I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
5
5
6
~
3
5
6
7
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
5
5
6
-
3
5
6
7
3556
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
A ma
chine l
ea
rni
ng
f
ra
mewo
rk
for
skin cancer
clas
si
ficatio
n
using
t
e
x
ture
desc
riptors
Sh
ik
ha
M
a
lik
1
,
Va
ibh
a
v
V.
Dix
it
2
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
G
.
H
.
R
a
i
so
n
i
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
M
a
n
a
g
e
me
n
t
,
S
a
v
i
t
r
i
b
a
i
P
h
u
l
e
P
u
n
e
U
n
i
v
e
r
si
t
y
,
P
u
n
e
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
T
e
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
,
R
a
si
k
l
a
l
M
.
D
h
a
r
i
w
a
l
S
i
n
h
g
a
d
Te
c
h
n
i
c
a
l
I
n
st
i
t
u
t
e
s
C
a
m
p
u
s,
S
a
v
i
t
r
i
b
a
i
P
h
u
l
e
P
u
n
e
U
n
i
v
e
r
si
t
y
,
P
u
n
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
9
,
2
0
2
5
R
ev
is
ed
J
u
n
2
3
,
2
0
2
6
Acc
ep
ted
J
u
l 9
,
2
0
2
6
S
k
in
c
a
n
c
e
r
re
m
a
in
s
a
c
rit
ica
l
h
e
a
lt
h
a
n
d
e
c
o
n
o
m
ic
c
o
n
c
e
rn
wo
rld
wi
d
e
.
Ti
m
e
ly
a
n
d
a
c
c
u
ra
te
d
iag
n
o
sis
i
s
c
ru
c
ial
fo
r
imp
r
o
v
in
g
m
o
r
talit
y
ra
te
o
f
p
a
ti
e
n
ts.
Alt
h
o
u
g
h
a
u
t
o
m
a
ted
m
a
c
h
in
e
lea
rn
i
n
g
(
M
L)
m
o
d
e
ls
a
ss
ist
d
o
c
t
o
rs
in
c
li
n
ica
l
a
ss
e
ss
m
e
n
t
o
f
sk
in
c
a
n
c
e
r
fro
m
d
e
rm
o
sc
o
p
ic
ima
g
e
s,
th
e
ir
p
e
rfo
rm
a
n
c
e
o
ften
su
ffe
rs
fro
m
e
x
trem
e
c
las
s
imb
a
lan
c
e
,
a
s
b
e
n
ig
n
ima
g
e
sa
m
p
les
g
re
a
tl
y
e
x
c
e
e
d
m
a
li
g
n
a
n
t
o
n
e
s.
Th
is
lea
d
s
to
fa
lse
d
e
t
e
c
ti
o
n
a
n
d
d
e
lay
e
d
d
ia
g
n
o
sis
o
f
sk
i
n
c
a
n
c
e
r.
To
o
v
e
rc
o
m
e
th
is
issu
e
,
th
e
stu
d
y
p
ro
p
o
se
s
a
n
e
f
ficie
n
t
a
n
d
l
ig
h
twe
ig
h
t
g
e
o
m
e
tri
c
tran
sfo
rm
a
ti
o
n
(G
T)
–
a
u
g
m
e
n
ted
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
(
S
VM)
fra
m
e
wo
rk
f
o
r
e
a
rly
d
iag
n
o
si
s
o
f
sk
in
c
a
n
c
e
r.
It
e
ffe
c
ti
v
e
ly
a
d
d
re
ss
e
s
th
e
c
las
s
imb
a
lan
c
e
issu
e
s
p
re
s
e
n
t
in
th
e
d
a
tas
e
ts
a
n
d
imp
r
o
v
e
s
t
h
e
d
e
tec
ti
o
n
o
f
p
o
sit
iv
e
c
a
se
s.
Th
e
n
o
v
e
l
p
ip
e
li
n
e
in
teg
ra
tes
p
re
p
ro
c
e
ss
in
g
,
m
o
r
p
h
o
l
o
g
y
p
re
se
rv
in
g
GT
,
ro
tatio
n
-
in
v
a
rian
t
tex
tu
re
fe
a
tu
re
e
x
trac
ti
o
n
,
fe
a
tu
re
v
a
li
d
a
ti
o
n
,
a
n
d
fe
a
tu
re
sc
a
li
n
g
f
o
r
p
e
rfo
rm
in
g
b
i
n
a
ry
c
las
sifica
ti
o
n
u
sin
g
a
n
o
p
ti
m
ize
d
S
VM
fra
m
e
wo
rk
.
T
h
is
fra
m
e
wo
rk
p
ro
v
id
e
s
b
a
lan
c
e
d
a
n
d
a
c
c
u
ra
te
les
io
n
c
las
sifica
ti
o
n
e
v
e
n
if
ima
g
e
sa
m
p
les
a
re
in
s
u
fficie
n
t.
Ex
p
e
rime
n
tal
re
su
lt
s
h
a
v
e
s
u
c
c
e
ss
fu
ll
y
a
c
h
iev
e
d
a
tru
e
p
o
sit
iv
e
ra
te
(TP
R)
o
f
9
3
%
o
n
P
H
2
a
n
d
8
1
.
1
%
o
n
In
tern
a
ti
o
n
a
l
S
k
i
n
Im
a
g
i
n
g
Co
l
lab
o
ra
ti
o
n
2
0
1
6
(
IS
IC
-
2
0
1
6
)
d
a
tas
e
t,
wh
ic
h
is
b
e
tt
e
r
th
a
n
c
o
n
v
e
n
ti
o
n
a
l
M
L
m
o
d
e
ls.
T
h
e
se
fin
d
i
n
g
s
p
r
o
v
e
t
h
a
t
p
ro
p
o
se
d
GT
-
S
VM
fra
m
e
wo
rk
is
a
l
ig
h
t
we
ig
h
t,
in
ter
p
re
tab
le,
a
n
d
c
o
m
p
u
tatio
n
a
ll
y
e
fficie
n
t
a
p
p
r
o
a
c
h
fo
r
e
a
rly
s
k
in
c
a
n
c
e
r
d
iag
n
o
sis.
F
u
t
u
re
d
e
v
e
l
o
p
m
e
n
ts
sh
a
ll
e
x
p
lo
re
m
u
lt
i
-
c
las
s
les
io
n
c
las
si
fica
ti
o
n
,
v
a
li
d
a
ti
o
n
a
c
ro
ss
d
i
v
e
rse
c
li
n
ica
l
d
a
tas
e
ts,
a
n
d
h
y
b
rid
fe
a
t
u
re
fu
si
o
n
.
K
ey
w
o
r
d
s
:
Geo
m
etr
ic
tr
an
s
f
o
r
m
atio
n
au
g
m
en
tatio
n
Gr
ay
lev
el
co
-
o
cc
u
r
r
en
ce
m
atr
ix
Sk
in
ca
n
ce
r
class
if
icatio
n
Su
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
T
ex
tu
r
e
f
ea
tu
r
es
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Sh
ik
h
a
Ma
lik
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
T
elec
o
m
m
u
n
icatio
n
G.
H.
R
aiso
n
i Co
lleg
e
o
f
E
n
g
i
n
ee
r
in
g
a
n
d
Ma
n
a
g
em
en
t,
Sav
itrib
ai
Ph
u
le
Pu
n
e
Un
iv
e
r
s
ity
New
Gate
No
.
1
2
0
0
,
Do
m
k
h
el
R
o
ad
,
W
ag
h
o
li,
Pu
n
e
4
1
2
2
0
7
,
Ma
h
ar
ash
tr
a,
I
n
d
ia
E
m
ail:
s
h
ik
h
a.
m
alik
.
p
h
d
etc@g
h
r
ce
m
.
r
aiso
n
i.n
et
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
is
in
g
s
k
in
ca
n
ce
r
in
cid
e
n
ce
is
a
cr
itical
h
ea
lth
an
d
ec
o
n
o
m
ic
co
n
ce
r
n
w
o
r
ld
wid
e.
T
h
e
d
is
ea
s
e
p
r
im
ar
ily
s
tem
s
f
r
o
m
th
e
e
p
i
d
er
m
is
(
o
u
ter
m
o
s
t)
s
tr
atu
m
o
f
th
e
s
k
in
.
T
h
is
o
cc
u
r
s
d
u
e
to
d
am
ag
e
t
o
ce
llu
lar
d
eo
x
y
r
ib
o
n
u
cleic
ac
id
(
DNA
)
th
at
lead
s
to
g
en
etic
m
u
tatio
n
s
an
d
u
n
r
eg
u
lated
ce
ll
m
u
ltip
licatio
n
.
Sk
in
ca
n
ce
r
ex
is
ts
as
ac
u
te
m
alig
n
an
t
m
elan
o
m
a
(
ME
L
)
an
d
th
e
m
o
r
e
f
r
e
q
u
en
t
n
o
n
-
m
elan
o
m
a
v
ar
ieties
s
u
ch
as
b
asal
ce
l
l
ca
r
cin
o
m
a
(
B
C
C
)
an
d
s
q
u
am
o
u
s
ce
ll
ca
r
cin
o
m
a
(
SC
C
)
.
Glo
b
al
C
an
ce
r
Statis
tics
r
ep
o
r
ted
n
ea
r
ly
1
9
.
3
m
illi
o
n
n
ew
p
atien
ts
an
d
o
v
er
1
0
m
illi
o
n
ca
n
ce
r
f
atalities
in
2
0
2
0
[
1
]
.
Me
lan
o
m
a
s
k
in
ca
n
c
er
h
as
th
e
h
ig
h
est
m
o
r
tality
r
ate
b
ec
au
s
e
it
m
et
astas
ize
s
an
d
s
ev
er
ely
a
f
f
ec
t
s
s
u
r
r
o
u
n
d
in
g
tis
s
u
es
[
2
]
.
E
p
id
em
io
lo
g
ical
d
ata
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
ma
ch
in
e
lea
r
n
i
n
g
fr
a
mewo
r
k
fo
r
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
u
s
in
g
textu
r
e
d
escr
ip
to
r
s
(
S
h
ikh
a
Ma
lik
)
3557
h
ig
h
lig
h
t
t
h
at
m
elan
o
m
a
in
cid
en
ce
s
ar
e
h
ig
h
er
i
n
m
e
n
th
an
i
n
wo
m
en
.
T
h
e
f
o
r
ec
asts
in
d
icate
th
at
if
th
e
s
am
e
cu
r
r
en
t tr
e
n
d
s
ex
is
t,
th
en
m
ela
n
o
m
a
ca
s
es c
o
u
ld
r
ea
ch
5
1
0
,
0
0
0
with
a
g
r
o
wth
r
ate
o
f
6
8
% b
y
2
0
4
0
[
3
]
.
I
n
a
clin
ical
s
ce
n
ar
io
,
th
e
d
ia
g
n
o
s
is
o
f
s
k
in
ca
n
ce
r
is
m
ad
e
b
y
d
e
r
m
ato
lo
g
is
ts
b
y
p
er
f
o
r
m
in
g
v
is
u
al
in
s
p
ec
tio
n
o
f
th
e
lesi
o
n
s
alo
n
g
with
th
e
in
d
iv
id
u
al’
s
h
is
to
r
y
,
ag
e,
lo
ca
tio
n
,
an
d
m
o
r
p
h
o
lo
g
ical
f
ea
tu
r
es
o
f
th
e
lesi
o
n
s
.
T
h
e
ac
cu
r
ac
y
o
f
c
lin
ical
d
iag
n
o
s
is
b
y
th
is
m
eth
o
d
,
h
o
wev
er
,
b
ec
o
m
es
lo
w
d
u
e
to
b
r
o
ad
m
o
r
p
h
o
lo
g
ical
v
ar
iatio
n
s
am
o
n
g
th
e
lesi
o
n
s
.
Ar
tific
ial
in
tel
lig
en
ce
(
AI
)
an
d
im
a
g
e
p
r
o
ce
s
s
in
g
to
o
ls
ca
n
b
e
v
a
l
u
a
b
l
e
o
p
t
i
o
n
s
f
o
r
a
s
s
i
s
t
i
n
g
d
e
r
m
a
t
o
l
o
g
i
s
t
s
i
n
b
e
t
t
e
r
,
m
o
r
e
r
e
l
i
a
b
l
e
,
a
n
d
t
i
m
e
l
y
d
i
a
g
n
o
s
i
s
o
f
s
k
i
n
c
a
n
c
e
r
[
4
]
,
[
5
]
.
E
v
en
with
th
ese
ad
v
an
ce
s
,
cl
ass
im
b
alan
ce
in
d
er
m
o
s
co
p
i
c
s
k
in
ca
n
ce
r
d
atasets
r
em
ain
s
th
e
k
ey
lim
itatio
n
d
u
r
in
g
au
to
m
ated
d
iag
n
o
s
is
,
as
th
e
b
en
i
g
n
lesi
o
n
s
am
p
les
g
r
ea
tly
ex
ce
ed
m
alig
n
a
n
t
o
n
es.
T
h
is
im
b
alan
ce
ex
is
ts
in
s
ev
er
al
p
u
b
lic
r
ep
o
s
i
to
r
ies
s
u
ch
as
I
n
ter
n
atio
n
al
S
k
in
I
m
ag
in
g
C
o
llab
o
r
atio
n
2
0
1
6
(
I
SIC
-
2
0
1
6
)
an
d
PH
2
,
wh
ich
ca
u
s
es
AI
m
o
d
els
to
p
r
ef
e
r
m
aj
o
r
ity
b
en
ig
n
class
es
an
d
p
r
o
d
u
ce
lo
w
tr
u
e
p
o
s
itiv
e
r
ates
(
T
PR
)
f
o
r
m
in
o
r
ity
m
alig
n
an
t
lesi
o
n
s
[
6
]
.
As
a
r
esu
lt,
th
e
r
is
in
g
n
u
m
b
er
o
f
f
alse
n
eg
ativ
e
ca
s
es
is
d
elay
in
g
tim
ely
tr
ea
tm
en
t
an
d
a
d
v
er
s
ely
af
f
ec
tin
g
th
e
s
u
r
v
iv
al
o
f
p
atien
ts
.
R
esear
ch
er
s
ex
p
lo
r
ed
s
ev
er
al
m
ac
h
in
e
lear
n
i
n
g
(
ML
)
m
eth
o
d
s
as
an
au
to
m
ate
d
ap
p
r
o
ac
h
to
s
k
in
ca
n
ce
r
d
ete
ctio
n
.
