I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
2
,
A
p
r
il
20
25
,
p
p
.
2
4
0
4
~
2
4
1
5
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
2
.
pp
2
4
0
4
-
2
4
1
5
2404
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
A comp
rehe
nsiv
e
a
na
ly
sis
of dif
fere
nt
mo
dels:
skin c
a
ncer
detec
tion
Am
rut
a
T
ho
ra
t
1
,
Cha
y
a
J
a
d
ha
v
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
T
e
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
E
n
g
i
n
e
e
r
i
n
g
,
D
r
.
D
.
Y
.
P
a
t
i
l
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
P
i
mp
r
i
,
P
u
n
e
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
D
r
.
D
.
Y
.
P
a
t
i
l
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
P
i
m
p
r
i
,
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
J
u
n
2
6
,
2
0
2
4
R
ev
is
ed
Dec
9
,
2
0
2
4
Acc
ep
ted
Dec
1
9
,
2
0
2
4
Du
e
to
fa
st
-
g
r
o
win
g
wo
rl
d
wid
e
a
ir
p
o
l
lu
ti
o
n
a
n
d
o
z
o
n
e
lay
e
r
d
e
str
u
c
ti
o
n
,
a
n
a
larm
in
g
n
u
m
b
e
r
o
f
p
e
o
p
le
a
re
fo
u
n
d
t
o
h
a
v
e
sk
i
n
c
a
n
c
e
r,
m
o
re
th
a
n
a
n
y
o
th
e
r
k
i
n
d
o
f
c
a
n
c
e
r
c
o
m
b
in
e
d
.
It
is
k
n
o
wn
to
b
e
o
n
e
o
f
th
e
d
e
a
d
li
e
st
m
a
li
g
n
a
n
c
ies
;
if
n
o
t
i
d
e
n
ti
fied
a
n
d
c
u
re
d
in
i
ts
e
a
rly
sta
g
e
s,
it
i
s
li
k
e
l
y
to
sp
re
a
d
to
o
th
e
r
b
o
d
y
p
a
rts.
Early
d
e
tec
ti
o
n
is
c
rit
ica
l
a
n
d
h
e
lp
s
p
re
v
e
n
t
c
a
n
c
e
r
fro
m
sp
re
a
d
in
g
.
T
h
is
a
ll
o
ws
fo
r
e
a
rly
d
e
c
isio
n
s
o
n
d
iag
n
o
stic
a
n
d
trea
tme
n
t
o
p
ti
o
n
s
.
Early
d
ia
g
n
o
sis
a
n
d
d
isc
o
v
e
ry
,
c
o
m
b
in
e
d
wit
h
th
e
ri
g
h
t
trea
tme
n
t,
c
a
n
sa
v
e
li
v
e
s.
In
th
is
p
a
p
e
r,
we
h
a
v
e
d
o
n
e
a
d
e
tailed
su
rv
e
y
o
n
v
a
rio
u
s
tec
h
n
iq
u
e
s
a
n
d
m
o
d
e
ls
d
e
v
e
lo
p
e
d
f
o
r
sk
i
n
c
a
n
c
e
r
d
e
tec
ti
o
n
a
n
d
a
lso
d
isc
u
ss
e
d
d
iffere
n
t
se
c
u
rit
y
-
re
late
d
issu
e
s.
T
h
is
wo
r
k
t
h
o
r
o
u
g
h
l
y
e
x
p
l
o
re
s
th
e
se
v
e
ra
l
ty
p
e
s o
f
m
o
d
e
ls u
t
il
ize
d
to
id
e
n
ti
fy
c
a
n
c
e
r
in
t
h
e
sk
i
n
.
K
ey
w
o
r
d
s
:
Alex
-
N
et
Dee
p
lear
n
in
g
E
f
f
icien
tNet
I
m
ag
eNe
t
Ma
ch
in
e
lear
n
in
g
Mo
b
ile
N
et
-
V2
R
esNet
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
:
Am
r
u
ta
T
h
o
r
at
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
T
elec
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
in
g
,
Dr
.
D.
Y.
Patil I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
Pimp
r
i,
Pu
n
e,
I
n
d
ia
E
m
ail:
am
r
u
tath
o
r
at
1
2
0
6
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
Natio
n
al
C
an
ce
r
I
n
s
titu
te
(
NC
I
)
p
r
ed
icts
th
at
b
y
2
0
4
0
,
th
er
e
m
ig
h
t
b
e
an
asto
u
n
d
in
g
2
9
.
5
m
illi
o
n
n
ew
ca
s
es
[
1
]
.
