
u
s
io
n
-
g
u
id
e
d
n
av
i
g
atio
n
s
y
s
tem
s
.
B
len
d
o
wsk
i
et
a
l.
[
1
9
]
p
r
o
p
o
s
ed
a
f
r
am
ewo
r
k
th
at
r
el
ies
s
o
lely
o
n
in
d
ep
en
d
en
tly
o
b
tain
ed
s
eg
m
en
tatio
n
lab
els
f
r
o
m
ea
c
h
m
o
d
ality
an
d
e
x
p
lo
its
an
ato
m
ical
s
h
ap
e
p
r
io
r
s
.
A
s
h
ap
e
-
co
n
s
tr
ain
ed
en
co
d
er
–
d
ec
o
d
er
s
eg
m
en
tatio
n
n
etwo
r
k
was
f
ir
s
t
tr
ain
ed
o
n
lab
ele
d
C
T
an
d
MRI
d
ata
with
o
u
t
s
k
ip
co
n
n
ec
tio
n
s
.
Su
b
s
eq
u
en
tly
,
a
n
iter
ativ
e
e
n
er
g
y
-
m
in
im
izatio
n
s
ch
e
m
e,
d
r
iv
en
b
y
th
e
n
etwo
r
k
’
s
ab
ilit
y
to
g
e
n
er
ate
in
ter
m
ed
iate
n
o
n
-
lin
ea
r
s
h
ap
e
r
ep
r
esen
tatio
n
s
,
was
in
tr
o
d
u
ce
d
to
en
h
a
n
ce
m
u
ltim
o
d
al
alig
n
m
en
t
i
n
th
e
p
r
esen
ce
o
f
la
r
g
e
d
e
f
o
r
m
atio
n
s
.
Mo
r
e
r
ec
en
tly
,
So
n
g
et
a
l
.
[
2
0
]
p
r
esen
ted
a
c
r
o
s
s
-
m
o
d
al
a
tten
tio
n
–
b
ased
ap
p
r
o
ac
h
th
at
co
r
r
elate
s
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
ex
tr
ac
ted
f
r
o
m
m
u
ltimo
d
al
im
ag
e
p
air
s
an
d
lev
er
ag
es
th
e
r
esu
ltin
g
co
r
r
elatio
n
s
to
r
ef
in
e
th
e
r
eg
is
tr
atio
n
p
r
o
ce
s
s
.
I
n
[
2
1
]
,
a
B
ay
esian
f
r
am
ew
o
r
k
was
d
ev
elo
p
e
d
t
o
co
n
s
tr
u
ct
m
u
lti
-
class
b
r
ain
atlases
an
d
to
m
o
d
el
lar
g
e
in
ter
-
s
u
b
ject
d
e
f
o
r
m
atio
n
s
b
etwe
en
m
u
ltimo
d
al
b
r
ain
im
ag
es
b
y
ex
p
l
o
itin
g
m
o
d
ality
-
s
p
ec
if
ic
n
e
u
r
o
a
n
ato
m
ical
in
f
o
r
m
atio
n
.
Fin
ally
,
a
d
ee
p
r
ei
n
f
o
r
ce
m
en
t
lear
n
in
g
–
b
ased
m
u
ltimo
d
al
r
eg
is
tr
atio
n
m
eth
o
d
was
in
tr
o
d
u
ce
d
in
[
2
2
]
,
wh
ich
aim
s
to
r
ed
u
ce
m
o
d
ality
d
is
cr
ep
an
ci
es
b
y
au
to
m
atica
lly
lear
n
in
g
r
o
b
u
s
t a
n
d
d
is
cr
im
in
a
tiv
e
f
ea
tu
r
e
r
e
p
r
esen
tatio
n
s
.
3.
M
E
T
H
O
D
I
n
th
is
s
ec
tio
n
,
we
p
r
esen
t
o
u
r
n
ew
m
eth
o
d
3D
-
MReg
Def
ViT
to
p
er
f
o
r
m
DDF
o
f
m
u
lti
-
m
o
d
al
m
ed
ical
im
ag
es.
Du
e
to
th
e
n
o
n
-
lin
ea
r
r
elatio
n
s
h
ip
s
b
etw
ee
n
in
p
u
t
im
ag
e
p
air
esp
ec
ia
lly
wh
en
u
tili
zin
g
im
ag
es
f
r
o
m
d
if
f
er
en
t
m
o
d
alit
ies,
th
e
r
eg
is
tr
atio
n
p
r
o
ce
d
u
r
e
r
em
ain
s
v
er
y
c
h
allen
g
in
g
.
T
h
e
u
s
e
o
f
lab
els
with
an
ato
m
ical
d
etails h
elp
s
m
ak
e
th
e
p
r
o
ce
s
s
s
m
o
o
th
er
.
Fig
u
r
e
1
illu
s
tr
ates
th
e
o
v
er
a
ll
wo
r
k
f
lo
w
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
.
At
th
e
b
eg
in
n
in
g
,
tr
ain
in
g
s
am
p
les
ar
e
r
an
d
o
m
ly
g
en
e
r
at
ed
b
y
s
elec
tin
g
eq
u
al
n
u
m
b
e
r
s
o
f
i
m
ag
e
p
air
s
to
g
eth
e
r
with
t
h
eir
co
r
r
esp
o
n
d
in
g
lab
els.
T
h
ese
s
am
p
les
ar
e
t
h
e
n
p
r
o
ce
s
s
ed
with
in
a
m
in
i
-
b
at
ch
o
p
tim
izatio
n
s
ch
em
e.
I
n
th
e
n
ex
t
s
tag
e,
a
ViT
r
ec
eiv
es
th
e
f
ix
ed
an
d
m
o
v
i
n
g
im
ag
e
p
air
as
in
p
u
t
an
d
esti
m
ates
th
e
af
f
in
e
tr
an
s
f
o
r
m
atio
n
m
atr
ix
,
wh
ich
is
s
u
b
s
eq
u
en
tly
u
s
ed
to
war
p
th
e
m
o
v
in
g
im
a
g
e
to
war
d
th
e
f
ix
ed
o
n
e.
W
e
r
ef
er
to
t
h
is
ViT
-
b
ased
m
o
d
u
le
f
o
r
3
D
af
f
in
e
r
e
g
is
tr
atio
n
as
3D
-
MR
eg
Def
ViT
.
Su
b
s
eq
u
e
n
tly
,
a
l
o
ca
l
-
n
et
tak
es
as
in
p
u
t
th
e
c
o
n
ca
ten
atio
n
o
f
th
e
af
f
in
ely
war
p
ed
m
o
v
in
g
im
ag
e
an
d
th
e
f
ix
ed
im
ag
e
t
o
p
r
e
d
ict
a
lo
ca
l
n
o
n
-
r
i
g
id
DDF,
wh
ich
is
r
eg
u
lar
ize
d
u
s
in
g
a
weig
h
te
d
b
e
n
d
in
g
en
e
r
g
y
ter
m
[
2
3
]
.
T
h
e
a
f
f
in
e
tr
a
n
s
f
o
r
m
atio
n
a
n
d
th
e
p
r
e
d
icted
l
o
ca
l
DDF
ar
e
th
en
co
m
b
in
ed
to
f
o
r
m
th
e
f
i
n
al
d
is
p
lace
m
en
t
f
ield
,
wh
ich
is
ap
p
lied
to
war
p
th
e
m
o
v
i
n
g
lab
els
to
alig
n
with
th
e
f
ix
ed
lab
els.
L
ab
el
co
r
r
esp
o
n
d
en
ce
is
ev
alu
ate
d
u
s
in
g
t
h
e
m
u
ltis
ca
le
Dice
m
etr
ic
[
2
4
]
.
Du
r
in
g
tr
ain
in
g
,
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
[
2
5
]
is
em
p
lo
y
ed
,
r
ely
in
g
o
n
an
u
n
b
iased
g
r
ad
ien
t
esti
m
ato
r
u
n
d
er
th
e
ass
u
m
p
tio
n
th
at
im
ag
es
an
d
lab
els
a
r
e
c
o
n
d
itio
n
ally
in
d
e
p
en
d
e
n
t,
wh
ich
en
ab
les
e
f
f
ec
tiv
e
m
in
im
i
za
tio
n
o
f
th
e
lo
s
s
f
u
n
ctio
n
.
L
et
an
d
d
en
o
te
t
h
e
f
ix
ed
a
n
d
m
o
v
in
g
tr
ain
in
g
im
ag
es,
an
d
an
d
th
eir
c
o
r
r
esp
o
n
d
in
g
lab
els.
T
h
e
f
o
llo
win
g
s
u
b
s
ec
tio
n
s
d
escr
ib
e
ea
ch
c
o
m
p
o
n
en
t
o
f
th
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
in
d
etail.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
r
o
s
ta
te
ma
g
n
etic
r
eso
n
a
n
ce
ima
g
in
g
/tr
a
n
s
r
ec
ta
l u
ltr
a
s
o
u
n
d
r
eg
is
tr
a
tio
n
…
(
Ha
n
a
e
Ma
h
mo
u
d
i
)
1191
Fig
u
r
e
1
.
Ov
e
r
v
iew
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
3
−
3
.
1
.
−
W
e
aim
to
o
b
tain
th
e
o
p
tim
a
l
af
f
in
e
m
atr
ix
to
alig
n
an
d
b
y
th
e
tr
an
s
f
o
r
m
er
en
co
d
er
[
1
2
]
as
illu
s
tr
ated
in
Fig
u
r
e
2
.
Af
ter
r
esizin
g
th
e
im
ag
es
to
th
e
s
am
e
s
ize,
th
e
co
n
ca
ten
atio
n
o
f
an
d
d
en
o
ted
b
y
is
d
iv
id
ed
in
to
v
ec
to
r
ized
3
D
p
atch
es
∈
ℝ
×
3
wh
er
e
=
×
×
3
,
an
d
(
,
,
)
in
d
icate
th
e
d
im
en
s
io
n
o
f
.
