I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14
,
No.
4
,
Augus
t
20
25
,
pp.
3182
~
3191
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
4
.
pp
31
82
-
3191
3182
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
U
si
n
g t
h
e
R
e
s
N
e
t
-
50
p
r
e
-
t
r
ai
n
e
d
m
od
e
l
t
o i
m
p
r
ov
e
t
h
e
c
la
ssi
f
ic
at
io
n
o
u
t
p
u
t
o
f
a n
o
n
-
image
k
id
n
e
y s
t
on
e
d
a
t
ase
t
Kaze
e
m
Oyebod
e
1
,
An
n
e
Ngoz
i
Od
o
h
2
1
D
e
pa
r
tm
e
nt
of
C
omput
e
r
a
nd I
nf
or
ma
ti
on
S
c
ie
nc
e
,
S
c
hool
of
S
c
ie
nc
e
a
nd T
e
c
hnol
ogy, P
a
n
-
A
tl
a
nt
ic
U
ni
ve
r
s
it
y, L
a
gos
, N
ig
e
r
ia
2
S
c
hool
of
M
e
di
a
a
nd C
omm
uni
c
a
ti
on, P
a
n
-
A
tl
a
nt
ic
U
ni
ve
r
s
it
y, L
a
gos
, N
ig
e
r
ia
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
ul
6,
2024
R
e
vis
e
d
Apr
4,
2025
Ac
c
e
pted
J
un
8,
2025
K
i
d
n
e
y
s
t
o
n
e
d
et
ec
t
i
o
n
b
a
s
ed
o
n
u
ri
n
e
s
amp
l
es
s
eem
s
t
o
b
e
a
co
s
t
-
effect
i
v
e
w
ay
o
f
d
e
t
ect
i
n
g
t
h
e
fo
rma
t
i
o
n
o
f
s
t
o
n
es
.
U
ri
n
e
feat
u
re
s
are
u
s
u
a
l
l
y
co
l
l
ect
e
d
fro
m
p
a
t
i
e
n
t
s
t
o
d
et
erm
i
n
e
i
f
t
h
ere
i
s
a
l
i
k
e
l
i
h
o
o
d
o
f
k
i
d
n
ey
s
t
o
n
e
f
o
rmat
i
o
n
.
T
h
ere
are
e
x
i
s
t
i
n
g
mach
i
n
e
l
ear
n
i
n
g
m
o
d
e
l
s
t
h
a
t
can
b
e
u
s
ed
t
o
c
l
as
s
i
f
y
i
f
a
s
t
o
n
e
e
x
i
s
t
s
i
n
t
h
e
k
i
d
n
ey
,
s
u
ch
as
t
h
e
s
u
p
p
o
rt
v
ect
o
r
mac
h
i
n
e
(SV
M)
a
n
d
d
eep
l
earn
i
n
g
(D
L
)
mo
d
el
s
.
W
e
p
ro
p
o
s
e
a
DL
n
et
w
o
rk
t
h
at
w
o
rk
s
w
i
t
h
a
p
re
-
t
ra
i
n
e
d
(Res
N
e
t
-
5
0
)
mo
d
el
,
mak
i
n
g
n
o
n
-
i
mag
e
u
ri
n
e
feat
u
r
e
s
w
o
r
k
w
i
t
h
an
i
mag
e
-
b
as
e
d
p
re
-
t
rai
n
ed
m
o
d
e
l
(Res
N
e
t
-
5
0
).
Si
x
u
ri
n
e
fea
t
u
res
c
o
l
l
ect
e
d
fro
m
p
at
i
en
t
s
are
p
ro
j
ect
e
d
o
n
t
o
1
7
2
,
8
0
0
n
eu
r
o
n
s
.
T
h
i
s
o
u
t
p
u
t
i
s
t
h
en
res
h
ap
e
d
i
n
t
o
a
2
4
0
b
y
2
4
0
b
y
3
t
e
n
s
o
rs
.
T
h
e
res
h
ap
e
d
o
u
t
p
u
t
s
erv
e
s
as
t
h
e
i
n
p
u
t
t
o
t
h
e
Res
N
e
t
-
5
0
.
T
h
e
o
u
t
p
u
t
o
f
t
h
i
s
i
s
t
h
en
s
e
n
t
i
n
t
o
a
b
i
n
ar
y
cl
as
s
i
f
i
er
t
o
d
e
t
ermi
n
e
i
f
a
k
i
d
n
e
y
s
t
o
n
e
ex
i
s
t
s
o
r
n
o
t
.
T
h
e
p
r
o
p
o
s
e
d
mo
d
el
i
s
b
en
c
h
mark
e
d
ag
a
i
n
s
t
t
h
e
SV
M,
X
G
Bo
o
s
t
,
an
d
t
w
o
v
ar
i
an
t
s
o
f
DL
n
e
t
w
o
rk
s
,
an
d
i
t
s
h
o
w
s
i
mp
r
o
v
e
d
p
erfo
rma
n
ce
u
s
i
n
g
t
h
e
A
U
C
-
RO
C,
A
cc
u
ra
cy
an
d
F1
-
s
c
o
re
met
ri
c
s
.
W
e
d
emo
n
s
t
rat
e
t
h
a
t
co
mb
i
n
i
n
g
n
o
n
-
i
ma
g
e
u
ri
n
e
feat
u
re
s
w
i
t
h
an
i
mag
e
-
b
as
e
d
p
re
-
t
ra
i
n
e
d
mo
d
e
l
i
mp
r
o
v
e
s
cl
as
s
i
f
i
cat
i
o
n
o
u
t
c
o
mes
,
h
i
g
h
l
i
g
h
t
i
n
g
t
h
e
p
o
t
en
t
i
a
l
o
f
i
n
t
eg
ra
t
i
n
g
h
et
er
o
g
e
n
eo
u
s
d
a
t
a
s
o
u
rce
s
fo
r
en
h
an
ce
d
p
red
i
ct
i
v
e
accu
rac
y
.
K
e
y
w
o
r
d
s
:
Ar
ti
f
icia
l
int
e
ll
igenc
e
D
e
e
p
lea
r
ning
Kidne
y
s
tone
P
r
e
-
tr
a
ined
model
S
uppor
t
ve
c
tor
mac
hine
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
:
Ka
z
e
e
m
Oye
bode
De
pa
r
tm
e
nt
of
C
omput
e
r
a
nd
I
nf
or
mation
S
c
ienc
e
,
S
c
hool
of
S
c
ienc
e
a
nd
T
e
c
hnology
P
a
n
-
Atlantic
Unive
r
s
it
y
L
a
gos
,
Nige
r
ia
E
mail:
koye
bode
@pa
u.
e
du.
ng
1.
I
NT
RODU
C
T
I
ON
Kidne
y
s
tone
is
a
u
r
ologi
c
a
l
c
ondit
ion
that
oc
c
ur
s
whe
n
ther
e
is
a
de
pos
it
o
f
a
c
id
s
a
lt
s
in
the
kidneys
,
whic
h
r
e
duc
e
s
their
f
unc
ti
on
[
1]
.
I
n
s
ome
c
a
s
e
s
,
it
blocks
the
f
low
o
f
ur
ine
a
nd
c
a
us
e
s
a
gonizing
pa
in
to
the
s
uf
f
e
r
e
r
[
1
]
.
I
t
is
be
li
e
ve
d
to
a
f
f
e
c
t
1
in
10
indi
vidua
ls
[
2]
,
with
it
s
pr
e
va
lenc
e
incr
e
a
s
ing
globally
[
2
]
,
[
3
]
.
C
a
s
e
s
of
kidney
s
tone
dis
e
a
s
e
ha
ve
be
e
n
on
the
r
is
e
a
s
r
e
por
ted
in
[
4
]
.
I
n
a
ddit
ion,
the
lac
k
o
f
a
c
c
e
s
s
to
c
a
r
e
in
r
ur
a
l
a
nd
r
e
mot
e
a
r
e
a
s
c
ontr
ibut
e
s
to
thi
s
incr
e
a
s
e
[
4
]
.
I
n
the
USA
c
a
s
e
s
of
Kidne
y
s
tones
r
os
e
f
r
om
3.
2
%
in
1980
to
10
%
in
2014
[
5
]
,
a
f
f
e
c
ti
ng
mo
r
e
men
than
wo
men
[
6
]
.
R
e
c
e
nt
s
tudi
e
s
f
r
om
Af
r
ica
[
7
]
a
nd
As
ia
[
8]
s
how
a
s
im
il
a
r
tr
e
nd
in
the
nu
mber
o
f
c
a
s
e
s
r
e
c
or
de
d
r
e
ga
r
ding
kidney
s
tones
.
T
h
is
de
ve
lo
pme
n
t
s
h
ows
th
a
t
t
he
r
e
is
a
ne
e
d
to
c
o
mb
a
t
k
id
ne
y
s
t
on
e
d
is
e
a
s
e
.
T
he
r
e
a
r
e
s
e
ve
r
a
l
met
ho
ds
a
va
il
a
b
le
to
de
te
c
t
ki
dn
e
y
s
t
one
s
.
T
he
m
os
t
c
o
mm
on
f
o
r
m
o
f
di
a
g
nos
is
is
a
c
o
mp
ut
e
d
to
mo
g
r
a
ph
y
(
C
T
)
i
ma
ge
s
c
a
n
[
2]
,
[
3
]
.
H
owe
ve
r
,
C
T
i
mag
e
s
c
a
ns
r
e
qu
i
r
e
ve
r
y
h
ig
h
r
a
d
ia
ti
on
a
n
d
a
r
e
e
x
pe
ns
i
ve
f
or
l
ow
-
i
nc
om
e
e
a
r
ne
r
s
,
e
s
pe
c
ia
l
ly
i
n
d
e
ve
l
op
in
g
c
ou
nt
r
ies
.
A
ls
o
,
t
he
s
e
i
ma
ge
s
ne
e
d
t
o
be
a
na
ly
z
e
d
b
y
a
s
p
e
c
ia
li
s
t
s
uc
h
a
s
a
r
a
di
ol
o
gis
t
or
a
ne
ph
r
ol
og
is
t
,
w
hi
c
h
r
e
qu
i
r
e
s
t
im
e
,
a
nd
is
la
bo
ur
-
i
n
te
ns
i
ve
[
9]
.
F
u
r
t
he
r
mo
r
e
,
a
C
T
s
c
a
n
ma
y
r
e
ve
a
l
a
k
id
ne
y
s
to
ne
,
b
ut
in
s
om
e
i
ns
t
a
n
c
e
s
,
a
s
pe
c
ia
li
s
t
m
a
y
no
t
no
t
ice
it
[
10
]
.
C
ons
e
qu
e
n
tl
y
,
c
om
pu
te
r
v
is
io
n
a
i
de
d
by
A
I
a
n
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Us
ing
R
e
s
N
e
t
-
50
pr
e
-
tr
ained
mode
l
to
impr
ov
e
the
c
las
s
if
ication
output
of
a
…
(
K
az
e
e
m
Oy
e
bode
)
3183
d
e
e
p
l
e
a
r
n
in
g
(
DL
)
,
is
now
us
e
d
to
a
u
to
ma
ti
c
a
ll
y
a
na
ly
z
e
a
nd
c
las
s
if
y
the
i
ma
ge
s
f
o
r
e
a
r
l
y
c
las
s
i
f
i
c
a
ti
on
[
3
]
,
[
10
]
,
[
1
1
]
.
How
e
v
e
r
,
s
ome
o
f
t
he
s
e
a
dv
a
nc
e
d
A
I
m
ode
ls
f
o
r
a
na
lyz
in
g
i
ma
ge
s
m
ig
ht
b
e
f
o
un
d
in
d
e
v
e
l
op
in
g
c
o
u
nt
r
ies
.
T
he
s
tudy
ther
e
f
o
r
e
f
oc
us
e
s
on
de
ve
lopi
ng
a
n
im
pr
ove
d
kidney
s
tone
de
tec
ti
on
s
ys
tem
that
leve
r
a
ge
s
a
pr
e
-
tr
a
ined
model.