W
o
r
ld
h
ea
lth
o
r
g
an
izatio
n
(
W
HO)
p
r
ed
icts
o
n
e
o
f
th
e
th
r
ee
m
o
s
t
d
ea
d
ly
ca
n
ce
r
s
th
at
ca
n
r
esu
lt
f
r
o
m
d
eo
x
y
r
ib
o
n
u
cl
eic
ac
id
(
DNA)
d
am
ag
e
is
s
k
i
n
ca
n
ce
r
.
T
h
e
u
n
co
n
tr
o
llab
le
g
r
o
win
g
o
f
tis
s
u
es,
is
ca
u
s
ed
b
y
th
is
d
am
ag
ed
D
NA,
an
d
it
is
cu
r
r
en
tly
ac
ce
le
r
atin
g
q
u
ick
l
y
.
T
h
is
ty
p
e
o
f
u
n
ev
en
g
r
o
wth
o
f
ce
ll
p
atter
n
s
ca
n
b
e
class
ed
as
eith
er
b
en
ig
n
o
r
m
alig
n
an
t
[
2
]
.
Sk
in
ca
n
ce
r
m
ay
also
b
e
ca
u
s
ed
b
y
u
ltra
v
io
let
(
UV
)
lig
h
t e
x
p
o
s
u
r
e,
a
wea
k
e
n
ed
im
m
u
n
e
s
y
s
tem
,
a
f
am
ily
h
is
to
r
y
,
an
d
o
t
h
er
f
ac
t
o
r
s
.
T
h
e
n
atu
r
e
o
f
h
u
m
a
n
s
k
in
is
in
cr
ed
ib
ly
co
m
p
le
x
.
T
h
e
s
k
in
s
h
ield
s
all
th
e
o
r
g
a
n
s
with
in
f
r
o
m
th
e
h
ar
s
h
ex
ter
n
al
en
v
ir
o
n
m
en
t.
I
t
g
u
ar
d
s
ag
ai
n
s
t
in
f
ec
tio
n
s
an
d
ass
is
t
s
in
tem
p
er
atu
r
e
r
eg
u
l
atio
n
.
As
s
h
o
wn
in
Fig
u
r
e
1
,
th
e
r
e
ar
e
th
r
ee
lay
e
r
s
o
f
s
k
in
,
wh
ic
h
c
o
n
s
is
t
h
y
p
o
d
er
m
is
ep
id
e
r
m
is
an
d
d
e
r
m
is
[
3
]
.
T
h
e
ep
id
er
m
is
an
d
d
er
m
is
ar
e
th
e
two
p
r
i
n
c
ip
al
m
em
b
r
a
n
e
d
iv
is
io
n
s
.
T
h
e
ep
id
er
m
is
is
th
e
s
k
in
'
s
to
p
lay
er
an
d
ac
ts
as
a
p
r
o
tectiv
e
b
ar
r
ie
r
b
y
k
ee
p
in
g
b
o
d
ily
f
lu
id
s
in
p
lace
an
d
p
r
ev
en
tin
g
b
ac
ter
ial
g
r
o
wth
.
T
h
e
s
k
in
'
s
ten
s
ile
s
tr
en
g
th
an
d
s
u
p
p
len
ess
ar
e
d
e
r
iv
ed
f
r
o
m
th
e
co
n
n
ec
tiv
e
tis
s
u
es
th
at
co
m
p
r
is
e
th
e
d
er
m
is
[
3
]
.
Fig
u
r
e
2
s
h
o
ws
th
e
d
if
f
er
en
t
ty
p
es
o
f
s
k
in
ca
n
ce
r
,
th
e
th
r
ee
m
ain
ca
teg
o
r
ies
o
f
s
k
in
ca
n
ce
r
s
ar
e:
i)
Sq
u
am
o
u
s
-
ce
ll
ca
r
cin
o
m
a
(
SC
C
)
,
ii)
b
asal
-
ce
ll
ca
r
cin
o
m
a
(
B
C
C
)
,
an
d
iii)
m
alig
n
an
t
m
elan
o
m
a
[
3
]
,
th
e
o
th
e
r
ty
p
es
a
r
e
ac
tin
ic
k
er
ato
s
is
(
AKI
E
C
)
h
as a
b
en
ig
n
n
atu
r
e
b
u
t c
an
b
ec
o
m
e
m
alig
n
a
n
t tu
r
n
in
g
in
to
a
ca
n
ce
r
o
u
s
lesi
o
n
.