T
h
en
,
ea
c
h
p
atch
is
m
ap
p
e
d
to
a
v
ec
to
r
b
y
ap
p
ly
in
g
a
lear
n
ab
le
lin
ea
r
e
m
b
ed
d
in
g
=
E:
=
[
1
;
2
;
…
…
;
]
, E
∈
ℝ
3
×
(
1
)
Fo
llo
win
g
th
at,
we
ad
d
ed
a
lea
r
n
ab
le
p
o
s
itio
n
al
em
b
ed
d
in
g
to
:
=
+
,
∈
ℝ
×
(
2
)
T
h
e
g
e
n
er
ated
p
atch
es,
ca
lled
to
k
en
s
,
ar
e
th
en
s
en
t
to
th
e
tr
an
s
f
o
r
m
er
en
co
d
er
,
wh
ich
u
s
e
s
th
em
to
ca
lcu
late
s
elf
-
atten
tio
n
(
SA)
.
T
h
e
SA
r
e
co
r
d
s
th
e
d
e
p
en
d
e
n
cies
b
etwe
en
in
p
u
t
s
eq
u
e
n
ce
t
o
k
en
s
.
T
h
e
ar
e
tr
a
n
s
lated
lin
ea
r
ly
in
to
th
e
q
u
e
r
y
(
)
,
k
ey
(
)
,
an
d
v
al
u
e
(
)
m
atr
ices b
y
m
u
ltip
ly
in
g
th
em
with
lear
n
e
d
weig
h
t m
atr
ices
W
:
{
Q
=
k
=
V
=
(
3
)
T
h
e
atten
tio
n
weig
h
ts
ar
e
t
h
en
o
b
tain
ed
b
y
ca
lcu
latin
g
th
e
at
ten
tio
n
s
co
r
es a
n
d
u
s
in
g
th
e
S
o
f
tMa
x
f
u
n
ctio
n
:
=
s
o
ftma
x
(
√
)
A
∈
ℝ
×
(
4
)
T
h
e
s
ca
led
d
o
t
-
p
r
o
d
u
ct
atten
ti
o
n
is
co
n
s
tr
u
cted
as
(
5
)
:
SA (
)
=
(
5
)
T
h
en
,
SA
is
ca
lcu
lated
s
ev
er
al
tim
es
in
p
ar
allel
with
d
if
f
er
e
n
t
lear
n
ed
weig
h
t
m
atr
ices,
cr
ea
tin
g
m
u
lti
-
h
ea
d
e
d
atten
tio
n
.
T
h
er
ef
o
r
e,
th
e
m
o
d
el
ca
n
f
o
cu
s
o
n
v
ar
io
u
s
p
iece
s
o
f
in
p
u
t
d
ata
an
d
ca
p
tu
r
e
m
u
ltip
le
in
ter
ac
tio
n
s
.
Fo
llo
win
g
,
th
e
r
esu
lts
ar
e
tr
a
n
s
m
itted
to
a
f
ee
d
-
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
.
At
th
e
en
d
o
f
ea
ch
SA
an
d
f
ee
d
-
f
o
r
war
d
,
lay
er
n
o
r
m
aliza
tio
n
an
d
r
esid
u
al
co
n
n
ec
tio
n
s
ar
e
ap
p
lied
.
T
h
e
p
r
o
d
u
ce
d
o
u
tp
u
t
o
f
th
e
tr
an
s
f
o
r
m
er
en
co
d
er
is
g
iv
e
n
as
in
p
u
t
to
th
e
class
if
icatio
n
h
ea
d
th
at
im
p
lem
en
ts
th
e
m
u
lti
-
lay
er
p
er
ce
p
tr
o
n
with
t
h
e
ac
tiv
atio
n
f
u
n
ctio
n
Gau
s
s
ian
e
r
r
o
r
lin
ea
r
u
n
it
(
GE
L
U)
.
Fin
al
ly
,
th
e
r
esu
ltin
g
is
a
s
et
o
f
1
2
-
d
eg
r
ee
-
of
-
f
r
ee
d
o
m
af
f
in
e
m
atr
ix
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
1
8
8
-
1
1
9
8
1192
Fig
u
r
e
2
.
VI
T
ar
ch
itectu
r
e
a
p
p
lied
to
3
D
m
ed
ical
im
a
g
es u
s
ed
in
o
u
r
ap
p
r
o
ac
h
3
.
2
.
L
o
ca
l
-
n
et
f
o
r
def
o
r
m
a
ble r
e
g
is
t
ra
t
io
n in
−
T
h
e
p
r
ed
icted
a
f
f
in
e
tr
a
n
s
f
o
r
m
atio
n
m
atr
ix
r
esu
ltin
g
b
y
3D
−
M
R
e
g
Aff
ViT
is
u
s
ed
to
d
ef
o
r
m
th
e
.
T
h
e
war
p
ed
m
o
v
in
g
im
a
g
e
is
r
ep
r
esen
ted
b
y
̂
.
Af
ter
b
ein
g
co
n
ca
ten
ated
,
an
d
̂
ar
e
f
ed
in
to
th
e
lo
ca
l
n
etwo
r
k
to
p
r
e
d
ict
th
e
lo
ca
l
DDF.
T
h
e
lo
ca
l
-
n
et
[
8
]
co
m
p
r
is
es
f
o
u
r
s
u
b
s
am
p
lin
g
b
l
o
ck
s
an
d
f
o
u
r
u
p
-
s
am
p
lin
g
b
lo
ck
s
,
in
co
r
p
o
r
atin
g
s
h
o
r
ten
e
d
co
n
n
ec
tio
n
s
at
d
if
f
er
en
t
lev
els
o
f
r
eso
lu
tio
n
.
B
y
b
eg
in
n
in
g
with
an
in
itial
3
2
ch
an
n
els,
th
e
n
etwo
r
k
s
tr
ateg
ically
d
o
u
b
les
an
d
h
alv
es
th
e
n
u
m
b
er
o
f
th
e
ch
an
n
els
d
u
r
in
g
s
u
b
s
am
p
lin
g
an
d
Up
-
s
am
p
lin
g
,
r
esp
ec
tiv
ely
.
E
ac
h
b
lo
ck
u
s
es
r
esid
u
al
n
etw
o
r
k
u
n
its
(
R
esNet)
wi
th
s
h
o
r
ten
ed
co
n
n
ec
tio
n
s
,
im
p
r
o
v
in
g
th
e
n
etwo
r
k
'
s
tr
ain
in
g
ca
p
ab
ilit
ies.
Su
b
s
am
p
lin
g
is
p
er
f
o
r
m
ed
th
r
o
u
g
h
co
n
v
o
l
u
tio
n
al
lay
er
s
with
s
tep
s
o
f
two
wh
ile
u
p
-
s
am
p
lin
g
u
s
es
tr
an
s
p
o
s
ed
co
n
v
o
lu
t
io
n
al
lay
er
s
.
T
h
e
o
u
tp
u
t
lay
er
o
f
th
e
n
etwo
r
k
in
itializes
with
an
ad
d
itio
n
al
c
o
n
v
o
lu
ti
o
n
o
p
er
atio
n
a
n
d
a
b
i
as
ter
m
,
d
elib
e
r
ately
lac
k
in
g
b
atch
n
o
r
m
aliza
tio
n
o
r
n
o
n
-
lin
ea
r
ac
tiv
atio
n
to
all
o
w
r
an
d
o
m
in
itializatio
n
with
ze
r
o
m
ea
n
an
d
m
in
im
al
v
ar
i
atio
n
.
At
last
,
lo
ca
l
DDF
is
co
m
p
u
ted
.
T
h
e
f
in
a
l
o
u
tp
u
t
DDF
r
esu
lts
f
r
o
m
co
m
b
in
in
g
th
e
af
f
in
e
tr
an
s
f
o
r
m
atio
n
,
p
r
o
d
u
ce
d
by
3D
−
M
R
e
g
Aff
ViT
,
an
d
th
e
lo
ca
l
DDF,
g
e
n
e
r
ated
b
y
lo
ca
l
-
n
et
.
Usi
n
g
th
is
co
m
b
in
e
d
DDF,
th
e
m
o
v
i
n
g
lab
el
is
d
ef
o
r
m
ed
in
to
t
h
e
f
ix
ed
lab
el
o
n
e
b
y
a
tr
ilin
ea
r
im
ag
e
r
esam
p
ler
.
T
o
av
o
id
o
v
e
r
-
f
itti
n
g
is
s
u
es,
b
in
ar
y
m
ask
s
o
f
la
b
elled
im
ag
es
h
av
e
b
ee
n
p
r
ep
r
o
ce
s
s
ed
u
s
in
g
a
o
n
e
-
s
id
ed
s
m
o
o
th
i
n
g
m
e
ch
an
is
m
f
o
llo
win
g
th
e
s
am
e
s
tr
ateg
y
i
n
[
8
]
.
Fo
r
t
h
e
in
f
er
e
n
ce
p
ar
t,
o
n
ly
t
h
e
u
n
l
ab
eled
im
ag
e
p
air
s
ar
e
u
s
ed
a
s
in
p
u
t
d
ata
f
o
r
th
e
m
o
d
el.
3
.
3
.
M
ulti
-
s
ca
le
dice
T
h
e
s
o
f
t
p
r
o
b
ab
ilis
tic
Dice
m
etr
ic
p
r
o
p
o
s
ed
b
y
Milletar
i
et
a
l.
[
2
4
]
f
o
r
m
s
th
e
f
o
u
n
d
atio
n
o
f
th
e
m
u
lti
-
s
ca
le
Dice
m
ea
s
u
r
e.