B
ut
why
is
th
is
im
po
r
tant?
T
h
is
r
e
s
e
a
r
c
h
is
im
po
r
tant
be
c
a
us
e
it
pr
opo
s
e
s
a
c
os
t
-
e
f
f
e
c
ti
ve
a
ppr
oa
c
h
–
a
s
it
only
r
e
quir
e
s
a
ur
ine
s
a
mpl
e
,
unli
k
e
other
methods
s
uc
h
a
s
the
C
T
s
c
a
n
or
X
-
r
a
y.
One
of
the
goa
ls
of
thi
s
r
e
s
e
a
r
c
h
is
a
ls
o
to
im
p
r
ove
the
a
c
c
ur
a
c
y
of
kidney
s
tone
de
tec
ti
on
by
f
us
ing
a
non
-
im
a
ge
f
e
a
tur
e
int
o
a
n
i
mage
-
ba
s
e
d
pr
e
-
tr
a
ined
model.
T
his
de
mons
tr
a
tes
that
pr
e
-
tr
a
ined
models
a
r
e
ve
r
s
a
ti
le.
T
he
ne
xt
que
s
ti
on
to
a
s
k
is
the
model's
de
ve
lopm
e
nt.
T
his
r
e
quir
e
s
pr
ojec
ti
ng
the
s
ix
f
e
a
tur
e
s
e
li
c
it
e
d
f
r
om
t
he
ur
ine
s
a
mpl
e
a
nd
then
s
e
nding
them
int
o
172
,
800
ne
ur
o
ns
.
Af
ter
that
,
the
output
of
thi
s
is
r
e
s
ha
pe
d
a
nd
f
e
d
int
o
the
R
e
s
Ne
t
-
50
[
12]
a
nd
then
to
a
f
ull
y
c
onne
c
ted
laye
r
,
a
nd
then
to
the
c
las
s
if
ica
ti
on
laye
r
.
T
he
pa
pe
r
pr
opos
e
s
a
s
olut
ion
that
leve
r
a
ge
s
DL
to
a
na
lyze
ur
ine
s
a
mpl
e
s
f
r
om
humans
.
T
his
a
ppr
oa
c
h
is
vit
a
l
be
c
a
us
e
medic
a
l
e
xpe
r
ts
a
r
e
r
a
r
e
ly
p
r
e
s
e
nt
i
n
r
e
mot
e
plac
e
s
,
e
xc
e
pt
in
the
c
it
y
c
e
ntr
e
.
T
he
r
e
f
or
e
,
a
nother
objec
ti
ve
of
thi
s
r
e
s
e
a
r
c
h
is
to
de
ve
lop
a
de
e
p
-
lea
r
ning
model
that
c
las
s
if
ies
whe
ther
a
pa
ti
e
nt
ha
s
kidn
e
y
s
tones
ba
s
e
d
on
the
pa
ti
e
nt's
ur
ine
a
na
ly
s
is
.
T
he
r
e
a
f
ter
,
the
pa
ti
e
nt
c
ould
be
r
e
f
e
r
r
e
d
to
the
c
it
y
c
e
ntr
e
f
o
r
f
ur
ther
a
na
lys
is
,
s
uc
h
a
s
a
C
T
s
c
a
n.
R
e
s
e
a
r
c
he
r
s
ha
ve
a
ls
o
de
ve
loped
a
DL
model
f
or
kidney
s
tone
de
tec
ti
on
a
s
f
ound
in
[
13]
a
nd
a
ls
o
f
o
r
c
las
s
if
ica
ti
on
of
kidney
im
a
ge
s
[
14]
.
DL
models
a
r
e
de
ve
loped
to
r
e
f
lec
t
the
wa
ys
ne
ur
ons
a
r
e
f
i
r
e
d
in
the
human
b
r
a
in
[
15]
.
T
he
y
a
r
e
a
s
ubs
e
t
of
mac
hine
lea
r
ning,
whic
h
is
the
de
ve
lopm
e
nt
of
a
r
ti
f
icia
l
int
e
ll
igenc
e
(
A
I
)
s
ys
tems
with
out
p
r
e
c
is
e
ly
pr
ogr
a
mm
ing
the
mac
hine
on
how
to
lea
r
n
[
15]
.
T
he
y
ha
ve
r
e
c
or
de
d
s
uc
c
e
s
s
e
s
in
di
f
f
e
r
e
nt
f
ields
s
uc
h
a
s
tr
a
ns
por
tation
[
16]
,
he
a
lt
h
[
17
]
,
a
nd
a
gr
icultu
r
e
[
18
]
with
outs
tanding
pe
r
f
or
manc
e
s
.
F
or
e
x
a
mpl
e
,
Alba
r
a
ka
ti
e
t
al
.
[
18
]
,
a
c
las
s
if
ica
ti
on
model
of
la
nd,
f
r
om
im
a
ge
s
s
e
nt
f
r
om
r
e
mot
e
s
e
ns
or
s
,
is
im
p
leme
nted
us
ing
the
“
s
e
lf
-
a
tt
e
nti
on
mec
ha
ni
s
m”
[
19]
a
nd
DL
ne
twor
k.
T
he
objec
ti
ve
wa
s
to
de
ter
mi
ne
wha
t
f
a
r
m
pr
oduc
e
c
ould
do
we
ll
in
a
given
a
r
e
a
.
I
n
the
he
a
lt
h
s
e
c
tor
,
a
utom
a
ted
diagnos
is
tool
s
a
r
e
be
c
omi
ng
mo
r
e
us
e
f
ul
to
medic
a
l
pr
a
c
ti
ti
one
r
s
.
F
or
e
xa
mpl
e
,
in
r
e
mot
e
a
r
e
a
s
in
de
ve
lopi
ng
c
ount
r
ies
whe
r
e
ther
e
is
a
de
a
r
th
o
f
medic
a
l
doc
tor
s
a
nd
li
mi
ted
medic
a
l
r
e
s
our
c
e
s
,
nur
s
e
s
who
a
r
e
not
s
pe
c
ialis
ts
,
c
a
n
us
e
thes
e
a
s
s
i
s
ti
ve
tool
s
to
pr
ovide
the
f
i
r
s
t
li
ne
of
diagnos
is
a
nd
s
ometi
mes
,
e
ve
n
a
n
a
lt
e
r
na
ti
ve
diagnos
ti
c
pr
oc
e
dur
e
f
o
r
kidney
s
tone
d
e
tec
ti
on.
E
mer
ge
nc
y
r
oom
s
taf
f
c
a
n
a
ls
o
de
ploy
it
a
s
a
c
he
a
pe
r
,
f
a
s
ter
a
nd
mo
r
e
a
c
c
ur
a
te
diagnos
ti
c
pr
oc
e
dur
e
[3
]
.
T
his
de
ve
lopm
e
nt
c
a
n
potentially
s
a
ve
li
ve
s
,
e
s
pe
c
ially
whe
n
the
c
ondit
ion
is
de
tec
ted
e
a
r
ly.
De
laye
d
tr
e
a
t
ment
c
a
n
lea
d
to
r
e
na
l
f
a
il
ur
e
s
[
10
]
.
F
or
im
pr
ove
d
c
las
s
if
ica
ti
on
pr
e
-
tr
a
ined
models
ha
v
e
be
e
n
us
e
d
in
thi
s
s
tudy.
A
pr
e
-
tr
a
ined
model
is
a
model
that
ha
s
be
e
n
tr
a
ined
on
thous
a
nds
of
im
a
ge
s
,
s
uc
h
that
the
model
c
ould
be
us
e
d
in
dif
f
e
r
e
nt
do
mains
to
s
olve
a
pr
oblem
,
whic
h
c
ould
be
a
n
im
a
ge
c
las
s
if
i
c
a
ti
on
or
s
e
gmenta
ti
on
p
r
oblem
[
20]
.
P
r
e
vious
us
e
s
of
the
pr
e
-
tr
a
ined
model
ha
ve
be
e
n
f
o
r
land
c
las
s
if
ica
t
ion
on
im
a
ge
s
[
20
]
a
nd
r
e
mot
e
s
e
ns
ing
[
21
]
.
I
n
he
a
lt
h,
p
re
-
tr
a
ined
models
ha
ve
be
e
n
us
e
d
f
or
the
c
las
s
if
ica
ti
on
of
br
a
in
tum
our
s
[
22]
.
I
n
a
ddit
ion,
pr
e
-
tr
a
ine
d
models
ha
ve
be
e
n
us
e
d
f
or
s
e
gmenta
ti
on
with
im
pr
ove
d
ou
tcome
s
,
a
s
s
e
e
n
in
[
23]
.
I
n
thi
s
r
e
s
e
a
r
c
h,
we
de
mons
tr
a
ted
that
p
r
e
-
tr
a
ined
models
c
ould
be
us
e
d
in
non
-
im
a
ge
tas
ks
.
P
a
r
ti
c
ular
ly,
we
de
mons
tr
a
ted
it
s
c
a
pa
c
it
y
f
or
k
idney
s
tone
de
tec
ti
on
us
ing
s
ix
f
e
a
tur
e
s
f
r
om
the
Ka
ggle
da
tas
e
t
f
ound
in
[
24]
.
F
ir
s
t,
the
s
ix
-
f
e
a
tur
e
da
tas
e
t
is
c
onn
e
c
ted
to
172
,
800
ne
ur
ons
,
f
or
mi
ng
the
ba
s
is
to
r
e
s
ha
pe
the
output
of
thes
e
ne
ur
ons
int
o
a
240
by
240
by
3
inp
ut
s
ha
pe
by
us
ing
the
s
e
lec
ted
R
e
s
Ne
t
-
50
pr
e
-
tr
a
in
e
d
model.
Af
ter
r
e
s
ha
ping,
we
then
pa
s
s
e
d
it
int
o
the
R
e
s
Ne
t
-
50
pr
e
-
tr
a
ined
model
,
a
nd
then
it
s
output
is
f
latt
e
ne
d
a
nd
s
ubs
e
que
ntl
y
f
e
d
int
o
the
2048
laye
r
ne
ur
ons
,
a
nd
onwa
r
d
to
the
c
las
s
if
ica
ti
on
laye
r
f
o
r
bina
r
y
c
las
s
if
ica
ti
on.
T
his
idea
is
ins
pir
e
d
by
a
uthor
s
in
[
25]
,
[
26]
.
Our
pr
opos
e
d
model
c
ont
r
ibut
e
s
to
s
c
ienc
e
in
two
wa
ys
–
the
int
r
oduc
ti
on
of
a
pr
e
-
tr
a
ined
model
in
a
non
-
im
a
ge
tas
k,
a
nd
the
tr
a
ns
f
or
mation
of
the
s
ix
input
f
e
a
t
ur
e
s
int
o
im
a
ge
-
li
ke
da
ta
that
c
a
n
be
a
c
c
omm
oda
ted
by
a
pr
e
-
tr
a
ined
model.
S
tones
a
r
e
not
the
only
d
is
e
a
s
e
s
of
the
kidney.
Ot
he
r
s
include
c
onge
nit
a
l
a
bnor
malit
ies
,
c
a
nc
e
r
,
a
nd
obs
tr
uc
ti
on
of
the
ur
inar
y
t
r
a
c
t.
Kidne
y
s
tones
a
r
e
of
va
r
ious
types
,
whic
h
includ
e
“
c
a
lcium
oxa
late
s
to
ne
s
,
ur
ic
a
c
id
s
tones
,
c
a
lcium
phos
pha
te
s
tones
a
nd
s
tr
uvit
e
s
tones
”
[
27]
.
Dif
f
e
r
e
nt
im
a
ging
pr
oc
e
s
s
ing
tec
hniq
ue
s
s
uc
h
a
s
C
T
s
c
a
ns
,
X
-
r
a
ys
a
nd
ult
r
a
s
ounds
a
r
e
incr
e
a
s
ingl
y
be
ing
us
e
d
f
or
the
de
tec
ti
on
of
int
e
r
na
l
or
ga
n
a
nd
ti
s
s
ue
dis
or
de
r
s
.