B
asal c
e
ll c
ar
cin
o
m
a
(
B
C
C
)
an
d
m
elan
o
m
a
(
ME
L
)
ar
e
ca
n
ce
r
o
u
s
.
B
en
ig
n
k
er
at
o
s
is
(
B
KL
)
,
d
er
m
ato
f
ib
r
o
m
a
(
DF)
,
v
ascu
lar
s
k
in
lesi
o
n
(
VASC
)
,
an
d
m
elan
o
c
y
tic
n
ev
i
(
NV)
a
r
e
n
o
n
-
ca
n
c
er
o
u
s
[
4
]
.
T
h
er
e
ar
e
s
tu
d
ies
f
o
r
t
h
e
a
u
to
m
ated
id
en
tific
atio
n
o
f
ca
n
ce
r
in
im
a
g
es
o
f
s
k
in
lesi
o
n
s
.
T
h
e
an
aly
s
is
o
f
th
ese
im
ag
es,
h
o
wev
er
,
is
v
er
y
d
if
f
icu
lt
d
u
e
to
s
o
m
e
s
ig
n
if
ican
t
ca
u
s
es,
s
u
ch
as
th
e
r
ef
lectio
n
o
f
lig
h
t
f
r
o
m
th
e
s
k
in
'
s
s
u
r
f
ac
e,
d
if
f
er
en
ce
s
in
th
e
co
lo
r
b
r
ig
h
tn
ess
o
f
th
e
lesi
o
n
s
,
an
d
t
h
e
lesi
o
n
s
'
v
ar
ied
s
izes a
n
d
f
o
r
m
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
co
mp
r
eh
en
s
ive
a
n
a
lysi
s
o
f
d
iffer
en
t m
o
d
els:
s
kin
ca
n
ce
r
d
etec
tio
n
(
A
mru
ta
Th
o
r
a
t
)
2405
Fig
u
r
e
1
.
L
a
y
er
s
o
f
s
k
in
an
d
b
asic th
r
ee
ty
p
es o
f
s
k
in
ca
n
ce
r
[
3
]
Fig
u
r
e
2
.
T
y
p
es o
f
s
k
i
n
ca
n
ce
r
s
[
4
]
T
h
e
n
at
u
r
e
o
f
h
u
m
a
n
s
k
in
is
i
n
cr
ed
ib
ly
co
m
p
licated
.
T
h
e
c
h
allen
g
in
g
p
ar
t
is
id
en
tif
y
in
g
s
k
in
ca
n
ce
r
f
r
o
m
clin
ical
p
h
o
to
g
r
ap
h
s
.
Ho
wev
er
,
f
o
r
p
h
y
s
ician
s
it
is
ex
tr
em
ely
d
if
f
icu
lt
to
m
a
n
u
ally
d
i
ag
n
o
s
e
s
k
in
is
s
u
es.
C
o
n
s
eq
u
en
tly
,
co
m
p
u
ter
ized
s
k
in
d
is
ea
s
e
p
r
ed
ictio
n
is
n
ee
d
ed
f
o
r
b
o
th
p
atien
ts
an
d
d
o
ct
o
r
s
.
T
h
er
ef
o
r
e,
it
is
im
p
o
r
tan
t
to
im
p
r
o
v
e
p
ath
o
lo
g
is
ts
'
ac
cu
r
ac
y
an
d
ex
p
er
tis
e
in
th
e
ea
r
ly
s
tag
es
th
r
o
u
g
h
ev
id
en
tial
au
to
m
atic
s
k
in
ca
n
ce
r
r
ec
o
g
n
itio
n
.
T
h
e
s
tu
d
y
ex
p
lo
r
es
ad
v
an
ce
d
t
ec
h
n
iq
u
es
an
d
f
ea
tu
r
es
d
esig
n
ed
f
o
r
t
h
e
d
etec
tio
n
a
n
d
class
if
icatio
n
o
f
s
k
in
ca
n
ce
r
.
I
t
em
p
h
asizes
s
ev
er
al
p
r
o
m
in
en
t
d
ee
p
lear
n
i
n
g
m
o
d
els,
in
clu
d
in
g
Alex
Net,
Mo
b
ileNet
-
V2
,
E
f
f
icien
tNet,
I
m
ag
eNe
t,
R
esNet,
VGG
-
1
6
,
Den
s
eNe
t,
I
n
ce
p
t
io
n
V3
,
an
d
im
ag
e
s
u
p
e
r
-
r
eso
l
u
tio
n
(
I
SR
)
,
wh
ich
h
av
e
b
ee
n
in
s
tr
u
m
e
n
tal
in
im
p
r
o
v
in
g
d
ia
g
n
o
s
tic
p
r
ec
is
io
n
.