T
h
is
in
clu
d
es
i
)
clin
ically
d
r
iv
en
m
eth
o
d
s
,
ii)
h
an
d
cr
af
ted
f
ea
tu
r
e
-
b
ased
class
if
icatio
n
,
an
d
iii)
d
ee
p
le
ar
n
in
g
ap
p
r
o
ac
h
es.
T
h
e
f
o
llo
win
g
s
ec
tio
n
r
ev
iews
k
ey
ad
v
a
n
ce
s
in
th
ese
ar
ea
s
,
h
ig
h
lig
h
tin
g
th
eir
s
h
o
r
tco
m
in
g
s
an
d
id
en
tify
i
n
g
th
e
r
esear
ch
g
ap
s
th
at
m
o
tiv
ated
th
e
d
ev
elo
p
m
e
n
t
o
f
th
e
p
r
o
p
o
s
ed
g
eo
m
etr
ic
tr
an
s
f
o
r
m
atio
n
(
GT
)
–
au
g
m
en
ted
s
u
p
p
o
r
t
v
ec
t
o
r
m
ac
h
in
e
(
SVM)
f
r
am
ewo
r
k
f
o
r
ca
n
ce
r
class
if
icatio
n
.
Du
r
in
g
clin
ical
d
iag
n
o
s
is
o
f
s
k
in
ca
n
ce
r
,
th
e
AB
C
D
r
u
le
is
f
o
llo
wed
,
wh
ich
m
ea
s
u
r
es
v
al
u
es
o
f
f
o
u
r
m
ain
p
a
r
am
eter
s
:
asy
m
m
etr
y
(
A)
i
n
p
ig
m
en
ted
lesi
o
n
s
,
ir
r
eg
u
lar
ities
in
b
o
r
d
er
s
(
B
)
,
v
ar
iatio
n
s
in
lesi
o
n
co
lo
r
s
(
C
)
,
a
n
d
d
iam
eter
(
D)
o
f
lesi
o
n
s
.
T
h
e
m
alig
n
an
t
le
s
io
n
s
co
m
m
o
n
l
y
h
a
v
e
ir
r
e
g
u
l
ar
s
h
ap
es,
u
n
ev
en
b
o
r
d
er
s
,
n
o
n
-
u
n
if
o
r
m
co
l
o
r
s
p
r
ea
d
,
a
n
d
a
m
in
im
u
m
6
m
m
d
iam
eter
.
T
h
ese
d
ia
g
n
o
s
is
cr
iter
ia
h
av
e
b
ee
n
co
m
m
o
n
l
y
im
p
lem
e
n
ted
i
n
s
t
u
d
ies
in
v
o
l
v
in
g
m
an
u
al
m
eth
o
d
s
o
f
s
k
in
ca
n
ce
r
d
etec
tio
n
.
An
AB
C
D
clin
ical
p
r
o
to
ty
p
e
s
y
s
tem
p
r
o
p
o
s
ed
in
[
7
]
ac
h
iev
ed
9
2
.
2
2
% sp
ec
if
icity
,
9
0
%
ac
cu
r
ac
y
,
a
n
d
8
5
% sen
s
itiv
ity
o
n
th
e
PH
2
d
ataset.
An
o
th
er
s
im
ilar
r
esear
ch
[
8
]
in
v
o
lv
e
d
th
e
AB
C
D
p
ar
am
eter
s
a
n
d
t
o
tal
d
er
m
o
s
co
p
y
s
co
r
e
(
T
DS)
ca
lcu
latio
n
s
,
ac
h
iev
in
g
8
4
%
a
cc
u
r
ac
y
an
d
6
0
.
5
% sen
s
itiv
ity
f
o
r
m
elan
o
m
a
d
etec
tio
n
.
B
o
th
ap
p
r
o
ac
h
es d
id
n
o
t
em
p
lo
y
ML
m
eth
o
d
s
an
d
u
tili
ze
d
o
n
ly
class
ical
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
th
at
p
r
o
v
i
d
ed
o
n
ly
ac
ce
p
tab
le
ac
cu
r
ac
y
.
T
h
is
m
a
k
es
it
d
if
f
icu
lt
f
o
r
th
e
m
o
d
el
to
wo
r
k
ef
f
ec
tiv
ely
o
n
a
wid
e
s
ca
le
with
co
m
p
le
x
d
atasets
.
T
h
ese
clin
ical
ap
p
r
o
ac
h
es
d
o
n
o
t
in
clu
d
e
au
to
m
atic
f
ea
tu
r
e
lear
n
in
g
a
n
d
lack
th
e
ab
ili
ty
to
h
an
d
le
class
im
b
alan
ce
o
r
h
eter
o
g
en
e
o
u
s
d
ata
s
am
p
les.
C
o
n
s
eq
u
e
n
tly
,
r
ec
en
t
s
tu
d
ies
h
av
e
f
o
c
u
s
ed
o
n
ad
v
a
n
ce
d
ap
p
r
o
ac
h
es
s
u
ch
as
ML
an
d
d
ee
p
lear
n
in
g
to
s
tr
en
g
th
e
n
cli
n
ical
p
er
f
o
r
m
an
ce
an
d
im
p
r
o
v
e
g
en
er
aliza
tio
n
o
n
n
ew
d
ata.
Featu
r
e
en
g
i
n
ee
r
in
g
tech
n
iq
u
es
h
av
e
b
ee
n
em
p
l
o
y
ed
in
s
ev
er
al
wo
r
k
s
f
o
r
t
h
e
e
x
tr
ac
tio
n
o
f
c
o
lo
r
,
s
h
ap
e,
an
d
tex
tu
r
e
attr
ib
u
tes.
W
ar
s
i
et
a
l
.
[
9
]
d
ev
elo
p
ed
a
t
h
r
ee
-
d
im
e
n
s
io
n
al
an
aly
s
is
to
o
l
th
at
ex
tr
ac
ts
co
lo
r
an
d
tex
tu
r
e
in
f
o
r
m
atio
n
f
r
o
m
th
e
PH
2
d
ataset
b
y
u
tili
zin
g
r
ed
,
g
r
ee
n
,
an
d
b
lu
e
(
R
GB
)
co
lo
r
ch
an
n
els.
T
h
e
y
class
if
ied
f
ea
tu
r
es
u
s
in
g
a
p
r
o
p
o
s
ed
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
(
ML
P)
alg
o
r
ith
m
.
T
h
e
r
esu
lts
o
f
th
e
s
im
u
latio
n
s
h
o
w
th
at
th
e
3
D
t
o
o
l
o
f
f
er
ed
im
p
r
o
v
ed
p
e
r
f
o
r
m
an
ce
c
o
m
p
ar
ed
with
ex
is
tin
g
b
en
ch
m
a
r
k
m
o
d
els.
Ho
wev
er
,
r
esear
ch
er
s
d
id
n
o
t
em
p
lo
y
alter
n
ativ
e
class
if
icatio
n
m
eth
o
d
s
f
o
r
co
m
p
ar
is
o
n
an
d
ev
alu
ate
d
th
em
o
n
a
s
in
g
le
d
ataset.
T
h
u
s
,
th
e
m
o
d
el
f
ailed
to
g
en
er
alize
ac
r
o
s
s
ch
an
g
in
g
im
ag
e
co
n
d
itio
n
s
an
d
p
o
p
u
latio
n
s
.
T
u
m
p
a
a
n
d
Kab
ir
[
1
0
]
p
r
o
p
o
s
ed
a
h
y
b
r
id
tex
t
u
r
e
-
f
ea
t
u
r
e
e
x
tr
ac
tio
n
f
r
o
m
I
SIC
an
d
PH
2
d
atasets
.
T
h
e
m
o
d
el
attain
ed
9
7
.
7
%
ac
cu
r
ac
y
w
ith
an
a
r
tific
ial
n
eu
r
al
n
et
wo
r
k
(
ANN)
class
if
ier
.
Alth
o
u
g
h
th
e
m
o
d
el
o
u
tp
er
f
o
r
m
ed
s
ev
er
al
ex
is
tin
g
class
if
ier
s
an
d
ex
tr
ac
ted
a
r
ic
h
s
et
o
f
tex
t
u
r
e
d
escr
ip
to
r
s
,
it
s
p
er
f
o
r
m
a
n
ce
was
n
o
t
co
m
p
a
r
ed
ag
ai
n
s
t
b
aselin
e
ML
alg
o
r
ith
m
s
.
An
e
n
s
em
b
le
f
r
am
ewo
r
k
p
r
o
p
o
s
ed
b
y
Alizad
eh
an
d
Ma
h
lo
o
jifa
r
[
1
1
]
in
teg
r
ated
c
o
n
v
o
lu
ti
o
n
al
n
e
u
r
al
n
etwo
r
k
(
C
NN)
,
Har
alick
,
an
d
lo
ca
l
b
i
n
ar
y
p
atter
n
(
L
B
P)
tex
tu
r
e
f
ea
tu
r
es.
T
h
is
h
y
b
r
id
f
ea
tu
r
e
m
o
d
el
p
er
f
o
r
m
e
d
b
etter
o
n
PH
2
an
d
I
SIC
-
2
0
1
9
d
a
tasets
b
u
t
at
tain
ed
in
s
u
f
f
icien
t
s
en
s
itiv
ity
o
f
5
2
%
o
n
I
SIC
-
2
0
1
6
.
T
h
is
lo
w
s
e
n
s
itiv
ity
o
n
I
SIC
-
2
0
1
6
is
d
u
e
to
s
ev
er
e
im
b
ala
n
ce
b
etwe
en
b
en
i
g
n
a
n
d
m
alig
n
a
n
t
lesi
o
n
s
.
Similar
ly
,
Yu
et
a
l
.
[
1
2
]
c
o
m
b
in
e
d
f
ea
t
u
r
es
f
r
o
m
m
u
ltip
le
C
NN
ar
ch
itectu
r
es
u
s
in
g
f
is
h
er
v
ec
to
r
en
co
d
i
n
g
to
en
h
a
n
ce
lesi
o
n
class
if
icat
io
n
.
Alth
o
u
g
h
GT
an
d
co
lo
r
n
o
r
m
aliza
tio
n
wer
e
u
s
ed
to
m
in
im
ize
class
im
b
alan
ce
,
t
h
e
m
o
d
el
s
till
d
em
o
n
s
tr
ated
r
ed
u
ce
d
s
en
s
itiv
ity
b
ec
au
s
e
o
f
h
ig
h
co
m
p
lex
ity
an
d
in
ad
e
q
u
ate
f
ea
tu
r
e
s
elec
tio
n
.
Haid
er
et
a
l
.
[
1
3
]
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
i
n
v
o
lv
e
d
ac
tiv
e
co
n
to
u
r
r
eg
io
n
o
f
in
ter
est
(
R
OI
)
ex
tr
ac
tio
n
,
L
B
P
tex
tu
r
e
d
escr
ip
to
r
,
k
-
n
ea
r
est
n
e
ig
h
b
o
r
(
K
NN)
,
a
n
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
d
ec
is
io
n
m
o
d
els
f
o
r
d
ete
ctio
n
o
f
p
ig
m
e
n
ted
lesi
o
n
s
i
n
PH
2
im
ag
es.
T
h
e
s
y
s
tem
was
s
u
p
er
io
r
to
m
an
u
a
l
lesi
o
n
ex
tr
ac
tio
n
b
u
t
d
id
n
o
t
p
er
f
o
r
m
well
o
n
au
to
m
atic
R
OI
ex
tr
ac
tio
n
.
T
h
ey
ass
es
s
ed
th
e
m
o
d
el
o
n
ly
o
n
th
e
PH
2
d
ataset
with
o
u
t
co
n
s
id
er
in
g
its
class
im
b
alan
ce
is
s
u
es
,
th
er
ef
o
r
e,
it
is
n
o
t
ca
p
ab
le
o
f
d
e
p
lo
y
m
e
n
t
in
h
e
alth
ca
r
e.
Featu
r
e
-
b
ased
class
if
icatio
n
h
as
th
e
p
o
ten
tial
to
b
e
ap
p
lied
ac
r
o
s
s
m
u
ltip
le
ap
p
licatio
n
d
o
m
ain
s
[
1
4
]
,
[
1
5
]
.
I
t
ca
n
im
p
r
o
v
e
t
h
e
o
u
tco
m
es
p
r
o
v
i
d
ed
m
ea
n
in
g
f
u
l
f
ea
t
u
r
es
ar
e
s
elec
ted
an
d
v
alid
ated
p
r
ec
is
ely
.
I
n
r
ec
en
t
y
ea
r
s
,
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es
h
av
e
b
ee
n
a
d
o
p
ted
wid
ely
as
an
au
to
m
ated
m
eth
o
d
to
ex
tr
ac
t
in
f
o
r
m
ativ
e
an
d
co
m
p
lex
p
atter
n
s
f
r
o
m
d
er
m
o
s
co
p
ic
im
ag
es.
Me
m
o
n
et
a
l
.
[
1
6
]
em
p
lo
y
ed
th
e
Xce
p
tio
n
alg
o
r
ith
m
f
o
r
th
e
p
r
ed
ictio
n
o
f
s
k
in
ca
n
ce
r
an
d
at
tain
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
s
co
r
e
o
f
9
3
%
o
n
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
5
6
-
3
5
6
7
3558
HAM
1
0
0
0
0
d
ataset.
Ho
wev
er
,
th
ey
wo
r
k
ed
o
n
ly
o
n
a
s
m
a
ll
s
et
o
f
th
e
HAM
1
0
0
0
0
d
ata
b
ase;
th
er
ef
o
r
e,
its
ap
p
licatio
n
is
lim
ited
in
ac
tu
al
clin
ical
s
ett
in
g
s
.
T
an
g
et
a
l
.
[
1
7
]
im
p
lem
e
n
ted
a
d
u
al
-
n
etwo
r
k
ar
ch
itectu
r
e
ca
lled
g
lo
b
al
-
p
ar
t
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
with
d
ata
-
tr
an
s
f
o
r
m
ed
en
s
em
b
le
lear
n
in
g
(
GP
-
C
NN
-
DT
E
L
)
th
at
co
m
p
letely
ca
p
tu
r
es
im
ag
e
in
f
o
r
m
atio
n
alo
n
g
with
p
r
ec
is
e
lesi
o
n
d
etails.