I
t is d
ef
in
ed
a:
(
,
)
=
2
∑
−
1
∑
+
∑
−
1
−
1
(
6
)
w
h
er
e
=
{
}
,
=
{
}
d
e
n
o
te
th
e
f
ix
ed
an
d
m
o
v
in
g
lab
el
m
ap
s
,
r
esp
ec
t
iv
ely
,
with
{
,
}
∈
[
0
,
1
]
an
d
=
1
,
…
…
,
,
o
v
er
I
v
o
x
els in
th
e
im
a
g
e.
T
h
e
m
u
lti
-
s
ca
le
Dice
m
etr
ic
is
th
en
d
ef
in
e
d
as:
=
1
∑
(
(
)
,
(
)
)
(
7
)
w
h
er
e
=
{
(
)
}
,
=
{
(
)
}
ar
e
a
p
air
o
f
b
in
ar
y
l
ab
el
m
ap
s
.
T
h
e
o
p
er
ato
r
d
en
o
tes
a
3
D
G
au
s
s
ian
s
m
o
o
th
in
g
f
ilter
with
is
o
tr
o
p
i
c
s
tan
d
ar
d
d
ev
iatio
n
σ
.
I
n
th
i
s
wo
r
k
,
σ
∈
{0
,
1
,
2
,
4
,
8
,
1
6
}
co
r
r
esp
o
n
d
in
g
to
=
6
d
if
f
er
en
t
s
ca
les.
T
h
e
ad
o
p
tio
n
o
f
th
e
m
u
lti
-
s
ca
le
Dice
m
etr
ic
en
h
an
ce
s
s
en
s
itiv
it
y
to
s
p
atial
r
elatio
n
s
h
ip
s
b
etwe
en
an
ato
m
ical
lab
els,
in
co
n
tr
ast
to
co
n
v
en
tio
n
al
m
ea
s
u
r
es
s
u
ch
as
Dice
,
J
ac
ca
r
d
,
an
d
cr
o
s
s
-
en
tr
o
p
y
,
wh
ich
d
o
n
o
t e
x
p
licitly
ac
co
u
n
t
f
o
r
s
p
atial
in
f
o
r
m
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
r
o
s
ta
te
ma
g
n
etic
r
eso
n
a
n
ce
ima
g
in
g
/tr
a
n
s
r
ec
ta
l u
ltr
a
s
o
u
n
d
r
eg
is
tr
a
tio
n
…
(
Ha
n
a
e
Ma
h
mo
u
d
i
)
1193
4.
E
XP
E
R
I
M
E
N
T
S
4
.
1
.
Da
t
a
s
et
s
a
nd
co
nfig
ura
t
io
n
W
e
ass
es
s
ed
th
e
p
r
o
p
o
s
ed
m
eth
o
d
u
s
in
g
p
r
o
s
tate
MRI/T
R
US
d
ata
f
r
o
m
th
e
s
m
ar
t
tar
g
e
t
®
clin
ical
tr
ials
[
2
6
]
.
T
h
e
d
ataset
in
cl
u
d
es
1
0
8
m
atch
e
d
p
air
s
o
f
T
2
-
weig
h
ted
MRI
a
n
d
T
R
US
v
o
l
u
m
es
ac
q
u
i
r
ed
f
r
o
m
7
6
p
atien
ts
.
B
ec
au
s
e
p
atien
ts
u
n
d
er
wen
t
d
if
f
er
e
n
t
clin
ical
in
ter
v
en
tio
n
s
,
ea
c
h
s
u
b
ject
co
u
ld
h
av
e
u
p
to
th
r
ee
s
ep
ar
ate
im
ag
in
g
s
ess
io
n
s
,
in
clu
d
in
g
b
io
p
s
y
p
r
o
ce
d
u
r
es,
th
er
ap
eu
tic
in
ter
v
en
tio
n
s
,
o
r
r
e
p
ea
ted
ec
h
o
g
r
ap
h
y
ac
q
u
is
itio
n
s
[
2
7
]
.
B
o
th
MRI
an
d
T
R
US
v
o
lu
m
es
wer
e
r
esam
p
led
to
an
is
o
t
r
o
p
ic
s
p
atial
r
eso
lu
tio
n
o
f
1
.
0
3
an
d
s
u
b
s
eq
u
en
tly
s
tan
d
ar
d
ized
to
ze
r
o
m
ea
n
an
d
u
n
it
v
ar
ian
c
e.
W
ith
r
esp
ec
t
to
th
e
an
ato
m
ical
an
n
o
tatio
n
s
,
p
r
o
s
tate
s
eg
m
en
tatio
n
s
wer
e
f
ir
s
t
m
an
u
ally
d
elin
ea
ted
b
y
a
m
ed
ical
s
tu
d
en
t
in
th
r
ee
o
r
th
o
g
o
n
al
v
iews
ax
ial,
s
ag
ittal,
an
d
co
r
o
n
al
an
d
th
en
r
ev
iewe
d
an
d
co
r
r
ec
ted
b
y
a
n
ex
p
e
r
ien
ce
d
u
r
o
lo
g
is
t.
Fo
r
t
h
e
T
R
US
m
o
d
ality
,
g
lan
d
c
o
n
to
u
r
s
wer
e
m
an
u
ally
r
ef
i
n
ed
b
ased
o
n
au
to
m
ati
ca
lly
esti
m
ated
p
r
o
s
tate
p
r
o
f
i
les
o
n
th
e
o
r
ig
in
al
u
ltra
s
o
u
n
d
s
lices
[
2
8
]
.
Pro
s
tat
e
s
eg
m
en
tatio
n
s
o
n
MRI
wer
e
p
r
o
v
id
ed
d
ir
ec
tl
y
as
p
ar
t
o
f
th
e
clin
ical
s
tu
d
y
p
r
o
to
co
ls
[
2
6
]
.
T
h
e
an
ato
m
ical
lab
els
wer
e
en
co
d
ed
as
b
i
n
ar
y
m
ask
s
an
d
s
m
o
o
th
ed
u
s
in
g
a
n
o
r
m
alize
d
in
v
er
s
e
d
is
tan
ce
tr
an
s
f
o
r
m
af
ter
r
esam
p
lin
g
to
m
atch
th
e
s
p
atial
d
im
en
s
io
n
s
o
f
th
e
c
o
r
r
esp
o
n
d
in
g
MRI
o
r
T
R
US
v
o
lu
m
es.
T
h
e
p
r
o
p
o
s
e
d
f
r
am
ewo
r
k
was
im
p
lem
en
ted
in
Py
th
o
n
with
th
e
T
en
s
o
r
Flo
w
lib
r
ar
y
an
d
tr
ain
ed
o
n
a
wo
r
k
s
tatio
n
eq
u
i
p
p
ed
with
a
2
4
GB
NVI
DI
A
Qu
ad
r
o
P6
0
0
0
GPU.
Op
tim
iz
atio
n
was
ca
r
r
ied
o
u
t
u
s
in
g
th
e
Ad
am
alg
o
r
ith
m
with
a
m
in
i
-
b
atch
s
ize
o
f
2
,
a
le
ar
n
in
g
r
ate
o
f
10
−
5
,
an
d
a
to
tal
o
f
1
0
0
0
tr
ain
in
g
iter
atio
n
s
.
4
.
2
.
E
v
a
lua
t
i
o
n
m
et
rics
I
n
th
is
wo
r
k
,
a
1
0
-
f
o
ld
c
r
o
s
s
-
v
alid
atio
n
s
tr
ateg
y
was
co
n
d
u
ct
ed
at
th
e
p
atien
t
lev
el.
I
n
ea
c
h
f
o
ld
,
d
ata
f
r
o
m
7
to
8
p
atien
ts
wer
e
h
el
d
o
u
t
f
o
r
test
in
g
,
wh
ile
t
h
e
r
e
m
ain
in
g
p
atien
t
s
ca
n
s
wer
e
u
s
ed
f
o
r
tr
ain
in
g
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
was
e
v
alu
ated
u
s
in
g
two
q
u
a
n
titativ
e
m
ea
s
u
r
es:
th
e
d
ice
s
im
ilar
ity
co
ef
f
icien
t
(
DSC
)
an
d
th
e
tar
g
et
r
eg
is
tr
atio
n
er
r
o
r
(
T
R
E
)
.
D
SC
is
d
eter
m
in
ed
b
etwe
en
th
e
b
in
ar
y
war
p
ed
an
d
f
ix
ed
lab
els r
ep
r
esen
tin
g
th
e
p
r
o
s
tate
g
lan
d
s
:
(
,
)
=
2
|
∩
|
|
|
+
|
|
(
8
)
w
h
er
e
is
f
ix
ed
lab
el
an
d
is
th
e
war
p
ed
m
o
v
in
g
lab
el.
T
R
E
is
co
m
p
u
te
d
as
th
e
r
o
o
t
m
ea
n
s
q
u
ar
e
o
f
th
e
s
p
atial
d
is
tan
ce
b
etwe
en
th
e
c
en
ter
s
o
f
m
ass
o
f
th
e
wa
r
p
ed
l
ab
els an
d
th
o
s
e
o
f
th
e
co
r
r
esp
o
n
d
in
g
f
ix
ed
lab
els.
(
,
)
=
√
1
∑
‖
−
‖
2
=
1
(
9
)
w
h
er
e
th
e
co
o
r
d
in
ates
o
f
a
p
o
in
t
in
f
ix
ed
lab
el
,
is
th
e
co
o
r
d
in
ates
o
f
a
p
o
in
t
i
n
th
e
war
p
ed
m
o
v
in
g
lab
el
an
d
is
th
e
to
tal
n
u
m
b
e
r
o
f
co
r
r
esp
o
n
d
i
n
g
p
o
in
ts
o
r
lan
d
m
ar
k
s
.
4
.
3
.
Resul
t
s
a
nd
d
is
cu
s
s
io
n
Fig
u
r
e
3
illu
s
tr
ates
th
e
r
eg
is
tr
atio
n
r
esu
lts
p
r
o
d
u
ce
d
b
y
3D
−
M
R
e
g
af
f
ViT
an
d
3D
−
M
R
e
g
d
ef
ViT
u
s
in
g
im
ag
e
s
lices f
r
o
m
two
r
ep
r
esen
tativ
e
test
ca
s
es (
C
ases
1
an
d
2
)
.