T
h
ese
m
o
d
els
s
h
o
wca
s
e
s
ig
n
if
ican
t
ad
v
an
ce
m
en
ts
in
ac
cu
r
ac
y
,
ef
f
icien
c
y
,
an
d
r
e
liab
ilit
y
,
co
n
tr
ib
u
tin
g
to
m
o
r
e
ef
f
ec
tiv
e
au
to
m
ate
d
s
k
in
ca
n
c
er
d
iag
n
o
s
is
.
Sk
in
ca
n
ce
r
im
p
ac
t
o
n
h
u
m
an
life
ca
n
b
e
p
r
o
f
o
u
n
d
.
Sk
in
ca
n
ce
r
ca
n
s
ig
n
if
ican
tly
af
f
ec
t
a
n
in
d
iv
id
u
al'
s
q
u
ality
o
f
life
.
W
h
ile
ea
r
ly
-
s
tag
e
s
k
in
ca
n
ce
r
s
ca
n
o
f
ten
b
e
tr
ea
ted
ef
f
e
ctiv
ely
,
ad
v
a
n
ce
d
ca
s
es,
esp
ec
ially
m
elan
o
m
a,
ca
n
b
e
life
-
th
r
ea
ten
in
g
.
T
h
e
d
is
ea
s
e
ca
n
lead
t
o
p
h
y
s
ical
an
d
p
s
y
c
h
o
lo
g
ical
b
u
r
d
en
s
,
in
clu
d
in
g
d
is
f
ig
u
r
em
e
n
t,
an
x
i
ety
,
an
d
r
ed
u
ce
d
life
e
x
p
ec
t
an
cy
.
Su
r
v
iv
al
r
ates
ca
n
b
e
im
p
r
o
v
e
d
b
y
ea
r
ly
d
etec
tio
n
an
d
tr
ea
tm
en
t.
Sk
in
test
s
f
o
r
m
a
n
u
al
d
iag
n
o
s
tic
m
eth
o
d
s
u
s
u
ally
in
clu
d
e
v
i
s
u
al
ex
am
in
atio
n
an
d
b
io
p
s
y
.
Sk
in
test
s
u
s
u
ally
in
clu
d
e:
−
Vis
u
al
ex
am
in
atio
n
:
Den
tis
ts
u
s
e
in
s
tr
u
m
en
ts
s
u
ch
as
d
er
m
o
s
co
p
es
to
ex
am
in
e
t
h
e
s
k
in
an
d
id
e
n
tify
ab
n
o
r
m
alities
b
ased
o
n
s
y
m
p
t
o
m
s
(
e.
g
.
,
asy
m
m
etr
y
,
ir
r
eg
u
lar
ity
,
ir
r
eg
u
lar
b
o
r
d
er
,
d
is
co
lo
r
atio
n
)
.
−
B
io
p
s
y
:
Per
f
o
r
m
a
b
io
p
s
y
if
a
lesi
o
n
is
p
r
esen
t.
A
s
m
all
p
ie
ce
o
f
s
k
in
is
co
llected
an
d
ex
am
in
ed
u
n
d
er
a
m
icr
o
s
co
p
e
to
d
etec
t th
e
p
r
ese
n
ce
o
f
ca
n
ce
r
ce
lls
.
Alth
o
u
g
h
s
u
cc
ess
f
u
l,
ce
ll d
iag
n
o
s
is
r
elie
s
h
ea
v
ily
o
n
th
e
k
n
o
wled
g
e
an
d
s
k
ills
o
f
a
d
er
m
ato
lo
g
is
t.
Misd
iag
n
o
s
is
o
r
d
elay
e
d
d
ia
g
n
o
s
is
is
p
o
s
s
ib
le,
esp
ec
ially
if
s
y
m
p
to
m
s
ar
e
v
a
g
u
e
o
r
aty
p
ic
al.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
4
0
4
-
2
4
1
5
2406
Ma
ch
in
e
lear
n
in
g
(
ML
)
an
d
d
ee
p
lear
n
in
g
(
DL
)
p
lay
a
tr
an
s
f
o
r
m
ativ
e
r
o
le
in
s
k
i
n
ca
n
ce
r
d
etec
tio
n
b
y
en
a
b
lin
g
f
aster
a
n
d
m
o
r
e
ac
cu
r
ate
d
iag
n
o
s
es.