E
n
s
em
b
le
lear
n
in
g
was
ap
p
lied
to
n
o
r
m
alize
d
co
lo
r
im
ag
es
to
im
p
r
o
v
e
r
o
b
u
s
tn
ess
.
T
h
e
m
o
d
el
ac
h
ie
v
ed
9
9
.
7
%
s
p
ec
i
f
icity
an
d
8
6
.
3
%
ac
cu
r
ac
y
,
b
u
t
it
h
as
l
o
w
s
en
s
itiv
ity
(
3
2
%),
wh
ic
h
in
d
icate
s
its
lim
ited
ab
ilit
y
to
d
etec
t
m
alig
n
an
t
lesi
o
n
s
.
T
h
is
is
ca
u
s
ed
b
y
class
im
b
alan
ce
a
n
d
s
tr
ict
r
ed
u
ctio
n
o
f
f
alse p
o
s
itiv
e
ca
s
es.
An
o
th
er
s
tu
d
y
b
y
W
u
et
a
l
.
[
1
8
]
p
r
o
p
o
s
ed
th
e
at
ten
ti
o
n
a
n
d
r
e
s
i
d
u
a
l
l
ea
r
n
in
g
d
en
s
e
ly
co
n
n
ec
te
d
co
n
v
o
lu
t
io
n
a
l
n
e
two
r
k
(
A
R
DT
-
Den
s
e
Ne
t
)
f
r
am
e
wo
r
k
co
n
s
i
s
t
in
g
o
f
c
o
m
p
l
ex
co
n
n
ec
t
io
n
s
,
at
ten
tio
n
m
ec
h
an
i
s
m
s
,
an
d
r
es
id
u
al
l
ea
r
n
in
g
f
o
r
r
el
iab
le
c
la
s
s
if
i
ca
t
io
n
o
f
m
e
lan
o
m
a.
E
v
en
af
ter
em
p
lo
y
in
g
im
b
a
lan
ce
m
i
tig
at
io
n
te
ch
n
i
q
u
es
,
th
e
m
o
d
el
s
t
il
l
u
n
d
e
r
p
er
f
o
r
m
s
d
u
e
to
th
e
o
v
er
f
it
tin
g
p
r
o
b
l
em
an
d
d
ee
p
-
lay
er
ar
ch
it
ec
t
u
r
e.
S
im
i
lar
wo
r
k
b
y
B
a
lah
a
an
d
Ha
s
s
an
[
1
9
]
im
p
lem
en
t
ed
A
t
ten
t
io
n
U
-
N
et
f
o
r
R
O
I
ex
tr
ac
tio
n
,
p
r
e
-
tr
a
in
ed
C
NN
cla
s
s
if
i
er
s
,
an
d
a
s
p
ar
r
o
w
s
e
ar
ch
o
p
tim
iz
at
io
n
(
S
SO)
a
lg
o
r
ith
m
o
n
th
e
P
H
2
d
ata
s
et
f
o
r
m
e
lan
o
m
a
id
en
t
if
i
ca
t
io
n
.
T
h
e
a
tt
en
t
io
n
U
-
N
et
wi
th
D
en
s
e
Ne
t
-
2
0
1
y
i
el
d
ed
9
5
%
a
cc
u
r
ac
y
,
8
6
%
s
en
s
it
iv
ity
,
9
3
%
s
p
ec
if
i
ci
ty
,
an
d
9
2
%
F1
-
s
co
r
e
.
Ho
wev
er
,
th
e
m
o
d
el
wa
s
h
ig
h
l
y
co
m
p
lex
d
u
e
to
d
en
s
e
ar
ch
it
ec
tu
r
e
s
an
d
co
m
p
lex
o
p
tim
iz
at
io
n
l
ay
e
r
s
.
T
h
e
h
ig
h
co
m
p
u
t
at
io
n
a
l
co
s
t
m
ad
e
it
l
e
s
s
o
p
er
a
ti
v
e
f
o
r
p
r
ac
ti
ca
l c
lin
ic
al
ap
p
li
ca
tio
n
s
.
T
h
e
ex
am
i
n
atio
n
o
f
p
r
io
r
r
ese
ar
ch
s
h
o
ws
th
at
m
o
s
t
wo
r
k
h
a
s
u
s
ed
im
b
alan
ce
d
d
atasets
s
u
ch
as
I
SIC
an
d
PH
2
f
o
r
tr
ain
in
g
th
eir
m
o
d
els.
T
h
is
lead
s
to
m
o
d
el
b
ias
an
d
h
en
ce
r
e
d
u
ce
s
its
ab
ilit
y
to
d
etec
t
m
alig
n
an
t
ca
s
es.
Fu
r
th
er
m
o
r
e,
m
an
y
s
tu
d
ies h
av
e
u
tili
ze
d
o
n
ly
a
s
in
g
le
d
ataset
with
in
s
u
f
f
icien
t
s
am
p
l
es
f
o
r
test
in
g
th
eir
m
o
d
el.
T
h
u
s
,
th
eir
m
o
d
el
p
er
f
o
r
m
a
n
ce
ca
n
n
o
t
b
e
v
alid
ated
ac
r
o
s
s
d
iv
e
r
s
e
clin
ical
s
ce
n
ar
io
s
.
An
o
th
er
co
m
m
o
n
lim
itatio
n
is
t
h
at
th
e
m
ajo
r
ity
o
f
r
esear
ch
is
f
o
c
u
s
ed
o
n
im
p
r
o
v
in
g
o
v
er
all
ac
cu
r
ac
y
,
wh
ich
is
an
in
ad
eq
u
ate
m
ea
s
u
r
e
f
o
r
im
b
al
an
ce
d
class
es.
T
h
er
e
is
v
er
y
l
im
ited
atten
tio
n
to
th
e
T
PR
i
n
th
e
liter
atu
r
e,
a
n
im
p
o
r
tan
t
ev
al
u
atio
n
m
etr
ic
f
o
r
r
eliab
le
m
elan
o
m
a
d
etec
tio
n
.
T
h
e
d
ee
p
lear
n
i
n
g
ap
p
r
o
ac
h
es
im
p
lem
en
ted
to
im
p
r
o
v
e
p
er
f
o
r
m
a
n
ce
ar
e
co
m
p
u
tatio
n
ally
co
m
p
le
x
an
d
l
ess
in
ter
p
r
etab
le.
T
h
is
m
ak
es
th
em
u
n
s
u
itab
le
f
o
r
r
eg
u
lar
clin
ical
p
r
ac
tice.
T
o
ad
d
r
ess
th
e
ab
o
v
e
-
d
escr
ib
e
d
ch
allen
g
es,
th
is
s
tu
d
y
p
r
o
p
o
s
es
an
ef
f
icien
t
au
to
m
ated
a
p
p
r
o
ac
h
f
o
r
class
if
y
in
g
s
k
in
ca
n
ce
r
t
h
at
is
r
elia
b
le,
in
ter
p
r
etab
le,
an
d
co
m
p
u
t
atio
n
ally
ef
f
icien
t
,
d
esig
n
ed
to
o
p
tim
ize
th
e
T
PR
an
d
ca
n
b
e
ef
f
ec
tiv
el
y
d
ep
l
o
y
e
d
in
clin
ical
en
v
i
r
o
n
m
e
n
ts
.
T
h
is
s
tu
d
y
’
s
m
ain
r
esear
ch
co
n
tr
ib
u
tio
n
s
ar
e
d
escr
ib
ed
as f
o
llo
ws:
i)
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
ap
p
lies
m
o
r
p
h
o
lo
g
y
p
r
eser
v
in
g
GT
to
th
e
m
in
o
r
ity
class
s
am
p
les
p
r
esen
t
in
th
e
I
SIC
-
2
0
1
6
an
d
PH
2
d
atasets
.
T
h
is
ef
f
ec
tiv
ely
tack
les
t
h
e
class
im
b
alan
ce
i
s
s
u
es
an
d
en
h
an
ce
s
th
e
T
PR
f
o
r
m
alig
n
an
t m
i
n
o
r
ity
c
ases
.
ii)
T
h
e
f
r
am
ewo
r
k
ef
f
ec
tiv
ely
ca
p
tu
r
es
d
is
tin
ct
im
ag
e
f
ea
tu
r
es
th
r
o
u
g
h
th
e
e
x
tr
ac
tio
n
o
f
r
o
t
atio
n
-
in
v
ar
ia
n
t
Har
alick
tex
tu
r
e
an
d
s
tatis
tical
d
escr
ip
to
r
s
f
r
o
m
th
e
g
r
ay
lev
e
l c
o
-
o
cc
u
r
r
en
ce
m
at
r
ix
(
GL
C
M)
ap
p
r
o
ac
h
.
iii)
I
t
im
p
lem
en
ts
h
y
p
o
th
esis
test
in
g
f
o
r
f
ea
t
u
r
e
v
alid
atio
n
a
n
d
ass
ess
es
th
e
s
tatis
tica
l
s
i
g
n
if
ican
ce
o
f
ex
tr
ac
ted
f
ea
tu
r
es.
T
h
is
m
ak
es
th
e
m
o
d
el
m
o
r
e
in
ter
p
r
etab
le
an
d
en
s
u
r
es
r
eliab
le
f
ea
t
u
r
e
r
e
p
r
esen
tatio
n
.
T
h
er
e
is
v
er
y
lim
ited
atten
tio
n
o
n
th
is
s
tatis
tical
asp
ec
t in
ea
r
lier
s
tu
d
ies.
iv
)
T
h
e
p
r
o
p
o
s
ed
GT
-
SVM
f
r
a
m
ewo
r
k
is
a
lig
h
tweig
h
t
a
n
d
in
ter
p
r
etab
le
f
r
am
ewo
r
k
th
at
p
r
esen
ts
an
ef
f
ec
tiv
e
alter
n
ativ
e
to
c
o
m
p
lex
d
ee
p
m
o
d
els
f
o
r
b
in
ar
y
class
if
icatio
n
o
f
s
k
in
lesi
o
n
s
u
n
d
e
r
lim
ited
d
ata
co
n
d
itio
n
s
.
v)
T
h
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
is
t
h
o
r
o
u
g
h
ly
ev
alu
ate
d
a
n
d
c
o
m
p
ar
ed
with
ex
is
tin
g
ML
ap
p
r
o
ac
h
es
u
s
in
g
I
SIC
-
2
0
1
6
a
n
d
PH
2
d
atasets
.
T
h
e
s
tr
u
ctu
r
e
o
f
t
h
e
r
em
ain
in
g
p
ap
er
is
as
f
o
llo
ws.
Sectio
n
2
p
r
o
v
id
es
a
d
etailed
ex
p
lan
a
tio
n
o
f
th
e
m
eth
o
d
o
l
o
g
y
an
d
im
p
lem
en
tat
io
n
s
tep
s
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
.
T
h
is
is
to
s
u
p
p
o
r
t
th
e
r
ep
r
o
d
u
ctio
n
o
f
t
h
e
ex
p
er
im
en
tal
r
esu
lts
f
o
r
f
u
r
th
er
ex
ten
d
in
g
r
esear
ch
wo
r
k
.
T
h
e
r
esu
lts
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el,
k
e
y
f
in
d
in
g
s
,
an
d
th
e
f
r
am
ewo
r
k
’
s
im
p
licatio
n
s
an
d
lim
itatio
n
s
ar
e
d
is
cu
s
s
ed
in
s
ec
tio
n
3
.
Sectio
n
4
co
n
c
lu
d
es
th
e
s
tu
d
y
b
y
b
r
ief
ly
p
r
esen
tin
g
th
e
clin
ical
u
s
ef
u
ln
ess
o
f
th
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
an
d
d
ir
ec
tio
n
s
f
o
r
f
u
t
u
r
e
r
esear
ch
.
2.
M
E
T
H
O
D
T
h
e
r
esear
ch
aim
s
to
im
p
r
o
v
e
th
e
ea
r
ly
d
etec
tio
n
o
f
m
alig
n
an
t
lesi
o
n
s
b
y
p
r
o
p
o
s
in
g
a
n
ef
f
icien
t
ap
p
r
o
ac
h
f
o
r
class
if
y
in
g
s
k
in
ca
n
ce
r
f
r
o
m
d
er
m
o
s
co
p
ic
im
ag
es.
T
o
ac
h
iev
e
th
is
g
o
al,
th
e
GT
-
SVM
f
r
am
ewo
r
k
is
s
u
g
g
ested
,
wh
i
ch
en
h
a
n
ce
s
th
e
m
in
o
r
ity
cla
s
s
es,
ca
p
tu
r
es
d
is
tin
ct
im
ag
e
f
ea
tu
r
es,
p
e
r
f
o
r
m
s
f
ea
tu
r
e
v
alid
atio
n
,
an
d
class
if
ies
lesi
o
n
s
u
s
in
g
an
o
p
tim
ized
class
if
ier
.
T
h
e
p
r
o
p
o
s
ed
p
ip
eli
n
e
in
v
o
lv
es
s
ix
k
e
y
s
tep
s
as
s
h
o
wn
in
Fig
u
r
e
1
.
I
t
co
m
p
r
is
es
i)
d
ata
co
llectio
n
,
ii)
p
r
e
-
p
r
o
ce
s
s
in
g
,
iii)
GT
,
iv
)
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
f
ea
tu
r
e
s
ca
lin
g
,
v
)
f
ea
tu
r
e
v
alid
atio
n
,
an
d
v
i)
b
in
ar
y
cla
s
s
if
icatio
n
u
s
in
g
GT
-
SVM.
T
h
e
en
tire
f
r
a
m
ewo
r
k
was
d
ev
elo
p
ed
u
s
in
g
s
ev
er
al
Py
th
o
n
p
ac
k
a
g
es
s
u
ch
as
Op
e
n
C
V,
Nu
m
Py
,
Pan
d
as,
an
d
s
c
ik
it
-
lear
n
to
en
s
u
r
e
co
n
s
is
ten
t m
eth
o
d
o
lo
g
y
a
n
d
r
e
p
r
o
d
u
cib
le
r
esu
lts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
ma
ch
in
e
lea
r
n
i
n
g
fr
a
mewo
r
k
fo
r
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
u
s
in
g
textu
r
e
d
escr
ip
to
r
s
(
S
h
ikh
a
Ma
lik
)
3559
Fig
u
r
e
1
.