T
he
s
e
im
a
ge
s
a
r
e
then
a
na
lyze
d
by
hum
a
n
s
pe
c
ialis
ts
,
whic
h
c
ould
s
ometim
e
s
lea
d
to
e
r
r
o
r
s
in
the
c
las
s
if
ica
ti
on
of
il
lnes
s
e
s
.
Als
o,
s
pe
c
kl
e
nois
e
is
pr
oduc
e
d
in
ult
r
a
s
ound
pictur
e
s
,
whic
h
incr
e
a
s
e
s
the
dif
f
iculty
in
the
manua
l
de
tec
ti
on
of
kidney
s
tones
.
T
hus
,
it
be
c
a
me
ne
c
e
s
s
a
r
y
to
de
ploy
a
utom
a
ted
tool
s
s
uc
h
a
s
im
a
ge
pr
oc
e
s
s
ing
togethe
r
with
mac
hine
lea
r
ning
a
lgor
it
h
ms
,
to
de
tec
t
a
nd
c
las
s
if
y
kidney
s
tones
[
10]
.
S
uppo
r
t
ve
c
tor
mac
hine
(
S
VM
)
c
a
n
be
us
e
d
to
c
las
s
if
y
kidney
s
to
ne
s
a
s
de
mons
tr
a
ted
in
[
10]
.
S
e
ve
r
a
l
de
e
p
-
lea
r
ning
models
a
r
e
be
ing
us
e
d
f
or
kidney
s
tone
de
tec
ti
on
us
ing
C
T
im
a
ge
s
.
I
n
their
s
tudy
[
28]
,
they
us
e
d
the
VG
G16
model
to
c
las
s
if
y
a
C
T
im
a
ge
,
a
nd
a
human
s
pe
c
ialis
t
to
e
ns
ur
e
the
a
c
c
ur
a
c
y
of
the
de
tec
ti
on
.
Anothe
r
DL
method
that
is
wi
de
ly
us
e
d
to
c
las
s
if
y
kidney
s
tones
us
ing
C
T
i
mage
s
is
c
onvolut
ional
ne
ur
a
l
ne
twor
ks
(
C
NN
s
)
[
9]
,
[
11]
,
[
29]
,
[
30]
.
S
ome
of
the
C
NN
va
r
iant
ne
twor
ks
include
I
nc
e
pti
onNe
t,
Google
Ne
t,
Ale
xNe
t
,
a
nd
I
mage
Ne
t
[
11]
.
I
r
uda
ya
r
a
j
[
1
]
us
e
d
f
our
DL
a
lgor
it
hms
f
or
c
las
s
if
ica
ti
on
(
VG
G16,
R
e
s
Ne
t
-
50V2,
M
obil
e
Ne
t
V2,
a
nd
I
nc
e
pti
onNe
tV3)
with
I
nc
e
pti
onNe
t
pr
od
uc
ing
the
mos
t
a
c
c
ur
a
te
r
e
s
ult
s
f
or
de
tec
ti
ng
ki
dne
y
s
tone
s
f
r
om
C
T
im
a
ge
s
.
W
hil
e
the
a
bove
s
tudi
e
s
s
h
owe
d
a
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
318
2
-
3191
3184
im
pr
ove
ment
ove
r
tr
a
dit
ional
methods
f
or
kidney
s
tone
de
tec
ti
on
by
us
ing
DL
a
lgor
it
hms
togethe
r
with
C
T
s
c
a
ns
,
ther
e
is
s
ti
ll
a
c
ha
ll
e
nge
r
e
ga
r
ding
the
huge
c
os
t
of
C
T
im
a
ge
s
f
or
pa
ti
e
nts
in
r
e
mot
e
a
r
e
a
s
of
Af
r
ica
a
nd
other
de
ve
lopi
ng
c
ountr
ies
.
W
h
i
le
ther
e
mi
ght
be
e
xpe
r
ts
a
nd
a
dva
nc
e
d
AI
C
T
im
a
ge
s
c
a
ns
,
they
a
r
e
us
ua
ll
y
f
ound
in
c
it
y
c
e
ntr
e
s
.
T
he
s
e
ga
ps
a
r
e
wha
t
o
ur
pr
o
pos
e
d
model
a
im
s
to
f
il
l.
T
h
is
r
e
s
e
a
r
c
h
a
i
ms
t
o
de
ve
l
op
a
DL
m
ode
l
t
ha
t
c
o
uld
po
te
nt
ia
ll
y
be
us
e
d
a
s
t
he
f
i
r
s
t
l
ine
of
di
a
g
nos
is
i
n
r
e
mo
te
a
r
e
a
s
wi
th
ou
t
r
e
qu
i
r
i
ng
a
s
pe
c
ia
li
s
t
.
How
e
ve
r
,
a
n
u
r
s
e
is
r
e
qu
i
r
e
d
.
T
he
n
ur
s
e
t
a
k
e
s
t
he
u
r
i
ne
s
a
m
p
le
o
f
a
p
a
t
ie
nt
a
n
d
the
n
c
a
r
r
ies
o
ut
th
e
r
e
qu
i
r
e
d
tes
ts
.
T
he
tes
ts
s
ho
ul
d
g
e
n
e
r
a
t
e
s
i
x
f
e
a
t
u
r
e
s
:
s
p
e
c
i
f
ic
g
r
a
vi
ty
,
ur
i
ne
pH
l
e
v
e
l
,
u
r
ine
os
mo
la
li
t
y
,
u
r
in
e
c
on
duc
t
iv
it
y
,
ur
e
a
,
a
n
d
u
r
i
ne
c
a
lci
u
m
.
T
h
e
s
e
f
e
a
t
u
r
e
s
a
r
e
the
n
pa
s
s
e
d
i
n
to
ou
r
d
e
v
e
l
op
e
d
A
I
mo
de
l
f
o
r
c
l
a
s
s
i
f
ica
t
io
n
.
I
f
the
mo
de
l
g
iv
e
s
a
p
os
i
t
ive
r
e
s
po
ns
e
,
th
e
pe
r
s
on
is
r
e
f
e
r
r
e
d
to
t
he
c
i
ty
c
e
n
t
r
e
f
o
r
a
c
om
p
r
e
h
e
ns
iv
e
AI
C
T
s
c
a
n
.
R
e
s
e
a
r
c
h
ha
s
b
e
e
n
u
nd
e
r
t
a
ke
n
on
t
he
i
de
nt
i
f
i
c
a
ti
on
o
f
k
id
ne
y
s
t
o
ne
s
v
ia
u
r
in
e
s
a
m
ples
.
A
n
e
xa
mp
le
is
wi
th
t
he
S
V
M
.
F
o
r
e
xa
m
p
le
,
B
a
l
bi
n
e
t
a
l
.
[
3
1
]
p
r
op
os
e
d
a
n
S
VM
s
ys
tem
t
ha
t
d
e
t
e
c
ts
c
a
l
c
i
um
in
u
r
i
ne
s
a
m
ple
s
.
T
he
c
o
mb
in
a
t
io
n
o
f
k
-
ne
a
r
e
s
t
ne
ig
hb
or
s
(
KNN
)
a
n
d
S
VM
wa
s
us
e
d
f
o
r
the
d
e
t
e
c
ti
on
o
f
s
to
ne
s
i
n
ki
dn
e
y
i
ma
ge
s
[
3
2]
.
A
b
r
a
ha
m
e
t
a
l
.
[
3
3
]
c
o
mp
a
r
e
d
t
he
p
e
r
f
o
r
m
a
nc
e
o
f
XG
B
oos
t
wi
th
l
og
is
t
ic
r
e
g
r
e
s
s
io
n
f
o
r
t
he
ide
n
ti
f
ica
ti
on
o
f
k
id
ne
y
s
t
on
e
s
us
i
n
g
pa
t
ien
ts
'
he
a
l
th
da
ta
a
nd
u
r
i
ne
s
a
mp
les
.
T
he
y
f
o
u
nd
o
u
t
t
ha
t
t
he
X
GB
o
os
t
ou
tp
e
r
f
o
r
m
e
d
the
lo
gis
t
ic
r
e
g
r
e
s
s
i
o
n
in
th
is
r
e
g
a
r
d
.
F
u
r
t
he
r
mo
r
e
,
Al
gha
m
di
a
nd
A
mo
u
di
[
3
4]
u
s
e
d
a
n
e
ns
e
mb
le
a
p
p
r
oa
c
h
th
a
t
e
nc
a
ps
ul
a
tes
t
he
r
a
n
d
om
f
o
r
e
s
t
(
R
F
)
f
o
r
ki
dne
y
s
to
ne
d
e
t
e
c
ti
on
.
2.
M
E
T
HO
D
I
n
de
ve
lopi
ng
the
pr
opos
e
d
model,
we
ha
d
to
s
e
e
k
a
publi
c
ly
a
va
il
a
ble
da
tas
e
t.
W
e
us
e
d
the
kidney
s
tone
da
tas
e
t
a
s
s
e
e
n
in
[
24]
.
T
he
da
tas
e
t
ha
s
s
ix
f
e
a
tur
e
s
:
s
pe
c
if
ic
gr
a
vit
ies
,
ur
ine
pH
leve
l,
u
r
ine
os
mol
a
li
ty,
ur
ine
c
onduc
ti
vit
y,
ur
e
a
,
a
nd
ur
ine
c
a
lcium.
T
he
s
e
ur
ine
f
e
a
tur
e
s
a
r
e
im
por
tant
f
o
r
the
f
or
mation
o
f
s
tones
in
the
kidney.
I
t
t
h
e
r
e
f
or
e
mea
ns
that
if
we
a
r
e
a
ble
to
de
ve
lop
a
nd
t
r
a
in
a
model
a
r
ound
thes
e
f
e
a
tur
e
s
,
o
ne
c
ould
a
utom
a
te
the
de
tec
ti
on
pr
oc
e
s
s
.
As
mentioned
e
a
r
li
e
r
,
the
p
r
opos
e
d
a
r
c
hit
e
c
tur
e
(
de
ve
loped
with
P
ytor
c
h)
us
e
s
a
DL
model
with
a
pr
e
-
t
r
a
ined
model.
I
n
thi
s
ins
tanc
e
,
we
us
e
d
the
R
e
s
Ne
t
-
50
pr
e
-
tr
a
ined
model.
On
e
r
e
a
s
on
we
c
hos
e
the
R
e
s
Ne
t
-
50
is
be
c
a
us
e
of
it
s
s
kip
c
on
ne
c
ti
ons
to
laye
r
s
e
a
r
li
e
r
to
mi
t
igate
the
va
nis
hing
gr
a
dient
pr
oblem
[
35
]
.
T
he
model
a
r
c
hit
e
c
tur
e
ha
s
be
e
n
i
ll
u
s
tr
a
ted
in
F
igur
e
1.
F
igur
e
1.
P
r
opos
e
d
c
las
s
if
ica
ti
on
model
T
he
pr
opos
e
d
model
a
s
s
hown
in
F
igur
e
1
ha
s
5
la
ye
r
s
L
1
to
L
5.
L
a
ye
r
1
(
L
1)
is
whe
r
e
the
6
f
e
a
tur
e
s
of
the
ur
ine
s
a
mpl
e
a
r
e
pa
s
s
e
d.
L
1
e
nc
a
ps
ulate
s
the
ba
tch
nor
maliza
ti
on
late
r
a
s
we
ll
a
s
the
r
e
c
ti
f
ied
li
ne
a
r
unit
(
R
e
L
U
)
laye
r
.
T
h
e
ba
tch
no
r
maliza
ti
on
no
r
malize
s
the
input
da
ta
s
uc
h
that
they
ha
ve
a
mea
n
of
z
e
r
o
to
a
c
hieve
a
n
e
f
f
icie
nt
tr
a
ini
ng
pr
o
c
e
s
s
[
36]
.
T
he
R
e
L
U
he
lps
lea
r
n
non
-
li
ne
a
r
a
tt
r
ibut
e
s
of
the
input
da
ta
[
37
]
.
T
he
output
of
L
1
is
f
e
d
int
o
a
nother
laye
r
,
whic
h
is
L
2,
with
1
72
,
800
(
3×
240×
240)
ne
u
r
ons
.