Dee
p
le
ar
n
in
g
m
o
d
els,
esp
ec
ially
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs),
ar
e
tr
ain
ed
o
n
lar
g
e
d
atasets
o
f
d
er
m
o
s
c
o
p
ic
im
ag
es
to
d
etec
t
an
d
cla
s
s
if
y
s
k
in
lesi
o
n
s
,
o
f
ten
s
u
r
p
ass
in
g
th
e
ac
cu
r
ac
y
o
f
h
u
m
a
n
ex
p
er
ts
.
T
h
ese
m
o
d
els as
s
is
t
d
er
m
ato
lo
g
is
ts
in
id
en
tify
in
g
ea
r
ly
-
s
tag
e
ca
n
ce
r
,
r
ed
u
cin
g
th
e
n
ee
d
f
o
r
i
n
v
asiv
e
b
io
p
s
ies
th
r
o
u
g
h
n
o
n
-
in
v
asiv
e
im
ag
e
an
aly
s
is
.
ML
a
lg
o
r
ith
m
s
ca
n
also
in
teg
r
ate
ad
d
itio
n
al
p
atien
t
d
ata
to
im
p
r
o
v
e
d
iag
n
o
s
is
an
d
p
r
ed
ict
ca
n
ce
r
r
is
k
.
B
y
au
g
m
en
tin
g
clin
ical
wo
r
k
f
lo
ws,
ar
tific
ial
in
tellig
e
n
ce
(
AI
)
d
r
iv
en
s
y
s
tem
s
en
h
an
ce
d
iag
n
o
s
tic
co
n
s
is
ten
cy
,
s
p
ee
d
,
an
d
o
v
er
all
p
atien
t o
u
tco
m
es.
T
h
is
tech
n
o
l
o
g
y
is
h
elp
in
g
to
r
ev
o
lu
tio
n
ize
s
k
in
ca
n
ce
r
s
cr
ee
n
i
n
g
an
d
ea
r
ly
d
etec
tio
n
.
T
h
ey
ca
n
im
p
r
o
v
e
s
k
in
d
iag
n
o
s
is
b
y
i)
im
ag
e
class
if
icatio
n
,
ii)
ea
r
ly
d
etec
tio
n
,
iii)
d
ec
is
io
n
s
u
p
p
o
r
t,
an
d
iv
)
au
t
o
m
ated
s
cr
ee
n
in
g
.
R
ec
en
t
tech
n
o
lo
g
ical
ad
v
a
n
ce
m
en
ts
h
av
e
f
o
c
u
s
ed
o
n
th
e
d
e
v
elo
p
m
en
t
o
f
d
ee
p
lear
n
in
g
an
d
m
ac
h
in
e
lear
n
in
g
,
r
esu
ltin
g
in
im
p
r
o
v
ed
ac
cu
r
a
cy
,
s
p
ee
d
,
an
d
s
ca
lab
ilit
y
f
o
r
ea
r
ly
d
etec
tio
n
an
d
en
h
an
ce
d
ef
f
icien
c
y
.
T
h
e
m
ai
n
co
n
tr
ib
u
tio
n
s
in
clu
d
e
a
d
etai
led
s
u
r
v
e
y
o
f
th
e
m
o
s
t
p
r
o
m
in
en
t
m
o
d
els
f
o
r
s
k
in
ca
n
ce
r
d
etec
tio
n
an
d
c
o
m
p
ar
at
iv
e
an
aly
s
is
o
f
th
e
ac
cu
r
ac
y
a
n
d
r
eliab
ilit
y
o
f
d
if
f
er
e
n
t m
o
d
els.
T
h
is
p
ap
er
t
h
o
r
o
u
g
h
ly
s
u
r
v
e
y
s
th
e
v
ar
i
o
u
s
m
o
d
els
an
d
tech
n
iq
u
es
d
e
v
elo
p
ed
f
o
r
s
k
in
ca
n
ce
r
d
etec
tio
n
,
in
clu
d
i
n
g
d
ee
p
lea
r
n
in
g
m
o
d
els
s
u
ch
as
Alex
Net,
Mo
b
ileNet
-
V2
,
E
f
f
icien
tNet,
I
m
ag
eNe
t,
R
esNet
,
VGG
-
1
6
,
Den
s
eNe
t,
an
d
I
n
c
ep
tio
n
V3
.
T
h
e
r
em
ain
in
g
s
ec
tio
n
s
o
f
th
is
p
ap
er
a
r
e
o
r
g
an
ized
as
f
o
llo
ws:
I
n
s
ec
tio
n
2
,
r
ele
v
an
t
m
et
h
o
d
s
u
s
ed
b
y
r
esear
ch
er
s
o
n
d
ee
p
lear
n
in
g
a
n
d
m
ac
h
in
e
lear
n
in
g
f
o
r
th
e
id
e
n
tific
atio
n
o
f
s
k
in
ca
n
ce
r
ar
e
s
u
m
m
ar
i
ze
d
.
