I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
E
lec
t
r
ical
an
d
Com
p
u
t
e
r
E
n
gin
e
e
r
in
g
(
I
JE
CE
)
Vol.
1
4
,
No.
5
,
Oc
tober
20
2
4
,
pp
.
5524
~
5533
I
S
S
N:
2088
-
8708
,
DO
I
:
10
.
11591/i
jec
e
.
v
1
4
i
5
.
pp
5
524
-
5533
5524
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
e
c
e
.
iaes
c
or
e
.
c
om
A
f
r
am
e
w
o
r
k
f
or
3D
r
ad
io
t
h
e
r
ap
y
d
ose
p
r
e
d
ic
t
io
n
u
si
n
g
t
h
e
d
e
e
p
l
e
a
r
n
i
n
g ap
p
r
oa
c
h
L
am
T
h
an
h
Hi
e
n
1
,
Ha
M
an
h
T
oan
2
,
Do
Na
n
g
T
oan
2
1
F
a
c
ul
ty
of
I
nf
or
ma
ti
on T
e
c
hnol
ogy, L
a
c
H
ong Unive
r
s
it
y, B
ie
n H
oa
, D
ong Na
i,
V
ie
tn
a
m
2
I
ns
ti
tu
te
of
I
nf
or
ma
ti
on T
e
c
hnol
ogy, Vie
tn
a
m Ac
a
de
my of
S
c
i
e
nc
e
a
nd
T
e
c
hnol
ogy, Ha
noi
, V
ie
tn
a
m
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
J
a
n
4,
2024
R
e
vis
e
d
J
un
20,
2024
Ac
c
e
pted
J
ul
2,
2024
Can
cer
i
s
k
n
o
w
n
as
a
d
a
n
g
er
o
u
s
d
i
s
eas
e
t
o
h
u
man
s
w
i
t
h
a
v
er
y
h
i
g
h
d
e
at
h
rat
e.
T
h
ere
are
a
l
o
t
o
f
can
cer
t
reat
me
n
t
me
t
h
o
d
s
t
h
a
t
h
av
e
b
een
s
t
u
d
i
e
d
an
d
ap
p
l
i
e
d
i
n
t
h
e
w
o
rl
d
.
O
n
e
o
f
t
h
e
mai
n
met
h
o
d
s
i
s
u
s
i
n
g
rad
i
at
i
o
n
b
eams
t
o
k
i
l
l
can
cer
cel
l
s
.
T
h
i
s
met
h
o
d
,
al
s
o
k
n
o
w
n
as
rad
i
o
t
h
er
ap
y
,
req
u
i
res
ex
p
ert
s
h
av
i
n
g
a
h
i
g
h
l
e
v
el
o
f
s
k
i
l
l
a
n
d
ex
p
eri
e
n
ce.
O
u
r
w
o
rk
f
o
cu
s
e
s
o
n
t
h
e
3
D
d
o
s
e
p
red
i
ct
i
o
n
p
ro
b
l
em
i
n
rad
i
o
t
h
era
p
y
b
y
p
ro
p
o
s
i
n
g
a
framew
o
r
k
ai
mi
n
g
t
o
creat
e
a
med
i
ca
l
i
n
t
e
l
l
i
g
e
n
t
s
y
s
t
em
f
o
r
t
h
i
s
p
ro
b
l
em.
T
o
d
o
t
h
at
,
w
e
creat
e
d
a
co
n
v
o
l
u
t
i
o
n
a
l
n
eu
ra
l
n
e
t
w
o
rk
b
a
s
ed
o
n
Res
N
et
a
n
d
U
-
N
et
t
o
g
e
n
erat
e
t
h
e
p
red
i
ct
e
d
rad
i
a
t
i
o
n
d
o
s
e.
T
o
i
m
p
ro
v
e
t
h
e
q
u
a
l
i
t
y
o
f
t
h
e
t
rai
n
i
n
g
p
h
as
e,
w
e
al
s
o
ap
p
l
i
e
d
s
o
me
d
at
a
p
ro
ce
s
s
i
n
g
t
ech
n
i
q
u
e
s
b
a
s
ed
o
n
t
h
e
ch
arac
t
eri
s
t
i
cs
o
f
t
h
e
3
D
co
m
p
u
t
ed
t
o
mo
g
rap
h
y
(CT
)
d
at
a.
T
h
e
ex
p
er
i
m
en
t
u
s
e
d
t
h
e
d
a
t
as
e
t
fro
m
p
a
t
i
e
n
t
s
w
h
o
w
ere
ca
n
cer
-
t
reat
e
d
w
i
t
h
ra
d
i
o
t
h
era
p
y
i
n
t
h
e
O
p
en
K
BP
co
mp
e
t
i
t
i
o
n
.
T
h
e
res
u
l
t
s
ach
i
ev
e
d
g
o
o
d
ev
a
l
u
a
t
i
n
g
met
r
i
cs
,
t
h
e
fi
rs
t
i
s
b
y
t
h
e
D
o
s
e
-
s
c
o
re
an
d
t
h
e
s
ec
o
n
d
i
s
b
y
t
h
e
d
o
s
e
-
v
o
l
u
me
h
i
s
t
o
g
ram
(D
V
H
)
s
co
re.
Fro
m
t
h
e
t
rai
n
i
n
g
re
s
u
l
t
,
w
e
b
u
i
l
t
t
h
e
med
i
cal
s
y
s
t
em
s
u
p
p
o
r
t
i
n
g
3
D
d
o
s
e
p
red
i
ct
i
o
n
an
d
v
i
s
u
a
l
i
z
i
n
g
t
h
e
re
s
u
l
t
as
s
l
i
ces
i
n
h
ea
t
map
fo
rm.
K
e
y
w
o
r
d
s
:
3D
dos
e
pr
e
diction
C
a
nc
e
r
tr
e
a
tm
e
nt
R
a
diot
he
r
a
py
R
e
s
N
et
U
-
N
et
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
Do
Na
ng
T
oa
n
I
ns
ti
tut
e
of
I
nf
o
r
mation
T
e
c
hnology,
Vie
tnam
Ac
a
de
my
of
S
c
ienc
e
a
nd
T
e
c
hnology
18
Hoa
ng
Quoc
Vie
t,
C
a
uGia
y
Dis
tr
ict,
Ha
noi
100
72,
Vie
tnam
E
mail:
dntoan@
ioi
t
.
a
c
.
vn
1.
I
NT
RODU
C
T
I
ON
C
a
nc
e
r
is
c
u
r
r
e
n
tl
y
o
ne
of
th
e
m
os
t
da
nge
r
ous
dis
e
a
s
e
s
f
o
r
pe
o
pl
e
a
ll
ove
r
th
e
wo
r
ld
.
I
n
20
20
,
th
e
W
o
r
l
d
H
e
a
lt
h
O
r
ga
n
iz
a
t
io
n
(
W
HO
)
[
1
]
ha
d
a
s
ta
ti
s
ti
c
s
r
e
p
o
r
t
p
r
e
s
e
n
ti
ng
ne
a
r
l
y
t
e
n
mi
l
li
on
de
a
t
hs
r
e
l
a
ted
t
o
c
a
nc
e
r
.
T
he
f
ig
u
r
e
m
e
a
n
t
n
e
a
r
ly
o
ne
-
s
ix
th
o
f
de
a
th
s
a
n
d
t
he
r
e
po
r
t
a
l
s
o
s
how
e
d
th
a
t
the
r
e
a
r
e
a
pp
r
ox
i
ma
te
ly
4
0
0
,
0
00
c
hi
ld
r
e
n
d
e
ve
l
op
in
g
c
a
nc
e
r
e
a
c
h
ye
a
r
.
Ac
c
o
r
di
ng
to
[
2]
in
20
19
a
nd
2
02
0
,
V
ie
tn
a
m
r
e
c
o
r
d
e
d
ne
a
r
l
y
18
3
t
h
ous
a
nd
ne
w
c
a
nc
e
r
pa
t
ie
nt
s
a
n
d
a
p
p
r
o
xi
ma
te
ly
123
t
ho
us
a
n
d
d
e
a
ths
.
M
a
ny
r
e
a
s
o
ns
m
a
y
c
a
us
e
c
a
nc
e
r
,
s
u
c
h
a
s
u
s
i
ng
a
lc
oh
ol
,
us
in
g
c
iga
r
e
t
tes
,
a
nd
un
he
a
l
th
y
l
i
f
e
r
o
u
ti
ne
s
.
F
r
o
m
t
he
me
dic
a
l
vi
e
w
,
c
a
n
c
e
r
s
t
a
r
ts
whe
n
t
he
r
e
is
a
n
a
pp
e
a
r
a
nc
e
o
f
c
a
nc
e
r
c
e
ll
s
,
a
n
d
t
he
n
the
y
w
ou
ld
g
r
o
w
to
c
r
e
a
te
ma
li
gna
n
t
t
um
o
r
s
.
T
h
e
c
a
nc
e
r
c
e
ll
s
w
ou
ld
i
n
va
d
e
a
nd
e
l
im
in
a
t
e
h
e
a
l
t
hy
o
r
ga
ns
in
th
e
h
um
a
n
bo
d
y
.
S
o
,
t
he
ma
in
i
de
a
in
ma
ny
c
a
nc
e
r
t
r
e
a
tm
e
n
ts
is
t
he
wa
y
t
o
de
s
tr
oy
c
a
nc
e
r
c
e
l
ls
a
nd
r
e
mo
ve
c
a
n
c
e
r
t
u
mo
r
s
.
An
d
one
o
f
t
he
mos
t
c
om
mo
n
m
e
t
ho
ds
is
r
a
d
io
th
e
r
a
py
.
R
a
d
io
the
r
a
p
y
t
r
ies
t
o
d
e
s
t
r
oy
c
a
nc
e
r
c
e
l
ls
b
y
us
in
g
t
he
po
we
r
o
f
r
a
dia
t
io
n
t
he
r
a
py
b
e
a
ms
i
n
t
r
e
a
t
me
nt
m
a
c
hi
ne
s
.
B
y
th
e
wa
y
,
w
e
c
a
n
p
r
e
v
e
n
t
t
he
g
r
o
wt
h
o
f
the
dis
e
a
s
e
.
A
n
a
d
va
nta
ge
o
f
r
a
di
a
t
io
n
t
he
r
a
p
y
is
t
ha
t
th
is
m
e
th
od
c
a
n
be
a
pp
l
ie
d
t
o
a
lm
os
t
a
l
l
pa
r
ts
o
f
the
hu
ma
n
b
od
y
,
bu
t
i
t
a
ls
o
is
a
c
om
pl
e
x
pr
oc
e
s
s
r
e
q
u
ir
i
n
g
h
ig
h
-
l
e
v
e
l
e
xpe
r
ts
[
3
]
.
T
his
m
e
t
ho
d
c
a
n
t
a
k
e
h
ou
r
s
t
o
c
a
lc
ul
a
t
e
th
e
r
a
di
ot
he
r
a
py
d
os
e
d
is
t
r
ib
u
ti
on
a
n
d
a
d
jus
t
t
o
ge
t
t
he
f
i
na
l
o
p
ti
ma
l
p
la
nn
in
g
.
T
o
do
th
a
t
,
p
r
a
c
t
it
io
ne
r
s
ne
e
d
to
be
s
ki
ll
f
ul
a
n
d
e
xpe
r
ie
nc
e
d
t
o
r
e
a
c
h
th
e
a
c
c
u
r
a
c
y
o
f
e
a
c
h
de
ta
il
o
f
t
he
wh
ole
p
r
o
c
e
s
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
fr
ame
w
or
k
for
3D
r
adiot
he
r
apy
dos
e
pr
e
diction
us
ing
the
de
e
p
lear
ning
appr
oac
h
(
L
am
T
hanh
Hi
e
n
)
5525
T
h
is
pa
p
e
r
c
o
nc
e
r
ns
d
os
e
p
r
e
d
ic
t
io
n
,
a
met
ho
d
to
h
e
l
p
mak
e
a
hi
gh
-
q
ua
li
t
y
p
la
n
q
uic
k
ly
.