T
he
p
r
ojec
ted
ne
ur
o
ns
f
r
om
the
output
of
L
1
a
r
e
r
e
s
ha
pe
d
in
L
2
a
nd
then
f
e
d
i
nto
the
pr
e
-
tr
a
ined
model
r
e
s
idi
ng
in
L
2.
T
he
outp
ut
of
the
pr
e
-
tr
a
ined
model
is
then
s
e
nt
ove
r
to
L
3
whe
r
e
it
unde
r
goe
s
no
r
maliza
ti
on
a
s
we
ll
a
s
the
a
c
ti
va
ti
on
laye
r
–
R
e
L
U.
T
he
output
is
s
e
nt
to
L
4
whe
r
e
the
s
igm
oid
gives
a
n
output
withi
n
the
0
-
1.
T
he
o
u
t
pu
t
i
s
s
e
nt
t
o
L
5
f
or
binar
y
c
las
s
if
ica
ti
on.
W
e
tr
a
ined
the
model
on
60
e
poc
hs
,
lea
r
ning
r
a
te
of
0.
0001
with
a
ba
tch
s
ize
of
8
.
T
he
t
r
a
ini
ng
da
tas
e
t
f
ound
in
the
Ka
ggle
[
24]
wa
s
us
e
d
f
or
thi
s
m
ode
l
(
tr
a
in.
c
s
v)
.
T
he
tr
a
in.
c
s
v
f
il
e
ha
s
414
da
ta
point
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Us
ing
R
e
s
N
e
t
-
50
pr
e
-
tr
ained
mode
l
to
impr
ov
e
the
c
las
s
if
ication
output
of
a
…
(
K
az
e
e
m
Oy
e
bode
)
3185
T
his
t
r
a
in.
c
s
v
is
f
u
r
ther
s
pli
t
int
o
thr
e
e
c
a
tegor
ies
(
s
ubs
e
ts
)
.
T
he
s
e
s
ubs
e
ts
a
r
e
the
tr
a
ini
ng
s
e
t,
the
v
a
li
da
ti
on
s
e
t,
a
nd
the
tes
t
s
e
t.
T
he
t
r
a
ini
ng
da
tas
e
t
ha
s
291
da
ta
point
s
,
with
I
D
va
lues
f
r
om
123
to
413,
a
nd
the
va
li
da
ti
on
da
tas
e
t
ha
s
82
da
ta
point
s
,
with
I
Ds
r
a
nging
f
r
o
m
0
to
81.
L
a
s
tl
y,
the
tes
ti
ng
da
tas
e
t
ha
s
41
da
ta
point
s
with
a
n
I
D
r
a
nge
of
82
to
122
.
F
r
om
the
da
tas
e
t,
it
is
c
lea
r
that
the
tr
a
ini
ng
da
tas
e
t
take
s
a
bout
70
%
of
the
e
n
ti
r
e
da
ta
point
s
,
the
va
li
da
ti
on
da
tas
e
t
take
s
20
%
a
nd
then
t
he
tes
ti
ng
da
ta
take
s
10
%
.
I
t
is
c
r
uc
ial
to
note
that
the
be
s
t
model
with
the
lea
s
t
va
li
da
ti
on
e
r
r
or
is
s
a
ve
d
a
nd
then
us
e
d
to
e
va
luate
the
10
pe
r
c
e
nt
f
r
om
the
tes
t
da
tas
e
t.
T
o
va
li
da
te
the
pe
r
f
or
manc
e
of
the
p
r
opos
e
d
model,
we
r
e
pli
c
a
ted
two
va
r
iants
of
the
pr
opos
e
d
mod
e
l,
while
e
xc
ludi
ng
the
pr
e
-
tr
a
ined
model.
T
h
is
is
il
lus
tr
a
ted
in
F
i
gu
r
e
2
a
nd
it
s
hows
the
f
ir
s
t
va
r
iant
of
F
igu
r
e
1.
T
he
f
ir
s
t
va
r
iant
a
s
s
hown
in
F
igur
e
2
omi
ts
the
p
r
e
-
tr
a
ined
model.
T
his
model
a
im
s
to
e
s
tablis
h
the
r
e
leva
nc
e
of
the
p
r
opos
e
d
model
.
T
he
s
e
c
ond
va
r
iant
im
pleme
nts
a
no
r
mal
DL
model
without
a
n
e
leva
ted
number
of
ne
ur
ons
a
s
s
e
e
n
in
the
p
r
opos
e
d
model.
T
his
s
e
c
ond
va
r
iant
is
il
lus
tr
a
ted
in
F
igu
r
e
3
a
nd
pr
ovides
a
ba
lanc
e
d
pe
r
s
pe
c
ti
ve
on
the
s
igni
f
ica
nc
e
of
the
pr
o
pos
e
d
model.
T
he
s
e
c
ond
va
r
iant
a
s
s
hown
in
F
igu
r
e
3
a
ls
o
ha
s
6
input
s
jus
t
li
ke
the
pr
opos
e
d
model.
How
e
ve
r
,
it
doe
s
not
ha
ve
the
pr
e
-
tr
a
ined
model
.
T
he
s
e
c
o
nd
laye
r
is
e
quipped
with
20
0
ne
ur
ons
,
the
thi
r
d
(
L
3)
250
,
a
nd
the
f
ou
r
th
(
100,
32,
a
nd
1)
.
T
his
va
r
iant
a
ls
o
loo
ks
s
im
il
a
r
to
the
model
pr
opos
e
d
in
[
13]
,
howe
ve
r
,
in
thi
e
r
s
,
the
DL
model
ha
d
two
lay
e
r
s
e
xc
ludi
ng
the
inp
ut
laye
r
.
T
he
output
is
then
pa
s
s
e
d
onto
L
4.
I
n
a
ddit
ion,
we
a
ls
o
im
pl
e
mente
d
the
S
VM
a
nd
XG
B
oos
t
m
ode
ls
to
s
tr
e
ngthen
our
a
r
gument
.
F
igur
e
2.
DL
's
f
ir
s
t
va
r
iant
(
va
r
iant
one
)
F
igur
e
3
.
DL
s
e
c
ond
va
r
iant
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
318
2
-
3191
3186
3.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
W
e
c
a
r
r
ied
out
the
e
xpe
r
im
e
nt
(
tes
t)
on
the
10
%
da
tas
e
t
he
ld
ba
c
k
(
41
da
ta
point
s
)
,
a
s
dis
c
us
s
e
d
e
a
r
li
e
r
.
T
he
s
e
da
ta
point
s
we
r
e
pa
s
s
e
d
int
o
the
t
r
a
ined
models
in
ba
tche
s
of
8.
T
he
pr
e
dictions
of
a
ll
the
c
ons
ider
e
d
models
we
r
e
e
va
luate
d
us
ing
the
the
a
c
c
ur
acy
in
(
1)
,
a
r
e
a
unde
r
the
r
e
c
e
iver
ope
r
a
ti
ng
c
ha
r
a
c
ter
is
ti
c
c
ur
ve
(
AUC
-
R
OC
)
[
38]
,
a
nd
the
F
1
-
s
c
or
e
metr
ics
in
(
2)
to
e
va
luat
e
the
models
put
f
or
wa
r
d
.
T
he
A
UC
-
R
OC
qua
nti
f
ies
the
pe
r
f
or
manc
e
of
a
binar
y
c
las
s
if
ica
ti
on
model.
A
s
c
or
e
of
1
s
hows
the
model
pe
r
f
or
manc
e
is
100
%
while
a
pe
r
f
or
manc
e
of
0
%
mea
ns
the
model
pe
r
f
or
med
poor
ly
[
38]
.
T
his
va
lue
AUC
-
R
OC
indi
c
a
tes
the
tr
a
de
of
f
be
twe
e
n
the
t
r
ue
pos
it
ive
r
a
te
(
T
P
R
)
in
(
3
)
a
nd
the
f
a
ls
e
pos
it
ive
r
a
te
(
F
P
R
)
in
(
4)
a
t
va
r
ious
c
las
s
if
ica
ti
on
thr
e
s
holds
.
T
he
F
1
-
s
c
or
e
in
(
2)
is
a
metr
ic
that
s
tr
ikes
a
ba
lanc
e
be
twe
e
n
a
model's
pr
e
c
is
ion
in
(
5)
a
nd
r
e
c
a
ll
in
(
6)
.
T
r
ue
pos
it
ive
(
TP
)
,
tr
ue
n
e
g
a
ti
ve
(
TN
)
,
f
a
ls
e
ne
ga
ti
ve
(
FN
)
,
a
nd
f
a
ls
e
pos
it
ive
(
FP
).
A
ccu
r
a
cy
=
TP
+
TN
TP
+
TN
+
FN
+
FP
×
100%
(
1)
F1
−
s
co
r
e
=
P
r
e
c
i
s
i
on
×
R
e
c
a
ll
P
r
e
c
i
s
i
on
+
R
e
c
a
ll
×
100%
(
2)
T
P
R
=
TP
(
TP
+
FN
)
(
3)
F
P
R
=
FP
(
FP
+
TN
)
(
4)
P
r
e
cis
io
n
=
TP
(
TP
+
FP
)
(
5)
Re
ca
l
l
=
TP
(
TP
+
FN
)
(
6)
I
n
F
igur
e
4,
the
‘
tr
a
ini
ng
vs
va
li
da
ti
on
los
s
gr
a
ph’
s
hows
the
model
ove
r
f
it
s
a
s
s
e
e
n
f
r
om
the
va
li
da
ti
on
los
s
–
a
ga
p
e
xis
ts
be
twe
e
n
the
t
r
a
ini
ng
los
s
a
nd
the
v
a
li
da
ti
on
los
s
.
One
r
e
a
s
on
f
or
thi
s
is
that
the
tr
a
ini
n
g
da
tas
e
t
is
s
mall,
a
nd
ther
e
f
or
e
,
it
mi
gh
t
be
gin
to
“
memor
iz
e
”
[
39]
the
tr
a
in
ing
da
tas
e
t,
lea
ding
to
ove
r
f
it
ti
ng
.
How
e
ve
r
,
be
c
a
us
e
the
model
us
e
s
a
pr
e
-
tr
a
ined
model,
ther
e
is
the
pos
s
ibi
li
ty
to
ge
ne
r
a
li
z
e
,
ther
e
by
pe
r
f
or
mi
ng
we
ll
on
the
tes
t
da
tas
e
t.
F
igur
e
4.
T
r
a
ini
ng
a
nd
va
li
da
ti
on
los
s
vs
tr
a
ini
ng
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y
on
the
pr
opos
e
d
model
T
his
is
e
vident
in
T
a
ble
1
,
whe
r
e
it
a
c
hieve
s
a
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
of
70
.
242%
outper
f
or
mi
ng
a
ll
other
c
ons
ider
e
d
models
,
s
uc
h
a
s
the
DL
va
r
iants
one
a
nd
two,
a
s
we
ll
a
s
XG
B
oo
s
t.
How
e
ve
r
,
it
f
a
il
e
d
to
outper
f
or
m
the
S
VM
-
thi
s
is
be
c
a
u
s
e
S
VM
pe
r
f
or
ms
we
ll
on
s
mall
da
tas
e
t
[
40]
.
T
he
va
r
iant
two
r
e
f
lec
ts
the
DL
model
de
ve
loped
in
[
13]
,
howe
ve
r
,
with
only
tw
o
laye
r
s
,
whe
r
e
a
s
va
r
iant
two
ha
s
4
laye
r
s
.
Nothwid
s
tanding
it
did
not
outper
f
or
m
the
pr
opos
e
d
model
.