Sectio
n
3
o
u
tlin
es
p
r
es
en
t
th
e
r
esu
lts
an
d
d
is
cu
s
s
i
o
n
o
f
th
e
m
o
d
els’
p
er
f
o
r
m
an
ce
u
s
in
g
v
ar
i
o
u
s
d
atasets
,
in
clu
d
in
g
co
m
p
ar
is
o
n
s
b
etwe
en
d
if
f
er
e
n
t
ar
ch
itect
u
r
es.
I
n
s
ec
tio
n
4
,
co
n
clu
d
es th
e
p
ap
er
with
f
u
tu
r
e
r
esear
ch
d
ir
ec
tio
n
s
.
2.
M
E
T
H
O
D
Sk
in
ca
n
ce
r
c
o
n
tin
u
es
to
b
e
o
n
e
o
f
th
e
m
o
s
t
wid
esp
r
ea
d
an
d
d
a
n
g
er
o
u
s
ty
p
es
o
f
ca
n
c
er
g
lo
b
ally
,
u
n
d
er
s
co
r
i
n
g
th
e
im
p
o
r
ta
n
ce
o
f
ea
r
ly
d
etec
tio
n
in
en
h
a
n
cin
g
s
u
r
v
iv
al
r
ates.
W
h
ile
tr
ad
itio
n
al
d
iag
n
o
s
tic
tech
n
iq
u
es,
in
clu
d
i
n
g
b
i
o
p
s
ies
an
d
d
er
m
o
s
co
p
ic
ev
alu
ati
o
n
s
,
ar
e
ef
f
ec
tiv
e,
th
e
y
ten
d
to
b
e
l
ab
o
r
-
in
te
n
s
iv
e
an
d
d
em
an
d
s
p
ec
ialized
k
n
o
wled
g
e.
R
ec
en
t
d
ev
elo
p
m
en
ts
in
AI
,
esp
ec
ially
in
th
e
r
ea
lm
s
o
f
d
ee
p
lear
n
in
g
an
d
m
ac
h
in
e
lear
n
in
g
,
p
r
esen
t
p
r
o
m
is
in
g
av
en
u
es
f
o
r
a
u
to
m
at
in
g
an
d
i
m
p
r
o
v
in
g
t
h
e
p
r
ec
is
io
n
o
f
s
k
in
ca
n
ce
r
d
etec
tio
n
.
2
.
1
.
Dee
p lea
rning
2
.
1
.
1
.
T
ra
ns
f
er
lea
rning
o
n o
rig
ina
l a
nd
a
ug
m
ent
ed
da
t
a
T
h
e
s
tu
d
y
ex
p
lo
r
ed
two
d
is
tin
ct
ca
teg
o
r
ies
o
f
tr
an
s
f
er
lea
r
n
i
n
g
(
T
L
)
m
eth
o
d
s
:
T
L
ap
p
lied
to
o
r
ig
in
al
d
ata
an
d
T
L
ap
p
lied
to
au
g
m
e
n
ted
d
ata,
u
s
in
g
th
e
PAD
-
UFES
-
2
0
d
ataset.
A
s
ig
n
if
ican
t
f
i
n
d
in
g
was
th
at
d
ata
au
g
m
en
tatio
n
en
h
an
ce
d
th
e
p
er
f
o
r
m
a
n
ce
o
f
m
o
d
els,
h
i
g
h
li
g
h
tin
g
its
im
p
o
r
tan
ce
in
s
k
in
ca
n
ce
r
class
if
icatio
n
task
s
.
Am
o
n
g
th
e
m
o
d
els ev
alu
ated
,
Alex
Net
d
em
o
n
s
tr
ated
t
h
e
b
est p
er
f
o
r
m
a
n
ce
,
s
h
o
wca
s
in
g
its
p
o
ten
tial f
o
r
in
teg
r
atio
n
in
to
m
o
b
ile
a
p
p
lic
atio
n
s
aim
ed
at
im
p
r
o
v
in
g
s
k
in
ca
n
ce
r
d
etec
tio
n
an
d
d
iag
n
o
s
is
[
5
]
.
2
.
1
.
2
.