T
he
r
e
w
e
r
e
s
ome
s
tu
di
e
s
t
ha
t
a
t
te
mp
te
d
t
o
a
p
pl
y
t
r
a
di
t
io
na
l
ma
c
h
in
e
-
le
a
r
n
in
g
mo
de
ls
t
o
t
his
pr
ob
le
m
.
I
n
2
01
6
,
M
c
I
nt
os
h
a
nd
P
ur
die
[
4]
a
pp
l
ied
th
e
r
e
g
r
e
s
s
io
n
f
o
r
e
s
ts
to
pr
e
di
c
t
d
os
e
.
R
e
s
u
lt
s
,
the
ove
r
a
l
l
dos
e
p
r
e
di
c
t
io
n
a
c
c
u
r
a
c
ies
r
e
a
c
h
e
d
78
.
68
%
,
64
.
76
%
,
a
n
d
86
.
83
%
r
e
s
pe
c
t
iv
e
ly
.
Ne
x
t
i
n
20
21
,
Z
ho
u
e
t
al
.
[
5
]
pr
e
s
e
nte
d
a
s
t
udy
us
in
g
a
s
up
po
r
t
ve
c
t
o
r
m
a
c
h
in
e
t
o
p
r
e
d
ic
t
d
os
e
s
in
th
e
c
a
s
e
o
f
c
e
r
v
ic
a
l
c
a
n
c
e
r
w
it
h
the
me
a
n
s
qu
a
r
e
d
e
r
r
o
r
a
n
d
G
a
m
ma
m
e
tr
ics
.
T
he
e
xp
e
r
i
me
nta
l
da
t
a
c
a
me
f
r
o
m
50
p
a
t
ie
nt
s
i
n
t
he
a
f
f
il
ia
te
d
H
os
p
it
a
l
o
f
S
ou
th
we
s
t
M
e
d
ic
a
l
Un
i
ve
r
s
i
ty
o
ve
r
th
e
2
-
y
e
a
r
s
p
e
r
i
od
be
twe
e
n
20
16
a
nd
20
18
.
T
oda
y,
the
de
e
p
lea
r
ning
a
ppr
oa
c
h
wa
s
a
ppli
e
d
s
uc
c
e
s
s
f
ull
y
in
many
s
tudi
e
s
of
the
medic
a
l
f
ield
with
va
r
ious
di
f
f
e
r
e
nt
da
ta
types
.
I
n
2021
,
F
ur
tad
o
[
6]
c
onduc
ted
a
s
tudy
to
s
e
gment
a
bdomi
na
l
o
r
ga
ns
in
magne
ti
c
r
e
s
ona
nc
e
im
a
ging
da
ta
us
ing
two
de
e
p
-
lea
r
ning
a
r
c
hit
e
c
tur
e
s
De
e
pL
a
bV3
a
nd
f
ull
y
c
onv
olut
ional
ne
twor
k
(
F
C
N)
.
Als
o
in
2021
,
C
he
n
e
t
al
.
[
7]
de
s
igned
S
que
e
z
e
-
a
nd
-
E
xc
it
a
ti
on
c
onvolut
ional
ne
ur
a
l
ne
twor
k
(
C
NN
)
to
c
las
s
if
y
lung
nodules
f
r
om
c
omput
e
d
t
omogr
a
phy
im
a
ge
s
.
I
n
2023
,
T
oa
n
e
t
al
.
[
8]
ha
d
r
e
s
e
a
r
c
h
us
ing
s
e
ve
r
a
l
c
onvolut
ion
ne
ur
a
l
ne
twor
ks
on
c
he
s
t
X
-
r
a
y
im
a
ge
s
f
or
tuber
c
ulos
is
diagnos
is
.
F
ur
ther
mor
e
,
r
e
c
e
nt
de
e
p
lea
r
ning
r
e
s
e
a
r
c
h
in
he
a
lt
hc
a
r
e
is
a
ls
o
or
iente
d
towa
r
ds
buil
ding
a
ppli
c
a
ble
s
ys
tems
.
F
or
e
xa
mpl
e
,
P
r
a
ba
ka
r
a
n
a
nd
S
e
lvar
a
j
[
9
]
pr
e
s
e
nted
a
n
int
e
ll
igent
he
a
lt
hc
a
r
e
s
ys
tem
us
ing
a
n
e
ns
e
mbl
e
a
pp
r
oa
c
h
to
diagnos
e
lung
dis
e
a
s
e
.
T
he
s
a
me
ye
a
r
,
Alqu
r
a
n
e
t
al.
[
10]
s
howe
d
a
hybr
id
a
utom
a
ted
s
ys
tem
to
c
las
s
if
y
li
ve
r
tum
or
s
on
magne
ti
c
r
e
s
ona
nc
e
im
a
ge
s
.
I
n
r
e
c
e
nt
ti
mes
,
th
e
d
e
e
p
-
lea
r
ni
ng
a
p
p
r
oa
c
h
ha
s
b
e
e
n
a
p
p
li
e
d
w
id
e
l
y
i
n
the
r
a
d
io
the
r
a
p
y
d
os
e
p
r
e
d
ic
ti
on
p
r
o
bl
e
m
.
I
n
20
21
,
a
3D
De
ns
e
-
U
-
N
e
t
m
o
de
l
wa
s
p
r
e
s
e
nt
e
d
by
L
iu
e
t
a
l
.
[
1
1
]
i
n
t
he
r
e
p
o
r
t
wi
t
h
r
e
s
pe
c
t
ive
ly
2
.
93
%
a
nd
2
.
4
2
%
o
f
th
e
a
ve
r
a
ge
de
v
iat
i
ons
f
r
om
th
e
m
a
x
im
u
m
a
n
d
a
ve
r
a
ge
dos
e
v
a
l
ue
s
of
o
r
g
a
ns
a
t
r
is
ks
(
OA
R
s
)
a
n
d
pl
a
n
ni
ng
t
a
r
ge
t
vo
lu
mes
(
P
T
Vs
)
.
T
he
a
ut
ho
r
s
e
x
pe
r
i
me
nt
e
d
o
n
the
d
a
ta
o
f
1
24
p
a
ti
e
n
ts
t
r
e
a
ted
w
it
h
T
o
mo
T
he
r
a
py
.
An
ot
he
r
s
t
ud
y
is
a
c
on
vo
lu
t
io
na
l
ne
u
r
a
l
ne
tw
o
r
k
o
f
A
hn
e
t
al
.
[
1
2]
.
T
h
e
i
r
a
r
c
h
it
e
c
t
u
r
e
w
a
s
ba
s
e
d
on
th
e
U
-
N
e
t
mo
de
l
a
n
d
a
d
j
us
t
e
d
to
f
i
t
the
dos
e
p
r
e
di
c
t
io
n
p
r
ob
le
m
i
n
r
a
d
iot
he
r
a
p
y
.
5
0
s
a
m
p
les
o
f
th
e
r
a
p
y
pl
a
ns
f
o
r
b
r
e
a
s
t
c
a
n
c
e
r
we
r
e
u
s
e
d
t
o
tes
t
th
a
t
m
e
t
ho
d
a
n
d
the
a
ut
ho
r
s
a
ls
o
c
o
nd
uc
te
d
the
c
om
pa
r
is
o
n
w
it
h
R
a
pi
d
P
l
a
n
™
.
T
he
wo
r
k
o
f
B
a
bi
e
r
e
t
a
l
.
[
1
3
]
w
it
h
t
he
g
e
ne
r
a
t
iv
e
a
d
ve
r
s
a
r
ia
l
n
e
t
wo
r
k
(
G
AN
)
m
o
de
l
wa
s
c
o
nd
uc
ted
on
a
n
no
ta
te
d
c
om
pu
te
d
tom
o
gr
a
p
hy
(
CT
)
im
a
g
e
s
f
r
o
m
1
27
s
a
mp
les
i
n
t
he
a
c
c
e
p
te
d
t
r
e
a
tm
e
n
t
pl
a
ns
f
o
r
o
r
o
pha
r
yn
ge
a
l
c
a
nc
e
r
c
a
s
e
s
.
I
n
a
no
th
e
r
a
t
tem
p
t
,
man
y
ve
r
s
io
ns
o
f
c
on
vo
lu
t
io
na
l
ne
u
r
a
l
n
e
t
wo
r
ks
we
r
e
t
e
s
t
e
d
i
n
th
e
wo
r
k
of
G
r
on
be
r
g
e
t
a
l
.
[
14
]
.
T
h
e
y
t
r
a
i
ne
d
a
nd
t
e
s
t
e
d
t
he
i
r
mo
de
l
on
t
he
d
a
t
a
o
f
t
h
e
O
pe
nKB
P
c
ha
l
len
ge
[
15
]
wi
t
h
s
o
me
d
if
f
e
r
e
n
t
op
ti
o
ns
o
f
h
yp
e
r
pa
r
a
me
te
r
s
a
n
d
a
ug
me
nt
a
t
io
n
me
th
ods
.
I
n
the
s
a
me
da
tas
e
t
,
S
o
o
mr
o
e
t
al
.
[
16
]
pu
bl
is
h
e
d
a
n
a
r
c
hi
t
e
c
t
u
r
e
c
o
mb
in
in
g
d
i
la
ted
De
ns
e
N
e
t
a
n
d
R
e
s
N
e
t
to
p
r
e
di
c
t
d
os
e
c
a
ll
e
d
D
e
e
p
Dos
e
Ne
t
.
I
n
e
x
pe
r
i
me
nts
,
s
om
e
lo
s
s
f
un
c
t
io
ns
w
e
r
e
us
e
d
a
nd
th
e
q
ua
li
ty
o
f
t
ha
t
mo
de
l
w
a
s
s
c
o
r
e
d
i
n
te
r
ms
o
f
d
os
e
-
s
c
o
r
e
a
n
d
d
os
e
-
v
o
lu
me
his
t
o
gr
a
m
(
DV
H
)
s
c
o
r
e
.
Z
i
mm
e
r
m
a
n
n
e
t
a
l
.
[
1
7
]
a
ppl
i
e
d
one
c
yc
le
lea
r
ni
ng
t
o
the
i
r
c
us
to
mi
z
e
d
U
-
N
e
t
ve
r
s
io
n
a
n
d
a
ls
o
e
x
pe
r
i
me
n
ted
o
n
t
he
da
tas
e
t
o
f
t
he
Op
e
n
KB
P
c
ha
ll
e
n
ge
.
I
n
t
ha
t
s
t
ud
y
,
s
om
e
c
h
a
n
ge
s
we
r
e
t
r
i
e
d
w
i
th
l
os
s
f
u
nc
ti
ons
a
nd
M
is
h
a
c
t
iva
t
io
n
f
un
c
t
io
n
,
a
n
d
a
t
th
a
t
t
i
me
,
th
e
i
r
r
e
s
ul
ts
r
a
nke
d
4
th
a
nd
2
nd
r
e
s
pe
c
t
i
ve
l
y
o
n
d
os
e
s
c
o
r
e
a
n
d
DV
H
s
c
o
r
e
i
n
the
Op
e
nK
B
P
c
ha
l
leng
e
.
T
h
is
pa
pe
r
is
in
te
r
e
s
te
d
in
th
e
r
a
di
a
t
io
n
th
e
r
a
p
y
d
os
e
p
r
e
di
c
t
io
n
p
r
o
ble
m
.
I
n
d
e
t
a
i
l
,
we
us
e
C
T
i
ma
ge
s
o
f
p
a
t
ie
nts
tr
e
a
ted
w
i
th
r
a
d
i
ot
he
r
a
py
i
n
ou
r
f
r
a
m
e
w
o
r
k
a
nd
a
pp
ly
t
he
d
e
e
p
lea
r
n
i
ng
a
p
p
r
o
a
c
h
.