R
o
ws a
–
d
co
r
r
esp
o
n
d
to
s
lices f
r
o
m
th
e
o
r
ig
in
al
MRI
im
ag
es,
th
e
f
ix
e
d
T
R
US
im
ag
es,
th
e
MRI
im
ag
es
war
p
ed
b
y
3D
−
M
R
e
g
d
ef
ViT
an
d
th
e
MRI
im
ag
es wa
r
p
ed
b
y
3D
−
M
R
e
g
d
ef
ViT
,
r
esp
ec
tiv
ely
.
I
n
o
u
r
s
tu
d
y
,
we
s
tar
ted
th
e
tr
ain
in
g
with
3D
−
M
R
e
g
af
f
ViT
to
p
r
ed
ict
th
e
af
f
in
e
tr
an
s
f
o
r
m
atio
n
.
T
h
en
,
we
u
s
ed
3D
−
M
R
e
g
D
ef
ViT
to
p
r
ed
ict
t
h
e
DDF.
W
e
co
m
p
ar
e
d
o
u
r
a
p
p
r
o
ac
h
i
n
clu
d
in
g
b
o
th
o
f
3D
−
M
R
e
g
af
f
ViT
an
d
3D
−
M
R
e
g
d
ef
ViT
with
C
NN
-
b
ased
m
eth
o
d
s
p
r
o
p
o
s
ed
in
[
8
]
,
[
9
]
u
s
in
g
th
e
Dice
s
co
r
es
an
d
th
e
T
R
E
m
ea
s
u
r
e
as
ev
alu
atio
n
m
etr
ics.
I
n
th
e
o
r
ig
in
al
wo
r
k
o
f
Hu
et
a
l
.
[
8
]
,
th
e
y
u
s
ed
t
h
e
g
lo
b
al
-
n
et,
lo
c
al
-
n
et
an
d
co
m
p
o
s
ite
-
n
et
m
eth
o
d
s
.
T
h
e
g
lo
b
al
-
n
et
m
eth
o
d
p
r
ed
icts
an
af
f
in
e
tr
an
s
f
o
r
m
atio
n
an
d
th
e
lo
ca
l
-
n
et
m
eth
o
d
esti
m
ates
a
n
o
n
-
r
ig
id
lo
ca
l
DDF,
wh
ile
th
e
C
o
m
p
o
s
ite
-
n
et
m
eth
o
d
is
co
m
p
o
s
ed
o
f
th
e
Glo
b
al
-
n
et
f
o
llo
wed
b
y
t
h
e
l
o
ca
l
-
n
et.
L
a
ter
,
in
th
eir
e
x
ten
d
e
d
s
tu
d
y
[
9
]
,
th
e
y
i
n
teg
r
ated
th
ese
n
et
wo
r
k
s
in
to
a
s
in
g
l
e
f
r
am
ewo
r
k
th
at
p
r
ed
icts
th
e
D
DF
as
f
in
al
o
u
tp
u
t.
All
th
ese
m
eth
o
d
s
ar
e
test
ed
in
th
e
s
am
e
co
n
d
itio
n
s
as
o
u
r
ap
p
r
o
ac
h
,
u
s
in
g
th
e
s
am
e
d
ata
s
et
an
d
th
e
s
am
e
n
u
m
b
er
s
o
f
i
m
ag
e
p
air
s
.
M
o
r
eo
v
er
,
we
em
p
lo
y
ed
an
a
d
ap
tiv
e
g
r
ad
ien
t d
escen
t
o
p
tim
izatio
n
p
r
o
ce
s
s
an
d
s
im
ilar
ex
ec
u
tio
n
en
v
ir
o
n
m
en
t a
s
in
[
9
]
.
T
ab
le
1
an
d
Fig
u
r
e
4
s
u
m
m
ar
ize
th
e
q
u
an
titativ
e
r
esu
lts
in
t
er
m
o
f
d
ice
m
etr
ic
an
d
T
ab
le
2
p
r
esen
ts
th
e
T
R
E
m
ea
s
u
r
es
f
o
r
all
th
e
co
m
p
ar
e
d
m
eth
o
d
s
.
Ou
r
p
r
o
p
o
s
ed
n
etwo
r
k
f
o
r
3D
−
M
R
e
g
af
f
ViT
,
y
ield
ed
a
m
ed
ian
DSC
o
f
0
.
9
0
,
with
5
t
h
an
d
9
5
th
p
er
ce
n
tiles
o
f
0
.
8
0
an
d
0
.
9
4
,
r
esp
ec
tiv
ely
.
I
n
p
ar
allel,
th
e
m
ed
ia
n
T
R
E
f
o
r
lan
d
m
ar
k
ce
n
tr
o
id
s
o
b
tain
ed
f
r
o
m
t
h
e
s
am
e
n
etwo
r
k
s
m
ea
s
u
r
ed
6
.
0
,
with
5
th
an
d
9
5
th
p
er
ce
n
tiles
o
f
2
.
9
an
d
1
0
.
9
,
r
esp
ec
tiv
el
y
.
Mo
r
eo
v
er
,
3D
−
M
R
e
g
d
ef
ViT
g
av
e
a
m
ed
ian
DSC
o
f
0
.
9
5
,
with
5
th
an
d
9
5
th
p
er
ce
n
tiles
o
f
0
.
8
3
a
n
d
0
.
9
7
,
r
esp
ec
tiv
ely
.
Simu
ltan
eo
u
s
ly
,
th
e
m
ed
ian
T
R
E
f
o
r
lan
d
m
ar
k
ce
n
tr
o
id
s
o
b
tain
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
1
8
8
-
1
1
9
8
1194
f
r
o
m
th
e
s
am
e
n
etwo
r
k
s
m
ea
s
u
r
ed
1
.
7
,
with
5
th
a
n
d
9
5
th
p
e
r
ce
n
tiles
o
f
0
.
5
an
d
6
.
6
,
r
esp
e
ctiv
ely
.
W
e
d
ed
u
ce
th
at
o
u
r
a
p
p
r
o
ac
h
ac
h
iev
es
th
e
h
ig
h
est
d
ice
s
co
r
e
an
d
lo
w
er
T
R
E
co
m
p
ar
ed
to
th
e
o
th
e
r
ap
p
r
o
ac
h
es
(
with
p
-
v
alu
e
<
0
.
0
0
1
)
.
B
y
t
h
e
f
o
llo
win
g
,
we
will d
is
cu
s
s
th
ese
r
esu
lts
in
d
etails.
Firstl
y
,
we
ex
p
lo
r
e
t
h
e
p
o
wer
o
f
t
h
e
ViT
in
teg
r
atio
n
in
th
e
a
f
f
in
e
r
eg
is
tr
atio
n
s
tep
.
W
e
co
m
p
ar
e
3D
−
M
R
e
g
af
f
ViT
(
th
e
f
ir
s
t
s
tep
o
f
o
u
r
n
etwo
r
k
)
with
Glo
b
al
-
Net,
w
h
ich
is
th
e
clo
s
est
m
eth
o
d
to
o
u
r
wo
r
k
s
in
ce
b
o
th
o
f
t
h
em
tack
le
th
e
af
f
in
e
tr
an
s
f
o
r
m
atio
n
task
.
As
r
ep
o
r
ted
in
T
ab
l
e
1
,
o
u
r
ap
p
r
o
ac
h
p
r
o
d
u
ce
s
9
0
%
in
ter
m
o
f
D
SC
v
alu
e
o
u
tp
er
f
o
r
m
in
g
Glo
b
al
-
Net
b
y
1
3
%.
I
n
a
d
d
itio
n
,
3D
−
M
R
e
g
af
f
ViT
o
u
tp
er
f
o
r
m
s
lo
ca
l
-
n
et,
C
o
m
p
o
s
ite
-
Net
an
d
th
o
s
e
g
en
er
ated
b
y
Hu
et
a
l.
[
9
]
with
8
%,
6
% a
n
d
2
% r
esp
ec
tiv
ely
,
ev
en
th
ese
n
etwo
r
k
s
h
av
e
b
ee
n
p
r
o
p
o
s
ed
to
h
an
d
le
d
ef
o
r
m
ab
le
r
eg
is
tr
atio
n
.
T
h
is
p
er
f
o
r
m
an
ce
o
f
3D
−
M
R
e
g
af
f
ViT
is
d
u
e
to
ViT
'
s
k
n
o
wn
a
b
ilit
y
to
m
o
d
el
l
o
n
g
-
ter
m
d
ata
d
ep
en
d
en
cies
u
s
in
g
S
A
m
ec
h
an
is
m
s
[
1
0
]
,
[
1
1
]
,
esp
ec
ially
in
af
f
in
e
r
eg
is
tr
atio
n
,
wh
ich
in
v
o
lv
es
th
e
co
r
r
ec
tio
n
o
f
lar
g
e
-
s
ca
le
s
p
atial
tr
an
s
f
o
r
m
atio
n
s
s
u
c
h
as
tr
a
n
s
latio
n
s
,
r
o
tatio
n
s
a
n
d
s
ca
lin
g
.
On
th
e
o
th
er
h
an
d
,
C
NNs
h
av
e
b
ee
n
v
er
y
ef
f
ec
tiv
e
in
r
eg
is
ter
in
g
d
ef
o
r
m
ab
le
m
ed
ical
im
ag
es
[
2
7
]
,
b
u
t
ar
e
less
s
u
ited
in
m
o
d
elin
g
a
n
d
lear
n
in
g
a
f
f
in
e
r
eg
is
tr
atio
n
.