Anot
he
r
r
e
a
s
on
f
or
thi
s
pe
r
f
or
manc
e
(
p
r
opos
e
d
mode
l)
is
that
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Us
ing
R
e
s
N
e
t
-
50
pr
e
-
tr
ained
mode
l
to
impr
ov
e
the
c
las
s
if
ication
output
of
a
…
(
K
az
e
e
m
Oy
e
bode
)
3187
r
e
s
ha
ping
the
s
ix
ur
ine
f
e
a
tur
e
s
a
nd
pa
s
s
ing
them
int
o
the
pr
e
-
tr
a
ined
model
ha
s
he
lped
it
lea
r
n
dis
ti
nc
t
f
e
a
tur
e
s
a
nd
textur
e
s
that
ult
im
a
tely
i
mpr
ove
d
the
c
las
s
if
ic
a
ti
on
outcome
.
T
he
R
e
s
Ne
t
-
50
playe
d
a
vit
a
l
r
ole.
Anothe
r
c
ontr
ibut
ing
f
a
c
tor
c
ould
be
that
the
im
a
ge
s
us
e
d
t
o
tr
a
in
the
p
r
e
-
tr
a
ined
model
(
R
e
s
Ne
t
-
50)
a
li
gne
d
with
the
f
e
a
tur
e
s
of
the
u
r
ine
s
a
mpl
e
.
T
his
de
ve
lopm
e
nt
ha
s
ult
im
a
tely
he
lped
im
pr
ove
it
s
c
las
s
if
ica
ti
on
outcome
.
T
his
im
pr
ove
ment
is
a
ls
o
r
e
f
lec
ted
in
T
a
ble
2
,
f
o
r
t
he
F
1
-
s
c
or
e
of
67.
29
%
,
a
s
we
ll
a
s
a
n
AU
C
-
R
OC
of
0.
70288
in
T
a
ble
3
,
a
s
il
lus
tr
a
ted
in
F
igur
e
5
.
T
a
bl
e
1
.
Ac
c
u
r
a
c
y
M
ode
ls
A
c
c
ur
a
c
y (
%
)
P
r
opos
e
d
70.242
V
a
r
ia
nt
one
63.412
V
a
r
ia
nt
t
w
o
64.3
S
V
M
70.73
X
G
boos
t
60.98
T
a
ble
2.
F1
-
s
c
or
e
M
ode
ls
F1
-
s
c
or
e
(
%
)
P
r
opos
e
d
67.29
V
a
r
ia
nt
one
53.91
V
a
r
ia
nt
t
w
o
53.5
S
V
M
64.71
X
G
boos
t
57.89
T
a
ble
3.
AU
C
-
R
OC
M
ode
ls
AUC
-
R
O
C
P
r
opos
e
d
0.70288
V
a
r
ia
nt
one
0.66426
V
a
r
ia
nt
t
w
o
0.7005
S
V
M
0.6786
X
G
boos
t
0.6238
F
igur
e
5
.
R
OC
-
AU
C
f
or
the
pr
opos
e
d
model
M
oving
on
to
va
r
iant
one
,
it
did
not
ove
r
-
f
it
,
a
s
il
l
us
tr
a
ted
in
F
igur
e
6
.
T
his
is
due
to
the
a
bs
e
nc
e
of
the
pr
e
-
tr
a
ined
model
–
f
a
r
f
e
we
r
ne
ur
ons
than
th
e
pr
opos
e
d,
howe
ve
r
e
nough
ne
ur
ons
to
c
a
ptur
e
c
ompl
e
x
pa
tt
e
r
ns
to
de
li
v
e
r
im
p
r
ove
d
pe
r
f
or
manc
e
.
T
he
d
e
ve
lopm
e
nt
ga
ve
a
n
a
c
c
ur
a
c
y
s
c
or
e
of
63.
412
in
T
a
ble
1,
a
n
F
1
-
s
c
or
e
of
53.
91
in
T
a
ble
2.
T
he
s
e
c
ond
va
r
iant
wide
ns
the
ga
p
be
twe
e
n
the
tr
a
ini
ng
los
s
a
nd
the
v
a
li
da
ti
on
los
s
in
F
igur
e
7
.
One
e
xplana
ti
on
f
or
thi
s
c
ould
be
that
ther
e
a
r
e
not
e
nough
ne
ur
ons
to
lea
r
n
the
c
ompl
e
xit
ies
of
the
tr
a
ini
ng
da
ta
-
thi
s
mea
ns
the
li
mi
ted
ne
ur
ons
may
be
gin
to
“
memor
ize
”
[
39]
the
t
r
a
ini
ng
da
ta
le
a
ding
to
the
ove
r
-
f
it
ti
ng
in
F
igu
r
e
7.
T
he
f
i
r
s
t
va
r
iant
ha
s
a
n
AUC
-
R
O
C
in
F
igur
e
8
of
0
.
66426
in
T
a
ble
3.
T
he
R
OC
AU
C
f
igur
e
f
or
the
s
e
c
ond
va
r
iant
is
il
lus
tr
a
ted
in
F
igur
e
9.
T
he
R
OC
AU
C
f
igur
e
s
of
S
VM
a
nd
XG
boos
t
a
r
e
il
lus
tr
a
ted
in
F
igu
r
e
s
10
a
nd
11
.
F
igur
e
6
.
T
r
a
ini
ng
a
n
d
va
l
idation
los
s
vs
tr
a
ini
ng
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y
on
the
f
i
r
s
t
va
r
iant
(
va
r
iant
one
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
318
2
-
3191
3188
F
igur
e
7.
T
r
a
ini
ng
a
n
d
va
li
da
ti
on
los
s
vs
tr
a
ini
ng
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y
on
the
s
e
c
ond
va
r
iant
(
va
r
ia
nt
two)
F
igur
e
8
.
R
OC
-
AU
C
f
or
v
a
r
iant
one
F
igur
e
9
.
R
OC
-
AU
C
f
or
v
a
r
iant
two
F
igur
e
10
.
R
OC
-
AU
C
f
or
S
VM
F
igur
e
1
1
.
R
OC
-
AU
C
f
or
XG
boos
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Us
ing
R
e
s
N
e
t
-
50
pr
e
-
tr
ained
mode
l
to
impr
ov
e
the
c
las
s
if
ication
output
of
a
…
(
K
az
e
e
m
Oy
e
bode
)
3189
F
r
om
the
r
e
s
ult
pr
e
s
e
nted
a
bove
,
it
is
e
vident
that
t
he
R
e
s
Ne
t
-
50
im
pr
ove
d
the
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
of
the
model
pr
opos
e
d
–
us
ing
the
a
c
c
ur
a
c
y,
F
1
-
s
c
or
e
a
s
we
ll
a
s
the
R
OC
-
AU
C
metr
ic.
T
he
R
OC
-
AU
C
ge
ne
r
a
tes
a
plot
of
the
TP
r
a
te
a
nd
the
FP
r
a
te.
I
t
de
picts
a
model's
s
c
or
e
c
a
r
d.
T
he
T
P
R
gives
how
of
ten
a
model
pr
e
dicts
a
kidney
s
tone
dis
e
a
s
e
,
while
the
F
P
R
pr
ovides
how
of
ten
a
model
gives
a
wr
ong
c
las
s
if
ica
ti
on
of
ne
ga
ti
ve
ins
tanc
e
s
a
s
pos
it
ive
[
38
]
.
I
t
is
wor
th
noti
ng
that
the
pr
opos
e
d
model
ga
ve
the
highes
t
R
OC
-
AU
C
s
c
or
e
of
0.
70288
.
T
his
indi
c
a
tes
th
a
t
the
model
is
0
.
70288
e
f
f
e
c
ti
ve
out
of
a
m
a
xim
um
s
c
or
e
of
1
in
d
is
ti
nguis
hing
if
a
pa
t
ient
ha
s
a
k
idney
s
tone
or
n
ot.
Othe
r
c
ons
ider
e
d
models
a
r
e
be
low
the
0.
70288
mar
k.
A
s
c
or
e
of
0.
70288
c
lea
r
ly
r
e
d
uc
e
s
the
e
xtent
to
whic
h
a
model
mak
e
s
mi
s
c
l
a
s
s
i
f
ica
ti
on.
T
he
r
e
f
or
e
,
it
is
s
a
f
e
r
to
de
ploy
s
uc
h
a
model
in
a
r
e
a
l
-
li
f
e
s
c
e
na
r
io
than
to
de
ploy
a
ny
of
the
other
c
o
ns
ider
e
d
models
.
W
e
de
ployed
the
pr
opos
e
d
mo
de
l
on
H
uggingf
a
c
e
a
s
s
e
e
n
f
r
om
the
UR
L
-
htt
ps
:/
/huggi
ngf
a
c
e
.
c
o/s
pa
c
e
s
/Ka
z
e
e
mkz/Kidne
y
s
tone_de
tec
ti
on
us
ing
S
tr
e
a
ml
it
a
nd
Gr
a
dio
[
41]
.
F
o
r
im
pr
ove
ment,
mor
e
da
tas
e
t
ne
e
ds
to
c
oll
e
c
ted
to
a
s
s
is
t
the
model
to
lea
r
n
the
c
ompl
e
x
pa
r
tt
e
r
ns
of
ur
ine
f
e
a
tur
e
s
.
T
his
would
potentially
ha
ve
im
pr
ov
e
d
the
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y.
4
.
CONC
L
USI
ON
T
his
pa
pe
r
p
r
opos
e
s
a
pr
e
-
tr
a
ined
model
f
or
kidney
s
tone
c
las
s
if
ica
ti
on
ba
s
e
d
on
s
ix
f
e
a
tur
e
s
:
s
pe
c
if
ic
gr
a
vit
y,
ur
ine
pH
leve
l,
ur
ine
os
mol
a
li
ty,
u
r
ine
c
o
nduc
ti
vit
y,
ur
e
a
,
a
nd
c
a
lcium.
T
he
s
e
f
e
a
tur
e
s
a
r
e
pr
ojec
ted
int
o
a
higher
-
dim
e
ns
ional
s
pa
c
e
of
172
,
800
a
nd
t
he
n
r
e
s
ha
pe
d
int
o
a
240
by
240
by
3
a
r
r
a
y
to
f
it
the
input
s
ha
pe
of
t
he
s
e
lec
ted
pr
e
-
tr
a
ined
model
-
R
e
s
Ne
t
-
5
0.
T
he
output
of
the
p
r
e
-
tr
a
ined
model
is
pa
s
s
e
d
thr
ough
a
nother
laye
r
with
2048
ne
ur
ons
,
then
to
a
lay
e
r
with
32
ne
ur
ons
,
a
nd
f
inally
to
a
s
ingl
e
ne
ur
on
f
or
c
las
s
if
ica
ti
on.
W
e
e
va
luate
d
the
p
r
opos
e
d
model
a
ga
ins
t
two
other
va
r
iants
of
DL
models
the
S
V
M
a
nd
the
XG
boos
t
models
.
Us
ing
the
R
OC
-
AU
C
,
a
nd
the
F
1
-
s
c
or
e
metr
ics
it
is
e
s
tablis
he
d
that
the
model
pr
opos
e
d
outper
f
or
med
other
c
ons
ider
e
d
models
.
T
he
e
va
lu
a
ti
on
de
mons
tr
a
tes
that
a
n
im
a
ge
-
ba
s
e
d
pr
e
-
tr
a
ine
d
model
(
pr
opos
e
d)
c
ould
e
nha
nc
e
the
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y
of
n
on
-
im
a
ge
-
ba
s
e
d
da
tas
e
t
s
highl
ight
ing
the
po
tential
of
int
e
gr
a
ti
ng
he
ter
oge
ne
ous
da
ta
s
our
c
e
s
f
or
e
nha
nc
e
d
pr
e
dictive
a
c
c
ur
a
c
y.
F
or
im
pr
ove
ment
,
mo
r
e
da
ta
ne
e
ds
to
be
c
oll
e
c
ted
to
a
s
s
is
t
the
model
to
lea
r
n
the
c
om
plex
pa
r
tt
e
r
ns
of
ur
ine
f
e
a
tur
e
s
.
T
his
c
ou
ld
potenti
a
ll
y
ha
ve
im
pr
ov
e
d
the
c
las
s
if
ica
ti
on
a
c
c
ur
a
c
y.
F
UN
DI
NG
I
NF
ORM
AT
I
ON
No
f
unding
invo
lved.