T
h
e
r
e
s
ul
ts
o
f
d
e
e
p
-
le
a
r
n
in
g
me
th
ods
a
r
e
h
ig
hl
y
r
e
la
te
d
to
t
he
t
r
a
i
n
in
g
da
tas
e
t
,
in
b
o
th
q
ua
n
t
it
y
a
nd
q
ua
li
t
y
.
P
a
r
ti
c
ul
a
r
l
y
f
or
m
e
d
ic
a
l
im
a
g
e
s
,
t
he
r
e
a
r
e
m
a
n
y
c
a
s
e
s
wh
e
r
e
i
t
is
d
if
f
i
c
u
lt
to
d
is
t
in
gu
is
h
be
tw
e
e
n
r
e
g
io
ns
o
f
in
te
r
n
a
l
o
r
g
a
ns
due
t
o
po
o
r
c
on
t
r
a
s
t
a
nd
di
f
f
e
r
e
n
t
c
on
f
ig
ur
a
t
io
ns
of
im
a
g
i
ng
mac
h
ines
.
M
o
r
e
ov
e
r
,
m
e
d
ica
l
im
a
ge
s
a
r
e
g
e
n
e
r
a
l
ly
m
o
r
e
s
c
a
r
c
e
th
a
n
r
e
gu
la
r
i
ma
ge
s
d
ue
to
p
r
iv
a
c
y
a
nd
s
e
c
u
r
it
y
c
o
ns
tr
a
i
nts
.
S
o
,
on
th
e
on
e
ha
nd
,
w
e
ne
e
d
n
o
r
m
a
l
iz
a
t
io
n
t
e
c
h
ni
qu
e
s
t
o
r
e
d
uc
e
no
is
e
t
o
ge
t
n
o
r
m
a
li
z
e
d
da
t
a
.
On
th
e
o
the
r
h
a
n
d
,
we
a
ls
o
ha
ve
t
o
de
a
l
w
i
th
t
he
s
c
a
r
c
i
ty
of
d
a
t
a
by
a
pp
ly
in
g
da
ta
a
u
gm
e
n
ta
ti
on
.
I
n
th
is
wo
r
k
,
o
u
r
s
t
ud
y
a
i
ms
to
b
ui
l
d
a
d
os
e
pr
e
di
c
t
io
n
s
ys
t
e
m
in
t
he
c
on
te
xt
o
f
r
a
d
i
ot
he
r
a
py
t
r
e
a
t
men
t
by
p
r
o
pos
i
ng
a
f
r
a
m
e
wo
r
k
c
on
ne
c
ti
ng
a
l
l
tas
ks
.
F
ir
s
tl
y
,
we
d
e
s
i
gn
e
d
a
c
u
s
t
om
ize
d
c
on
vo
lu
t
io
n
n
e
u
r
a
l
n
e
t
wo
r
k
f
o
r
d
os
e
p
r
e
d
ic
t
io
n
a
n
d
a
pp
li
e
d
s
om
e
p
r
e
p
r
o
c
e
s
s
in
g
a
nd
a
ug
me
nt
a
t
io
n
t
e
c
hn
iq
ue
s
to
e
nh
a
nc
e
t
his
mo
de
l
.
S
e
c
o
nd
ly
,
we
a
na
lyz
e
d
th
e
p
r
e
di
c
te
d
r
e
s
ul
ts
i
n
m
a
ny
f
a
c
e
s
,
u
s
i
ng
s
t
a
n
da
r
d
m
e
t
r
ics
s
u
c
h
a
s
d
os
e
-
s
c
or
e
a
n
d
DV
H
-
s
c
o
r
e
,
v
is
ua
l
iz
in
g
the
pr
e
d
ic
te
d
r
e
s
ul
ts
a
n
d
c
o
mp
a
r
in
g
t
he
m
t
o
t
he
g
r
ou
nd
t
r
u
th
d
a
t
a
,
p
r
o
vi
di
ng
r
u
nn
i
ng
t
i
me
i
n
f
o
r
ma
ti
on
.
L
a
s
t
ly
,
w
e
bu
i
lt
a
c
om
p
let
e
s
ys
te
m
to
s
up
p
or
t
3D
r
a
di
ot
he
r
a
py
dos
e
p
r
e
di
c
t
io
n
a
nd
vis
ua
li
z
e
th
e
r
e
s
u
lt
i
n
di
f
f
e
r
e
nt
d
im
e
ns
io
ns
.
2.
M
E
T
HO
D
T
h
is
s
e
c
t
i
on
de
s
c
r
ib
e
s
h
ow
we
c
on
du
c
t
e
d
ou
r
r
e
s
e
a
r
c
h
a
n
d
t
he
e
x
pe
r
i
me
nt
a
l
p
r
o
c
e
d
u
r
e
.
I
n
s
i
de
,
w
e
w
ou
l
d
p
r
e
s
e
n
t
t
he
p
r
o
pos
e
d
f
r
a
me
wo
r
k
a
nd
t
he
n
d
e
t
a
il
t
he
c
o
mp
on
e
n
ts
in
t
his
.
T
h
e
y
i
nc
lu
de
da
ta
p
r
e
pr
o
c
e
s
s
in
g
a
nd
a
u
gm
e
n
ta
ti
on
tec
hn
i
que
s
,
th
e
c
us
t
om
c
o
nv
ol
ut
io
na
l
n
e
u
r
a
l
n
e
t
wo
r
k
,
a
n
d
the
dos
e
v
is
u
a
l
iz
a
t
io
n
t
e
c
hn
i
que
.
2.
1.
P
r
op
os
e
d
f
r
am
e
wor
k
T
he
f
r
a
mew
or
k
include
s
the
tr
a
ini
ng
a
nd
the
tes
ti
ng
pha
s
e
s
a
s
in
F
igur
e
1.
T
he
tr
a
ini
ng
pha
s
e
a
im
s
to
im
pr
ove
the
qua
li
ty
o
f
the
model
ba
s
e
d
on
the
tr
a
ini
ng
da
ta.
I
n
thi
s
wo
r
k,
the
tr
a
ini
ng
da
ta
incl
ude
s
C
T
im
a
ge
s
,
a
nnotate
d
inf
or
mation
a
bout
P
T
Vs
a
nd
O
AR
s
.
OA
R
s
a
nd
P
T
Vs
a
r
e
mas
ks
to
point
out
the
a
nnotate
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
1
4
,
No.
5
,
Oc
tober
2
02
4
:
5524
-
5533
5526
r
e
gions
.
S
o,
the
input
da
ta
ha
s
11
c
ha
nne
ls
with
C
T
im
a
ge
s
,
OA
R
s
,
a
nd
P
T
Vs
.
T
he
inpu
t
da
ta
would
be
pr
e
pr
oc
e
s
s
e
d
a
nd
a
ugmente
d
b
e
f
or
e
pus
hing
to
the
model
in
the
tr
a
ini
ng
tas
k.
Dur
ing
the
tr
a
ini
ng
p
ha
s
e
,
the
s
tate
of
the
model's
we
ight
s
would
be
e
va
luate
d
a
t
e
a
c
h
e
poc
h
with
s
ome
metr
ics
.
He
r
e
,
the
dos
e
s
c
or
e
a
nd
the
DV
H
s
c
or
e
a
r
e
two
indi
c
a
tor
s
us
e
d
f
or
s
c
or
ing
the
model's
c
he
c
kpoint
s
to
c
hoos
e
the
be
s
t
s
tate
.
T
he
tes
ti
ng
pha
s
e
a
im
s
to
the
pr
e
diction
f
unc
ti
on
of
the
s
tudy's
pr
oblem
whic
h
c
a
n
be
a
ppli
e
d
in
the
medic
a
l
s
of
twa
r
e
of
r
a
diot
he
r
a
py.
He
r
e
,
the
input
da
ta
a
ls
o
include
s
C
T
im
a
ge
s
,
OA
R
s
,
a
nd
P
T
Vs
.
T
he
input
da
ta
would
be
p
r
e
pr
oc
e
s
s
e
d
a
nd
then
pus
he
d
int
o
the
t
r
a
ined
model
to
ge
t
the
pr
e
dicte
d
dos
e
.
T
his
dos
e
would
be
a
ppli
e
d
to
the
c
ons
tr
a
int
a
bout
the
va
l
ue
r
a
nge
to
ge
t
the
f
inal
r
e
s
ult
.
T
his
dos
e
a
ls
o
would
be
dis
playe
d
a
s
a
he
a
tm
a
p
to
he
lp
pe
ople
e
a
s
il
y
unde
r
s
tand.
F
igur
e
1.
T
he
pr
opos
e
d
f
r
a
mew
or
k
2.
2.
Dat
a
p
r
oc
e
s
s
in
g
t
e
c
h
n
iq
u
e
s
T
he
da
ta
us
e
d
in
thi
s
wor
k
we
r
e
C
T
im
a
ge
s
of
p
a
ti
e
nts
who
ha
d
he
a
d
-
a
nd
-
ne
c
k
c
a
nc
e
r
a
nd
we
r
e
tr
e
a
ted
by
r
a
diot
he
r
a
py
t
r
e
a
tm
e
nt.
B
e
f
o
r
e
the
de
e
p
lea
r
ning
tas
k,
we
tr
ied
to
a
pply
s
ome
da
ta
pr
oc
e
s
s
ing
tec
hniques
to
C
T
im
a
ge
s
.
T
his
s
tep
would
he
lp
nor
malize
im
a
ge
da
ta
to
im
pr
ove
the
f
inal
r
e
s
ult
s
.
I
n
c
ompar
is
on
with
nor
mal
im
a
ge
s
,
be
c
a
us
e
of
the
dif
f
e
r
e
nc
e
be
twe
e
n
c
a
ptur
ing
mac
hines
,
C
T
im
a
ge
da
ta
ha
s
s
ome
pa
r
ti
c
ular
c
ha
r
a
c
ter
is
ti
c
s
.
P
ixel
va
lues
a
r
e
r
e
late
d
to
the
mate
r
ial
o
f
c
a
ptu
r
e
d
objec
ts
,
s
uc
h
a
s
b
one
,
a
nd
high
-
f
a
t
ti
s
s
ue
s
.
T
his
inf
or
mation
is
r
e
f
e
r
e
nc
e
d
in
Houns
f
ield
unit
s
[
18]
.
F
or
a
ugmenta
ti
on
,
in
the
c
a
s
e
of
thi
s
s
tudy,
the
d
a
ta
we
r
e
3D
C
T
s
c
a
ns
.
I
n
C
T
im
a
ge
s
,
the
e
mpt
y
s
pa
c
e
us
ua
ll
y
ge
ts
va
lues
c
los
e
to
z
e
r
o.
B
e
c
a
us
e
we
a
r
e
only
int
e
r
e
s
ted
in
the
body
in
the
C
T
i
mage
,
we
would
a
pply
tr
a
ns
f
or
mations
wi
th
t
r
a
ns
lation
a
nd
f
li
pping
a
nd
no
c
ha
nge
in
the
s
c
a
le
o
f
the
C
T
im
a
g
e
.
I
n
thi
s
wa
y,
the
3D
c
ontent
of
the
human
body
is
s
ti
ll
p
r
e
s
e
r
ve
d.
Anothe
r
r
e
a
s
on
is
that
the
im
a
ge
-
c
a
ptur
ing
pos
e
is
f
ixed
due
to
medic
a
l
indi
c
a
ti
ons
.
T
his
mea
ns
that
o
bjec
ts
in
C
T
im
a
ge
s
of
ten
ha
ve
s
table
s
ize
.