W
h
en
it
co
m
es
to
m
u
ltimo
d
al
r
e
g
is
tr
atio
n
o
f
m
ed
ical
im
a
g
es,
wh
ich
r
eq
u
ir
es
d
etailed
lo
ca
l
atten
tio
n
as
well
as
g
lo
b
al
u
n
d
er
s
tan
d
in
g
,
af
f
in
e
r
eg
is
tr
atio
n
p
r
o
v
es
in
s
u
f
f
icien
t.
T
h
is
is
wh
y
,
in
o
u
r
s
tu
d
y
,
we
o
p
ted
f
o
r
th
e
ap
p
licatio
n
o
f
af
f
in
e
tr
an
s
f
o
r
m
atio
n
as
an
in
itial
s
tep
u
s
in
g
ViT
(
3D
−
M
R
e
g
af
f
ViT
)
,
f
o
llo
wed
b
y
a
C
NN
n
etwo
r
k
to
ac
h
iev
e
d
ef
o
r
m
ab
le
r
e
g
is
tr
atio
n
.
T
h
is
co
m
b
i
n
atio
n
r
ep
r
esen
ted
b
y
o
u
r
3D
−
M
R
e
g
d
ef
ViT
led
to
s
ig
n
if
ican
tly
b
etter
r
esu
lts
th
an
th
e
ex
clu
s
iv
e
u
s
e
o
f
C
NN
o
r
ViT
alo
n
e.
As
d
ep
icted
in
T
ab
le
1
an
d
T
ab
le
2
,
th
e
Dice
an
d
T
R
E
s
co
r
es
o
f
3D
−
M
R
e
g
d
ef
ViT
,
r
ea
ch
ed
9
5
%
an
d
1
7
.
8
%
r
esp
ec
tiv
ely
.
T
h
is
r
ep
r
esen
ts
a
5
%
im
p
r
o
v
em
en
t
o
f
th
e
r
esu
lts
o
b
tain
ed
b
y
lo
ca
l
-
n
et,
a
7
%
im
p
r
o
v
em
e
n
t
o
f
th
o
s
e
p
r
o
d
u
c
ed
b
y
Hu
et
a
l.
[
9
]
,
in
ter
m
s
o
f
Dice
v
alu
e.
C
o
n
ce
r
n
in
g
th
e
T
R
E
s
co
r
e,
a
5
%
im
p
r
o
v
em
e
n
t
o
n
th
e
r
esu
lts
o
b
tain
ed
b
y
lo
ca
l
-
n
et,
an
d
a
7
%
im
p
r
o
v
em
e
n
t
o
n
th
o
s
e
g
en
er
at
ed
b
y
Hu
et
a
l.
[
9
]
h
av
e
b
ee
n
r
ea
ch
e
d
.
T
h
e
er
r
o
r
p
r
o
d
u
ce
d
is
th
e
lo
west,
wh
ich
in
d
icate
s
th
at
o
u
r
m
eth
o
d
i
s
m
o
r
e
ef
f
ec
tiv
e
at
m
in
im
izin
g
r
e
g
is
tr
atio
n
er
r
o
r
s
th
an
o
th
e
r
ap
p
r
o
ac
h
es.
A
m
in
im
al
T
R
E
in
d
icate
s
a
b
e
tter
s
p
atial
m
atch
b
etwe
en
th
e
war
p
e
d
im
ag
e
an
d
th
e
r
ef
e
r
en
ce
im
a
g
e,
u
n
d
er
li
n
in
g
th
e
s
u
p
er
io
r
p
er
f
o
r
m
a
n
c
e
o
f
o
u
r
m
et
h
o
d
i
n
ter
m
s
o
f
r
e
g
is
tr
atio
n
ac
cu
r
ac
y
co
m
p
a
r
ed
t
o
o
th
e
r
ap
p
r
o
ac
h
es.
T
h
e
o
b
tain
ed
r
esu
lts
o
f
3D
−
M
R
e
g
d
ef
ViT
im
p
r
o
v
e
s
ig
n
if
ican
tly
th
e
DDF
co
m
p
ar
e
d
t
o
d
ef
o
r
m
ab
le
C
NN
-
b
ased
m
eth
o
d
s
.
T
h
is
f
ac
t
ca
n
b
e
ex
p
lain
ed
b
y
th
e
f
o
cu
s
o
f
C
NN
ar
ch
itectu
r
es o
n
lo
ca
l in
f
o
r
m
atio
n
s
an
d
n
e
g
lect
g
lo
b
al
im
ag
e
co
n
tex
t.
I
n
co
n
tr
ast,
th
e
u
s
e
o
f
ViT
in
o
u
r
m
o
d
el
in
t
h
e
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
o
lv
es
th
e
li
m
itatio
n
s
o
f
c
o
n
v
o
lu
tio
n
al
o
p
er
atio
n
s
an
d
led
to
ca
p
tu
r
e
g
lo
b
al
an
ato
m
ical
s
tr
u
ctu
r
es
d
ep
en
d
e
n
cies
to
b
e
p
r
o
ce
s
s
ed
in
th
e
s
ec
o
n
d
n
etwo
r
k
an
d
g
en
e
r
ate
ac
cu
r
ate
DDF.
W
e
n
o
tice
th
a
t
it
is
d
if
f
icu
lt
to
c
o
n
d
u
ct
s
af
e
co
m
p
a
r
is
o
n
with
o
th
er
wo
r
k
s
p
r
o
p
o
s
ed
i
n
th
e
liter
atu
r
e
o
f
3
D
-
MReg
b
ec
au
s
e
o
f
th
e
lack
o
f
av
ailab
le
b
en
c
h
m
ar
k
s
an
d
m
o
s
t
o
f
p
u
b
lis
h
ed
p
ap
er
s
in
th
e
f
ield
d
ea
l w
ith
s
p
ec
if
ic
m
ed
ical
ap
p
licatio
n
s
an
d
p
r
iv
ate
d
atasets
.
Fig
u
r
e
3
.
E
x
am
p
le
o
f
im
ag
e
s
l
ices f
r
o
m
two
test
ca
s
es,
1
an
d
2
.
R
o
ws a,
b
,
c
,
a
n
d
d
s
h
o
w
th
e
s
lices e
x
tr
ac
ted
f
r
o
m
th
e
o
r
ig
in
al
MRI
im
a
g
es,
th
e
co
r
r
esp
o
n
d
in
g
f
ix
e
d
T
R
US im
ag
es,
th
e
s
lices g
en
er
ated
b
y
3
−
f
r
o
m
t
h
e
war
p
ed
m
o
v
i
n
g
MRI
im
ag
es,
an
d
th
e
s
lices g
en
er
a
ted
b
y
3
−
f
r
o
m
th
e
war
p
ed
m
o
v
in
g
MRI
im
ag
es,
r
esp
ec
tiv
ely
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
r
o
s
ta
te
ma
g
n
etic
r
eso
n
a
n
ce
ima
g
in
g
/tr
a
n
s
r
ec
ta
l u
ltr
a
s
o
u
n
d
r
eg
is
tr
a
tio
n
…
(
Ha
n
a
e
Ma
h
mo
u
d
i
)
1195
T
ab
le
1
.
T
h
e
r
esu
lts
o
f
th
e
c
u
r
r
en
t stu
d
y
in
ter
m
s
o
f
DSC
an
d
th
eir
co
m
p
ar
is
o
n
with
a
n
o
th
er
r
ec
en
t r
esear
ch
D
S
C
M
e
d
i
a
n
P
e
r
c
e
n
t
i
l
e
s
[
5
t
h
,
9
5
t
h
]
G
l
o
b
a
l
-
n
e
t
[
8
]
0
.
7
7
[
0
.
4
6
,
0
.
8
7
]
L
o
c
a
l
-
n
e
t
[
8
]
0
.
8
2
[
0
.
4
6
,
0
.
9
1
]
C
o
m
p
o
si
t
e
-
n
et
[
8
]
H
u
e
t
a
l
.
[
9
]
0
.
8
4
0
.
8
8
[
0
.
8
0
,
0
.
9
4
]
[
0
.
7
7
,
0
.
9
1
]
−
−
0
.
9
0
0
.
9
5
[
0
.
8
0
,
0
.
9
4
]
[
0
.
8
3
,
0
.
9
7
]
Fig
u
r
e
4
.
T
u
k
ey
b
o
x
p
lo
ts
d
is
p
lay
in
g
th
e
cr
o
s
s
-
v
alid
atio
n
r
es
u
lts
f
o
r
th
e
d
i
f
f
er
en
t
m
eth
o
d
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
1
8
8
-
1
1
9
8
1196
T
ab
le
2
.
T
h
e
r
esu
lts
o
f
th
e
c
u
r
r
en
t stu
d
y
in
ter
m
o
f
T
R
E
an
d
th
eir
co
m
p
a
r
is
o
n
with
o
th
e
r
r
e
ce
n
t r
esear
ch
TR
E
M
e
d
i
a
n
P
e
r
c
e
n
t
i
l
e
s
[
5
t
h
,
9
5
t
h
]
G
l
o
b
a
l
-
n
e
t
[
8
]
5
.
8
[
2
.
3
,
1
2
.
0
]
L
o
c
a
l
-
n
e
t
[
8
]
4
.
5
[
1
.
9
,
1
1
.
3
]
C
o
m
p
o
si
t
e
-
n
e
t
[
8
]
H
u
e
t
a
l
.
[
9
]
4
.
7
3
.
6
[
2
.
3
,
1
2
.
0
]
[
1
.
6
,
1
0
.
0
]
−
−
6
.
0
1
.
7
[
2
.
9
,
1
0
.
9
]
[
0
.
5
,
6
.
6
]
5.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
p
r
esen
ts
d
ef
o
r
m
ab
le
m
u
ltimo
d
al
r
eg
is
tr
atio
n
o
f
m
ed
ical
im
ag
es
b
ased
o
n
ViT
an
d
C
NN
(
3D
−
M
R
e
g
d
ef
ViT
)
.