AU
T
HO
R
CONT
RI
B
U
T
I
ONS
S
T
AT
E
M
E
N
T
T
his
jour
na
l
us
e
s
the
C
ontr
ibut
o
r
R
oles
T
a
xo
nomy
(
C
R
e
diT
)
to
r
e
c
ognize
indi
vidual
a
uthor
c
ontr
ibut
ions
,
r
e
duc
e
a
utho
r
s
hip
dis
putes
,
a
nd
f
a
c
il
it
a
te
c
oll
a
bor
a
ti
on.
Nam
e
of
Au
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Ka
z
e
e
m
Oye
bode
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Anne
Ngoz
i
Odoh
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
li
z
a
ti
on
M
:
M
e
th
odol
ogy
So
:
So
f
twa
r
e
Va
:
Va
li
da
ti
on
Fo
:
Fo
r
ma
l
a
na
ly
s
is
I
:
I
nve
s
ti
ga
ti
on
R
:
R
e
s
our
c
e
s
D
:
D
a
ta
C
ur
a
ti
on
O
:
W
r
it
in
g
-
O
r
ig
in
a
l
D
r
a
f
t
E
:
W
r
it
in
g
-
R
e
vi
e
w
&
E
di
ti
ng
Vi
:
Vi
s
ua
li
z
a
ti
on
Su
:
Su
pe
r
vi
s
io
n
P
:
P
r
oj
e
c
t
a
dmi
ni
s
tr
a
ti
on
Fu
:
Fu
ndi
ng a
c
qui
s
it
io
n
CONF
L
I
CT
OF
I
NT
E
RE
S
T
S
T
AT
E
M
E
N
T
Author
s
s
tate
no
c
onf
li
c
t
of
int
e
r
e
s
t.
DA
T
A
AV
AI
L
A
B
I
L
I
T
Y
T
he
da
ta
that
s
uppor
ts
the
f
ind
ings
of
th
i
s
s
tudy
is
ope
nly
a
va
il
a
ble
in
Ka
ggle
a
t
htt
ps
:/
/kaggle
.
c
om/
c
ompetit
ions
/pl
a
ygr
ound
-
s
e
r
ies
-
s
3e
12.
RE
F
E
RE
NC
E
S
[
1]
A
.
A
.
I
r
uda
ya
r
a
j,
“
K
id
ne
y
s
to
ne
de
t
e
c
ti
on
us
in
g
de
e
p
le
a
r
ni
ng
me
th
odol
ogi
e
s
,”
M
.Sc
.
P
r
oj
e
c
t
,
S
c
hool
of
C
omput
in
g
,
N
a
ti
ona
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
318
2
-
3191
3190
C
ol
le
ge
of
I
r
e
la
nd,
D
ubl
in
, I
r
e
la
nd,
2022.
[
2]
O
.
S
a
bunc
u
a
nd
B
.
B
il
ge
ha
n,
“
P
e
r
f
or
ma
nc
e
e
va
lu
a
ti
on
f
or
va
r
io
us
de
e
p
le
a
r
ni
ng
(
D
L
)
me
th
ods
a
ppl
ie
d
to
ki
dne
y
s
to
ne
di
s
e
a
s
e
s
,
”
in
2021
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
F
or
th
c
om
in
g
N
e
tw
or
k
s
and
Sus
ta
in
abi
li
t
y
in
A
I
oT
E
r
a,
F
oN
e
S
-
A
I
oT
2021
,
2021,
pp.
1
–
3
, doi
:
10.1109/F
oN
e
S
-
A
I
oT
54873.2021.00010.
[
3]
Y
.
Y
.
L
iu
,
Z
.
H
.
H
ua
ng,
a
nd
K
.
W
.
H
u
a
ng,
“
D
e
e
p
le
a
r
ni
ng
mode
l
f
or
c
omput
e
r
-
a
id
e
d
di
a
gnos
is
of
ur
ol
it
hi
a
s
is
de
t
e
c
ti
on
f
r
om
ki
dne
y
–
ur
e
te
r
–
bl
a
dde
r
i
ma
ge
s
,”
B
io
e
ngi
ne
e
r
in
g
, vol
. 9, no. 12,
2022, doi:
10.3390/bi
oe
ngi
ne
e
r
in
g9120811.
[
4]
M
.
A
lm
us
a
f
e
r
e
t
al
.
,
“
U
nve
il
in
g
th
e
bur
de
n
of
ne
phr
ol
it
hi
a
s
is
in
lo
w
-
a
nd
lo
w
e
r
-
mi
ddl
e
-
in
c
ome
c
ount
r
ie
s
:
a
r
e
vi
e
w
o
n
it
s
pr
e
s
e
nt
a
ti
on,
r
is
k
f
a
c
to
r
s
,
tr
e
a
tm
e
nt
pr
a
c
ti
c
e
s
,
a
nd
f
ut
ur
e
di
r
e
c
ti
o
ns
,”
Soc
ié
té
I
nt
e
r
nat
io
nal
e
d’
U
r
ol
ogi
e
J
our
nal
,
vol
.
5,
no
.
5,
pp. 361
–
370, Oc
t.
2024, doi:
10.3390/s
iu
j5
050055.
[
5]
T
.
A
le
li
gn
a
nd
B
.
P
e
tr
os
,
“
K
id
ne
y
s
to
ne
di
s
e
a
s
e
:
a
n
upda
te
on
c
ur
r
e
nt
c
onc
e
pt
s
,”
A
dv
anc
e
s
in
U
r
ol
ogy
,
vol
.
6,
no.
2,
pp.
28
–
35,
2018, doi:
10.1155/2018/
3068365.
[
6]
C
.
D
.
S
c
a
le
s
,
A
.
C
.
S
mi
th
,
J
. M
.
H
a
nl
e
y, a
nd
C
.
S
.
S
a
ig
a
l,
“
P
r
e
va
le
nc
e
of
ki
dne
y
s
to
n
e
s
in
th
e
U
ni
te
d
S
ta
te
s
,”
E
ur
op
e
an
U
r
ol
ogy
,
vol
. 62, no. 1, pp. 160
–
165, J
ul
. 2012, doi:
10.1016/j
.e
ur
ur
o.20
12.03.052.
[
7]
N
.
B
or
uma
ndni
a
e
t
al
.
,
“
L
ongi
tu
di
na
l
tr
e
nd
of
ur
ol
it
hi
a
s
is
in
c
id
e
nc
e
r
a
te
s
a
mong
w
or
ld
c
ount
r
ie
s
dur
in
g
pa
s
t
de
c
a
de
s
,
”
B
M
C
U
r
ol
ogy
, vol
. 23, no. 1, Oc
t.
2023, doi:
10.1186/s
12894
-
023
-
01336
-
0.
[
8]
J
.
L
i,
Y
.
Z
ha
o,
Z
.
X
io
ng,
a
nd
G
.
Y
a
ng,
“
G
lo
ba
l,
r
e
gi
ona
l,
a
nd na
ti
ona
l
in
c
id
e
nc
e
a
nd
di
s
a
bi
li
ty
-
a
dj
us
te
d
li
f
e
-
gl
oba
l,
r
e
gi
ona
l,
a
nd,
na
ti
ona
l
in
c
id
e
nc
e
a
nd
di
s
a
bi
li
ty
-
a
dj
us
te
d
li
f
e
-
ye
a
r
s
f
or
ur
ol
it
hi
a
s
is
in
195
c
ount
r
ie
s
a
nd
te
r
r
it
or
ie
s
,
1990
–
2019:
r
e
s
ul
ts
f
r
om
th
e
gl
oba
l
bur
de
n of
di
s
e
a
s
e
s
tu
d,”
J
our
nal
of
C
li
ni
c
al
M
e
di
c
in
e
, v
ol
. 12,
no. 3, J
a
n. 2023, doi:
10.3390/j
c
m12031048.
[
9]
A
.
C
a
gl
a
ya
n,
M
.
O
.
H
or
s
a
na
li
,
K
.
K
oc
a
dur
du,
E
.
I
s
ma
il
ogl
u,
a
nd
S
.
G
une
yl
i,
“
D
e
e
p
le
a
r
ni
ng
mode
l
-
a
s
s
is
te
d
de
te
c
ti
on
of
ki
d
ne
y
s
to
ne
s
on
c
omput
e
d
to
mogr
a
phy,”
I
nt
e
r
nat
io
nal
B
r
az
il
ia
n
J
our
nal
of
U
r
ol
ogy
,
vol
.
48
,
no.
5,
pp.
830
–
839,
2022,
doi
:
10.1590/S
1677
-
5538.I
B
J
U
.2022.0132.
[
10]
P
.
K
.
M
in
,
D
.
B
ha
tt
a
c
ha
r
yya
,
B
.
C
.
M
in
,
T
.
H
.
K
i
m,
a
nd
K
.
M
it
o,
“
E
nha
nc
e
d
ki
dne
y
s
to
ne
id
e
nt
if
ic
a
ti
on
us
in
g
ul
t
r
a
s
onogr
a
phi
c
im
a
ge
s
in
im
a
ge
pr
oc
e
s
s
in
g,”
I
nt
e
r
nat
io
nal
J
our
nal
of
I
nt
e
ll
ig
e
nt
Sy
s
te
m
s
and
A
ppl
ic
at
io
ns
in
E
ngi
ne
e
r
in
g
,
vol
.
12,
no.
4,
pp. 477
–
484, 2024.
[
11]
B
.
R
e
ube
n
a
nd
C
.
N
a
r
ma
dha
,
“
E
f
f
e
c
ti
ve
ki
dne
y
s
to
ne
pr
e
di
c
ti
on
ba
s
e
d
on
opt
im
iz
e
d
Y
O
L
O
v7
s
e
gm
e
nt
a
ti
on
a
nd
de
e
p
le
a
r
n
in
g
c
la
s
s
if
ic
a
ti
on,”
I
nt
e
r
nat
io
nal
J
ou
r
nal
of
I
nt
e
l
li
ge
nt
Sy
s
te
m
s
and
A
ppl
ic
at
io
ns
i
n E
ngi
ne
e
r
in
g
, vol
. 12, no. 1, pp. 183
–
192, 2024.
[
12]
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
ng
f
o
r
im
a
ge
r
e
c
ogni
ti
on,”
in
P
r
oc
e
e
di
ngs
of
th
e
I
E
E
E
C
om
put
e
r
Soc
ie
ty
C
onf
e
r
e
nc
e
on C
om
put
e
r
V
is
io
n and P
at
te
r
n R
e
c
ogni
ti
on
, 2016, pp. 770
–
778
, doi
:
10.1109/C
V
P
R
.2016.90.
[
13]
A
.
J
a
r
ghon
a
nd
S
.
S
.
A
bu
-
N
a
s
e
r
,
“
P
r
e
di
c
ti
ng
ki
dne
y
s
to
ne
pr
e
s
e
nc
e
f
r
om
ur
in
e
a
na
ly
s
is
:
a
ne
ur
a
l
ne
twor
k
a
ppr
oa
c
h
us
in
g
J
N
N
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
A
c
ade
m
ic
I
nf
or
m
at
io
n Sy
s
te
m
s
R
e
s
e
a
r
c
h
, vol
. 7, no. 9, pp. 32
–
39, 2023.
[
14]
J
.
R
ubi
a
,
S
.
S
hi
bi
,
B
.
L
in
c
y,
J
.
P
.
C
a
th
e
r
in
,
V
ig
ne
s
h
w
a
r
a
n,
a
nd
E
.
N
it
hi
la
,
“
A
ut
oma
ti
c
ki
dne
y
di
s
e
a
s
e
pr
e
di
c
ti
on
u
s
in
g
de
e
p
l
e
a
r
n
in
g
te
c
hni
que
s
,”
I
ndone
s
ia
n
J
ou
r
nal
of
E
le
c
tr
ic
al
E
ngi
ne
e
r
in
g
a
nd
C
om
put
e
r
Sc
ie
nc
e
,
vol
.
36,
no.
3,
pp.
1798
–
1806,
2024,
doi
:
10.11591/i
je
e
c
s
.v36.i3.pp1798
-
1806.