T
he
f
ir
s
t
tec
hnique
is
f
li
pping:
As
de
s
c
r
ibed
in
F
i
gur
e
2,
thi
s
tec
hnique
will
r
otate
im
a
ge
s
r
a
ndoml
y
by
a
n
a
ngle
e
qua
li
ng
180
de
gr
e
e
s
.
S
o
,
the
im
a
ge
r
e
s
ult
will
ha
ve
a
di
f
f
e
r
e
nt
pos
e
ve
r
s
us
or
igi
n.
I
n
t
his
c
a
s
e
,
we
c
a
n
quickly
a
pply
thi
s
tec
hnique
a
nd
do
not
ne
e
d
to
c
a
lcula
te
im
a
ge
c
ontent
a
t
ga
ps
in
the
c
or
ne
r
r
e
gions
of
3D
volu
me.
T
he
s
e
c
ond
is
t
r
a
ns
lation:
As
de
s
c
r
ibed
in
F
igu
r
e
3
,
thi
s
tec
hnique
will
tr
a
ns
late
the
C
T
im
a
ge
with
a
r
a
ndom
dis
plac
e
ment
a
nd
not
los
e
a
ny
body
pa
r
ts
.
T
his
tec
hnique
is
a
ls
o
s
im
ple
a
nd
f
a
s
t
to
pe
r
f
or
m.
F
igur
e
2.
I
mage
f
li
pping
e
xa
mpl
e
:
the
o
r
igi
na
l
i
mage
(
lef
t)
a
nd
the
f
li
ppe
d
im
a
ge
(
r
ight
)
F
igur
e
3.
I
mage
tr
a
ns
lation
e
xa
mpl
e
:
the
or
igi
na
l
im
a
ge
(
lef
t)
a
nd
the
tr
a
ns
late
d
im
a
ge
(
r
ight
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
fr
ame
w
or
k
for
3D
r
adiot
he
r
apy
dos
e
pr
e
diction
us
ing
the
de
e
p
lear
ning
appr
oac
h
(
L
am
T
hanh
Hi
e
n
)
5527
2.
3.
T
h
e
c
u
s
t
om
c
on
volu
t
ion
al
n
e
u
r
al
n
e
t
wor
k
I
n
th
is
wor
k
,
we
de
s
igned
a
n
a
r
c
hit
e
c
tur
e
that
is
i
ns
pir
e
d
by
the
s
tudy
o
f
R
onne
be
r
ge
r
e
t
al
.
[
19
]
in
2015.
W
e
a
ls
o
us
e
d
a
r
e
s
idual
block
pr
e
s
e
nted
by
He
e
t
al
.
[
20
]
in
thei
r
a
r
c
hit
e
c
tu
r
e
f
o
r
th
e
im
a
ge
c
las
s
if
ica
ti
on
pr
oblem.
F
ig
u
r
e
4
s
hows
the
de
s
ign
of
a
r
e
s
idual
block.
I
n
thi
s
wor
k,
the
s
tr
uc
tu
r
e
ha
s
two
pa
r
ts
a
s
de
s
c
r
ib
e
d
in
F
igur
e
5.
T
he
f
i
r
s
t
is
the
down
-
s
a
mpl
ing
pa
r
t
a
nd
the
s
e
c
ond
is
the
up
-
s
a
mpl
ing
pa
r
t
.
I
n
thi
s
a
r
c
hit
e
c
tur
e
,
r
e
s
idual
blocks
would
be
us
e
d
in
both
pa
r
ts
.
I
n
F
igur
e
5,
the
input
da
ta
is
a
t
the
be
ginni
ng
of
the
down
-
s
a
mpl
ing
pa
r
t
a
s
de
s
c
r
ibed
on
the
lef
t
-
top
s
ide,
a
nd
the
output
da
ta
is
a
t
the
e
nd
of
the
up
-
s
a
mpl
ing
pa
r
t
a
s
de
s
c
r
ibed
on
the
r
ight
-
top
s
ide.
T
he
s
ize
of
the
input
da
ta
is
128×
128×
128×
11.
T
he
f
ir
s
t
thr
e
e
number
s
a
r
e
the
s
ize
of
the
C
T
input
im
a
ge
a
nd
the
las
t
number
is
the
a
mount
of
c
ha
nne
ls
.
Among
the
11
c
ha
nne
ls
,
the
r
e
is
one
c
ha
nne
l
f
o
r
the
C
T
input
im
a
ge
,
s
e
ve
n
c
ha
nne
ls
f
or
OA
R
s
,
a
nd
th
r
e
e
c
ha
nne
ls
f
or
P
T
Vs
.
R
e
s
idual
blocks
a
r
e
de
s
igned
of
two
c
onvolut
ion
laye
r
s
.
E
a
c
h
laye
r
ha
s
the
s
a
me
3×
3×
3
s
li
ding
window
s
ize
,
1
s
tr
ide,
a
nd
a
ls
o
is
pa
dde
d
to
ke
e
p
s
ize
.
L
e
a
kyR
e
L
U
f
unc
ti
ons
a
r
e
s
e
t
f
oll
owing
c
onvolut
ion
laye
r
s
.
B
e
s
ides
,
in
the
up
-
s
a
mpl
ing
pa
r
t,
a
de
c
onvolut
ion
laye
r
is
us
e
d
a
f
ter
e
a
c
h
r
e
s
idual
b
lock.
T
he
e
nd
o
f
the
a
r
c
hit
e
c
tur
e
is
a
dde
d
with
a
R
e
L
U
f
unc
ti
on.
T
his
f
unc
ti
on
he
lps
ke
e
p
e
a
c
h
pixel
f
r
om
ne
ga
ti
ve
va
lues
be
c
a
us
e
of
the
mea
ning
of
r
a
diati
on
dos
e
s
.
At
las
t,
the
ou
tput
da
ta
s
ize
is
128×
128×
128.
F
igur
e
4
.
R
e
s
idual
block
F
igur
e
5
.
Our
c
us
tom
C
NN
a
r
c
hit
e
c
tur
e
2.
4.
Dos
e
vis
u
ali
z
at
ion
I
n
thi
s
wo
r
k,
the
output
dos
e
is
in
3D
r
a
s
ter
f
o
r
m
a
t.
T
hus
,
a
ll
r
e
s
ult
s
would
be
dis
playe
d
a
s
im
a
ge
s
of
e
a
c
h
s
li
c
e
.
B
e
c
a
us
e
the
C
T
im
a
ge
ha
s
one
c
ha
n
ne
l,
the
dis
playe
d
im
a
ge
is
gr
a
ys
c
a
le.
B
ut
human
vis
ion
is
not
s
e
ns
it
ive
to
the
c
ha
nge
s
in
gr
a
y
leve
ls
,
we
o
f
ten
dis
ti
nguis
h
we
ll
in
c
olor
s
.
S
o,
we
us
e
the
c
olor
map
tec
hnique
to
e
nha
nc
e
the
view
o
f
the
r
e
s
ult
s
.
I
n
im
pleme
ntation,
the
256
-
c
olor
look
-
up
table
is
a
ppli
e
d
to
dis
playe
d
im
a
ge
s
.
E
a
c
h
dos
e
da
ta
s
li
c
e
ne
e
ds
to
be
tr
a
ns
f
or
med
int
o
the
g
r
a
ys
c
a
le
to
be
a
ble
to
a
pply
the
c
olor
map
tec
hnique.
I
n
the
e
xpe
r
i
ment,
we
c
ode
d
thi
s
modul
e
with
the
s
uppor
t
of
M
a
tpl
otl
ib
[
21]
.
M
a
tpl
otl
ib
is
the
P
y
thon
l
ibr
a
r
y
that
s
uppor
ts
s
ome
opti
ons
of
c
olor
maps
.
An
e
xa
mpl
e
o
f
dos
e
vis
ua
li
z
a
ti
on
wa
s
de
s
c
r
ibed
in
F
ig
ur
e
6.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
1
4
,
No.
5
,
Oc
tober
2
02
4
:
5524
-
5533
5528
F
igur
e
6
.
Dos
e
vis
ua
li
z
a
ti
on
of
a
n
e
xa
mpl
e
in
the
d
a
tas
e
t:
the
C
T
da
ta
(
le
f
t)
,
the
vis
ua
li
z
a
ti
on
r
e
s
ult
(
r
i
ght)
3.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
3.
1.
E
xp
e
r
im
e
n
t
s
e
t
u
p
3.
1.
1.
Dat
as
e
t
T
he
da
tas
e
t
us
e
d
in
our
wor
k
wa
s
c
oll
e
c
ted
f
r
om
340
pa
ti
e
nts
who
ha
d
he
a
d
-
a
nd
-
ne
c
k
c
a
nc
e
r
a
nd
we
r
e
tr
e
a
ted
by
r
a
diot
he
r
a
py
t
r
e
a
tm
e
nt.
T
his
da
ta
ha
s
a
s
our
c
e
f
r
om
the
ope
n
-
a
c
c
e
s
s
da
tab
a
s
e
T
C
I
A.
T
C
I
A
c
ontains
medic
a
l
im
a
ge
s
s
uppor
ti
ng
c
a
nc
e
r
r
e
s
e
a
r
c
h
a
nd
be
longs
to
the
Unive
r
s
it
y
of
Ar
ka
ns
a
s
.
T
h
e
a
uthor
s
pus
he
d
a
ll
da
ta
int
o
the
da
ta
-
c
lea
ning
s
tep.
E
a
c
h
pa
ti
e
nt's
dos
e
inf
luenc
e
matr
ix
wa
s
obtaine
d
us
in
g
6
M
V
s
tep
-
a
nd
-
s
hoot
int
e
ns
it
y
modul
a
ted
r
a
diation
the
r
a
py
(
I
M
R
T
)
.
T
his
tas
k
wa
s
c
onduc
ted
with
e
q
uis
pa
c
e
d
c
oplana
r
f
ields
a
t
0°,
40°,
80°
,
120°
,
160°,
200°
,
2
40°,
280°,
a
nd
320°
.
All
f
ields
we
r
e
divi
de
d
int
o
5
×
5
mm
be
a
ml
e
ts
.
T
he
Ope
nKB
P
c
ha
ll
e
nge
o
r
ga
nize
d
thi
s
da
ta
f
o
r
the
c
ompetit
ion
.
S
o,
the
da
tas
e
t
wa
s
o
r
ga
nize
d
with
the
tr
a
ini
ng
s
e
t,
the
tes
ti
ng
s
e
t
a
nd
the
va
li
d
a
ti
on
s
e
t
with
s
ize
s
of
200
,
40,
a
nd
100
r
e
s
pe
c
ti
ve
ly.
E
a
c
h
pa
ti
e
nt's
r
e
c
or
d
include
s
a
C
T
im
a
ge
with
a
s
ize
o
f
128×
128×
128,
a
nnotations
f
o
r
o
r
ga
ns
a
t
r
is
k,
a
n
notations
f
or
planning
tar
ge
t
volum
e
s
,
a
nd
inf
o
r
mation
a
b
out
the
tr
e
a
ted
r
a
diot
he
r
a
py
dos
e
dis
tr
ibut
ion.
I
n
de
tail,
or
ga
ns
a
t
r
is
k
c
ons
is
t
s
pinal
c
or
d,
br
a
in
s
tem,
lef
t
pa
r
oti
d
gland
,
r
ight
pa
r
ot
id
gland,
lar
ynx
,
mandi
ble,
a
nd
e
s
opha
gus
.
P
lanning
tar
ge
t
volum
e
s
c
ons
is
t
of
5
6
Gy,
63
Gy,
a
nd
70
Gy
r
e
gions
.
F
ig
ur
e
7
s
ho
ws
a
da
ta
s
a
mpl
e
in
thi
s
da
tas
e
t.
F
i
gu
r
e
7
.