I
t
u
s
es
3
D
m
ed
ical
im
a
g
es
in
clu
d
in
g
MRI
an
d
T
R
US,
co
n
s
id
er
ed
r
esp
ec
tiv
ely
as
m
o
v
in
g
a
n
d
r
ef
e
r
en
ce
im
ag
es,
an
d
th
eir
lab
els.
I
n
th
is
p
r
o
ce
s
s
,
im
ag
es a
n
d
lab
els ar
e
p
r
ep
r
o
ce
s
s
ed
b
ef
o
r
e
u
s
e:
MRI
an
d
T
R
US
im
ag
es
ar
e
r
e
s
ized
an
d
n
o
r
m
alize
d
,
th
e
n
th
eir
lab
els
ar
e
r
e
-
d
im
e
n
s
io
n
ed
to
th
e
s
am
e
s
ize
as
im
ag
es
an
d
s
m
o
o
th
ed
u
s
in
g
in
v
er
s
e
d
is
tan
ce
tr
a
n
s
f
o
r
m
.
T
h
e
wh
o
le
ar
c
h
itectu
r
e
ta
k
es
th
e
co
n
ca
ten
atio
n
o
f
MRI
an
d
T
R
US
im
ag
es
as
in
p
u
t
to
th
e
ViT
ar
ch
itectu
r
e
to
p
r
ed
ict
th
e
af
f
in
e
tr
an
s
f
o
r
m
ati
o
n
m
atr
ix
th
at
will
b
e
ap
p
lied
to
th
e
m
o
v
in
g
im
a
g
e.
T
h
e
n
,
th
e
r
esu
ltin
g
im
a
g
e
will
b
e
co
n
ca
ten
ated
with
th
e
f
ix
ed
im
ag
e
an
d
f
ed
to
th
e
lo
ca
l
n
etwo
r
k
to
p
r
ed
ict
th
e
lo
ca
l
DDF.
T
h
e
f
in
al
s
tep
co
n
s
is
ts
o
f
co
m
b
in
in
g
th
e
tr
a
n
s
f
o
r
m
atio
n
m
atr
ix
with
th
e
lo
ca
l
DDF
to
p
r
ed
ict
th
e
o
u
tp
u
t
o
f
th
e
DDF,
wh
ich
is
th
en
ap
p
lied
to
th
e
m
o
v
i
n
g
lab
els.
T
h
e
r
esu
lts
ar
e
ev
alu
ated
u
s
in
g
th
e
d
ice
m
etr
ic
ca
lcu
lated
b
etwe
en
th
e
m
o
v
in
g
d
e
f
o
r
m
e
d
an
d
f
ix
ed
tar
g
ets.
Ou
r
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
ac
h
ie
v
ed
p
r
o
m
is
in
g
r
esu
lts
co
m
p
ar
ed
to
p
r
e
v
io
u
s
wo
r
k
s
u
s
in
g
p
u
r
e
C
NN
ar
c
h
itectu
r
es.
As
f
u
tu
r
e
wo
r
k
,
we
will e
x
p
lo
r
e
o
th
e
r
v
a
r
ian
ts
o
f
ViT
s
an
d
test
o
th
er
d
atasets
in
th
e
f
ield
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
au
th
o
r
s
s
in
ce
r
ely
th
an
k
th
e
an
o
n
y
m
o
u
s
r
ev
iewe
r
s
f
o
r
th
eir
co
n
s
tr
u
ctiv
e
co
m
m
en
ts
an
d
in
s
ig
h
tf
u
l
s
u
g
g
esti
o
n
s
,
wh
ich
s
ig
n
if
ican
t
ly
im
p
r
o
v
ed
th
is
m
an
u
s
cr
ip
t.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
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
Data
av
ailab
ilit
y
is
n
o
t
ap
p
li
ca
b
le
to
th
is
p
ap
er
as
n
o
n
e
w
d
ata
wer
e
cr
ea
ted
o
r
an
aly
ze
d
in
th
is
s
tu
d
y
.
RE
F
E
R
E
NC
E
S
[
1
]
Q
.
Zh
e
n
g
,
C
.
L
i
u
,
a
n
d
J
.
C
h
a
n
g
,
“
N
o
n
-
r
i
g
i
d
r
e
g
i
st
r
a
t
i
o
n
o
f
me
d
i
c
a
l
i
m
a
g
e
s
b
a
s
e
d
o
n
S
2
1
(
∆
mn
(
2
)
)
n
o
n
-
t
e
n
so
r
p
r
o
d
u
c
t
B
-
sp
l
i
n
e
,
”
Vi
su
a
l
C
o
m
p
u
t
i
n
g
f
o
r I
n
d
u
s
t
ry
,
B
i
o
m
e
d
i
c
i
n
e
,
a
n
d
Ar
t
,
v
o
l
.
5
,
n
o
.
1
,
p
.
5
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
2
4
9
2
-
0
2
2
-
0
0
1
0
1
-
8.
[
2
]
J.
Y
i
n
g
e
t
a
l
.
,
“
Tw
o
f
u
l
l
y
a
u
t
o
ma
t
e
d
d
a
t
a
-
d
r
i
v
e
n
3
D
w
h
o
l
e
-
b
r
e
a
st
se
g
m
e
n
t
a
t
i
o
n
s
t
r
a
t
e
g
i
e
s
i
n
M
R
I
f
o
r
M
R
-
b
a
sed
b
r
e
a
s
t
d
e
n
si
t
y
u
si
n
g
i
m
a
g
e
r
e
g
i
s
t
r
a
t
i
o
n
a
n
d
U
-
N
e
t
w
i
t
h
a
f
o
c
u
s o
n
r
e
p
r
o
d
u
c
i
b
i
l
i
t
y
,
”
V
i
s
u
a
l
C
o
m
p
u
t
i
n
g
f
o
r
I
n
d
u
s
t
ry,
Bi
o
m
e
d
i
c
i
n
e
,
a
n
d
Art
,
v
o
l
.
5
,
n
o
.
1
,
p
.
2
5
,
O
c
t
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
2
4
9
2
-
0
2
2
-
0
0
1
2
1
-
4.
[
3
]
G
.
H
e
r
mo
s
i
l
l
o
,
C
.
C
h
e
f
d
’
H
o
t
e
l
,
a
n
d
O
.
F
a
u
g
e
r
a
s,
“
V
a
r
i
a
t
i
o
n
a
l
met
h
o
d
s
f
o
r
mu
l
t
i
mo
d
a
l
i
ma
g
e
ma
t
c
h
i
n
g
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
Vi
s
i
o
n
,
v
o
l
.
5
0
,
n
o
.
3
,
p
p
.
3
2
9
–
3
4
3
,
D
e
c
.
2
0
0
2
,
d
o
i
:
1
0
.
1
0
2
3
/
A
:
1
0
2
0
8
3
0
5
2
5
8
2
3
.
[
4
]
J.
B
.
A
.
M
a
i
n
t
z
,
P
.
A
.
v
a
n
d
e
n
El
s
e
n
,
a
n
d
M
.
A
.
V
i
e
r
g
e
v
e
r
,
“
3
D
m
u
l
t
i
mo
d
a
l
i
t
y
m
e
d
i
c
a
l
i
m
a
g
e
r
e
g
i
s
t
r
a
t
i
o
n
u
s
i
n
g
mo
r
p
h
o
l
o
g
i
c
a
l
t
o
o
l
s
,
”
I
m
a
g
e
a
n
d
Vi
s
i
o
n
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
9
,
n
o
.
1
–
2
,
p
p
.
5
3
–
6
2
,
J
a
n
.
2
0
0
1
,
d
o
i
:
1
0
.
1
0
1
6
/
S
0
2
6
2
-
8
8
5
6
(
0
0
)
0
0
0
5
1
-
2.
[
5
]
F
.
M
a
e
s,
A
.
C
o
l
l
i
g
n
o
n
,
D
.
V
a
n
d
e
r
m
e
u
l
e
n
,
G
.
M
a
r
c
h
a
l
,
a
n
d
P
.
S
u
e
t
e
n
s,
“
M
u
l
t
i
m
o
d
a
l
i
t
y
i
m
a
g
e
r
e
g
i
s
t
r
a
t
i
o
n
b
y
max
i
mi
z
a
t
i
o
n
o
f
mu
t
u
a
l
i
n
f
o
r
ma
t
i
o
n
,
”
I
EEE
T
r
a
n
sa
c
t
i
o
n
s
o
n
Me
d
i
c
a
l
I
m
a
g
i
n
g
,
v
o
l
.
1
6
,
n
o
.
2
,
p
p
.
1
8
7
–
1
9
8
,
A
p
r
.
1
9
9
7
,
d
o
i
:
1
0
.
1
1
0
9
/
4
2
.
5
6
3
6
6
4
.
[
6
]
S
.
B
h
a
r
a
t
i
,
M
.
M
o
n
d
a
l
,
P
.
P
o
d
d
e
r
,
a
n
d
V
.
B
.
P
r
a
s
a
t
h
,
“
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
med
i
c
a
l
i
m
a
g
e
r
e
g
i
s
t
r
a
t
i
o
n
:
a
c
o
mp
r
e
h
e
n
si
v
e
r
e
v
i
e
w
,
”
a
rXi
v
p
re
p
r
i
n
t
a
rXi
v
:
2
2
0
4
.
1
1
3
4
1
,
2
0
2
2
.
[
7
]
B
.
D
.
d
e
V
o
s
,
F
.
F
.
B
e
r
e
n
d
s
e
n
,
M
.
A
.
V
i
e
r
g
e
v
e
r
,
M
.
S
t
a
r
i
n
g
,
a
n
d
I
.