Flo
w
d
ia
g
r
am
o
f
th
e
p
r
o
p
o
s
ed
GT
-
SVM
f
r
am
ewo
r
k
f
o
r
s
k
i
n
ca
n
ce
r
class
if
icatio
n
2
.
1
.
Da
t
a
a
cquis
it
io
n
T
h
is
wo
r
k
s
elec
ted
two
b
en
c
h
m
ar
k
d
e
r
m
o
s
co
p
ic
im
ag
e
d
ata
s
ets:
I
SIC
-
2016
[
2
0
]
a
n
d
PH
2
[
2
1
]
.
T
h
e
I
SIC
-
2
0
1
6
d
ataset
is
g
ath
er
e
d
f
r
o
m
th
e
I
SIC
ar
ch
iv
e
d
ata
b
ase.
1
,
2
7
9
d
er
m
o
s
co
p
ic
im
ag
es
o
f
th
is
d
ataset
co
m
p
r
is
e
1
,
0
3
1
n
o
n
-
m
elan
o
m
a
(
b
en
ig
n
)
an
d
2
4
8
m
ela
n
o
m
a
(
m
alig
n
a
n
t)
lesi
o
n
s
.
T
h
e
PH
2
d
ataset
i
s
d
ev
elo
p
e
d
b
y
th
e
Der
m
ato
lo
g
y
Dep
ar
tm
en
t
,
Po
r
tu
g
al
Ho
s
p
ital,
an
d
co
m
p
r
is
es
2
0
0
d
er
m
o
s
co
p
y
im
ag
es
o
f
r
eso
lu
tio
n
7
6
8
×
5
6
0
.
1
6
0
b
en
i
g
n
s
am
p
les
o
f
th
is
d
ataset
co
n
tain
8
0
co
m
m
o
n
n
ev
i
an
d
8
0
aty
p
ical
n
e
v
i,
a
n
d
4
0
im
ag
es
ar
e
c
o
n
f
ir
m
e
d
m
elan
o
m
a
ca
s
es.
T
h
is
d
ataset
in
clu
d
es
s
eg
m
e
n
tatio
n
m
ask
s
,
an
n
o
tatio
n
s
,
an
d
d
er
m
o
s
co
p
ic
f
ea
tu
r
es
v
e
r
if
ied
b
y
e
x
p
er
ts
an
d
is
p
er
f
e
ctly
s
u
itab
le
f
o
r
s
eg
m
en
tatio
n
as
well
as
class
if
icatio
n
.
T
h
e
s
tatis
tical
d
ata
o
f
b
o
th
d
at
asets
ar
e
p
r
esen
ted
in
T
ab
le
1
.
T
ab
le
1
.
I
SIC
-
2
0
1
6
a
n
d
PH
2
d
ataset
s
tati
s
tic
s
D
a
t
a
s
e
t
To
t
a
l
i
m
a
g
e
s
B
e
n
i
g
n
sam
p
l
e
s
M
a
l
i
g
n
a
n
t
s
a
m
p
l
e
s
I
S
I
C
-
2
0
1
6
1
,
2
7
9
1
,
0
3
1
2
4
8
P
H
2
2
0
0
1
6
0
40
2
.
2
.
P
re
-
pro
ce
s
s
ing
T
h
e
p
r
e
-
p
r
o
ce
s
s
in
g
s
tag
e
is
th
e
f
ir
s
t
im
p
o
r
tan
t
s
tep
r
eq
u
ir
ed
to
r
em
o
v
e
im
ag
e
d
ef
ec
ts
an
d
s
u
s
tain
th
e
m
o
r
p
h
o
lo
g
ical
f
ea
tu
r
es
o
f
lesi
o
n
s
.
I
t
was
im
p
lem
en
ted
in
th
r
ee
p
h
ases
,
co
n
s
is
tin
g
o
f
im
ag
e
r
esizin
g
,
f
o
llo
wed
b
y
n
o
is
e
r
em
o
v
al
an
d
u
n
wan
t
ed
h
air
a
r
tifa
cts
elim
in
atio
n
.
All
th
ese
wer
e
ca
r
ef
u
lly
im
p
l
em
en
ted
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
t
h
e
im
ag
es
wh
ile
p
r
eser
v
in
g
th
e
lesi
o
n
’
s
m
o
r
p
h
o
lo
g
ical
f
ea
tu
r
es
[
2
2
]
.
Fi
g
u
r
e
2
p
r
esen
ts
th
e
s
eq
u
en
ce
o
f
p
r
ep
r
o
ce
s
s
in
g
o
p
e
r
atio
n
s
ap
p
lied
to
b
en
ig
n
an
d
m
alig
n
an
t sam
p
les.
T
h
e
f
ir
s
t
s
tep
i
n
p
r
e
-
p
r
o
ce
s
s
in
g
was
a
r
esizin
g
o
p
er
atio
n
i
n
wh
ich
im
ag
es
o
f
v
ar
y
in
g
r
eso
lu
tio
n
s
wer
e
m
o
d
if
ied
to
m
ain
tain
co
n
s
is
ten
cy
in
th
e
d
atasets
.
All
im
ag
e
s
am
p
les
wer
e
s
tan
d
ar
d
i
ze
d
to
an
in
p
u
t
s
ize
o
f
2
2
4
×2
2
4
p
ix
els.
T
h
is
cr
e
ated
a
u
n
if
o
r
m
s
p
atial
s
ize
an
d
m
in
im
ized
th
e
p
r
o
ce
s
s
in
g
tim
e
r
eq
u
ir
ed
f
o
r
f
ea
tu
r
e
an
aly
s
is
an
d
class
if
icatio
n
.
T
h
is
s
tep
was p
er
f
o
r
m
ed
u
s
in
g
th
e
Op
en
C
V
lib
r
ar
y
with
th
e
c
v
2
.
r
esize(
)
f
u
n
ctio
n
.
Af
ter
r
esizin
g
,
f
ilter
in
g
was
p
er
f
o
r
m
ed
to
r
ed
u
ce
p
i
x
el
-
lev
el
n
o
is
e
with
o
u
t
m
o
d
i
f
y
in
g
m
ea
n
in
g
f
u
l
tex
tu
r
e
in
f
o
r
m
atio
n
.
I
n
s
tead
o
f
a
m
ed
ian
f
ilter
,
a
Gau
s
s
ian
s
m
o
o
th
in
g
f
ilter
with
a
3
×3
k
er
n
el
was p
r
ef
er
r
ed
as
it
b
etter
p
r
eser
v
es
lesi
o
n
ed
g
es
an
d
tex
tu
r
e
v
a
r
iatio
n
s
,
wh
ich
ar
e
ess
en
tial
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
Gau
s
s
ian
p
ar
am
eter
s
wer
e
a
u
to
m
atica
lly
tu
n
ed
o
n
th
e
b
asis
o
f
k
er
n
el
s
ize.
T
h
i
s
p
r
eser
v
ed
clin
ically
m
ea
n
in
g
f
u
l
tex
tu
r
e
in
f
o
r
m
ati
o
n
an
d
e
f
f
ec
tiv
ely
r
e
d
u
ce
d
h
ig
h
-
f
r
eq
u
e
n
cy
n
o
is
e.
T
h
e
c
v
2
.
Gau
s
s
ian
B
lu
r
(
)
f
u
n
ctio
n
a
v
ailab
le
in
th
e
Op
en
C
V
lib
r
ar
y
was u
s
ed
to
im
p
le
m
en
t th
e
Gau
s
s
ian
f
ilter
in
g
o
p
er
atio
n
.
Hair
ar
tifa
cts
r
em
o
v
al
was
th
e
n
ex
t
im
p
o
r
tan
t
p
ar
t
in
p
r
ep
r
o
ce
s
s
in
g
f
o
r
clea
r
v
is
u
aliza
tio
n
o
f
lesi
o
n
b
o
r
d
er
s
,
wh
ich
wer
e
h
id
d
e
n
b
y
u
n
wan
ted
h
air
lin
es
i
n
d
er
m
o
s
co
p
ic
im
a
g
es.
T
h
e
h
air
ar
tifa
cts
af
f
ec
t
th
e
ac
cu
r
ate
ex
tr
a
ctio
n
a
n
d
class
if
icatio
n
o
f
f
ea
tu
r
es.
T
h
e
h
air
ar
tifa
ct
r
em
o
v
al
p
r
o
ce
s
s
was
p
er
f
o
r
m
ed
in
m
u
ltip
le
m
o
r
p
h
o
lo
g
ical
s
tag
es,
wh
ich
in
v
o
lv
ed
g
r
ay
s
ca
le
co
n
v
er
s
io
n
,
b
lack
-
h
at
m
o
r
p
h
o
lo
g
y
,
a
d
a
p
tiv
e
th
r
esh
o
ld
in
g
,
an
d
im
ag
e
in
p
ain
tin
g
.
First,
th
e
d
er
m
o
s
co
p
ic
im
ag
es
wer
e
co
n
v
er
ted
in
to
g
r
ay
s
ca
le.
T
h
is
was
d
o
n
e
b
y
s
elec
tin
g
th
e
p
ar
am
eter
cv
2
.
C
OL
OR
_
R
G
B
2
GR
AY,
wh
ich
is
av
ailab
le
in
th
e
cv
2
.
cv
tC
o
l
o
r
(
)
f
u
n
ctio
n
f
r
o
m
th
e
Op
en
C
V
lib
r
ar
y
.
T
h
is
o
p
er
atio
n
h
elp
e
d
in
r
em
o
v
in
g
u
n
n
ec
ess
ar
y
co
lo
r
in
f
o
r
m
atio
n
an
d
r
ed
u
ce
d
th
e
co
m
p
u
tatio
n
al
co
s
t.
T
h
e
B
la
ck
-
h
at
m
o
r
p
h
o
l
o
g
ical
o
p
er
ati
o
n
was
th
e
n
ex
t
s
tep
to
b
e
im
p
lem
en
ted
v
ia
9
×9
ellip
tical
k
er
n
els.
T
h
e
p
ar
am
ete
r
s
f
o
r
th
is
o
p
er
atio
n
wer
e
s
elec
ted
with
th
e
h
elp
o
f
cv
2
.
g
etStru
ctu
r
in
g
E
lem
en
t
(
)
a
n
d
cv
2
.
m
o
r
p
h
o
lo
g
y
E
x
(
)
f
u
n
ctio
n
s
.
T
h
is
s
tep
s
ep
ar
ated
th
e
b
ac
k
g
r
o
u
n
d
a
n
d
h
air
lin
es
with
o
u
t
af
f
ec
tin
g
th
e
lesi
o
n
d
etails.
Ad
a
p
tiv
e
th
r
esh
o
ld
in
g
was
th
en
ap
p
lied
to
d
etec
t th
in
h
air
lin
es
v
er
y
p
r
ec
is
ely
.
T
h
e
o
p
er
atio
n
wa
s
im
p
lem
en
ted
u
s
in
g
th
e
cv
2
.
ad
ap
tiv
eT
h
r
esh
o
ld
(
)
f
u
n
ctio
n
.
Fin
ally
,
T
elea
’
s
in
p
ain
tin
g
alg
o
r
ith
m
was
ap
p
lied
to
r
esto
r
e
th
e
ar
ea
s
c
o
v
er
ed
b
y
th
e
m
ask
b
y
esti
m
atin
g
th
em
u
s
in
g
n
eig
h
b
o
r
in
g
p
ix
el
v
alu
es.
T
h
i
s
r
ec
o
n
s
tr
u
cted
th
e
n
atu
r
al
a
p
p
ea
r
an
ce
o
f
lesi
o
n
s
.
T
h
e
cv
2
.
in
p
ain
t(
)
f
u
n
ctio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
5
6
-
3
5
6
7
3560
with
th
e
cv
2
.
I
NPAI
NT
_
T
E
L
E
A
m
eth
o
d
was
u
s
ed
to
e
x
ec
u
te
th
is
s
tep
.
T
h
e
T
elea
in
p
ai
n
tin
g
tech
n
iq
u
e
was
em
p
lo
y
ed
as it is c
o
m
p
u
tatio
n
ally
ef
f
icien
t a
n
d
p
r
eser
v
es th
e
o
r
ig
in
al
ap
p
ea
r
an
ce
o
f
th
e
lesi
o
n
s
.
Fig
u
r
e
2
.
Seq
u
en
ce
o
f
p
r
e
-
p
r
o
ce
s
s
in
g
o
p
er
atio
n
s
ap
p
lied
to
b
en
ig
n
a
n
d
m
alig
n
an
t sam
p
les
2
.
3
.
G
eo
m
e
t
ric
t
ra
ns
f
o
r
m
a
t
i
o
ns
T
h
e
PH
2
an
d
I
SIC
-
2
0
1
6
d
atasets
ar
e
b
o
th
s
ig
n
if
ican
tly
im
b
a
lan
ce
d
,
as
b
en
ig
n
class
es
ar
e
h
ig
h
er
th
an
m
alig
n
an
t
o
n
es.
T
h
er
e
is
ap
p
r
o
x
im
ately
a
4
:1
b
en
ig
n
-
to
-
m
alig
n
an
t
r
atio
in
b
o
th
d
atasets
.
T
h
er
e
ar
e
1
6
0
b
e
n
ig
n
(
n
eg
ativ
e
ca
s
es),
an
d
4
0
m
alig
n
an
t
(
p
o
s
itiv
e
ca
s
es)
s
am
p
les
in
th
e
PH
2
d
ata
s
et,
an
d
I
SIC
-
2
0
1
6
co
m
p
r
is
es
o
f
1
,
0
3
1
b
e
n
ig
n
a
n
d
2
4
8
m
alig
n
an
t
s
am
p
les.
As
o
b
s
er
v
ed
in
r
ep
ea
ted
ex
p
er
im
en
ts
,
class
if
icatio
n
m
o
d
els
alwa
y
s
f
av
o
r
ed
th
e
m
a
jo
r
ity
(
b
e
n
ig
n
)
ca
s
es
d
u
e
to
t
h
is
im
b
alan
ce
.
T
h
is
r
esu
lted
in
r
ed
u
ce
d
s
en
s
itiv
ity
wh
ile
d
etec
tin
g
m
alig
n
an
t
le
s
io
n
s
.