D
a
t
a
e
x
a
m
pl
e
:
C
T
i
mag
e
(
l
e
f
t
)
,
t
a
r
ge
t
r
e
gio
ns
vis
ua
li
z
a
t
i
on
(
mi
dd
le
)
,
a
nd
dos
e
v
is
u
a
l
iza
t
io
n
(
r
i
g
ht
)
3.
1.
2.
E
valu
at
io
n
m
e
t
r
ics
T
o
e
va
luate
r
e
s
ult
s
,
we
ne
e
d
metr
ics
.
T
his
e
xpe
r
i
ment
us
e
d
the
DV
H
-
s
c
or
e
a
nd
the
dos
e
-
s
c
or
e
to
a
na
lyze
the
dos
e
-
pr
e
dicte
d
r
e
s
ult
s
.
All
metr
ics
a
r
e
c
a
lcula
ted
be
twe
e
n
the
tr
ue
dos
e
a
nd
the
pr
e
dicte
d
dos
e
.
T
he
dos
e
-
s
c
or
e
is
c
a
lcula
ted
u
s
ing
the
volum
e
of
int
e
r
e
s
t
r
e
c
e
ivi
ng
r
a
diation
ther
a
py
with
the
mea
n
of
the
a
bs
olut
e
dif
f
e
r
e
nc
e
be
twe
e
n
the
tr
ue
va
l
ue
a
nd
the
pr
e
dicte
d
va
lue.
T
he
dos
e
-
s
c
or
e
ne
e
ds
to
be
c
a
lcula
ted
in
e
a
c
h
voxe
l.
T
he
f
o
r
mul
a
o
f
the
dos
e
-
s
c
or
e
is
de
s
c
r
ibed
a
s
(
1)
:
α
p
=
‖
s
P
−
s
̂
P
‖
1
|
v
P
|
(
1)
T
he
s
c
or
e
α
p
is
c
a
lcula
ted
f
or
pa
ti
e
nt
p.
I
n
de
tail,
pa
ti
e
nt
p
ha
s
the
volum
e
tr
e
a
ted
r
a
diot
he
r
a
py
v
P
with
the
s
ize
of
|
v
P
|
.
About
dos
e
s
,
s
P
is
the
pr
e
dicte
d
dos
e
,
a
nd
la
s
tl
y,
s
̂
P
is
the
tr
ue
dos
e
.
T
he
DV
H
s
c
or
e
is
c
a
lcula
ted
a
s
the
a
bs
olut
e
dis
tanc
e
be
twe
e
n
the
DV
H
va
lues
of
the
tr
ue
dos
e
a
nd
the
pr
e
dicte
d
dos
e
.
F
or
OA
R
,
the
tas
k
is
to
e
s
ti
mate
the
dos
e
a
ve
r
a
ge
a
nd
the
dos
e
maxi
mi
z
a
ti
on
c
or
r
e
s
ponding
to
or
ga
ns
.
S
im
il
a
r
ly
,
f
or
P
T
V,
the
tas
k
is
to
e
s
ti
mate
the
dos
e
c
or
r
e
s
ponding
to
th
e
tar
ge
t
volum
e
s
a
t
1
pe
r
c
e
nt,
95
pe
r
c
e
nt,
a
nd
99
pe
r
c
e
nt
r
a
tes
.
T
he
DV
H
-
s
c
or
e
is
c
a
lcula
ted
a
s
(
2)
:
β
P
=
‖
D
(
s
P
)
−
D
(
s
̂
p
)
‖
1
(
2)
T
he
s
c
or
e
β
P
is
c
a
lcula
ted
f
or
pa
ti
e
nt
p.
About
the
pr
e
dicte
d
r
e
s
ult
,
s
P
is
the
pr
e
dicte
d
dos
e
with
D
(
s
P
)
is
the
DV
H
va
lue
of
s
P
.
About
the
g
r
ound
tr
u
th,
s
̂
p
is
the
tr
ue
dos
e
a
nd
D
(
s
̂
p
)
is
the
DV
H
va
lue
of
s
̂
p
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
fr
ame
w
or
k
for
3D
r
adiot
he
r
apy
dos
e
pr
e
diction
us
ing
the
de
e
p
lear
ning
appr
oac
h
(
L
am
T
hanh
Hi
e
n
)
5529
3.
1.
3.
T
r
ain
in
g
c
on
f
igu
r
at
io
n
I
nput
da
ta
c
ons
is
ts
of
11
c
ha
nne
ls
with
e
a
c
h
c
ha
n
ne
l
mea
s
ur
ing
128×
128×
128,
a
nd
the
output
da
ta
s
ize
is
128×
128×
128×
1.
F
or
input
,
the
f
i
r
s
t
c
ha
nne
l
is
a
C
T
im
a
ge
,
the
ne
xt
7
c
ha
nne
ls
c
ontain
OA
R
inf
or
mation,
a
nd
the
las
t
3
c
ha
nne
ls
c
ontain
P
T
V
inf
or
mation
.
T
r
a
ini
ng
tas
k
wa
s
c
onduc
ted
us
ing
Ada
m's
a
lgor
it
hm
[
22
]
.
Our
s
tudy
wa
s
im
pleme
nted
in
P
ython
langua
ge
a
nd
us
ing
the
de
e
p
-
lea
r
ning
platf
or
m
T
e
ns
or
F
low
[
23
]
.
T
he
e
xpe
r
im
e
nt
s
of
twa
r
e
r
u
n
on
both
the
Google
C
olab
s
ys
tem
[
24
]
a
nd
the
Ka
ggle
s
ys
tem
[
25]
,
whic
h
a
r
e
s
uppor
ted
by
the
c
omput
a
ti
on
pow
e
r
of
NV
I
DI
A
gr
a
phics
p
r
oc
e
s
s
ing
unit
s
(
GPUs
)
.
T
he
los
s
f
unc
ti
on
is
the
mea
n
o
f
a
bs
olut
e
e
r
r
or
.
I
ns
ide,
we
s
e
t
up
the
lea
r
ning
r
a
te
by
1
−
3
,
the
de
c
a
y
by
1
−
4
,
the
mom
e
ntum
β
1
by
0.
9,
a
nd
the
mom
e
ntum
β
2
by
0.
99.
T
he
f
o
r
mul
a
of
the
mea
n
a
bs
olut
e
e
r
r
or
los
s
f
u
nc
ti
on
is
c
a
lcula
ted
a
s
(
3)
:
L
=
1
n
∑
|
y
i
−
y
̂
i
|
n
i
=
1
(
3)
I
n
whic
h,
is
the
number
o
f
s
a
mpl
e
s
,
is
the
p
r
e
dict
e
d
dos
e
,
a
nd
̂
is
the
tr
ue
dos
e
of
s
a
mpl
e
.
I
n
the
e
xpe
r
im
e
nt
,
we
de
s
ign
one
mo
r
e
c
a
s
e
o
f
m
ode
l
tr
a
ini
ng
to
c
lar
i
f
y
the
e
f
f
e
c
t
of
pr
e
pr
oc
e
s
s
ing
a
nd
a
ugmenta
ti
on
tec
hniques
.
T
his
c
a
s
e
us
e
d
or
igi
na
l
da
ta,
whic
h
ha
d
no
a
ugmenta
ti
on
a
nd
pr
e
pr
oc
e
s
s
ing.
T
he
de
tail
wa
s
de
s
c
r
ibed
in
T
a
ble
1
.
I
n
our
hypot
he
s
is
,
the
r
e
s
ult
s
f
r
om
c
a
s
e
2
s
hould
be
be
tt
e
r
tha
n
c
a
s
e
1.
T
he
r
e
s
ult
f
igur
e
s
would
be
the
e
videnc
e
p
r
ovi
ng
the
e
f
f
e
c
t
o
f
p
r
e
pr
oc
e
s
s
ing
a
nd
a
ugmenta
ti
o
n
in
the
e
xpe
r
im
e
nt.
F
or
tr
a
ini
ng
a
nd
s
c
or
ing
the
model
,
we
us
e
d
the
tr
a
ini
ng
s
e
t
with
200
volum
e
s
a
mpl
e
s
a
nd
the
tes
ti
ng
s
e
t
with
100
volum
e
s
a
mpl
e
s
.
T
a
ble
1
.
T
wo
c
a
s
e
s
of
model
tr
a
ini
ng
T
r
a
in
in
g c
a
s
e
D
e
s
c
r
ip
ti
on
1
U
s
in
g a
n or
ig
in
a
l
da
ta
s
e
t
2
U
s
in
g a
da
ta
s
e
t
w
hi
c
h w
a
s
a
ppl
ie
d a
ugme
nt
a
ti
on a
nd pr
e
pr
oc
e
s
s
in
g
3.
2.
Re
s
u
lt
an
d
e
valu
a
t
ion
T
o
be
gin
with,
we
inves
ti
ga
ted
the
tr
a
ini
ng
r
e
s
ult
s
of
two
c
a
s
e
s
in
thi
s
s
tudy
.
T
he
li
ne
gr
a
phs
of
the
los
s
f
unc
ti
ons
a
r
e
pr
e
s
e
nted
in
F
igur
e
8.
I
n
ge
n
e
r
a
l,
the
tr
a
ini
ng
los
s
e
s
we
r
e
r
e
duc
e
d
dur
ing
the
tr
a
ini
ng
pe
r
iod.
I
n
c
a
s
e
2,
the
los
s
va
lue
wa
s
0.
607552
a
t
th
e
f
ir
s
t
e
poc
h,
de
c
r
e
a
s
e
d
to
0.
316148
a
t
the
thi
r
d
e
poc
h,
a
nd
c
onti
nue
d
to
de
c
li
ne
a
t
a
lowe
r
r
a
te.
T
he
tr
a
ini
ng
l
os
s
va
lue
in
c
a
s
e
1
wa
s
a
ls
o
r
e
duc
e
d
but
it
ha
d
s
o
me
s
mall
f
luctua
ti
ons
in
the
li
ne
gr
a
ph.
S
pe
c
if
ica
ll
y
,
the
lo
we
s
t
va
lue
a
lwa
y
s
be
longed
to
the
los
s
va
lu
e
of
c
a
s
e
2.
I
t
r
e
f
lec
ts
our
theor
e
ti
c
a
l
hypothes
is
be
c
a
us
e
in
c
a
s
e
2,
the
model
wa
s
tr
a
ined
with
be
tt
e
r
tr
a
ini
ng
da
t
a
,
whic
h
wa
s
pe
r
f
or
med
pr
e
p
r
oc
e
s
s
ing
a
nd
a
ugmenting.
F
igur
e
8.
T
he
tr
a
ini
ng
r
e
s
ult
s
of
two
c
a
s
e
s
T
he
ne
xt
tas
k
is
to
e
va
luate
the
tr
a
ined
models
in
tr
a
ini
ng
c
a
s
e
s
.
T
he
de
tailed
r
e
s
ult
s
a
r
e
s
hown
in
T
a
ble
2.
T
o
do
th
is
,
we
us
e
d
a
100
-
s
a
mpl
e
tes
t
s
e
t
a
nd
2
mea
s
ur
e
s
of
DV
H
-
s
c
or
e
a
nd
do
s
e
-
s
c
or
e
.
E
a
s
y
to
s
e
e
,
c
a
s
e
1
is
wor
s
e
than
c
a
s
e
2
in
both
mea
s
ur
e
s
.
About
DV
H
-
s
c
or
e
,
thi
s
s
c
or
e
in
c
a
s
e
1
is
18.
931728
while
in
c
a
s
e
2
is
2.
011907.
About
dos
e
-
s
c
or
e
,
thi
s
s
c
or
e
in
c
a
s
e
1
is
11
.
307606
while
in
c
a
s
e
2
is
3.
111
332.
All
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
1
4
,
No.
5
,
Oc
tober
2
02
4
:
5524
-
5533
5530
f
igur
e
s
point
e
d
out
that
the
r
e
s
ult
in
c
a
s
e
2
is
be
tt
e
r
than
the
r
e
s
ult
s
in
c
a
s
e
1
.