I
šg
u
m,
“
En
d
-
to
-
e
n
d
u
n
s
u
p
e
r
v
i
se
d
d
e
f
o
r
ma
b
l
e
i
ma
g
e
r
e
g
i
s
t
r
a
t
i
o
n
w
i
t
h
a
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
,
”
i
n
I
n
t
e
r
n
a
t
i
o
n
a
l
W
o
rks
h
o
p
o
n
D
e
e
p
L
e
a
r
n
i
n
g
i
n
M
e
d
i
c
a
l
I
m
a
g
e
A
n
a
l
y
s
i
s
,
2
0
1
7
,
p
p
.
2
0
4
–
212
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
6
7
5
5
8
-
9
_
2
4
.
[
8
]
Y
.
H
u
e
t
a
l
.
,
“
L
a
b
e
l
-
d
r
i
v
e
n
w
e
a
k
l
y
-
s
u
p
e
r
v
i
s
e
d
l
e
a
r
n
i
n
g
f
o
r
m
u
l
t
i
m
o
d
a
l
d
e
f
o
r
m
a
r
l
e
i
m
a
g
e
r
e
g
i
s
t
r
a
t
i
o
n
,
”
i
n
2
0
1
8
I
E
E
E
1
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
S
y
m
p
o
s
i
u
m
o
n
B
i
o
m
e
d
i
c
a
l
I
m
a
g
i
n
g
(
I
S
B
I
2
0
1
8
)
,
I
E
E
E
,
A
p
r
.
2
0
1
8
,
p
p
.
1
0
7
0
–
1
0
7
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
S
B
I
.
2
0
1
8
.
8
3
6
3
7
5
6
.
[
9
]
Y
.
H
u
e
t
a
l
.
,
“
W
e
a
k
l
y
-
s
u
p
e
r
v
i
se
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
mu
l
t
i
mo
d
a
l
i
ma
g
e
r
e
g
i
s
t
r
a
t
i
o
n
,
”
Me
d
i
c
a
l
I
m
a
g
e
A
n
a
l
y
s
i
s
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
r
o
s
ta
te
ma
g
n
etic
r
eso
n
a
n
ce
ima
g
in
g
/tr
a
n
s
r
ec
ta
l u
ltr
a
s
o
u
n
d
r
eg
is
tr
a
tio
n
…
(
Ha
n
a
e
Ma
h
mo
u
d
i
)
1197
v
o
l
.
4
9
,
p
p
.
1
–
1
3
,
O
c
t
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
m
e
d
i
a
.
2
0
1
8
.
0
7
.
0
0
2
.
[
1
0
]
A
.
V
a
sw
a
n
i
e
t
a
l
.
,
“
A
t
t
e
n
t
i
o
n
i
s
a
l
l
y
o
u
n
e
e
d
,
”
A
d
v
a
n
c
e
s
i
n
n
e
u
r
a
l
i
n
f
o
rm
a
t
i
o
n
p
r
o
c
e
ssi
n
g
s
y
st
e
m
s
,
v
o
l
.
3
0
,
2
0
1
7
.
[
1
1
]
Y
.
S
a
l
i
n
i
a
n
d
J
.
H
a
r
i
K
i
r
a
n
,
“
V
i
T:
q
u
a
n
t
i
f
y
i
n
g
c
h
e
st
x
-
r
a
y
i
ma
g
e
s
u
si
n
g
v
i
s
i
o
n
t
r
a
n
sf
o
r
m
e
r
&
X
A
I
t
e
c
h
n
i
q
u
e
,
”
S
N
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
4
,
n
o
.
6
,
p
.
7
5
4
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s4
2
9
7
9
-
0
2
3
-
0
2
2
0
4
-
2.
[
1
2
]
A
.
D
o
so
v
i
t
s
k
i
y
e
t
a
l
.
,
“
A
n
i
m
a
g
e
i
s
w
o
r
t
h
1
6
x
1
6
w
o
r
d
s
:
t
r
a
n
sf
o
r
mers
f
o
r
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
a
t
sc
a
l
e
,
”
i
n
a
r
Xi
v
p
re
p
r
i
n
t
a
rXi
v
:
2
0
1
0
.
1
1
9
2
9
,
2
0
2
1
.
[
1
3
]
H
.
R
a
ma
d
a
n
,
D
.
E
l
B
o
u
r
a
k
a
d
i
,
A
.
Y
a
h
y
a
o
u
y
,
a
n
d
H
.
Ta
i
r
i
,
“
M
e
d
i
c
a
l
i
mag
e
r
e
g
i
st
r
a
t
i
o
n
i
n
t
h
e
e
r
a
o
f
t
r
a
n
s
f
o
r
mers
:
a
r
e
c
e
n
t
r
e
v
i
e
w
,
”
I
n
f
o
rm
a
t
i
c
s
i
n
Me
d
i
c
i
n
e
U
n
l
o
c
k
e
d
,
v
o
l
.
4
9
,
p
.
1
0
1
5
4
0
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
m
u
.
2
0
2
4
.
1
0
1
5
4
0
.
[
1
4
]
T.
C
.
W
.
M
o
k
a
n
d
A
.
C
.
S
.
C
h
u
n
g
,
“
A
f
f
i
n
e
me
d
i
c
a
l
i
m
a
g
e
r
e
g
i
s
t
r
a
t
i
o
n
w
i
t
h
c
o
a
r
se
-
to
-
f
i
n
e
v
i
s
i
o
n
t
r
a
n
sf
o
r
mer,
”
i
n
2
0
2
2
I
EE
E/
C
VF
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
si
o
n
a
n
d
P
a
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
(
C
V
PR)
,
I
EEE,
Ju
n
.
2
0
2
2
,
p
p
.
2
0
8
0
3
–
2
0
8
1
2
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
5
2
6
8
8
.
2
0
2
2
.
0
2
0
1
7
.
[
1
5
]
G
.
H
a
s
k
i
n
s
e
t
a
l
.
,
“
Le
a
r
n
i
n
g
d
e
e
p
si
mi
l
a
r
i
t
y
met
r
i
c
f
o
r
3
D
M
R
–
TR
U
S
i
mag
e
r
e
g
i
st
r
a
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
C
o
m
p
u
t
e
r
Assi
st
e
d
R
a
d
i
o
l
o
g
y
a
n
d
S
u
rg
e
ry
,
v
o
l
.
1
4
,
n
o
.
3
,
p
p
.
4
1
7
–
4
2
5
,
M
a
r
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
5
4
8
-
0
1
8
-
1
8
7
5
-
7.
[
1
6
]
M
.
S
i
mo
n
o
v
s
k
y
,
B
.
G
u
t
i
é
r
r
e
z
-
B
e
c
k
e
r
,
D
.
M
a
t
e
u
s
,
N
.
N
a
v
a
b
,
a
n
d
N
.
K
o
m
o
d
a
k
i
s
,
“
A
d
e
e
p
m
e
t
r
i
c
f
o
r
mu
l
t
i
mo
d
a
l
r
e
g
i
s
t
r
a
t
i
o
n
,
”
i
n
I
n
t
e
r
n
a
t
i
o
n
a
l
c
o
n
f
e
re
n
c
e
o
n
m
e
d
i
c
a
l
i
m
a
g
e
c
o
m
p
u
t
i
n
g
a
n
d
c
o
m
p
u
t
e
r
-
a
ssi
s
t
e
d
i
n
t
e
r
v
e
n
t
i
o
n
,
2
0
1
6
,
p
p
.
1
0
–
18
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
4
6
7
2
6
-
9
_
2
.
[
1
7
]
O
.
Z
e
t
t
i
n
i
g
e
t
a
l
.
,
“
M
u
l
t
i
m
o
d
a
l
i
m
a
g
e
-
g
u
i
d
e
d
p
r
o
st
a
t
e
f
u
s
i
o
n
b
i
o
p
s
y
b
a
s
e
d
o
n
a
u
t
o
m
a
t
i
c
d
e
f
o
r
m
a
b
l
e
r
e
g
i
s
t
r
a
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
A
ss
i
s
t
e
d
R
a
d
i
o
l
o
g
y
a
n
d
S
u
r
g
e
ry
,
v
o
l
.
1
0
,
n
o
.
1
2
,
p
p
.
1
9
9
7
–
2
0
0
7
,
D
e
c
.
2
0
1
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
5
4
8
-
015
-
1233
-
y.
[
1
8
]
Y
.
W
a
n
g
e
t
a
l
.
,
“
M
u
l
t
i
m
o
d
a
l
r
e
g
i
s
t
r
a
t
i
o
n
o
f
u
l
t
r
a
so
u
n
d
a
n
d
M
R
i
m
a
g
e
s
u
si
n
g
w
e
i
g
h
t
e
d
s
e
l
f
-
si
mi
l
a
r
i
t
y
s
t
r
u
c
t
u
r
e
v
e
c
t
o
r
,
”
C
o
m
p
u
t
e
rs
i
n
B
i
o
l
o
g
y
a
n
d
Me
d
i
c
i
n
e
,
v
o
l
.
1
5
5
,
p
.
1
0
6
6
6
1
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
m
p
b
i
o
me
d
.
2
0
2
3
.
1
0
6
6
6
1
.
[
1
9
]
M
.
B
l
e
n
d
o
w
sk
i
,
N
.
B
o
u
t
e
l
d
j
a
,
a
n
d
M
.
P
.
H
e
i
n
r
i
c
h
,
“
M
u
l
t
i
mo
d
a
l
3
D
m
e
d
i
c
a
l
i
m
a
g
e
r
e
g
i
st
r
a
t
i
o
n
g
u
i
d
e
d
b
y
sh
a
p
e
e
n
c
o
d
e
r
–
d
e
c
o
d
e
r
n
e
t
w
o
r
k
s,”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
C
o
m
p
u
t
e
r
Assi
s
t
e
d
R
a
d
i
o
l
o
g
y
a
n
d
S
u
rg
e
ry
,
v
o
l
.
1
5
,
n
o
.
2
,
p
p
.
2
6
9
–
2
7
6
,
F
e
b
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
5
4
8
-
0
1
9
-
0
2
0
8
9
-
8.
[
2
0
]
X
.
S
o
n
g
e
t
a
l
.