T
o
s
o
lv
e
th
is
is
s
u
e,
GT
wer
e
ap
p
lied
to
en
h
a
n
ce
m
alig
n
an
t
m
in
o
r
ity
im
ag
es
with
o
u
t
m
o
d
if
y
in
g
th
e
co
u
n
t
o
f
b
en
ig
n
im
ag
es.
T
h
es
e
tr
an
s
f
o
r
m
atio
n
s
im
p
r
o
v
e
d
d
ataset
d
iv
er
s
ity
an
d
en
s
u
r
ed
th
at
lesi
o
n
b
o
u
n
d
ar
ies,
p
ig
m
e
n
t
n
etwo
r
k
s
,
an
d
s
tr
u
ct
u
r
al
p
atter
n
s
ar
e
clin
ically
v
ali
d
f
o
r
b
ala
n
ce
d
a
n
d
r
eliab
le
s
k
in
lesi
o
n
d
etec
tio
n
[
2
3
]
.
T
h
e
GT
ap
p
lied
wer
e
r
o
tatio
n
s
r
an
d
o
m
ly
v
a
r
y
in
g
b
e
twee
n
0
°
an
d
2
5
°,
v
er
tical
an
d
h
o
r
izo
n
tal
tr
an
s
latio
n
u
p
to
2
0
%
o
f
im
ag
e
s
ize,
r
an
d
o
m
s
h
ea
r
in
g
lim
ited
to
2
0
%,
h
o
r
izo
n
tal
f
lip
p
in
g
,
a
n
d
zo
o
m
in
g
u
p
to
2
0
%.
T
h
ese
tr
an
s
f
o
r
m
atio
n
s
wer
e
im
p
lem
en
ted
in
T
e
n
s
o
r
Flo
w/Ker
as
with
th
e
s
u
p
p
o
r
t
o
f
im
ag
e
d
ata
g
en
er
at
o
r
m
o
d
u
les.
As
s
h
o
wn
in
Fig
u
r
e
3
,
ea
ch
m
alig
n
a
n
t
im
ag
e
g
en
er
ated
f
o
u
r
n
ew
au
g
m
en
ted
v
er
s
io
n
s
.
Fig
u
r
e
3
(
a)
s
h
o
ws
a
s
am
p
le
im
ag
e
f
r
o
m
I
SIC
-
2
0
1
6
,
an
d
Fig
u
r
e
3
(
b
)
p
r
esen
ts
its
f
o
u
r
n
ew
au
g
m
e
n
ted
v
er
s
io
n
s
r
esu
l
tin
g
f
r
o
m
GT
.
T
h
ese
tr
an
s
f
o
r
m
atio
n
s
ef
f
ec
tiv
ely
ex
p
a
n
d
ed
th
e
tr
ain
in
g
d
atasets
with
o
u
t
alter
in
g
th
e
m
o
r
p
h
o
lo
g
ical
d
etails
o
f
s
k
in
lesi
o
n
s
.
C
o
n
s
eq
u
en
tly
,
th
e
m
in
o
r
ity
class
s
am
p
le
s
q
u
ad
r
u
p
led
f
r
o
m
2
4
8
to
9
9
2
f
o
r
I
SIC
-
2
0
1
6
an
d
f
r
o
m
4
0
to
1
6
0
s
am
p
les
in
th
e
P
H
2
d
ataset
af
ter
th
e
ap
p
licatio
n
o
f
GT
.
T
ab
le
2
p
r
e
s
en
ts
th
e
s
tatis
tics
o
f
b
o
th
d
ata
s
ets af
ter
th
e
ap
p
licatio
n
o
f
GT
.
(
a)
(
b
)
Fig
u
r
e
3
.
I
ll
u
s
tr
atio
n
o
f
GT
o
p
er
atio
n
s
of
(
a
)
I
SIC
-
2
0
1
6
s
am
p
le
im
ag
e
an
d
(
b
)
f
o
u
r
o
u
tp
u
ts
af
ter
GT
T
ab
le
2
.
Statis
tics
o
f
I
SIC
-
2
0
1
6
an
d
PH
2
d
atasets
af
ter
ap
p
lic
atio
n
o
f
GT
D
a
t
a
s
e
t
M
a
l
i
g
n
a
n
t
i
ma
g
e
s
(
B
e
f
o
r
e
t
r
a
n
sf
o
r
mat
i
o
n
)
M
a
l
i
g
n
a
n
t
i
ma
g
e
s
(
A
f
t
e
r
t
r
a
n
sf
o
r
m
a
t
i
o
n
)
B
e
n
i
g
n
i
m
a
g
e
s
To
t
a
l
I
S
I
C
-
2
0
1
6
2
4
8
9
9
2
1
,
0
3
1
2
,
0
2
3
PH
2
40
1
6
0
1
6
0
3
2
0
2
.
4
.
F
e
a
t
ure
ex
t
r
a
ct
io
n a
nd
f
ea
t
ure
s
ca
lin
g
Af
ter
GT
,
f
ea
tu
r
e
ex
tr
ac
tio
n
was
p
er
f
o
r
m
ed
b
y
ex
tr
a
ctin
g
tex
tu
r
e
an
d
s
tatis
tical
f
ea
tu
r
e
s
f
r
o
m
th
e
GL
C
M
m
eth
o
d
p
r
o
p
o
s
ed
b
y
Har
alick
[
2
4
]
.
GL
C
M
ap
p
r
o
a
ch
is
a
s
ec
o
n
d
-
o
r
d
e
r
s
tatis
tic
al
tex
tu
r
e
an
aly
s
is
Hai
r
r
e
m
oval
op
e
r
at
io
n
I
n
p
u
t
im
age
s
Nois
e
Re
m
oval
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
ma
ch
in
e
lea
r
n
i
n
g
fr
a
mewo
r
k
fo
r
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
u
s
in
g
textu
r
e
d
escr
ip
to
r
s
(
S
h
ikh
a
Ma
lik
)
3561
tech
n
iq
u
e
th
at
m
ea
s
u
r
es
th
e
r
elatio
n
s
h
ip
b
etwe
en
n
eig
h
b
o
r
in
g
p
ix
el
p
air
s
in
th
e
s
p
atial
d
o
m
ain
.
T
h
is
ap
p
r
o
ac
h
ca
p
tu
r
es
m
o
r
e
d
etai
led
tex
tu
r
e
in
f
o
r
m
atio
n
th
a
n
f
ir
s
t
-
o
r
d
er
h
is
to
g
r
am
m
eth
o
d
s
.
I
t
ev
alu
ates
h
o
w
f
r
eq
u
e
n
tly
s
p
ec
if
ic
in
ten
s
ity
p
air
s
o
cc
u
r
at
a
d
ef
in
e
d
d
is
tan
ce
(
)
an
d
o
r
ien
tatio
n
(
)
th
er
eb
y
r
ep
r
esen
tin
g
s
tr
u
ctu
r
al
an
d
m
o
r
p
h
o
lo
g
ical
ch
ar
ac
ter
is
tics
o
f
lesi
o
n
s
.
T
h
e
s
tep
-
by
-
s
tep
p
r
o
ce
d
u
r
e
f
o
r
co
n
s
tr
u
ctin
g
t
h
e
GL
C
M
is
d
etailed
in
Alg
o
r
ith
m
1
.
Alg
o
r
ith
m
1
.
GL
C
M
alg
o
r
ith
m
1
: I
n
itialize
g
r
ay
le
v
els
,
s
p
atial
d
is
tan
ce
an
d
an
g
le
.
2
: I
n
itialize
∗
GL
C
M
m
atr
ix
P.
3
: f
o
r
ea
c
h
p
ix
el
(
x
,
y
)
in
in
p
u
t
im
ag
e
I
:
co
m
p
u
te
n
eig
h
b
o
r
(
′
,
′
)
b
ased
o
n
s
an
d
an
g
le
α
let
p
ix
el
(
,
)
in
ten
s
ity
=
i a
n
d
p
ix
e
l
(
′
,
′
)
in
ten
s
ity
=
j
,
∈
{
0
,
1
,
…
−
1
}
in
cr
em
en
t
b
y
1
4
: N
o
r
m
alize
r
ep
r
esen
ts
p
r
o
b
ab
ilit
y
m
atr
ix
5
: E
x
tr
ac
t f
ea
tu
r
es f
r
o
m
n
o
r
m
alize
d
GL
C
M
I
n
th
is
s
tu
d
y
,
s
tatis
tical
f
e
atu
r
es,
in
clu
d
in
g
m
ax
im
u
m
p
r
o
b
a
b
ilit
y
,
m
ea
n
,
v
ar
ian
c
e,
s
tan
d
ar
d
d
ev
iatio
n
,
s
k
ewn
ess
,
an
d
k
u
r
to
s
is
,
to
g
eth
er
with
Ha
r
alick
tex
tu
r
e
f
ea
tu
r
es
s
u
c
h
as
co
r
r
elatio
n
,
c
o
n
tr
ast,
en
er
g
y
,
h
o
m
o
g
en
eity
,
en
tr
o
p
y
,
d
is
s
im
ilar
ity
,
an
d
an
g
u
la
r
s
ec
o
n
d
m
o
m
en
t,
wer
e
d
er
iv
ed
f
r
o
m
th
e
n
o
r
m
alize
d
GL
C
M
.
T
h
e
GL
C
M
m
atr
ices
wer
e
co
m
p
u
ted
at
f
o
u
r
o
r
ie
n
tatio
n
s
(
=
0
°,
4
5
°,
9
0
°,
an
d
1
3
5
°)
with
=
1
p
ix
el,
an
d
th
e
av
er
ag
e
o
f
th
e
r
esu
ltin
g
f
ea
tu
r
e
v
alu
es
was
u
s
ed
to
o
b
tain
a
r
o
tati
o
n
-
in
v
a
r
ian
t
tex
tu
r
e
r
ep
r
esen
tatio
n
.
T
h
is
m
eth
o
d
g
u
ar
an
teed
th
at
th
e
ex
tr
ac
te
d
f
ea
tu
r
es
r
em
ain
s
tab
le
with
r
esp
ec
t
to
lesi
o
n
o
r
ien
tatio
n
,
th
e
r
eb
y
im
p
r
o
v
i
n
g
th
e
m
o
d
el’
s
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
.
T
ab
le
3
s
u
m
m
ar
i
ze
s
th
e
1
3
s
elec
ted
tex
tu
r
e
an
d
s
tatis
tical
d
escr
ip
t
o
r
s
ex
tr
ac
ted
f
o
r
class
if
icatio
n
.
T
ab
le
3
.
B
r
ief
d
escr
ip
tio
n
o
f
1
3
ex
tr
ac
ted
s
tatis
tical
an
d
Har
alick
f
ea
tu
r
es
F
e
a
t
u
r
e
s
D
e
f
i
n
i
t
i
o
n
C
a
l
c
u
l
a
t
i
o
n
m
e
t
h
o
d
M
a
x
i
m
u
m
p
r
o
b
a
b
i
l
i
t
y
(
P
m
a
x
)
M
e
a
su
r
e
o
f
t
h
e
p
e
a
k
r
e
s
p
o
n
s
e
(
)
=
,
M
e
a
n
(
µ
)
I
n
f
o
r
mat
i
o
n
a
b
o
u
t
t
h
e
a
v
e
r
a
g
e
g
r
a
y
l
e
v
e
l
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
i
n
t
h
e
i
ma
g
e
(
)
=
∑
−
1
,
=
0
V
a
r
i
a
n
c
e
(
σ
2
)
A
mo
u
n
t
o
f
d
e
v
i
a
t
i
o
n
o
f
t
h
e
g
r
a
y
l
e
v
e
l
f
r
o
m t
h
e
a
v
e
r
a
g
e
v
a
l
u
e
(
2
)
=
∑
(
−
)
2
−
1
,
=
0
S
t
a
n
d
a
r
d
d
e
v
i
a
t
i
o
n
(
σ)
C
l
e
a
r
m
e
a
s
u
r
e
o
f
v
a
r
i
a
b
i
l
i
t
y
i
n
p
i
x
e
l
i
n
t
e
n
si
t
y
v
a
l
u
e
s
(
)
=
√
S
k
e
w
n
e
ss (
γ
)
Q
u
a
n
t
i
f
i
e
s
t
h
e
p
i
x
e
l
i
n
t
e
n
s
i
t
y
v
a
r
i
a
t
i
o
n
r
e
l
a
t
i
v
e
t
o
t
h
e
me
a
n
(
)
=
∑
(
−
)
3
3
−
1
,
=
0
K
u
r
t
o
si
s
(
k
)
Ev
a
l
u
a
t
e
t
h
e
p
e
a
k
i
n
e
ss
o
f
a
p
i
x
e
l
i
n
t
e
n
si
t
y
d
i
s
t
r
i
b
u
t
i
o
n
c
o
m
p
a
r
e
d
t
o
a
st
a
n
d
a
r
d
n
o
r
m
a
l
d
i
st
r
i
b
u
t
i
o
n
(
)
=
∑
(
−
)
4
4
−
1
,
=
0
-
3
C
o
r
r
e
l
a
t
i
o
n
(
ρ
)
Q
u
a
n
t
i
f
i
e
s
t
h
e
d
e
g
r
e
e
o
f
c
o
r
r
e
l
a
t
i
o
n
a
mo
n
g
a
p
i
x
e
l
a
n
d
i
t
s
n
e
i
g
h
b
o
r
i
n
g
p
i
x
e
l
s
(
)
=
∑
(
−
)
(
−
)
−
1
,
=
0
C
o
n
t
r
a
st
(
C
)
M
e
a
su
r
e
s
t
h
e
l
o
c
a
l
d
i
f
f
e
r
e
n
c
e
s
o
r
v
a
r
i
a
t
i
o
n
s
i
n
p
i
x
e
l
i
n
t
e
n
si
t
i
e
s
(
)
=
∑
(
−
)
2
−
1
,
=
0
En
e
r
g
y
o
r
u
n
i
f
o
r
mi
t
y
(
E)
M
e
a
su
r
e
s
h
o
w
e
v
e
n
l
y
o
r
h
o
m
o
g
e
n
e
o
u
s
l
y
t
h
e
i
n
t
e
n
si
t
i
e
s
o
f
t
h
e
p
i
x
e
l
s
i
n
a
n
i
mag
e
a
r
e
d
i
s
t
r
i
b
u
t
e
d
(
)
=
∑
2
−
1
,
=
0
H
o
mo
g
e
n
e
i
t
y
(
H
)
I
n
d
i
c
a
t
e
s
h
o
w
c
l
o
s
e
l
y
t
h
e
a
r
r
a
n
g
e
me
n
t
o
f
e
l
e
m
e
n
t
s
i
n
t
h
e
G
L
C
M
a
l
i
g
n
s
w
i
t
h
i
t
s
d
i
a
g
o
n
a
l
(
)
=
∑
1
+
(
−
)
2
−
1
,
=
0
En
t
r
o
p
y
(
E)
Ev
a
l
u
a
t
e
s
t
h
e
r
a
n
d
o
m
n
e
ss
o
r
c
o
m
p
l
e
x
i
t
y
i
n
p
i
x
e
l
i
n
t
e
n
s
i
t
i
e
s
(
)
=
−
∑
(
)
−
1
,
=
0
D
i
ssi
m
i
l
a
r
i
t
y
(
D
)
M
e
a
su
r
e
s
t
h
e
a
v
e
r
a
g
e
a
b
so
l
u
t
e
i
n
t
e
n
si
t
y
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
a
d
j
a
c
e
n
t
p
i
x
e
l
s
(
)
=
∑
|
−
|
−
1
,
=
0
A
n
g
u
l
a
r
se
c
o
n
d
m
o
me
n
t
(
A
S
M
)
H
i
g
h
e
r
A
S
M
d
e
mo
n
s
t
r
a
t
e
s
u
n
i
f
o
r
m
p
i
x
e
l
p
a
i
r
d
i
st
r
i
b
u
t
i
o
n
=
(
)
2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
5
6
-
3
5
6
7
3562
T
h
e
ex
tr
ac
ted
1
3
s
tatis
tical
a
n
d
tex
tu
r
e
f
ea
t
u
r
e
s
ets
wer
e
s
av
ed
as
a
co
m
m
a
-
s
ep
ar
ated
v
alu
e
(
.
csv
)
f
ile
f
o
r
f
u
r
t
h
er
a
n
aly
s
is
.