T
his
c
onc
lus
ion
a
ls
o
c
onf
or
ms
to
our
hypo
thes
is
whe
n
s
e
tt
ing
up
tr
a
ini
ng
c
a
s
e
s
.
T
a
ble
2.
E
va
luation
of
two
c
a
s
e
s
in
the
tes
ti
ng
pha
s
e
T
r
a
in
in
g c
a
s
e
D
os
e
-
s
c
or
e
DVH
-
s
c
or
e
1
11.307606
18.931728
2
3.111332
2.011907
T
o
ge
t
c
lea
r
e
r
,
we
c
onduc
ted
a
n
a
na
lys
is
of
the
pr
e
dicte
d
r
e
s
ult
in
one
s
a
mpl
e
.
I
n
de
tail
,
the
a
na
lys
is
wa
s
pe
r
f
or
med
on
the
r
e
s
ult
of
ou
r
c
us
tom
mode
l
that
wa
s
tr
a
ined
with
p
r
e
pr
oc
e
s
s
e
d
a
nd
a
ugmente
d
da
ta,
whic
h
mea
ns
c
a
s
e
2.
T
he
r
a
ndoml
y
c
hos
e
n
s
a
mp
le
is
s
a
mpl
e
258
in
the
da
tas
e
t.
T
he
c
or
r
e
s
pondi
ng
dos
e
-
s
c
or
e
is
2.
592035
a
nd
the
c
or
r
e
s
ponding
DV
H
-
s
c
or
e
is
1.
805577
.
W
e
f
oc
us
e
d
on
the
vis
ua
li
z
a
ti
on
of
the
pr
e
dicte
d
r
e
s
ult
.
And
be
c
a
us
e
the
input
da
ta
is
C
T
s
li
c
e
s
a
nd
the
output
is
the
3D
volum
e
o
f
dos
e
s
,
w
e
would
a
na
lyze
whe
n
view
ing
s
li
c
e
s
of
da
ta.
B
y
us
ing
vis
ua
li
z
a
ti
on
a
s
mentioned
in
s
ub
s
e
c
ti
on
2.
4,
we
would
obs
e
r
ve
e
a
c
h
s
li
c
e
of
pr
e
dicte
d
dos
e
a
s
a
he
a
tm
a
p,
a
s
dis
playe
d
in
F
igur
e
9
.
I
n
F
igur
e
9,
the
higher
r
a
diation
dos
e
int
e
ns
it
y
would
c
or
r
e
s
pond
to
the
br
igh
ter
c
olor
in
the
he
a
tm
a
p,
a
nd
a
ls
o
vice
ve
r
s
a
.
I
n
thi
s
wa
y,
we
c
a
n
obs
e
r
ve
the
dis
tr
ibut
ion
of
r
a
diation
dos
e
di
r
e
c
tl
y
with
our
e
ye
s
.
T
he
ne
xt
tas
k
is
to
r
e
pr
e
s
e
nt
the
c
ompar
is
on
be
twe
e
n
the
tr
ue
dos
e
a
nd
the
pr
e
dicte
d
dos
e
,
a
s
de
s
c
r
ibed
in
F
igur
e
10.
F
igur
e
9
.
T
h
r
e
e
view
s
of
the
pr
e
diction
r
e
s
ult
o
f
s
a
mpl
e
258:
the
C
T
im
a
ge
s
(
f
ir
s
t
r
ow)
a
nd
the
dos
e
vis
ua
li
z
a
ti
ons
(
s
e
c
ond
r
ow)
F
igur
e
10:
T
he
c
ompar
is
on
of
the
pr
e
diction
r
e
s
ult
s
with
the
gr
ound
tr
uth
da
ta
o
f
s
a
mpl
e
258:
the
gr
o
und
tr
uth
(
lef
t)
,
the
pr
e
dicte
d
dos
e
(
mi
dd
le)
,
a
nd
the
d
if
f
e
r
e
nc
e
im
a
ge
(
r
ight
)
I
n
F
ig
u
r
e
10
,
the
f
ir
s
t
two
c
olum
ns
a
r
e
the
pr
e
dicte
d
dos
e
s
a
nd
the
tr
ue
dos
e
s
whic
h
a
r
e
dis
playe
d
a
s
he
a
tm
a
ps
.
T
he
las
t
c
olum
n
is
the
dif
f
e
r
e
nc
e
im
a
ge
s
be
twe
e
n
the
pr
e
dicte
d
dos
e
s
a
nd
the
tr
ue
dos
e
s
.
E
a
c
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
fr
ame
w
or
k
for
3D
r
adiot
he
r
apy
dos
e
pr
e
diction
us
ing
the
de
e
p
lear
ning
appr
oac
h
(
L
am
T
hanh
Hi
e
n
)
5531
dif
f
e
r
e
nc
e
im
a
ge
is
c
ompu
ted
by
s
ubtr
a
c
ti
ng
the
c
or
r
e
s
ponding
pr
e
dicte
d
dos
e
f
r
om
the
c
o
r
r
e
s
ponding
tr
ue
dos
e
a
nd
ge
tt
ing
the
a
bs
olut
e
va
lue
of
the
s
ubtr
a
c
ti
on
r
e
s
ult
.
T
he
dif
f
e
r
e
nc
e
im
a
ge
s
a
r
e
a
ls
o
s
hown
a
s
he
a
tm
a
ps
a
nd
e
a
s
il
y
to
s
e
e
,
va
lues
of
pixels
in
the
dif
f
e
r
e
nc
e
im
a
ge
s
a
r
e
low
ove
r
the
major
it
y
of
the
a
r
e
a
.
About
the
c
ompar
is
on
of
c
a
s
e
1
a
nd
c
a
s
e
2,
we
a
ls
o
pr
e
s
e
nt
in
F
ig
ur
e
11
.
F
igur
e
11
.
T
he
c
ompar
is
on
be
twe
e
n
c
a
s
e
s
of
s
a
mpl
e
258:
c
a
s
e
1
(
le
f
t)
a
nd
c
a
s
e
2
(
r
ight
)
,
e
a
c
h
c
olum
n
include
s
the
pr
e
dicte
d
dos
e
vis
ua
li
z
a
ti
on
a
nd
the
di
f
f
e
r
e
nc
e
im
a
ge
I
n
F
ig
u
r
e
11
,
we
obs
e
r
ve
the
pr
e
dicte
d
dos
e
s
in
t
he
f
ir
s
t
r
ow
a
nd
the
di
f
f
e
r
e
nc
e
im
a
ge
s
,
whic
h
a
r
e
the
r
e
s
ult
of
a
c
ompar
is
on
be
twe
e
n
the
p
r
e
dicte
d
dos
e
s
to
the
gr
ound
tr
uth
in
the
s
e
c
ond
r
ow.
I
n
de
tail,
inf
or
mation
a
bout
c
a
s
e
1
is
p
r
e
s
e
nted
in
the
f
ir
s
t
c
olum
n
a
nd
in
f
or
mation
a
bout
c
a
s
e
2
is
p
r
e
s
e
nted
in
the
s
e
c
ond
c
olum
n.
W
he
n
obs
e
r
ving
the
s
e
c
ond
r
ow,
we
c
a
n
s
e
e
that
c
a
s
e
2
is
be
tt
e
r
than
c
a
s
e
1
be
c
a
us
e
the
dif
f
e
r
e
nc
e
im
a
ge
in
c
a
s
e
2
ha
s
lowe
r
int
e
ns
it
y
tha
n
c
a
s
e
1
in
ge
ne
r
a
l
.
T
he
ne
xt
tas
k
is
the
inf
or
mati
on
on
the
r
unning
ti
me
a
s
de
s
c
r
ibed
in
T
a
ble
3.
I
n
de
tail,
w
e
pr
e
s
e
nted
the
r
unning
ti
me
of
one
e
poc
h
in
the
tr
a
ini
ng
pha
s
e
a
nd
the
r
unning
ti
me
of
one
s
a
mpl
e
pr
e
dict
ion.
About
p
r
e
diction,
we
s
uppli
e
d
ti
me
in
f
or
mati
on
whe
n
r
unning
in
both
the
GPU
mac
hine
a
nd
the
c
e
ntr
a
l
p
r
oc
e
s
s
ing
unit
(
C
P
U
)
mac
hine.
T
o
e
s
ti
mate
the
r
unning
ti
me
in
c
a
s
e
s
,
we
us
e
d
th
e
Google
C
olab
platf
o
r
m
s
uppor
ted
by
T
e
s
la
K80
GPU
a
nd
our
pe
r
s
ona
l
laptop
with
the
s
im
ple
c
onf
igur
a
ti
on
o
f
I
ntel
I
5
with
1
.
60
GH
z
.
As
pr
e
s
e
nted
in
T
a
ble
3,
the
r
unning
ti
me
in
the
C
P
U
mac
hine
is
not
long
a
nd
it
is
a
c
c
e
ptable
in
ou
r
r
e
a
l
te
s
t.
T
his
inf
or
mation
is
he
lpf
ul
f
or
us
be
c
a
us
e
GPU
mac
hines
a
r
e
r
a
ther
e
xpe
ns
ive
a
nd
no
t
e
a
s
y
to
be
r
e
a
dy
f
or
de
ploym
e
nt
in
ge
ne
r
a
l.
Anothe
r
e
s
ti
mation
is
the
c
ompar
is
on
with
the
r
e
s
ult
o
f
De
e
pDos
e
Ne
t
in
t
he
s
tudy
[
16]
.
I
n
that
r
e
s
e
a
r
c
h,
the
a
uthor
s
e
xpe
r
im
e
nted
with
s
ome
dif
f
e
r
e
nt
los
s
f
unc
ti
ons
.
I
n
thi
s
s
tudy,
we
pe
r
f
or
med
a
c
ompar
is
on
be
twe
e
n
our
r
e
s
ult
in
c
a
s
e
2
with
the
r
e
s
ult
o
f
De
e
pDos
e
Ne
t
in
the
s
a
me
c
ontext
e
xpe
r
im
e
nt
of
the
los
s
f
unc
ti
on
-
the
mea
n
a
bs
olut
e
e
r
r
or
los
s
.
All
de
tails
a
r
e
s
hown
in
T
a
ble
4
.
T
a
ble
3
.
R
unning
ti
me
inf
or
mat
ion
C
a
s
e
M
a
c
hi
ne
A
ppr
oxi
ma
te
t
im
e
O
ne
e
poc
h t
r
a
in
in
g of
c
a
s
e
1
G
P
U
10.37 mi
nut
e
s
O
ne
e
poc
h t
r
a
in
in
g of
c
a
s
e
2
G
P
U
30.45 mi
nut
e
s
P
r
e
di
c
ti
ng one
s
a
mpl
e
of
c
a
s
e
2
G
P
U
1.24 s
e
c
ond
P
r
e
di
c
ti
ng one
s
a
mpl
e
of
c
a
s
e
2
C
P
U
11.15 s
e
c
onds
T
a
ble
4.
C
ompar
is
on
with
the
De
e
pDos
e
Ne
t
in
c
a
s
e
of
the
s
a
me
los
s
f
unc
ti
on
M
ode
l
D
os
e
-
s
c
or
e
DVH
-
s
c
or
e
O
ur
mode
l
(
c
a
s
e
2)
3.111332
2.011907
T
he
D
e
e
p
D
os
e
N
e
t
3.5
2.3
F
r
om
the
theo
r
y
o
f
the
De
e
pDos
e
Ne
t,
S
oomr
o
e
t
al.
[
1
6
]
a
ls
o
a
ppli
e
d
the
r
e
s
idual
blocks
.
How
e
ve
r
,
our
model
wa
s
de
s
igned
with
mor
e
laye
r
s
.