[
15]
Y
. L
e
C
un, Y. B
e
ngi
o, a
nd G
. H
in
to
n, “
D
e
e
p l
e
a
r
ni
ng,”
N
at
u
r
e
,
vol
. 521, no. 7553, pp. 436
–
444, 2015
,
doi
:
10.1038/natur
e
14539
.
[
16]
X
. Y
in
, G
. W
u, J
. W
e
i,
Y
. S
he
n, H
. Q
i,
a
nd B
. Y
in
, “
D
e
e
p l
e
a
r
ni
ng on tr
a
f
f
ic
p
r
e
di
c
ti
on:
me
th
ods
, a
na
ly
s
is
, a
nd f
ut
ur
e
di
r
e
c
t
io
ns
,”
I
E
E
E
T
r
ans
ac
ti
ons
on
I
nt
e
ll
ig
e
nt
T
r
ans
por
ta
ti
on
Sy
s
te
m
s
,
vol
.
23,
no.
6,
pp.
4927
–
4943,
J
un.
2022,
doi
:
10.1109/T
I
T
S
.2021.3054840.
[
17]
Q
.
R
a
n
a
nd
D
.
H
a
n,
“
D
e
s
ig
n
of
me
nt
a
l
he
a
lt
h
c
ons
ul
ta
ti
on
s
ys
te
m
ba
s
e
d
on
de
e
p
le
a
r
ni
ng
a
lg
or
it
hm,”
in
2023
I
nt
e
r
n
at
io
nal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
Sc
ie
nc
e
and
A
ut
om
at
io
n
T
e
c
hnol
ogy
(
C
SA
T
)
,
O
c
t.
2023,
pp.
257
–
262
,
doi
:
10.1109/C
S
A
T
61646.2023.00074.
[
18]
H
.
M
.
A
lb
a
r
a
ka
ti
e
t
al
.
,
“
A
nove
l
de
e
p
le
a
r
ni
ng
a
r
c
hi
te
c
tu
r
e
f
or
a
gr
ic
ul
tu
r
e
la
nd
c
ove
r
a
nd
la
nd
us
e
c
la
s
s
if
ic
a
ti
on
f
r
o
m
r
e
m
ot
e
s
e
ns
in
g
im
a
ge
s
ba
s
e
d
on
ne
twor
k
-
le
ve
l
f
us
io
n
of
s
e
lf
-
a
tt
e
nt
io
n
a
r
c
hi
te
c
tu
r
e
,”
I
E
E
E
J
our
nal
of
Se
le
c
te
d
T
opi
c
s
in
A
ppl
ie
d
E
ar
th
O
bs
e
r
v
at
io
n
s
and R
e
m
ot
e
Se
ns
in
g
, vol
. 17, pp. 6338
–
6353, 202
4, doi:
10.1109/J
S
T
A
R
S
.2024.3369950.
[
19]
A
.
V
a
s
w
a
ni
e
t
al
.
,
“
A
tt
e
nt
io
n
is
a
ll
you
ne
e
d,”
in
31s
t
C
onf
e
r
e
nc
e
on
N
e
ur
al
I
nf
or
m
at
io
n
P
r
oc
e
s
s
in
g
Sy
s
te
m
s
(
N
I
P
S
2017)
,
2
017,
pp. 1
–
11.
[
20]
B
. Z
ha
o, B
. H
ua
ng, a
nd Y
. Z
hong, “
T
r
a
ns
f
e
r
l
e
a
r
ni
ng w
it
h f
ul
ly
pr
e
tr
a
in
e
d de
e
p c
onvolut
io
n ne
twor
ks
f
or
l
a
nd
-
us
e
c
la
s
s
if
ic
a
ti
on,”
IE
E
E
G
e
os
c
ie
nc
e
and R
e
m
ot
e
Se
n
s
in
g L
e
tt
e
r
s
, vol
. 14, no. 9, pp. 1436
–
1440, S
e
p. 2017, doi:
10.1109/L
G
R
S
.2017.2691013.
[
21]
Z
.
H
ua
ng,
M
.
Z
ha
ng,
Y
.
G
ong,
Q
.
L
iu
,
a
nd
Y
.
W
a
ng,
“
G
e
ne
r
ic
knowle
dge
boos
te
d
pr
e
tr
a
in
in
g
f
or
r
e
mot
e
s
e
ns
in
g
im
a
ge
s
,”
I
E
E
E
T
r
ans
ac
ti
ons
on G
e
o
s
c
ie
nc
e
and R
e
m
ot
e
Se
ns
in
g
, vol
. 62, pp. 1
–
13, 2024, doi:
10.1109/T
G
R
S
.2024.3354031.
[
22]
K
.
M
uha
mm
a
d,
S
.
K
ha
n,
J
.
D
e
l
S
e
r
,
a
nd
V
.
H
.
C
.
de
A
lb
uqu
e
r
que
,
“
D
e
e
p
le
a
r
ni
ng
f
or
mul
ti
gr
a
de
br
a
in
tu
mor
c
la
s
s
if
ic
a
ti
on
in
s
m
a
r
t
he
a
lt
hc
a
r
e
s
ys
te
ms
:
a
pr
os
pe
c
ti
ve
s
ur
ve
y,”
I
E
E
E
T
r
ans
ac
ti
ons
on
N
e
ur
al
N
e
t
w
or
k
s
and
L
e
ar
ni
ng
S
y
s
te
m
s
,
vol
.
32,
no.
2,
pp.
5
07
–
522, F
e
b. 2021, doi:
10.1109/T
N
N
L
S
.2020.2995800.
[
23]
W
.
C
a
i,
L
.
X
ie
,
W
.
Y
a
ng,
Y
.
L
i,
Y
. G
a
o,
a
nd
T
.
W
a
ng,
“
D
F
T
N
e
t:
dua
l
-
pa
th
f
e
a
tu
r
e
t
r
a
ns
f
e
r
ne
twor
k
f
o
r
w
e
a
kl
y
s
upe
r
vi
s
e
d
me
di
c
a
l
im
a
ge
s
e
gme
nt
a
ti
on,”
I
E
E
E
/A
C
M
T
r
ans
ac
ti
ons
on
C
om
put
at
io
nal
B
io
lo
gy
and
B
io
in
fo
r
m
at
ic
s
,
vol
.
20,
no.
4,
pp.
2530
–
2540, J
ul
.
2023, doi:
10.1109/T
C
B
B
.2022.3198284.
[
24]
W
.
R
e
a
d
e
a
nd
A
.
C
how
,
“
B
in
a
r
y
c
la
s
s
if
ic
a
ti
on
w
it
h
a
ta
bul
a
r
k
id
ne
y
s
t
one
pr
e
di
c
ti
on
da
ta
s
e
t,
”
K
aggl
e
,
2023.
[
O
nl
in
e
]
.
A
va
il
a
bl
e
:
ht
tp
s
:/
/k
a
ggl
e
.c
om/
c
ompe
ti
ti
ons
/p
la
ygr
ound
-
s
e
r
ie
s
-
s
3e
12
[
25]
A
. S
ha
r
ma
, E
. V
a
ns
,
D
. S
hi
ge
mi
z
u,
K
. A
.
B
or
oe
vi
c
h, a
nd
T
.
T
s
unoda
, “
D
e
e
pI
n
s
ig
ht
:
a
me
th
odol
ogy to
t
r
a
ns
f
or
m a
non
-
im
a
ge
da
ta
to
a
n
im
a
ge
f
or
c
onvolut
io
n
ne
ur
a
l
ne
twor
k
a
r
c
hi
te
c
tu
r
e
,”
Sc
ie
nt
if
ic
R
e
por
ts
,
vol
.
9,
no.
1,
A
ug.
2019,
do
i:
10.1038/s
41598
-
0
19
-
477
65
-
6.
[
26]
Y
.
Z
hu
e
t
al
.
,
“
C
onv
e
r
ti
ng
ta
bul
a
r
da
ta
in
to
im
a
ge
s
f
or
de
e
p
le
a
r
ni
ng
w
it
h
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
,”
S
c
ie
nt
if
ic
R
e
por
ts
,
vol
.
11, no. 1, M
a
y 2021, doi:
10.1038/s
41598
-
021
-
90923
-
y.
[
27]
W
.
Z
hu,
R
.
Z
hou,
Y
.
Y
a
o,
T
.
C
a
mpb
e
ll
,
R
.
J
a
in
,
a
nd
J
.
L
uo,
“
S
e
gP
r
ompt
:
us
in
g
s
e
gme
nt
a
ti
on
ma
p
a
s
a
be
tt
e
r
pr
ompt
to
f
in
e
tu
ne
de
e
p mode
ls
f
or
ki
dne
y s
to
ne
c
la
s
s
if
ic
a
ti
on,”
i
n
P
r
oc
e
e
di
ngs
of
M
ac
hi
ne
L
e
ar
ni
ng R
e
s
e
ar
c
h
, 2023, vol. 227, pp.
1680
–
1690.
[
28]
B
.
M
a
noj
,
N
.
M
oha
n,
S
.
K
.
S
,
a
nd
S
.
K
.
P
,
“
A
ut
oma
te
d
de
te
c
ti
on
of
ki
dne
y
s
to
ne
us
in
g
de
e
p
le
a
r
ni
ng
mode
ls
,”
in
2022
2nd
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
I
nt
e
ll
ig
e
nt
T
e
c
hnol
ogi
e
s
(
C
O
N
I
T
)
,
H
ubl
i,
I
ndi
a
,
J
un.
2022,
pp.
1
–
5
,
doi
:
10.1109/C
O
N
I
T
55038.2022.9847894.
[
29]
K
.
M
.
B
la
c
k,
H
.
L
a
w
,
A
.
A
ld
oukhi,
J
.
D
e
ng,
a
nd
K
.
R
. G
h
a
ni
,
“
D
e
e
p
le
a
r
ni
ng
c
omput
e
r
vi
s
io
n
a
lg
or
it
hm f
or
de
te
c
ti
ng
ki
dne
y
s
t
one
c
ompos
it
io
n,”
B
J
U
I
nt
e
r
nat
io
nal
, vol
. 125, no. 6, pp. 920
–
924, J
un. 2020, doi:
10.1111/bj
u.15035.
[
30]
P
.
S
.
R
a
me
s
h,
S
.
P
a
te
l,
K
.
D
.
B
a
ma
ne
,
Y
.
U
s
ha
R
a
ni
,
M
.
T
iwa
r
i,
a
nd
T
.
K
a
r
th
ik
e
ya
n
,
“
A
ut
oma
ti
c
ki
dne
y
s
to
ne
de
te
c
ti
on
us
in
g
d
e
e
p
le
a
r
ni
ng me
th
od,”
J
our
nal
of
A
dv
anc
e
d Z
ool
ogy
, vol
. 44, no. 4,
pp. 100
–
109, Nov. 2023, doi:
10.17762/j
a
z
.v44iS
4.2176.
[
31]
J
.
J
.
R
.
B
a
lb
in
,
G
.
V
.
M
a
gw
il
i,
L
.
D
.
V
a
li
e
nt
e
,
D
.
L
.
B
.
G
a
w
a
r
a
n,
N
.
E
.
R
.
L
uma
pa
s
,
a
n
d
A
.
M
.
U
ma
li
,
“
D
e
te
c
ti
on
a
nd
id
e
nt
if
ic
a
t
io
n
of
tr
ip
le
phos
pha
te
c
r
ys
ta
ls
a
nd
c
a
lc
iu
m
oxa
la
te
c
r
ys
ta
l
s
in
hum
a
n
ur
in
e
s
e
di
me
nt
us
in
g
ha
a
r
f
e
a
tu
r
e
,
a
d
a
pt
iv
e
boos
ti
ng,
a
nd
s
up
por
t
ve
c
to
r
ma
c
hi
ne
vi
a
ope
n
C
V
,”
in
P
r
oc
e
e
di
ngs
of
th
e
2020
10t
h
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
B
io
m
e
di
c
al
E
ngi
ne
e
r
in
g
and
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
Us
ing
R
e
s
N
e
t
-
50
pr
e
-
tr
ained
mode
l
to
impr
ov
e
the
c
las
s
if
ication
output
of
a
…
(
K
az
e
e
m
Oy
e
bode
)
3191
T
e
c
hnol
ogy
, S
e
p. 2020, pp. 34
–
39
, doi
:
10.1145/3397391.33
97415.