Co
m
pa
riso
n o
f
d
er
m
o
s
co
pic
a
nd
s
m
a
rt
ph
o
ne
im
a
g
e
s
f
o
r
CNN
-
ba
s
ed
det
ec
t
io
n
T
h
is
s
tu
d
y
ex
am
in
ed
th
e
d
i
ag
n
o
s
tic
ac
cu
r
ac
y
o
f
d
e
r
m
o
s
co
p
ic
im
ag
es
(
DI
)
v
er
s
u
s
s
m
ar
tp
h
o
n
e
-
ca
p
tu
r
ed
im
a
g
es
(
SI)
u
s
in
g
a
d
u
al
C
NN
with
s
o
n
if
icatio
n
f
o
r
n
o
n
-
m
elan
o
m
a
s
k
in
ca
n
ce
r
(
NM
SC
)
.
R
esu
lt
s
s
h
o
wed
th
at
DI
o
u
tp
er
f
o
r
m
e
d
SI
in
ac
c
u
r
ac
y
,
s
en
s
itiv
ity
,
a
n
d
AUC,
s
u
g
g
esti
n
g
telem
ed
ici
n
e
ap
p
r
o
ac
h
es
m
ay
r
eq
u
ir
e
d
er
m
o
s
co
p
ic
im
ag
es
f
o
r
o
p
tim
al
r
esu
lts
.
Me
th
o
d
s
in
clu
d
ed
a
p
r
e
p
r
o
ce
s
s
in
g
p
ip
elin
e
with
h
ai
r
r
em
o
v
al,
d
ata
au
g
m
en
tatio
n
,
an
d
r
esizin
g
,
with
E
f
f
icien
tNet
B
4
y
ield
in
g
th
e
b
est
r
esu
lts
(
F1
-
s
co
r
e:
8
7
%,
T
o
p
-
1
ac
cu
r
ac
y
:
8
7
.
9
1
%)
[
6
]
,
[
7
]
.
2
.
1
.
3
.
Sk
in lesi
o
n c
la
s
s
if
ica
t
io
n us
ing
R
es
N
et
,
X
ce
ptio
n,
a
nd
D
ens
e
N
et
T
h
e
s
tu
d
y
u
tili
ze
d
R
esNet,
Xce
p
tio
n
,
an
d
Den
s
eNe
t
m
o
d
els
to
class
if
y
s
k
in
le
s
io
n
s
u
s
in
g
th
e
HAM
1
0
0
0
0
d
ataset.
B
y
em
p
lo
y
in
g
a
weig
h
te
d
en
s
em
b
le
tec
h
n
iq
u
e,
t
h
e
s
tu
d
y
ac
h
iev
ed
a
n
ac
cu
r
ac
y
o
f
8
5
.
8
%,
o
u
tp
er
f
o
r
m
in
g
th
e
in
d
iv
id
u
al
m
o
d
els.
T
h
ese
f
in
d
in
g
s
u
n
d
er
s
co
r
e
th
e
ef
f
ec
tiv
en
ess
o
f
en
s
em
b
le
m
eth
o
d
s
in
en
h
an
cin
g
th
e
p
er
f
o
r
m
a
n
ce
o
f
s
k
in
lesi
o
n
class
if
icatio
n
ta
s
k
s
,
p
r
o
v
id
in
g
a
r
o
b
u
s
t
ap
p
r
o
ac
h
f
o
r
im
p
r
o
v
in
g
d
iag
n
o
s
tic
ac
cu
r
ac
y
[
8
]
.
2
.
1
.
4
.
Cha
lleng
es in sk
in
im
a
g
e
cla
s
s
if
ica
t
io
n
T
h
e
s
tu
d
y
em
p
h
asized
th
e
c
o
m
p
lex
ity
o
f
s
k
in
im
ag
e
class
if
icatio
n
d
u
e
t
o
th
e
h
i
g
h
v
a
r
i
ab
ilit
y
in
lesi
o
n
ap
p
ea
r
an
ce
an
d
ch
ar
ac
t
er
is
tics
.
A
C
NN
im
p
lem
en
ted
in
T
en
s
o
r
Flo
w
ac
h
iev
ed
an
a
cc
u
r
ac
y
o
f
8
1
.
2
4
%,
s
h
o
wca
s
in
g
its
p
o
ten
tial
d
esp
ite
th
e
ch
allen
g
es.
Ho
wev
er
,
tr
an
s
f
er
lear
n
in
g
m
o
d
els
in
Py
T
o
r
ch
,
in
clu
d
in
g
W
id
e
R
e
s
Net1
0
1
,
R
esNet5
0
,
Den
s
eNe
t1
2
1
,
an
d
VGG1
9
,
s
ig
n
if
ican
tly
im
p
r
o
v
e
d
ac
cu
r
ac
y
,
r
an
g
i
n
g
f
r
o
m
9
6
.