T
h
e
f
ea
tu
r
e
v
al
u
es
wer
e
n
o
r
m
aliz
ed
to
a
s
tan
d
ar
d
s
ca
le
u
s
in
g
th
e
m
in
-
m
ax
s
ca
lin
g
ap
p
r
o
ac
h
.
T
h
is
en
s
u
r
ed
all
f
e
atu
r
es
co
n
tr
ib
u
te
u
n
if
o
r
m
l
y
in
class
if
icatio
n
.
T
h
e
m
in
-
m
ax
s
ca
ler
f
u
n
ctio
n
was
av
ailab
le
in
th
e
Py
th
o
n
s
cik
i
t
-
lear
n
p
ac
k
ag
e.
Af
ter
n
o
r
m
a
lizatio
n
,
all
f
ea
tu
r
es
wer
e
co
n
v
er
ted
t
o
v
al
u
es
b
etwe
en
[
0
,
1
]
.
T
h
e
m
in
-
m
ax
s
ca
lin
g
f
u
n
ctio
n
f
o
r
a
g
iv
e
n
f
e
atu
r
e
in
th
e
d
ataset
is
ex
p
r
ess
ed
b
y
(
1
)
a
n
d
(
2
)
.
=
−
−
(
1
)
=
∗
(
−
)
+
(
2
)
W
h
er
e
is
th
e
f
in
al
n
o
r
m
alize
d
f
ea
tu
r
e
v
al
u
e
v
ar
y
in
g
in
r
an
g
e
b
etwe
en
[
m
in
,
m
ax
]
.
2
.
5
.
F
e
a
t
ure
v
a
lid
a
t
i
o
n
An
in
d
ep
en
d
en
t
t
-
test
[
2
5
]
wa
s
p
er
f
o
r
m
ed
to
e
x
am
in
e
th
e
p
r
ed
ictio
n
ab
ilit
y
o
f
th
e
ex
tr
ac
te
d
f
ea
tu
r
es.
I
t m
ea
s
u
r
ed
th
e
m
ea
n
d
if
f
er
e
n
ce
s
b
etwe
en
b
en
ig
n
an
d
m
alig
n
an
t c
ateg
o
r
ies.
T
h
is
s
tatis
tica
l v
alid
atio
n
m
eth
o
d
en
s
u
r
ed
th
at
o
n
ly
im
p
o
r
tan
t a
n
d
u
n
iq
u
e
f
ea
t
u
r
es im
p
ac
ted
th
e
class
if
icat
io
n
.
T
h
er
e
is
v
er
y
lim
ited
atten
tio
n
o
n
th
is
s
tati
s
tical
asp
ec
t in
ea
r
lier
s
tu
d
ies.
T
h
e
h
y
p
o
th
eses
wer
e
d
ef
in
e
d
as f
o
llo
ws:
–
Nu
ll h
y
p
o
th
esis
(
H
0
)
:
th
e
m
ea
n
f
ea
tu
r
e
v
alu
es we
r
e
eq
u
al
in
b
o
th
d
ia
g
n
o
s
tic
class
es.
–
Alter
n
ativ
e
h
y
p
o
th
esis
(
H
a
)
:
th
er
e
ex
is
ted
a
s
tatis
tically
m
ea
n
in
g
f
u
l d
if
f
er
e
n
ce
b
etwe
en
th
e
m
ea
n
f
ea
tu
r
e
v
alu
es o
f
class
es.
T
h
e
t
-
s
tatis
tic
is
ca
lcu
lated
u
s
i
n
g
th
e
(
3
)
.
=
1
̅
̅
̅
̅
−
2
̅
̅
̅
̅
.
√
1
1
+
1
2
(
3
)
T
h
e
p
o
o
led
s
tan
d
ar
d
d
ev
iatio
n
is
co
m
p
u
ted
u
s
in
g
th
e
(
4
)
.
=
√
(
1
−
1
)
1
2
+
(
2
−
1
)
2
2
1
+
2
−
2
(
4
)
W
h
er
e
1
̅
̅
̅
r
ep
r
esen
t
t
h
e
s
am
p
le
m
ea
n
o
f
th
e
b
en
i
g
n
f
ea
tu
r
e
class
an
d
2
̅
̅
̅
r
ep
r
esen
ts
th
e
m
alig
n
a
n
t
class
s
am
p
le
m
ea
n
,
1
an
d
2
d
en
o
te
th
eir
r
esp
e
ctiv
e
s
am
p
le
s
izes a
n
d
1
2
an
d
2
2
ar
e
th
e
co
r
r
esp
o
n
d
in
g
s
am
p
le
v
a
r
ian
ce
s
.
T
h
e
in
d
ep
en
d
en
t
t
-
test
was
ca
r
r
ied
o
u
t
u
s
in
g
t
h
e
ttes
t_
in
d
(
)
f
u
n
ctio
n
f
r
o
m
t
h
e
SciPy
lib
r
ar
y
.
R
esu
lts
s
h
o
wed
s
tatis
tically
m
ea
n
in
g
f
u
l
d
if
f
er
e
n
ce
s
(
p
<
0
.
0
5
)
f
o
r
1
2
am
o
n
g
th
e
1
3
f
ea
tu
r
es,
ex
ce
p
t
f
o
r
en
tr
o
p
y
.
Alth
o
u
g
h
en
tr
o
p
y
d
id
n
o
t
m
e
et
th
e
s
ig
n
if
ican
ce
th
r
esh
o
ld
(
p
>
0
.
0
5
)
,
it
was
s
elec
ted
b
e
ca
u
s
e
it
ef
f
ec
tiv
ely
ca
p
tu
r
ed
r
elev
an
t
tex
tu
r
e
p
atter
n
v
ar
iatio
n
s
in
d
er
m
o
s
co
p
ic
im
ag
es.
Am
o
n
g
all
f
ea
tu
r
es,
co
r
r
elatio
n
(
t
=
1
6
.
4
5
)
an
d
h
o
m
o
g
e
n
eity
(
t
=
1
5
.
7
7
)
d
em
o
n
s
tr
ated
th
e
h
ig
h
est
t
-
s
co
r
es,
wh
ich
p
r
o
v
ed
th
ey
h
av
e
a
s
tr
o
n
g
ab
ilit
y
to
d
if
f
er
e
n
tiate
th
e
c
lass
es.
Ov
er
all,
th
e
t
-
test
r
e
s
u
lts
v
alid
ated
th
at
th
e
e
x
tr
ac
ted
f
ea
tu
r
es
ar
e
s
tatis
t
ically
s
ig
n
if
ican
t a
n
d
s
u
i
tab
le
f
o
r
th
e
p
r
o
p
o
s
ed
GT
-
SV
M
f
r
am
ewo
r
k
.
2
.
6
.
B
ina
ry
cla
s
s
if
ica
t
io
n us
ing
g
eo
m
e
t
ric
t
ra
ns
f
o
r
m
a
t
io
n
–
a
ug
m
ent
ed
s
up
po
rt
v
ec
t
o
r
m
a
chine
T
o
ex
ec
u
te
b
in
ar
y
class
if
icati
o
n
u
s
in
g
th
e
p
r
o
p
o
s
ed
GT
-
S
VM
f
r
am
ewo
r
k
,
th
e
v
alid
ate
d
f
ea
tu
r
e
s
et
was
s
ep
ar
ated
in
to
a
tr
ain
an
d
test
s
et
at
a
s
p
lit
r
atio
o
f
8
0
:2
0
.
T
h
e
tr
ain
s
u
b
s
et
was
ap
p
lied
to
o
p
tim
ize
th
e
class
if
ier
’
s
p
ar
am
eter
s
,
wh
ile
th
e
test
d
ata
was
m
ain
ly
em
p
l
o
y
ed
f
o
r
ass
ess
in
g
th
e
m
o
d
el’
s
p
er
f
o
r
m
a
n
ce
.
T
h
e
GT
-
SVM
alg
o
r
ith
m
class
if
ied
ea
ch
s
am
p
le
as
a
b
en
ig
n
o
r
m
alig
n
an
t
ca
teg
o
r
y
d
ep
e
n
d
in
g
o
n
th
e
s
tatis
tical
ly
ex
tr
ac
ted
s
ig
n
if
ica
n
t
f
ea
t
u
r
es.
GT
-
SVM
is
d
ev
el
o
p
ed
f
r
o
m
th
e
c
o
n
v
e
n
tio
n
al
SVM
m
o
d
el,
wh
ich
is
p
r
o
v
en
ef
f
ec
tiv
e
f
o
r
b
o
th
class
if
icatio
n
an
d
r
eg
r
ess
io
n
o
p
er
atio
n
s
.
SVM
f
o
cu
s
es
o
n
d
eter
m
in
in
g
th
e
m
o
s
t
s
u
itab
le
h
y
p
er
p
lan
e
th
at
m
a
x
im
izes th
e
s
ep
ar
atio
n
g
a
p
b
etwe
en
d
is
tin
ct
class
es
[
2
6
]
.
Giv
en
a
tr
ain
s
et
{
(
x
,
y
)
}
=
1
,
wh
er
e
x
∈
ℝ
r
ep
r
esen
ts
an
m
-
d
im
en
s
io
n
al
f
ea
tu
r
e
m
atr
ix
an
d
y
∈
{
−
1
,
1
}
ar
e
two
-
class
lab
els,
th
e
o
p
ti
m
izatio
n
p
r
o
b
lem
is
f
o
r
m
u
late
d
as in
(
5
)
.
,
,
1
2
∥
∥
2
+
∑
=
1
lim
ited
b
y
(
.
+
)
≥
1
−
,
≥
0
,
=
1
,
.
.
.
(
5
)
W
h
er
e
in
d
icate
s
th
e
weig
h
t
p
ar
am
eter
s
,
is
th
e
o
f
f
s
et
f
ac
to
r
,
ar
e
s
lack
v
ar
ia
b
les
f
o
r
h
an
d
lin
g
m
is
class
if
ied
in
s
tan
ce
s
,
an
d
is
a
p
en
alty
co
ef
f
icien
t
r
e
g
u
latin
g
th
e
b
alan
ce
b
etwe
en
m
a
x
im
i
zin
g
m
ar
g
in
a
n
d
m
in
im
izin
g
class
if
icatio
n
er
r
o
r
.
T
o
h
a
n
d
le
n
o
n
lin
ea
r
s
ep
a
r
atio
n
s
,
r
ad
ial
b
asis
f
u
n
ctio
n
(
R
B
F)
k
er
n
el
was
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
ma
ch
in
e
lea
r
n
i
n
g
fr
a
mewo
r
k
fo
r
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
u
s
in
g
textu
r
e
d
escr
ip
to
r
s
(
S
h
ikh
a
Ma
lik
)
3563
ap
p
lied
in
th
e
GT
-
SVM
f
r
am
ewo
r
k
.
T
h
is
tech
n
iq
u
e
im
p
licitly
p
r
o
jects
th
e
in
p
u
t
f
ea
tu
r
es
in
to
a
lar
g
e
f
ea
tu
r
e
s
p
ac
e
u
s
in
g
th
e
(
6
)
.
(
,
)
=
(
−
∥
−
∥
2
)
(
6
)
W
h
er
e
is
a
h
y
p
er
p
ar
am
eter
th
at
d
eter
m
in
es
th
e
k
er
n
el’
s
s
en
s
itiv
ity
to
p
air
wis
e
d
is
tan
ce
s
b
etwe
en
f
ea
tu
r
e
v
ec
to
r
s
.
an
d
.
Gr
id
s
ea
r
ch
a
p
p
r
o
ac
h
with
f
iv
e
-
f
o
ld
c
r
o
s
s
-
v
alid
atio
n
(
cv
=5
)
was
u
s
ed
to
o
p
tim
ize
h
y
p
er
p
ar
am
eter
s
b
y
im
p
lem
en
tin
g
s
cik
it
-
lear
n
’
s
Gr
id
Sear
ch
C
V
o
p
er
atio
n
.
T
h
e
b
est
co
n
f
ig
u
r
atio
n
was
o
b
tain
ed
with
C
=
1
,
0
0
0
,
γ
=
1
,
an
d
th
e
R
B
F
k
er
n
el.
T
h
is
y
ield
ed
s
u
p
er
io
r
class
if
icatio
n
ac
cu
r
ac
y
b
y
ef
f
ec
tiv
ely
b
alan
cin
g
m
ar
g
in
wid
th
a
n
d
e
r
r
o
r
m
in
i
m
izatio
n
.