,
“
C
r
o
ss
-
mo
d
a
l
a
t
t
e
n
t
i
o
n
f
o
r
m
u
l
t
i
-
m
o
d
a
l
i
m
a
g
e
r
e
g
i
st
r
a
t
i
o
n
,
”
M
e
d
i
c
a
l
I
m
a
g
e
A
n
a
l
y
si
s
,
v
o
l
.
8
2
,
p
.
1
0
2
6
1
2
,
N
o
v
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
me
d
i
a
.
2
0
2
2
.
1
0
2
6
1
2
.
[
2
1
]
P
.
Lo
r
e
n
z
e
n
,
M
.
P
r
a
s
t
a
w
a
,
B
.
D
a
v
i
s,
G
.
G
e
r
i
g
,
E.
B
u
l
l
i
t
t
,
a
n
d
S
.
J
o
s
h
i
,
“
M
u
l
t
i
-
m
o
d
a
l
i
ma
g
e
s
e
t
r
e
g
i
st
r
a
t
i
o
n
a
n
d
a
t
l
a
s
f
o
r
m
a
t
i
o
n
,
”
Me
d
i
c
a
l
I
m
a
g
e
A
n
a
l
y
s
i
s
,
v
o
l
.
1
0
,
n
o
.
3
,
p
p
.
4
4
0
–
4
5
1
,
Ju
n
.
2
0
0
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
me
d
i
a
.
2
0
0
5
.
0
3
.
0
0
2
.
[
2
2
]
K
.
M
a
e
t
a
l
.
,
“
M
u
l
t
i
mo
d
a
l
i
ma
g
e
r
e
g
i
st
r
a
t
i
o
n
w
i
t
h
d
e
e
p
c
o
n
t
e
x
t
r
e
i
n
f
o
r
c
e
m
e
n
t
l
e
a
r
n
i
n
g
,
”
i
n
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
Me
d
i
c
a
l
i
m
a
g
e
c
o
m
p
u
t
i
n
g
a
n
d
c
o
m
p
u
t
e
r
-
a
ss
i
st
e
d
i
n
t
e
r
v
e
n
t
i
o
n
,
2
0
1
7
,
p
p
.
2
4
0
–
2
4
8
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
6
6
1
8
2
-
7
_
2
8
.
[
2
3
]
D
.
R
u
e
c
k
e
r
t
,
L.
I
.
S
o
n
o
d
a
,
C
.
H
a
y
e
s,
D
.
L.
G
.
H
i
l
l
,
M
.
O
.
Le
a
c
h
,
a
n
d
D
.
J.
H
a
w
k
e
s,
“
N
o
n
r
i
g
i
d
r
e
g
i
st
r
a
t
i
o
n
u
si
n
g
f
r
e
e
-
f
o
r
m
d
e
f
o
r
mat
i
o
n
s
:
a
p
p
l
i
c
a
t
i
o
n
t
o
b
r
e
a
s
t
M
R
i
ma
g
e
s
,
”
I
EE
E
T
r
a
n
sa
c
t
i
o
n
s
o
n
Me
d
i
c
a
l
I
m
a
g
i
n
g
,
v
o
l
.
1
8
,
n
o
.
8
,
p
p
.
7
1
2
–
7
2
1
,
1
9
9
9
,
d
o
i
:
1
0
.
1
1
0
9
/
4
2
.
7
9
6
2
8
4
.
[
2
4
]
F
.
M
i
l
l
e
t
a
r
i
,
N
.
N
a
v
a
b
,
a
n
d
S
.
-
A
.
A
h
ma
d
i
,
“
V
-
N
e
t
:
f
u
l
l
y
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
f
o
r
v
o
l
u
me
t
r
i
c
med
i
c
a
l
i
ma
g
e
seg
m
e
n
t
a
t
i
o
n
,
”
i
n
2
0
1
6
F
o
u
rt
h
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
3
D
V
i
s
i
o
n
(
3
D
V)
,
I
EEE,
O
c
t
.
2
0
1
6
,
p
p
.
5
6
5
–
571
,
d
o
i
:
1
0
.
1
1
0
9
/
3
D
V
.
2
0
1
6
.
7
9
.
[
2
5
]
I
.
G
o
o
d
f
e
l
l
o
w
,
Y
.
B
e
n
g
i
o
,
A
.
C
o
u
r
v
i
l
l
e
,
a
n
d
Y
.
B
e
n
g
i
o
,
D
e
e
p
l
e
a
r
n
i
n
g
,
v
o
l
.
1
,
n
o
.
2
.
C
a
mb
r
i
d
g
e
:
M
I
T
p
r
e
ss
C
a
mb
r
i
d
g
e
,
2
0
1
6
.
[
2
6
]
I
.
D
o
n
a
l
d
s
o
n
e
t
a
l
.
,
“
M
P
3
3
-
2
0
t
h
e
s
mart
t
a
r
g
e
t
b
i
o
p
s
y
t
r
i
a
l
:
a
p
r
o
s
p
e
c
t
i
v
e
p
a
i
r
e
d
b
l
i
n
d
e
d
t
r
i
a
l
w
i
t
h
r
a
n
d
o
m
i
s
a
t
i
o
n
t
o
c
o
m
p
a
r
e
v
i
s
u
a
l
-
e
st
i
m
a
t
i
o
n
a
n
d
i
m
a
g
e
-
f
u
s
i
o
n
t
a
r
g
e
t
e
d
p
r
o
s
t
a
t
e
b
i
o
p
s
i
e
s,
”
J
o
u
r
n
a
l
o
f
U
ro
l
o
g
y
,
v
o
l
.
1
9
7
,
n
o
.
4
S
,
p
p
.
e
4
2
5
--
e
4
2
5
,
A
p
r
.
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
u
r
o
.
2
0
1
7
.
0
2
.
1
0
1
6
.
[
2
7
]
L.
U
n
i
v
e
r
si
t
y
C
o
l
l
e
g
e
,
“
S
mart
Ta
r
g
e
t
-
a
mag
n
e
t
i
c
r
e
s
o
n
a
n
c
e
i
ma
g
e
t
o
u
l
t
r
a
so
u
n
d
f
u
s
i
o
n
s
y
st
e
m
f
o
r
t
a
r
g
e
t
e
d
p
r
o
s
t
a
t
e
i
n
t
e
r
v
e
n
t
i
o
n
:
b
i
o
p
sy
.
”
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
c
l
i
n
i
c
a
l
t
r
i
a
l
s.
g
o
v
/
st
u
d
y
/
N
C
T0
2
3
4
1
6
7
7
[
2
8
]
N
.
G
h
a
v
a
mi
e
t
a
l
.
,
“
A
u
t
o
ma
t
i
c
sl
i
c
e
seg
me
n
t
a
t
i
o
n
o
f
i
n
t
r
a
o
p
e
r
a
t
i
v
e
t
r
a
n
sr
e
c
t
a
l
u
l
t
r
a
s
o
u
n
d
i
ma
g
e
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,”
i
n
M
e
d
i
c
a
l
I
m
a
g
i
n
g
2
0
1
8
:
I
m
a
g
e
-
G
u
i
d
e
d
Pro
c
e
d
u
r
e
s,
R
o
b
o
t
i
c
I
n
t
e
r
v
e
n
t
i
o
n
s,
a
n
d
M
o
d
e
l
i
n
g
,
S
P
I
E,
M
a
r
.
2
0
1
8
,
d
o
i
:
1
0
.
1
1
1
7
/
1
2
.
2
2
9
3
3
0
0
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
H
a
n
a
e
Ma
h
m
o
u
d
i
o
b
tain
e
d
a
m
a
ste
r’s
d
e
g
re
e
in
b
u
sin
e
ss
in
telli
g
e
n
c
e
a
n
d
in
telli
g
e
n
t
v
isi
o
n
fr
o
m
S
id
i
M
o
h
a
m
e
d
Be
n
Ab
d
e
ll
a
h
Un
i
v
e
rsity
,
M
o
ro
c
c
o
,
i
n
2
0
2
0
.
No
w
sh
e
is
a
P
h
.
D
.
s
tu
d
e
n
t
a
t
S
id
i
M
o
h
a
m
m
e
d
Be
n
Ab
d
e
ll
a
h
Un
i
v
e
rsity
,
M
o
ro
c
c
o
,
2
0
2
0
.
He
r
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
a
rti
ficia
l
in
tel
li
g
e
n
c
e
a
n
d
m
e
d
ica
l
ima
g
in
g
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
h
a
n
a
e
.
m
a
h
m
o
u
d
i@u
sm
b
a
.
a
c
.
m
a
.
H
ib
a
Ra
m
a
d
a
n
re
c
e
iv
e
d
h
e
r
P
h
.
D.
d
e
g
re
e
i
n
2
0
1
8
fro
m
t
h
e
Un
iv
e
rsit
y
S
id
i
M
o
h
a
m
e
d
Be
n
Ab
d
e
ll
a
h
(USM
B
A),
F
e
z
,
M
o
r
o
c
c
o
.
C
u
rre
n
tl
y
,
s
h
e
is
a
n
a
ss
istan
t
p
r
o
fe
ss
o
r
a
t
S
id
i
M
o
h
a
m
e
d
Be
n
A
b
d
e
ll
a
h
Un
i
v
e
rsity
M
o
r
o
c
c
o
.
He
r
re
se
a
rc
h
in
t
e
re
sts
a
re
in
v
i
d
e
o
a
n
a
l
y
sis
,
in
m
e
d
ica
l
ima
g
in
g
,
i
n
v
isu
a
l
in
fo
rm
a
ti
o
n
re
tr
iev
a
l
a
n
d
p
a
tt
e
r
n
re
c
o
g
n
it
i
o
n
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
h
i
b
a
.
ra
m
a
d
a
n
@u
sm
b
a
.
a
c
.
m
a
.
Evaluation Warning : The document was created with Spire.PDF for Python.