I
n
a
dd
it
ion,
S
oomr
o
e
t
al
.
[
1
6
]
did
not
pe
r
f
o
r
m
a
ugmenting
da
ta
a
nd
they
a
ls
o
did
not
talk
a
bout
pr
e
pr
oc
e
s
s
ing
da
t
a
.
F
inally,
we
buil
t
a
medic
a
l
s
ys
tem
f
or
pr
e
dictin
g
dos
e
a
s
de
s
c
r
ibed
in
F
igur
e
12
.
T
his
s
of
twa
r
e
would
outp
ut
the
dos
e
pr
e
diction
a
nd
dis
play
it
a
s
a
he
a
tm
a
p
on
the
a
ppli
c
a
ti
on
c
a
nva
s
.
B
e
s
ides
,
we
a
ls
o
s
uppor
t
view
ing
dos
e
da
ta
in
di
f
f
e
r
e
nt
dim
e
ns
ions
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
1
4
,
No.
5
,
Oc
tober
2
02
4
:
5524
-
5533
5532
F
igur
e
12
.
T
he
a
ppli
c
a
ti
on
int
e
r
f
a
c
e
s
4.
CONC
L
USI
ON
C
a
nc
e
r
is
a
da
nge
r
ous
dis
e
a
s
e
s
o
r
e
quir
e
s
a
lot
o
f
a
tt
e
mpt
s
to
pr
e
ve
nt
it
.
One
of
the
lea
ding
c
a
nc
e
r
tr
e
a
tm
e
nts
is
r
a
diot
he
r
a
py
whic
h
a
ppli
e
s
r
a
diati
on
ther
a
py
be
a
ms
to
e
li
mi
na
te
c
a
nc
e
r
c
e
ll
s
.
T
h
is
s
tudy
inves
ti
ga
ted
the
a
pplyi
ng
of
the
de
e
p
lea
r
n
ing
a
pp
r
oa
c
h
to
r
a
diot
he
r
a
py
dos
e
pr
e
diction
.
W
e
f
ound
th
a
t
it
wa
s
ne
c
e
s
s
a
r
y
f
or
s
uit
a
ble
de
e
p
a
r
c
hit
e
c
tur
e
a
nd
da
ta
-
pr
oc
e
s
s
ing
tec
hniques
.
T
he
pr
opos
e
d
method
in
t
his
s
tudy
us
e
s
a
ne
twor
k
a
r
c
hit
e
c
tur
e
f
or
3D
dos
e
pr
e
d
iction
ba
s
e
d
on
R
e
s
-
U
-
N
e
t
3D
a
nd
pr
e
p
r
oc
e
s
s
ing
a
nd
a
ugmenta
ti
on
tec
hniques
ba
s
e
d
on
a
na
lys
is
of
da
ta
c
ha
r
a
c
ter
is
ti
c
s
.
Our
s
tudy
de
mons
tr
a
tes
the
pe
r
f
or
manc
e
whe
n
c
ompar
e
d
to
the
De
e
pDos
e
Ne
t
s
tudy
whic
h
doe
s
not
a
pply
da
ta
-
pr
oc
e
s
s
ing
tec
hniques
.
L
a
s
tl
y,
we
a
ls
o
de
ve
loped
the
3D
dos
e
pr
e
diction
a
ppli
c
a
ti
on
s
uppor
ti
ng
dos
e
vis
ua
li
z
a
ti
on.
T
his
pa
pe
r
obtain
e
d
s
ome
s
uc
c
e
s
s
but
to
be
a
ble
to
bu
il
d
a
medic
a
l
int
e
ll
ige
nt
s
ys
tem
that
c
a
n
wor
k
e
f
f
e
c
ti
ve
ly
in
r
e
a
li
ty,
we
a
ls
o
ha
ve
to
do
a
lot
of
wor
k.
T
he
s
uc
c
e
s
s
of
thi
s
wor
k
is
o
ne
of
the
f
i
r
s
t
s
teps
in
our
wa
y
of
pr
e
ve
nti
ng
c
a
nc
e
r
in
the
int
e
ll
igent
medic
a
l
indus
tr
y
.
AC
KNOWL
E
DGE
M
E
NT
S
T
he
r
e
s
e
a
r
c
h
c
ontent
in
thi
s
pa
pe
r
is
s
pons
or
e
d
by
the
I
ns
ti
tut
e
of
I
nf
or
mation
T
e
c
hnology
(
Vie
t
Na
m
Ac
a
de
my
of
S
c
ienc
e
a
nd
T
e
c
hnology)
unde
r
the
pr
ojec
t
c
ode
“
C
S
24.
01”
.
RE
F
E
RE
NC
E
S
[
1]
W
H
O
,
“
C
a
nc
e
r
,”
W
or
ld
H
e
al
th
O
r
gani
z
at
io
n
,
2022.
ht
tp
s
:/
/w
w
w
.w
ho.i
nt
/n
e
w
s
-
r
oom/
f
a
c
t
-
s
he
e
ts
/d
e
ta
il
/c
a
n
c
e
r
(
a
c
c
e
s
s
e
d
D
e
c
.
12,
2023)
.
[
2]
W
H
O
,
“
C
a
nc
e
r
to
d
a
y
,
”
W
o
r
l
d
H
e
al
t
h
O
r
g
an
iz
at
io
n
,
ht
tp
s
:/
/
g
c
o
.i
a
r
c
.f
r
/t
od
a
y/
d
a
ta
/f
a
c
t
s
h
e
e
t
s
/
po
pu
l
a
ti
on
s
/7
04
-
vi
e
t
-
n
a
m
-
f
a
c
t
-
s
h
e
e
t
s
.p
df
(
a
c
c
e
s
s
e
d
D
e
c
.
1
2,
2
02
3)
.
[
3]
S
.
G
ia
nf
a
ld
oni
,
R
.
G
ia
nf
a
ld
oni
,
U
.
W
ol
l
in
a
,
J
.
L
ot
ti
,
G
.
T
c
he
r
ne
v,
a
nd
T
.
L
ot
ti
,
“
A
n
ove
r
vi
e
w
on
r
a
di
ot
he
r
a
py:
f
r
om
it
s
hi
s
to
r
y
to
it
s
c
ur
r
e
nt
a
ppl
ic
a
ti
ons
in
de
r
ma
to
lo
gy,”
O
pe
n
A
c
c
e
s
s
M
ac
e
do
ni
an
J
our
nal
of
M
e
di
c
al
Sc
ie
nc
e
s
,
vol
.
5,
no.
4,
pp.
521
–
525,
J
ul
.
2017, doi:
10.3889/oamjm
s
.2017.122.
[
4]
C
.
M
c
I
nt
os
h
a
nd
T
.
G
.
P
ur
di
e
,
“
C
ont
e
xt
ua
l
a
tl
a
s
r
e
gr
e
s
s
io
n
f
or
e
s
ts
:
mul
ti
pl
e
-
a
tl
a
s
-
ba
s
e
d
a
ut
oma
te
d
do
s
e
pr
e
di
c
ti
on
in
r
a
di
a
ti
on
th
e
r
a
py,”
I
E
E
E
T
r
ans
ac
ti
ons
on M
e
di
c
al
I
m
agi
ng
, vol
. 35, no. 4, pp. 1000
–
1012, Apr
. 2016, doi:
10.1109/T
M
I
.2015.2505188.
[
5]
P
. Z
hou
e
t
al
.
, “
S
uppor
t
ve
c
to
r
ma
c
hi
ne
mode
l
pr
e
di
c
ts
dos
e
f
or
or
ga
ns
a
t
r
is
k i
n hi
gh
-
dos
e
r
a
te
br
a
c
hyt
he
r
a
py of
c
e
r
vi
c
a
l
c
a
nc
e
r
,”
F
r
ont
ie
r
s
i
n O
nc
ol
ogy
, vol
. 11,
J
ul
. 2021, doi:
10.3389/f
onc
.2021.619384.
[
6]
P
.
F
ur
ta
do,
“
M
a
gne
ti
c
r
e
s
ona
nc
e
s
e
que
nc
e
s
:
e
xpe
r
im
e
nt
a
l
a
s
s
e
s
s
me
nt
of
a
c
hi
e
ve
me
nt
s
a
nd
li
mi
ta
ti
ons
,”
J
our
nal
of
A
dv
anc
e
s
in
I
nf
or
m
at
io
n T
e
c
hnol
ogy
, vol
. 12, no. 1, pp. 66
–
70, 2021, doi:
10.12720/j
a
it
.12.1.66
-
70.
[
7]
Y
.
C
he
n,
W
.
D
u,
X
.
D
ua
n,
Y
.
M
a
,
a
nd
H
.
Z
ha
ng,
“
S
que
e
z
e
-
a
nd
-
e
xc
it
a
ti
on
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
f
or
c
la
s
s
if
ic
a
ti
o
n
of
ma
li
gna
nt
a
nd
be
ni
gn
lu
ng
nodule
s
,”
J
ou
r
nal
of
A
dv
anc
e
s
in
I
nf
or
m
at
io
n
T
e
c
hnol
ogy
,
vol
.
12,
no.
2,
pp.
153
–
158,
2021,
doi
:
10.12720/j
a
it
.12.2.153
-
158.
[
8]
H
.
M
.
T
oa
n,
L
.
T
.
H
ie
n,
N
.
D
.
V
in
h,
a
nd
D
.
N
.
T
oa
n,
“
D
e
te
c
ti
ng
tu
be
r
c
ul
os
is
f
r
om
V
ie
tn
a
me
s
e
X
-
R
a
y
im
a
gi
ng
us
in
g
tr
a
ns
f
e
r
le
a
r
ni
ng a
ppr
oa
c
h,”
C
om
put
e
r
s
, M
at
e
r
ia
ls
&
C
ont
in
ua
, vol
. 74, no. 3, pp. 5001
–
5016, 2023, doi:
10.32604/cmc
.2023.033429.
[
9]
P
r
a
ba
ka
r
a
n
a
nd
P
.
S
e
lv
a
r
a
j,
“
A
dva
nc
e
I
oT
in
te
ll
ig
e
nt
he
a
lt
hc
a
r
e
s
ys
te
m
f
or
lu
ng
d
is
e
a
s
e
c
la
s
s
if
ic
a
ti
on
us
in
g
e
ns
e
mbl
e
te
c
hni
que
s
,”
C
om
put
e
r
Sy
s
te
m
s
Sc
ie
nc
e
and E
ngi
ne
e
r
in
g
, vol
.
46, no. 2, pp. 2141
–
2157, 2023, doi:
10.32604/cs
s
e
.2023.03421
0.
[
10]
H
.
A
lq
ur
a
n
e
t
al
.
,
“
L
iv
e
r
tu
mo
r
de
c
is
io
n
s
uppor
t
s
ys
te
m
on
huma
n
ma
gne
ti
c
r
e
s
ona
nc
e
im
a
ge
s
:
a
c
ompa
r
a
ti
ve
s
tu
dy,”
C
om
p
ut
e
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
A
fr
ame
w
or
k
for
3D
r
adiot
he
r
apy
dos
e
pr
e
diction
us
ing
the
de
e
p
lear
ning
appr
oac
h
(
L
am
T
hanh
Hi
e
n
)
5533
Sy
s
te
m
s
S
c
ie
nc
e
and E
ngi
ne
e
r
in
g
, vol
. 46, no. 2, pp. 1653
–
1671, 2023, doi:
10.32604/cs
s
e
.2023.033861.
[
11]
Y
.
L
iu
e
t
al
.
,
“
D
os
e
pr
e
di
c
ti
on
us
in
g
a
th
r
e
e
-
di
me
ns
io
na
l
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
f
o
r
na
s
opha
r
ynge
a
l
c
a
r
c
in
oma
w
it
h
to
mot
he
r
a
py,”
F
r
ont
ie
r
s
i
n O
nc
ol
ogy
, vol
. 11, Nov. 2021, doi:
10.3389/f
onc
.2021.752007.