[
32]
J
.
V
e
r
ma
,
M
.
N
a
th
,
P
.
T
r
ip
a
th
i,
a
nd
K
.
K
.
S
a
in
i,
“
A
na
ly
s
is
a
nd
id
e
nt
if
ic
a
ti
on
of
ki
dne
y
s
to
ne
us
in
g
K
th
ne
a
r
e
s
t
ne
ig
hbour
(
K
N
N
)
a
nd
s
uppor
t
ve
c
to
r
ma
c
hi
ne
(
S
V
M
)
c
la
s
s
if
ic
a
ti
on
te
c
hni
qu
e
s
,”
P
at
te
r
n
R
e
c
ogni
ti
on
and
I
m
age
A
nal
y
s
is
,
vol
.
27,
no
.
3,
pp. 574
–
580, J
ul
. 2017, doi:
10.1134/S
1054661817030294.
[
33]
A
. A
br
a
ha
m, N. L
. K
a
vou
s
s
i,
W
. S
ui
, C
.
B
e
ja
n,
J
. A
.
C
a
pr
a
,
a
n
d R
. H
s
i,
“
M
a
c
hi
ne
l
e
a
r
ni
ng pr
e
di
c
ti
on
of
ki
dne
y s
to
ne
c
ompos
i
ti
on
us
in
g
e
le
c
tr
oni
c
he
a
lt
h
r
e
c
or
d
-
de
r
iv
e
d
f
e
a
tu
r
e
s
,”
J
our
nal
o
f
E
ndour
ol
ogy
,
vol
.
36,
no.
2,
pp.
243
–
250,
F
e
b.
2022,
doi
:
10.1089/e
nd.2021.0211.
[
34]
H
.
A
lg
ha
mdi
a
nd
G
.
A
moudi
,
“
U
s
in
g
ma
c
hi
ne
le
a
r
ni
ng
f
or
non
-
in
va
s
iv
e
de
te
c
ti
on
of
ki
dne
y
s
to
ne
s
ba
s
e
d
on
la
bor
a
to
r
y
te
s
t
r
e
s
u
lt
s
:
a
c
a
s
e
s
tu
dy f
r
om a
S
a
udi
A
r
a
bi
a
n hos
pi
ta
l,
”
D
ia
gnos
ti
c
s
, vol
.
14, no. 13, J
un. 2024, doi:
10.3390/di
a
gnos
ti
c
s
14131343.
[
35]
Z
. D
ua
n, F
. W
a
ng, B
.
W
a
ng, G
. L
uo, a
nd Z
. J
ia
ng, “
A
n
a
da
pt
e
d
R
e
s
N
e
t
-
50 a
r
c
hi
te
c
tu
r
e
f
or
pr
e
di
c
ti
ng f
lo
w
f
ie
ld
s
of
a
n unde
r
w
a
te
r
ve
hi
c
le
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 12, pp. 66398
–
66407, 2024, doi:
10.1109/AC
C
E
S
S
.2024.3399077.
[
36]
K
.
H
e
,
X
.
Z
ha
ng,
S
.
R
e
n,
a
nd
J
.
S
un,
“
D
e
lv
in
g
de
e
p
in
to
r
e
c
ti
f
ie
r
s
:
s
ur
pa
s
s
in
g
huma
n
-
le
ve
l
pe
r
f
or
ma
nc
e
on
I
ma
g
e
N
e
t
c
l
a
s
s
if
ic
a
ti
on,”
in
2015 I
E
E
E
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on C
om
put
e
r
V
is
io
n (
I
C
C
V
)
, D
e
c
. 2015, pp. 1026
–
1034
, doi
:
10.1109/I
C
C
V
.2015.12
3.
[
37]
V
.
M
.
V
a
r
ga
s
,
P
.
A
.
G
ut
ié
r
r
e
z
,
J
.
B
a
r
be
r
o
-
G
óme
z
,
a
nd
C
.
H
e
r
vá
s
-
M
a
r
tí
ne
z
,
“
A
c
ti
va
ti
on
f
unc
ti
on
s
f
or
c
onvolut
io
na
l
ne
ur
a
l
n
e
two
r
ks
:
pr
opos
a
ls
a
nd
e
xpe
r
im
e
nt
a
l
s
tu
dy,”
I
E
E
E
T
r
ans
ac
ti
ons
on
N
e
ur
al
N
e
tw
o
r
k
s
and
L
e
ar
ni
ng
Sy
s
t
e
m
s
,
vol
.
34,
no.
3,
pp. 1478
–
1488,
M
a
r
. 2023, doi:
10.1109/T
N
N
L
S
.2021.3105444.
[
38]
J
.
A
.
H
a
nl
e
y
a
nd
B
. J
.
M
c
N
e
il
,
“
A
me
th
od
of
c
ompa
r
in
g
th
e
a
r
e
a
s
unde
r
r
e
c
e
iv
e
r
ope
r
a
ti
ng
c
ha
r
a
c
t
e
r
is
ti
c
c
ur
ve
s
de
r
iv
e
d
f
r
om
t
he
s
a
me
c
a
s
e
s
,
”
R
adi
ol
ogy
, vol
. 148, no. 3, pp. 839
–
843, 1983, doi:
10.1148/r
a
di
ol
ogy.148.3.6878708.
[
39]
C
. Z
ha
ng, B
. R
e
c
ht
, S
. B
e
ngi
o,
M
. H
a
r
dt
, a
nd
O
. V
in
ya
l
s
, “
U
nd
e
r
s
ta
ndi
ng de
e
p l
e
a
r
ni
ng r
e
qui
r
e
s
r
e
th
in
ki
ng ge
ne
r
a
li
z
a
ti
on,”
i
n
5t
h
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on L
e
ar
ni
ng R
e
pr
e
s
e
nt
at
io
ns
, I
C
L
R
2017
-
C
onf
e
r
e
nc
e
T
r
a
c
k
P
r
oc
e
e
di
ngs
, 2017.
[4
0]
C.
-
W
. H
s
u, C
.
-
C
. C
ha
ng, a
nd C
.
-
J
. L
in
, “
A
pr
a
c
ti
c
a
l
gui
de
t
o s
uppor
t
ve
c
to
r
c
la
s
s
if
ic
a
ti
on,”
B
J
U
i
nt
e
r
nat
io
nal
, vol
. 101, no. 1
, pp.
1396
–
1400, 2008.
[
41]
K
.
O
ye
bode
,
“
K
id
ne
y
s
to
ne
de
te
c
ti
on
not
e
-
f
or
r
e
s
e
a
r
c
h
pur
pos
e
onl
y
-
,”
H
uggi
ng
F
ac
e
-
Spac
e
s
.
[
O
nl
in
e
]
.
A
va
il
a
bl
e
:
ht
tp
s
:/
/h
uggi
ngf
a
c
e
.c
o/
s
pa
c
e
s
/Ka
z
e
e
mkz
/Ki
dne
y
s
to
ne
_de
te
c
ti
o
n
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Ka
zeem
O
y
ebo
de
h
o
l
d
s
a
Ph
.
D
.
i
n
co
m
p
u
t
er
en
g
i
n
e
eri
n
g
w
i
t
h
a
s
p
eci
a
l
i
za
t
i
o
n
i
n
co
mp
u
t
er
v
i
s
i
o
n
.
H
e
al
s
o
rece
i
v
e
d
a
Mas
t
er
o
f
Sci
en
ce
(
M.
Sc.
)
i
n
s
o
ft
w
are
en
g
i
n
eer
i
n
g
fro
m
t
h
e
U
n
i
v
er
s
i
t
y
o
f
t
h
e
W
e
s
t
o
f
E
n
g
l
a
n
d
.
H
e
co
m
p
l
e
t
ed
h
i
s
p
o
s
t
d
o
ct
o
ral
re
s
earch
at
T
s
h
w
a
n
e
U
n
i
v
er
s
i
t
y
o
f
T
ec
h
n
o
l
o
g
y
,
So
u
t
h
A
fr
i
ca,
i
n
2
0
1
8
.
H
e
h
as
d
e
v
el
o
p
e
d
s
o
l
u
t
i
o
n
s
i
n
t
h
e
fi
e
l
d
o
f
co
mp
u
t
er
v
i
s
i
o
n
fo
r
s
e
i
s
m
i
c
v
o
l
u
me
an
al
y
s
i
s
an
d
s
ecu
ri
t
y
ap
p
l
i
cat
i
o
n
s
.
H
e
t
each
e
s
at
t
h
e
Sch
o
o
l
o
f
Sci
en
ce
an
d
T
ech
n
o
l
o
g
y
a
t
Pan
-
A
t
l
a
n
t
i
c
U
n
i
v
er
s
i
t
y
,
L
ag
o
s
,
N
i
g
eri
a,
w
h
ere
h
e
al
s
o
co
n
d
u
ct
s
res
earch
.
H
i
s
res
earch
areas
o
f
i
n
t
eres
t
i
n
c
l
u
d
e
co
mp
u
t
e
r
v
i
s
i
o
n
a
n
d
i
mag
e
p
ro
ce
s
s
i
n
g
.
H
e
can
b
e
co
n
t
act
e
d
at
emai
l
:
k
o
y
eb
o
d
e@
p
a
u
.
ed
u
.
n
g
.
A
nne
N
g
o
zi
Od
o
h
h
o
l
d
s
a
Bach
e
l
o
r
o
f
Sci
e
n
ce
(B.
Sc.
)
i
n
co
m
p
u
t
er
s
c
i
en
ce
fr
o
m
t
h
e
U
n
i
v
ers
i
t
y
o
f
Ben
i
n
,
N
i
g
eri
a.
Sh
e
al
s
o
rece
i
v
e
d
a
Ma
s
t
er
o
f
B
u
s
i
n
e
s
s
A
d
mi
n
i
s
t
ra
t
i
o
n
(MB
A
)
an
d
a
D
o
c
t
o
ra
t
e
i
n
Bu
s
i
n
es
s
A
d
mi
n
i
s
t
ra
t
i
o
n
fr
o
m
t
h
e
L
a
g
o
s
Bu
s
i
n
es
s
Sch
o
o
l
an
d
t
h
e
U
n
i
v
ers
i
t
y
o
f
Brad
f
o
rd
,
U
n
i
t
e
d
K
i
n
g
d
o
m
re
s
p
ec
t
i
v
el
y
.
Sh
e
i
s
c
u
rr
en
t
l
y
i
s
a
s
e
n
i
o
r
facu
l
t
y
memb
e
r
at
t
h
e
Sch
o
o
l
o
f
Med
i
a
an
d
Co
mm
u
n
i
cat
i
o
n
,
Pan
-
A
t
l
an
t
i
c
U
n
i
v
ers
i
t
y
i
n
L
ag
o
s
,
N
i
g
er
i
a,
w
h
ere
s
h
e
faci
l
i
t
a
t
es
co
u
rs
e
s
i
n
d
i
g
i
t
al
b
u
s
i
n
e
s
s
an
d
t
ec
h
n
o
l
o
g
y
man
ag
eme
n
t
,
n
e
w
med
i
a
an
d
mark
e
t
i
n
g
co
mmu
n
i
ca
t
i
o
n
at
t
h
e
g
rad
u
at
e
a
n
d
u
n
d
erg
ra
d
u
a
t
e
l
e
v
el
s
.
H
er
res
earc
h
i
n
t
ere
s
t
s
i
n
cl
u
d
e
d
i
g
i
t
a
l
t
ran
s
fo
rma
t
i
o
n
,
art
i
f
i
ci
a
l
i
n
t
el
l
i
g
en
ce
,
an
d
i
n
f
o
rmat
i
o
n
s
y
s
t
ems
.
Sh
e
can
b
e
co
n
t
act
e
d
at
emai
l
:
ao
d
o
h
@
p
a
u
.
ed
u
.
n
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.