T
h
e
f
i
n
al
GT
-
SVM
f
r
am
ewo
r
k
was
tr
ain
e
d
o
n
m
o
r
p
h
o
lo
g
y
p
r
eser
v
i
n
g
an
d
r
o
tatio
n
-
in
v
a
r
ian
t
tex
tu
r
e
f
ea
tu
r
es
an
d
d
em
o
n
s
tr
ated
r
eliab
le
d
is
cr
im
in
atio
n
b
e
twee
n
b
en
ig
n
an
d
m
alig
n
an
t le
s
io
n
s
.
3.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S
AND
D
I
SC
USS
I
O
NS
All
co
m
p
u
tatio
n
al
ex
p
er
im
e
n
ts
wer
e
ca
r
r
ied
o
u
t
o
n
a
W
in
d
o
ws
1
0
o
p
er
atin
g
s
y
s
tem
eq
u
ip
p
ed
with
a
2
.
5
GHz
p
r
o
ce
s
s
o
r
an
d
8
GB
o
f
R
A
M.
T
h
is
en
s
u
r
ed
s
t
ab
le
an
d
ef
f
icien
t
p
er
f
o
r
m
an
ce
th
r
o
u
g
h
o
u
t
th
e
ex
p
er
im
en
ts
.
T
h
is
co
n
f
ig
u
r
atio
n
en
ab
le
d
th
e
s
m
o
o
t
h
ex
ec
u
tio
n
o
f
ML
wo
r
k
f
lo
ws
an
d
h
elp
ed
with
q
u
ick
e
r
tr
ain
in
g
a
n
d
p
r
e
d
ictio
n
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
.
T
h
e
Py
t
h
o
n
3
.
1
1
i
n
th
e
Sp
y
d
er
en
v
ir
o
n
m
en
t
a
n
d
Go
o
g
le
C
o
lab
was
u
s
ed
f
o
r
p
er
f
o
r
m
in
g
ex
p
er
im
en
ts
b
o
t
h
lo
ca
lly
an
d
in
th
e
clo
u
d
.
T
h
e
s
tan
d
a
r
d
Py
th
o
n
p
ac
k
ag
es
lik
e
Nu
m
Py
,
Pan
d
as,
Op
en
C
V,
an
d
s
cik
it
-
lear
n
wer
e
ap
p
lied
to
p
er
f
o
r
m
th
e
s
im
u
latio
n
w
o
r
k
.
T
h
e
p
er
f
o
r
m
a
n
ce
p
a
r
am
eter
s
m
ea
s
u
r
ed
wer
e
ac
c
u
r
ac
y
,
p
r
e
cisi
o
n
,
T
PR
,
s
p
ec
if
icity
,
an
d
F1
-
s
co
r
e
.
T
h
eir
b
r
ief
d
escr
ip
tio
n
is
p
r
o
v
id
ed
in
T
ab
le
4
.
T
h
e
tr
u
e
p
o
s
itiv
es
(
T
P)
m
ea
n
co
r
r
ec
tly
class
if
ied
m
alig
n
an
t
(
p
o
s
itiv
e)
s
am
p
les,
wh
ile
f
al
s
e
p
o
s
i
tiv
es
(
FP
)
d
e
n
o
te
in
c
o
r
r
ec
tly
id
en
tifie
d
b
en
i
g
n
(
n
eg
ativ
e)
s
am
p
les.
C
o
n
v
er
s
ely
,
tr
u
e
n
e
g
ativ
es
(
T
N)
r
ep
r
esen
t
ac
cu
r
ately
r
ec
o
g
n
ized
b
e
n
ig
n
s
am
p
les,
an
d
f
a
ls
e
n
eg
ativ
es
(
FN)
in
d
icate
in
co
r
r
ec
t c
lass
if
icatio
n
o
f
m
alig
n
an
t sam
p
les.
T
ab
le
4
.
Su
m
m
a
r
y
o
f
p
e
r
f
o
r
m
an
ce
p
ar
am
eter
s
M
e
t
r
i
c
C
a
l
c
u
l
a
t
i
o
n
C
l
i
n
i
c
a
l
m
e
a
n
i
n
g
A
c
c
u
r
a
c
y
+
+
+
+
M
e
a
su
r
e
s
o
v
e
r
a
l
l
p
r
e
d
i
c
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
P
r
e
c
i
s
i
o
n
+
C
a
p
a
b
i
l
i
t
y
t
o
r
e
d
u
c
e
f
a
l
s
e
p
o
si
t
i
v
e
s
TPR
+
M
e
a
su
r
e
t
h
e
d
e
t
e
c
t
i
o
n
r
a
t
e
o
f
t
h
e
p
o
s
i
t
i
v
e
c
l
a
ss
S
p
e
c
i
f
i
c
i
t
y
+
M
e
a
su
r
e
s t
h
e
c
o
r
r
e
c
t
b
e
n
i
g
n
c
a
ses
.
F1
-
sc
o
r
e
2
∗
∗
+
B
a
l
a
n
c
e
s
mo
d
e
l
’
s
p
r
e
c
i
s
i
o
n
a
n
d
r
e
c
a
l
l
v
a
l
u
e
s
3
.
1
.
K
ey
f
ind
ing
s
a
nd
inte
rp
re
t
a
t
io
n
T
h
e
class
if
icatio
n
p
er
f
o
r
m
an
c
e
o
f
th
e
p
r
o
p
o
s
ed
GT
-
SVM
f
r
am
ewo
r
k
o
n
th
e
I
SIC
-
2
0
1
6
an
d
PH
2
d
atasets
was
co
m
p
ar
ed
with
t
h
e
ex
is
tin
g
ML
alg
o
r
ith
m
s
,
m
ain
ly
lo
g
is
tic
r
eg
r
ess
io
n
(
L
R
)
,
r
an
d
o
m
f
o
r
est
(
R
F),
K
NN,
ML
P,
an
d
ex
tr
em
e
g
r
ad
ien
t
b
o
o
s
tin
g
(
XGBo
o
s
t)
,
a
n
d
p
r
esen
te
d
in
T
a
b
le
5
.
T
h
e
s
im
u
latio
n
r
esu
lts
h
ig
h
lig
h
t
th
at
th
e
GT
-
SVM
f
r
am
ewo
r
k
tr
ain
ed
o
n
1
3
ex
tr
ac
ted
f
ea
tu
r
es
s
h
o
wed
s
u
p
er
io
r
class
if
icatio
n
p
er
f
o
r
m
an
ce
co
m
p
ar
ed
to
e
x
is
tin
g
b
aselin
e
m
eth
o
d
s
,
with
an
ac
cu
r
ac
y
s
co
r
e
o
f
8
0
% o
n
I
SIC
-
2
0
1
6
an
d
9
5
.
3
%
o
n
th
e
PH
2
d
ataset.
T
h
e
T
PR
o
f
8
1
.
1
%
o
n
I
SIC
-
2
0
1
6
a
n
d
9
3
%
o
n
PH
2
in
d
icate
s
th
at
th
e
f
r
am
ewo
r
k
is
co
r
r
ec
tly
id
en
tify
i
n
g
p
o
s
itiv
e
ca
s
es.
T
h
e
s
p
ec
if
icity
s
co
r
e
o
f
7
8
.
1
%
o
n
I
SIC
-
2
0
1
6
an
d
9
7
.
2
%
o
n
PH
2
d
atasets
co
n
f
ir
m
s
p
r
ec
is
e
d
etec
tio
n
o
f
b
en
ig
n
ca
s
es.
T
h
e
p
r
ec
is
io
n
v
alu
es
o
f
8
2
%
an
d
9
6
.
3
%
r
ev
ea
l
th
at
th
e
m
o
d
el
h
as
f
ewe
r
f
alse
-
p
o
s
itiv
e
p
r
ed
ictio
n
s
,
wh
ile
th
e
F1
-
s
co
r
e
o
f
8
1
.
3
%
an
d
9
5
%
in
d
icate
a
n
e
f
f
ec
tiv
e
b
alan
ce
b
etwe
e
n
p
r
ec
is
io
n
an
d
r
ec
all
p
ar
am
ete
r
s
.
No
tab
ly
,
th
e
GT
-
SVM
f
r
am
ewo
r
k
s
u
cc
ess
f
u
lly
attain
ed
h
ig
h
s
en
s
itiv
ity
with
o
u
t
co
m
p
r
o
m
is
in
g
s
p
ec
if
icity
.
T
h
is
en
ab
les
ac
cu
r
ate
id
en
tific
atio
n
o
f
p
o
s
itiv
e
ca
s
es
w
ith
f
ewe
r
f
alse
p
r
ed
ictio
n
s
.
T
h
e
T
PR
v
alu
es
ac
r
o
s
s
b
o
th
d
atasets
w
er
e
co
n
s
is
ten
tly
h
ig
h
,
p
ar
ticu
l
ar
ly
th
e
9
3
%
ac
h
iev
ed
o
n
PH
2
.
T
h
is
co
n
f
ir
m
ed
th
at
m
o
r
p
h
o
lo
g
y
p
r
eser
v
in
g
GT
ef
f
ec
tiv
ely
allev
iated
th
e
class
im
b
alan
ce
is
s
u
e,
wh
ich
r
esu
lted
in
h
ig
h
T
PR
.
Ov
er
all,
th
e
f
r
am
ewo
r
k
ac
h
iev
ed
its
k
ey
o
b
jectiv
e
o
f
en
h
an
cin
g
m
alig
n
a
n
t
lesi
o
n
d
etec
tio
n
a
n
d
is
in
ter
p
r
etab
le
an
d
co
m
p
u
tatio
n
ally
ef
f
icien
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
5
6
-
3
5
6
7
3564
T
ab
le
5
.
C
o
m
p
a
r
is
o
n
o
f
GT
-
SVM
with
co
n
v
en
tio
n
al
m
o
d
els
f
o
r
class
if
icatio
n
D
a
t
a
s
e
t
M
o
d
e
l
s
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
TPR
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
F1
-
sc
o
r
e
(
%)
I
S
I
C
-
2
0
1
6
LR
76
78
7
7
.
9
73
7
7
.
7
RF
7
5
.
1
7
7
.
8
7
6
.
3
7
3
.
7
77
K
N
N
7
4
.
5
76
7
7
.
9
7
0
.
3
7
6
.
9
M
LP
76
7
7
.
6
7
7
.
9
72
7
7
.
7
X
G
B
o
o
st
7
7
.
1
7
9
.
5
7
8
.
3
7
5
.
7
7
8
.
8
P
r
o
p
o
se
d
G
T
-
S
V
M
80
82
8
1
.
1
7
8
.
1
8
1
.
3
PH
2
LR
91
8
9
.
2
8
9
.
2
92
8
9
.
2
RF
91
8
4
.
3
92
8
6
.
1
90
K
N
N
91
87
89
7
0
.
3
90
M
LP
9
2
.
1
90
93
92
9
1
.
2
X
G
B
o
o
st
91
8
4
.
3
9
2
.
2
8
6
.
1
90
P
r
o
p
o
se
d
G
T
-
S
V
M
9
5
.
3
9
6
.
3
93
9
7
.
2
95
T
h
e
r
ec
eiv
er
o
p
er
atin
g
ch
ar
a
cter
is
tic
(
R
OC
)
cu
r
v
es
ar
e
p
l
o
tted
in
Fig
u
r
e
4
at
d
if
f
er
en
t
th
r
esh
o
ld
v
alu
es
to
f
u
r
th
er
ass
ess
th
e
m
o
d
el’
s
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
.
Fig
u
r
e
4
(
a)
d
is
p
lay
s
th
e
R
OC
cu
r
v
e
ac
r
o
s
s
th
e
I
SIC
-
2
0
1
6
d
ataset,
a
n
d
Fi
g
u
r
e
4
(
b
)
d
is
p
lay
s
th
e
R
OC
cu
r
v
e
co
r
r
esp
o
n
d
in
g
to
th
e
PH
2
d
ataset.
I
t
attain
ed
a
h
ig
h
er
ar
ea
u
n
d
er
th
e
cu
r
v
e
(
AUC
)
s
co
r
e
o
f
0
.
9
9
o
n
th
e
PH
2
d
ataset,
th
at
p
r
o
v
e
th
at
th
e
m
o
d
el
h
as
a
s
tr
o
n
g
ca
p
ab
ilit
y
o
f
id
en
tif
y
in
g
m
alig
n
an
t
ca
s
es
co
r
r
ec
tly
.
I
n
co
m
p
a
r
is
o
n
,
th
e
AUC
o
f
0
.
8
2
o
n
th
e
I
SIC
-
2
0
1
6
d
ataset
was
s
lig
h
tly
lo
wer
b
e
ca
u
s
e
im
ag
e
s
am
p
les
in
th
is
d
ataset
ar
e
h
ig
h
ly
d
iv
er
s
e,
wh
ich
ch
allen
g
es
th
e
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
el.
Ov
er
all,
th
ese
r
es
u
lts
co
n
f
ir
m
e
d
th
at
t
h
e
GT
-
S
VM
f
r
am
ewo
r
k
ca
n
ef
f
ec
tiv
ely
d
is
tin
g
u
is
h
all
clas
s
es o
n
th
e
PH
2
d
ataset
wh
ile
m
ain
tain
in
g
h
i
g
h
s
en
s
itiv
ity
a
n
d
s
p
ec
if
icity
.
(
a)
(
b
)
Fig
u
r
e
4
.
R
OC
p
lo
t o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ew
o
r
k
ac
r
o
s
s
(
a)
I
S
I
C
-
2
0
1
6
d
ataset
an
d
(
b
)
PH
2
d
ataset
3
.
2
.
Co
m
pa
ra
t
iv
e
ev
a
lua
t
io
n wit
h e
x
is
t
ing
m
et
ho
ds
T
h
e
p
r
o
p
o
s
ed
GT
-
SVM
class
if
icatio
n
f
r
am
ewo
r
k
was
ev
alu
ated
ag
ain
s
t
ex
is
tin
g
s
tu
d
i
es
o
n
th
e
I
SIC
-
2
0
1
6
an
d
PH
2
d
atasets
.
T
ab
le
6
p
r
esen
ts
a
T
PR
co
m
p
ar
is
o
n
ac
r
o
s
s
th
e
I
SIC
-
2
0
1
6
d
ataset.