[
12]
S
. H
. A
hn
e
t
al
.
, “
D
e
e
p l
e
a
r
ni
ng
me
th
od f
o
r
p
r
e
di
c
ti
on of
pa
ti
e
nt
-
s
pe
c
if
ic
dos
e
di
s
tr
ib
ut
io
n i
n
br
e
a
s
t
c
a
nc
e
r
,”
R
adi
at
io
n
O
nc
ol
ogy
,
vol
. 16, no. 1, De
c
. 2021, doi:
10.1186/s
13014
-
021
-
01864
-
9.
[
13]
A
.
B
a
bi
e
r
,
R
.
M
a
hmood,
A
.
L
.
M
c
N
iv
e
n,
A
.
D
ia
m
a
nt
,
a
nd
T
.
C
.
Y
.
C
ha
n,
“
K
now
le
dge
‐
ba
s
e
d
a
ut
oma
te
d
pl
a
nni
ng
w
it
h
th
r
e
e
‐
di
me
ns
io
na
l
ge
ne
r
a
ti
ve
a
dve
r
s
a
r
ia
l
ne
twor
ks
,”
M
e
di
c
al
P
hy
s
ic
s
, vol
. 47, no. 2, pp. 297
–
306, F
e
b. 2020, doi:
10.1002/m
p.1389
6.
[
14]
M
.
P
.
G
r
onbe
r
g,
S
.
S
.
G
a
y,
T
.
J
.
N
e
th
e
r
to
n,
D
.
J
.
R
he
e
,
L
.
E
.
C
our
t,
a
nd
C
.
E
.
C
a
r
de
na
s
,
“
D
os
e
pr
e
di
c
ti
on
f
or
he
a
d
a
nd
n
e
c
k
r
a
di
ot
he
r
a
py
us
in
g
a
th
r
e
e
‐
di
me
ns
io
na
l
de
ns
e
di
la
te
d
U
‐
N
e
t
a
r
c
hi
te
c
tu
r
e
,”
M
e
di
c
al
P
hy
s
ic
s
,
vol
.
48,
no.
9,
pp.
5567
–
5573,
S
e
p.
2021, doi:
10.1002/m
p.14827.
[
15]
A
.
B
a
bi
e
r
e
t
al
.
,
“
O
pe
nK
B
P
:
th
e
ope
n‐
a
c
c
e
s
s
knowle
dg
e
‐
ba
s
e
d
pl
a
nni
ng
gr
a
nd
c
ha
ll
e
nge
a
nd
da
ta
s
e
t,
”
M
e
di
c
al
P
hy
s
ic
s
,
vol
.
48,
no. 9, pp. 5549
–
5561, S
e
p. 2021, doi:
10.1002/m
p.14845.
[
16]
M
.
H
.
S
oomr
o,
V
.
G
.
L
.
A
lv
e
s
,
H
.
N
our
z
a
de
h,
a
nd J
.
V
.
S
ie
be
r
s
,
“
D
e
e
pD
o
s
e
N
e
t:
a
de
e
p
le
a
r
ni
ng
mode
l
f
or
3D
dos
e
pr
e
di
c
ti
o
n
in
r
a
di
a
ti
on t
he
r
a
py,”
ar
X
iv
pr
e
pr
in
t
a
r
X
iv
:
2111.00077
, 2021.
[
17]
L
.
Z
im
me
r
ma
nn,
E
.
F
a
us
tm
a
nn,
C
.
R
a
m
s
l,
D
.
G
e
or
g,
a
nd
G
.
H
e
il
e
ma
nn,
“
D
o
s
e
pr
e
di
c
ti
on
f
or
r
a
di
a
ti
on
th
e
r
a
py
us
in
g
f
e
a
tu
r
e
‐
ba
s
e
d l
os
s
e
s
a
nd one
c
yc
le
l
e
a
r
ni
ng,”
M
e
di
c
al
P
hy
s
ic
s
, vol
. 48,
no. 9, pp. 5562
–
5566, S
e
p. 2021, doi:
10.1002/m
p.14774.
[
18]
R
a
di
opa
e
di
a
, “
H
oun
s
f
ie
ld
uni
t,
”
R
adi
opae
di
a
,
2024. htt
ps
:/
/r
a
di
opa
e
di
a
.or
g/
a
r
ti
c
le
s
/h
ouns
f
ie
ld
-
uni
t
(
a
c
c
e
s
s
e
d F
e
b. 19, 2024)
.
[
19]
O
.
R
onne
be
r
ge
r
,
P
.
F
is
c
h
e
r
,
a
nd T
.
B
r
ox,
“
U
-
N
e
t:
c
onvolut
io
na
l
ne
twor
ks
f
or
bi
ome
di
c
a
l
im
a
ge
s
e
gme
nt
a
ti
on,”
in
M
e
di
c
al
I
m
age
C
om
put
in
g and C
om
put
e
r
-
A
s
s
i
s
te
d I
nt
e
r
v
e
nt
io
n
–
M
I
C
C
A
I
2015
, 2015, pp. 234
–
241.
[
20]
K
.
H
e
,
X
.
Z
ha
ng,
S
.
R
e
n,
a
nd
J
.
S
un,
“
D
e
e
p
r
e
s
id
ua
l
le
a
r
ni
n
g
f
or
im
a
ge
r
e
c
ogni
ti
on,”
P
r
oc
e
e
di
ngs
of
th
e
I
E
E
E
c
onf
e
r
e
nc
e
on
c
om
put
e
r
v
is
io
n and patt
e
r
n r
e
c
ogni
ti
on
, pp. 770
–
778, 2016.
[
21]
M
a
tp
lo
tl
ib
,
“
C
hoos
in
g
C
ol
or
ma
ps
in
M
a
tp
lo
tl
ib
,
”
M
a
tp
l
ot
li
b
,
ht
tp
s
:/
/m
a
tp
lo
tl
ib
.o
r
g/
s
ta
bl
e
/t
ut
or
ia
ls
/c
ol
or
s
/c
ol
or
ma
ps
.ht
ml
(
a
c
c
e
s
s
e
d F
e
b. 19, 2024)
.
[
22]
D
. P
. K
in
gma
a
nd J
. B
a
, “
A
da
m:
a
me
th
od f
or
s
to
c
ha
s
ti
c
opt
im
iz
a
ti
on,”
ar
X
iv
pr
e
p
r
in
t
ar
X
iv
:
1412.6980
, 2014.
[
23]
M
. A
ba
di
e
t
al
.
, “
T
e
ns
or
F
lo
w
:
la
r
ge
-
s
c
a
le
ma
c
hi
ne
l
e
a
r
ni
ng on
he
te
r
oge
ne
ous
s
y
s
te
ms
,
”
P
r
e
pr
in
t
ar
X
iv
:
1603.04467
, 2016.
[
24]
“
W
e
lc
ome
t
o
C
ol
a
b
,
”
C
ol
ab
,
ht
tp
s
:/
/c
ol
a
b.r
e
s
e
a
r
c
h.google
.
c
om
(
a
c
c
e
s
s
e
d F
e
b. 19, 2024)
.
[
25]
“
L
e
ve
l
up w
it
h t
he
l
a
r
ge
s
t
A
I
& M
L
c
omm
uni
ty
,
”
K
aggl
e
,
ht
tp
s
:/
/ww
w
.ka
ggl
e
.c
om
(a
c
c
e
s
s
e
d F
e
b. 19, 2024
)
.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
La
m
Tha
nh
Hi
en
j
o
i
n
e
d
an
M.
Sc.
i
n
ap
p
l
i
e
d
i
n
f
o
rmat
i
cs
at
t
h
e
IN
N
O
T
E
C
H
In
s
t
i
t
u
t
e,
Fran
ce,
an
d
rece
i
v
e
d
a
d
eg
ree
i
n
2
0
0
4
.
In
2
0
1
7
,
h
e
earn
ed
a
P
h
.
D
.
d
eg
ree
a
t
t
h
e
V
i
e
t
n
am
A
cad
em
y
o
f
Sc
i
en
ce
a
n
d
T
ec
h
n
o
l
o
g
y
.
N
o
w
,
h
e
w
o
r
k
s
a
t
L
ac
H
o
n
g
U
n
i
v
er
s
i
t
y
i
n
t
h
e
h
ead
ma
s
t
er's
r
o
l
e.
H
i
s
s
t
u
d
i
e
s
i
n
t
ere
s
t
s
re
l
at
e
t
o
mach
i
n
e
l
earn
i
n
g
,
co
m
p
u
t
er
v
i
s
i
o
n
,
an
d
d
ee
p
l
earn
i
n
g
.
H
e
can
b
e
co
n
t
a
ct
e
d
at
emai
l
:
l
t
h
i
en
@
l
h
u
.
e
d
u
.
v
n
.
Ha
M
a
nh
To
a
n
l
earn
e
d
ap
p
l
i
ed
ma
t
h
ema
t
i
c
s
an
d
i
n
fo
rma
t
i
c
s
at
t
h
e
Co
l
l
e
g
e
o
f
Sci
en
ce,
V
i
et
n
am
N
at
i
o
n
al
U
n
i
v
er
s
i
t
y
,
H
a
n
o
i
,
an
d
rec
ei
v
e
d
a
d
eg
ree
i
n
2
0
0
9
.
In
2
0
1
5
,
h
e
earn
ed
an
M.
Sc.
d
eg
ree
at
t
h
e
U
n
i
v
e
rs
i
t
y
o
f
E
n
g
i
n
eer
i
n
g
an
d
T
ec
h
n
o
l
o
g
y
,
V
i
e
t
n
am
N
at
i
o
n
al
U
n
i
v
er
s
i
t
y
,
H
an
o
i
.
N
o
w
,
h
e
i
s
a
res
earch
er
at
t
h
e
V
i
et
n
ames
e
A
cad
em
y
o
f
Sci
en
ce
an
d
T
ech
n
o
l
o
g
y
.
H
i
s
s
t
u
d
i
es
i
n
t
eres
t
s
rel
a
t
e
t
o
mac
h
i
n
e
l
earn
i
n
g
,
co
mp
u
t
er
v
i
s
i
o
n
,
an
d
d
ee
p
l
earn
i
n
g
.
H
e
can
b
e
co
n
t
a
ct
e
d
at
emai
l
:
h
m
t
o
a
n
@
i
o
i
t
.
ac.
v
n
.
D
o
N
a
n
g
To
a
n
s
t
u
d
i
e
d
ap
p
l
i
e
d
mat
h
emat
i
cs
an
d
i
n
f
o
rmat
i
cs
at
H
an
o
i
U
n
i
v
ers
i
t
y
an
d
rece
i
v
e
d
a
d
eg
ree
i
n
1
9
9
0
.
I
n
2
0
0
1
,
h
e
ear
n
ed
a
Ph
.
D
.
d
eg
ree
at
t
h
e
V
i
e
t
n
am
A
ca
d
emy
o
f
Sci
en
ce
an
d
T
ec
h
n
o
l
o
g
y
.
N
o
w
,
h
e
i
s
an
as
s
o
c
i
at
e
p
ro
f
es
s
o
r
at
t
h
e
V
i
e
t
n
ame
s
e
A
ca
d
emy
o
f
Sci
en
ce
an
d
T
ech
n
o
l
o
g
y
.
H
i
s
s
t
u
d
i
e
s
i
n
t
eres
t
s
rel
a
t
e
t
o
mach
i
n
e
l
earn
i
n
g
,
co
mp
u
t
er
v
i
s
i
o
n
,
an
d
v
i
r
t
u
a
l
real
i
t
y
.
H
e
ca
n
b
e
co
n
t
ac
t
ed
a
t
emai
l
:
d
n
t
o
an
@
i
o
i
t
.
ac.
v
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.