Simu
latio
n
r
esu
lts
d
em
o
n
s
tr
ated
th
at
t
h
e
GT
-
SVM
f
r
am
ewo
r
k
ac
h
iev
e
d
a
T
PR
o
f
8
1
.
1
% o
n
I
SIC
-
2
0
1
6
an
d
o
u
tp
er
f
o
r
m
ed
m
o
s
t
ex
is
tin
g
b
en
ch
m
a
r
k
m
o
d
els.
T
h
e
en
s
em
b
le
m
o
d
el
p
r
o
p
o
s
ed
b
y
Alizad
eh
an
d
Ma
h
lo
o
jifa
r
[
1
1
]
d
id
n
o
t
ef
f
ec
tiv
ely
m
in
im
ize
class
im
b
alan
ce
,
wh
ich
r
esu
lted
i
n
lo
w
T
PR
.
Alth
o
u
g
h
th
e
m
o
d
els
p
r
o
p
o
s
ed
in
[
1
2
]
,
[
1
7
]
i
n
co
r
p
o
r
ated
GT
an
d
d
ee
p
class
if
icatio
n
ar
ch
itectu
r
es,
th
eir
f
o
c
u
s
was
m
o
r
e
o
n
im
p
r
o
v
in
g
s
p
ec
if
icity
r
ath
er
th
an
b
alan
ce
d
d
etec
tio
n
.
I
n
co
n
tr
ast,
th
e
A
R
DT
-
Den
s
eNe
t m
o
d
el
p
r
o
p
o
s
ed
b
y
W
u
et
a
l
.
[
1
8
]
ac
h
iev
ed
a
s
lig
h
tly
h
ig
h
e
r
T
PR
o
f
8
1
.
6
%,
b
u
t
th
e
d
ee
p
an
d
co
m
p
le
x
d
esig
n
o
f
t
h
e
m
o
d
el
m
ad
e
it
less
in
ter
p
r
etab
le
an
d
c
o
m
p
u
tatio
n
ally
in
ef
f
icien
t.
Un
lik
e
th
es
e
m
o
d
els,
th
e
p
r
o
p
o
s
ed
GT
-
SVM
f
r
am
ewo
r
k
is
lig
h
tweig
h
t,
in
ter
p
r
etab
le,
an
d
co
m
p
u
tatio
n
ally
e
f
f
icien
t
t
h
at
is
m
ain
tain
in
g
h
ig
h
s
en
s
itiv
ity
wh
ile
ef
f
ec
tiv
el
y
ad
d
r
ess
in
g
class
im
b
alan
ce
.
T
ab
le
7
d
ep
icts
a
co
m
p
ar
is
o
n
o
f
GT
-
SVM
an
d
o
th
e
r
b
aselin
e
s
tu
d
ies
o
n
th
e
PH
2
d
ataset.
T
h
e
AB
C
D
m
eth
o
d
im
p
lem
e
n
ted
in
[
7
]
,
[
8
]
d
id
n
o
t
in
co
r
p
o
r
ate
an
y
ML
alg
o
r
ith
m
s
an
d
was
o
n
ly
b
a
s
ed
o
n
co
n
v
e
n
tio
n
al
im
ag
e
p
r
o
ce
s
s
in
g
.
T
h
u
s
,
th
eir
m
o
d
els we
r
e
less
ca
p
ab
le
o
f
wo
r
k
in
g
r
eliab
ly
in
p
h
y
s
ical
clin
ical
en
v
ir
o
n
m
en
ts
.
Haid
er
et
a
l
.
[
1
3
]
ac
h
iev
e
d
a
T
PR
o
f
8
9
.
2
%
b
y
ap
p
ly
i
n
g
ac
tiv
e
co
n
to
u
r
s
eg
m
e
n
tatio
n
,
L
B
P
d
escr
ip
to
r
s
,
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
ma
ch
in
e
lea
r
n
i
n
g
fr
a
mewo
r
k
fo
r
s
kin
ca
n
ce
r
cla
s
s
ifica
tio
n
u
s
in
g
textu
r
e
d
escr
ip
to
r
s
(
S
h
ikh
a
Ma
lik
)
3565
an
SVM
class
if
ier
.
Ho
wev
er
,
th
ey
wo
r
k
e
d
o
n
im
b
alan
ce
d
d
ata,
wh
ich
m
ad
e
t
h
e
m
o
d
el
clin
ically
u
n
r
eliab
le
.
An
o
p
tim
ized
f
r
am
ewo
r
k
p
r
o
p
o
s
ed
b
y
B
alah
a
an
d
Hass
an
[
1
9
]
co
m
b
i
n
ed
Den
s
eNe
t
-
2
0
1
ar
ch
itectu
r
e
an
d
th
e
SS
O
alg
o
r
ith
m
.
I
t
attain
ed
9
4
.
7
5
%
ac
cu
r
ac
y
an
d
9
2
.
7
4
%
T
PR
.
Ho
wev
er
,
th
e
f
r
am
ewo
r
k
is
n
o
t
co
m
p
u
tatio
n
ally
ef
f
icien
t
b
ec
a
u
s
e
it
in
clu
d
es
m
u
ltip
le
d
ee
p
l
ea
r
n
in
g
an
d
o
p
tim
izatio
n
lay
e
r
s
.
I
n
co
m
p
a
r
is
o
n
,
th
e
p
r
o
p
o
s
ed
GT
-
SVM
m
o
d
el
ac
h
iev
ed
9
5
.
3
%
ac
cu
r
ac
y
,
9
3
%
T
PR
,
an
d
9
7
.
2
%
s
p
ec
if
icity
with
o
u
t
th
e
in
teg
r
atio
n
o
f
an
y
co
m
p
lex
la
y
er
.
T
h
er
ef
o
r
e,
its
p
er
f
o
r
m
a
n
ce
is
s
u
p
er
io
r
with
lo
w
co
m
p
u
tatio
n
al
co
s
t.
T
h
e
m
o
d
el
is
ca
p
ab
le
o
f
d
etec
tin
g
m
alig
n
a
n
t
lesi
o
n
s
r
eliab
ly
b
y
ef
f
e
ctiv
ely
m
in
im
izin
g
f
a
ls
e
p
o
s
itiv
es.
T
h
is
v
alid
ates it
as a
r
o
b
u
s
t a
n
d
i
n
ter
p
r
etab
le
ap
p
r
o
ac
h
f
o
r
au
to
m
ated
d
etec
tio
n
o
f
s
k
in
ca
n
ce
r
.
T
ab
le
6
.
C
o
m
p
a
r
ativ
e
T
PR
ass
ess
m
en
t o
f
th
e
p
r
o
p
o
s
ed
f
r
am
e
wo
r
k
ag
ain
s
t b
aselin
es o
n
I
SI
C
-
2016
M
o
d
e
l
TPR
(
%)
C
N
N
+
H
a
r
a
l
i
c
k
f
e
a
t
u
r
e
s
[
1
1
]
52
M
u
l
t
i
-
C
N
N
a
n
d
S
V
M
[
1
2
]
6
0
.
1
GP
-
C
N
N
-
D
TEL
[
1
7
]
32
A
R
D
T
-
D
e
n
s
e
N
e
t
[
1
8
]
8
1
.
6
P
r
o
p
o
se
d
G
T
-
S
V
M
8
1
.
1
T
ab
le
7
.
C
o
m
p
a
r
is
o
n
o
f
th
e
G
T
-
SVM
f
r
am
ewo
r
k
a
g
ain
s
t b
a
s
elin
es o
n
th
e
PH
2
d
ataset
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
TPR
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
A
B
C
D
[
7
]
90
85
9
2
.
2
2
A
B
C
D
[
8
]
8
4
.
0
0
6
0
.
5
0
8
9
.
5
0
LB
P
a
n
d
S
V
M
[
1
3
]
8
9
.
2
8
9
.
2
84
D
e
n
seN
e
t
-
2
0
1
a
n
d
S
S
O
[
1
9
]
9
4
.
7
5
9
2
.
7
4
9
6
.
2
0
P
r
o
p
o
se
d
G
T
-
S
V
M
9
5
.
3
93
9
7
.
2
3
.
3
.
P
r
a
ct
ica
l i
m
pli
ca
t
io
ns
a
nd
f
uture
s
co
pe
T
h
e
GT
-
SVM
f
r
am
ewo
r
k
e
x
h
ib
ited
a
r
o
b
u
s
t,
in
ter
p
r
etab
le,
an
d
co
m
p
u
tatio
n
ally
ef
f
icien
t
d
iag
n
o
s
tic
ap
p
r
o
ac
h
f
o
r
t
h
e
b
in
ar
y
class
if
icatio
n
o
f
d
e
r
m
o
s
co
p
ic
im
a
g
es
f
o
r
th
e
d
etec
tio
n
o
f
s
k
in
c
an
ce
r
.
T
h
e
f
i
n
d
in
g
s
r
ev
ea
led
th
at
m
o
d
el
r
esu
lts
ar
e
h
ig
h
ly
s
u
p
er
io
r
co
m
p
ar
ed
with
s
ev
er
al
b
aselin
e
ap
p
r
o
ac
h
e
s
.
I
t
is
lig
h
tweig
h
t
an
d
p
r
o
v
id
es
h
ig
h
T
PR
,
wh
ich
en
h
a
n
ce
s
its
ab
ilit
y
f
o
r
d
ep
l
o
y
m
en
t
i
n
p
h
y
s
ical
h
ea
lth
ca
r
e
p
r
ac
tice.
Ho
wev
e
r
,
th
e
cu
r
r
e
n
t
f
r
a
m
ewo
r
k
is
s
u
itab
le
o
n
ly
f
o
r
b
in
a
r
y
class
if
icatio
n
an
d
ca
n
n
o
t b
e
v
alid
ated
ac
r
o
s
s
d
iv
er
s
e
clin
ical
d
atasets
.
T
o
f
u
r
th
er
en
h
an
ce
its
g
en
er
aliza
tio
n
an
d
th
e
clin
ical
ap
p
licab
ilit
y
o
f
th
is
f
r
am
ewo
r
k
,
f
u
tu
r
e
d
ev
elo
p
m
e
n
ts
s
h
all
ex
p
lo
r
e
th
e
m
o
d
el
f
o
r
th
e
class
if
icatio
n
o
f
m
u
ltip
le
lesi
o
n
ca
te
g
o
r
ies,
in
co
r
p
o
r
atin
g
ad
d
itio
n
al
h
y
b
r
i
d
f
ea
tu
r
e
d
es
cr
ip
to
r
s
s
u
ch
as
c
o
lo
r
,
s
h
ap
e
,
an
d
d
ee
p
lear
n
in
g
f
ea
t
u
r
es,
an
d
u
s
in
g
d
iv
er
s
e
clin
ical
d
atasets
will im
p
r
o
v
e
its
g
en
er
aliza
tio
n
an
d
clin
ical
ap
p
licab
ilit
y
.
4.
CO
NCLU
SI
O
N
T
h
e
GT
-
SVM
f
r
am
ewo
r
k
p
r
e
s
en
ted
in
th
is
s
tu
d
y
is
a
r
o
b
u
s
t
an
d
in
ter
p
r
eta
b
le
ML
ap
p
r
o
a
ch
f
o
r
th
e
b
in
ar
y
class
if
icatio
n
o
f
d
er
m
o
s
co
p
ic
im
a
g
es
f
o
r
s
k
in
ca
n
ce
r
d
etec
tio
n
.
T
h
e
r
o
tati
o
n
-
in
v
a
r
ian
t
te
x
tu
r
e
d
escr
ip
to
r
s
d
er
iv
ed
f
r
o
m
th
e
GL
C
M
ap
p
r
o
ac
h
an
d
m
o
r
p
h
o
lo
g
y
p
r
eser
v
in
g
GT
h
av
e
b
ee
n
im
p
lem
en
ted
f
o
r
th
e
GT
-
SVM
f
r
am
ewo
r
k
.
T
h
e
f
r
a
m
ewo
r
k
im
p
r
o
v
ed
t
h
e
T
PR
b
y
ad
d
r
ess
in
g
class
im
b
alan
ce
is
s
u
es
in
an
ef
f
ec
tiv
e
way
.
T
h
e
f
r
am
ew
o
r
k
ac
h
iev
ed
a
T
PR
o
f
8
1
.
1
%
o
n
I
SIC
-
2
0
1
6
an
d
9
3
%
o
n
PH
2
with
ac
cu
r
ac
y
v
alu
es
o
f
8
0
%
an
d
9
5
.
3
%,
r
esp
ec
tiv
el
y
.
C
o
m
p
ar
ed
t
o
co
m
p
lex
d
ee
p
lea
r
n
in
g
m
o
d
els
av
ailab
le
f
o
r
th
e
class
if
icatio
n
o
f
d
er
m
o
s
co
p
ic
im
a
g
es,
th
e
GT
-
SVM
f
r
am
ewo
r
k
is
a
lig
h
twe
ig
h
t,
in
ter
p
r
eta
b
le,
an
d
co
m
p
u
tatio
n
ally
ef
f
icien
t
f
r
am
ewo
r
k
with
h
ig
h
s
en
s
itiv
ity
an
d
with
o
u
t
co
m
p
r
o
m
is
in
g
s
p
ec
if
icity
.
T
h
is
f
r
am
ewo
r
k
is
ab
le
to
d
etec
t
m
alig
n
an
t
ca
n
ce
r
s
in
an
e
f
f
ic
ien
t
way
;
th
er
ef
o
r
e,
it
is
s
u
itab
le
f
o
r
t
h
e
tim
ely
ass
ess
m
en
t
o
f
s
k
in
c
an
ce
r
i
n
p
h
y
s
ical
clin
ical
en
v
ir
o
n
m
en
ts
.
T
o
f
u
r
th
er
s
u
p
p
o
r
t
th
e
c
lin
ical
ap
p
licatio
n
s
o
f
th
is
f
r
am
ewo
r
k
,
f
u
t
u
r
e
d
ev
elo
p
m
e
n
ts
s
h
all
ex
p
lo
r
e
m
u
lti
-
class
lesi
o
n
class
if
icati
o
n
,
v
alid
atio
n
ac
r
o
s
s
d
iv
er
s
e
clin
ical
d
atasets
,
an
d
h
y
b
r
id
f
ea
tu
r
e
f
u
s
io
n
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.