4
0
% to
9
9
.
0
4
%,
h
ig
h
lig
h
ti
n
g
th
eir
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
in
ad
d
r
ess
in
g
th
ese
co
m
p
lex
iti
es
[
9
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
co
mp
r
eh
en
s
ive
a
n
a
lysi
s
o
f
d
iffer
en
t m
o
d
els:
s
kin
ca
n
ce
r
d
etec
tio
n
(
A
mru
ta
Th
o
r
a
t
)
2407
2
.
1
.
5
.
O
ptim
ized
CNN
m
o
del
wit
h
RM
S
pro
p
a
nd
ADA
M
A
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
m
o
d
el
p
r
e
-
tr
ain
e
d
o
n
d
er
m
o
s
co
p
ic
im
ag
es
was
f
u
r
th
er
r
e
f
in
ed
u
s
in
g
h
ig
h
way
C
NN
f
ea
tu
r
es.
T
h
e
o
p
tim
izatio
n
p
r
o
ce
s
s
em
p
lo
y
e
d
b
o
th
R
MSp
r
o
p
a
n
d
ADAM
,
two
wid
el
y
u
s
ed
alg
o
r
ith
m
s
f
o
r
d
ee
p
lear
n
in
g
task
s
.
No
tab
ly
,
th
e
ADAM
o
p
tim
izer
o
u
t
p
er
f
o
r
m
ed
R
MSp
r
o
p
,
ac
h
iev
in
g
a
tr
ain
in
g
ac
cu
r
ac
y
o
f
9
0
% a
n
d
a
v
alid
atio
n
ac
cu
r
ac
y
o
f
8
2
%
[
1
0
]
.
2
.
1
.
6
.
F
ea
t
ure
s
elec
t
io
n us
ing
g
enet
ic
a
lg
o
rit
h
m
a
nd
pa
rt
i
cle
s
wa
rm
o
ptim
iza
t
io
n
T
h
e
s
tu
d
y
u
tili
ze
d
E
f
f
icien
tN
etB
0
C
NN
f
ea
tu
r
es
to
en
h
an
ce
th
e
ac
cu
r
ac
y
o
f
class
if
icat
io
n
task
s
.
Featu
r
e
s
elec
tio
n
was
co
n
d
u
ct
ed
u
s
in
g
g
en
etic
al
g
o
r
ith
m
(
G
A)
an
d
p
ar
ticle
s
war
m
o
p
tim
izatio
n
(
PS
O)
,
two
p
o
wer
f
u
l
o
p
tim
izatio
n
tech
n
iq
u
es
wid
ely
ap
p
lied
in
m
ac
h
in
e
lear
n
in
g
.
Fo
llo
win
g
th
is
p
r
o
c
ess
,
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
class
if
icatio
n
was p
er
f
o
r
m
ed
,
ac
h
iev
i
n
g
an
im
p
r
ess
iv
e
ac
cu
r
ac
y
o
f
8
9
.
1
7
%
[
1
1
]
.
2
.
1
.
7
.
Dee
p lea
rning
in dia
g
no
s
ing
pig
m
ent
ed
nev
i
T
h
e
s
tu
d
y
u
tili
ze
d
C
NNs
to
c
lass
if
y
D
er
m
o
s
co
p
ic
im
ag
es
o
f
p
ig
m
e
n
ted
n
ev
i,
aim
i
n
g
to
d
is
tin
g
u
is
h
b
etwe
en
b
en
ig
n
an
d
m
alig
n
a
n
t le
s
io
n
s
.
T
h
e
u
s
e
o
f
C
NN
s
p
r
o
v
id
ed
a
r
o
b
u
s
t f
r
am
ewo
r
k
f
o
r
an
aly
zin
g
co
m
p
lex
p
atter
n
s
an
d
tex
tu
r
es
with
in
th
e
im
ag
es,
wh
ich
ar
e
cr
itical
f
o
r
ac
cu
r
ate
class
if
icatio
n
.
B
y
lev
er
ag
in
g
th
ese
d
ee
p
lear
n
in
g
m
o
d
els,
th
e
s
tu
d
y
h
ig
h
lig
h
ted
th
e
p
o
ten
tial
o
f
C
NNs
to
o
u
tp
er
f
o
r
m
tr
a
d
itio
n
al
d
ia
g
n
o
s
tic
m
eth
o
d
s
in
ter
m
s
o
f
ac
cu
r
ac
y
an
d
ef
f
icien
c
y
[
1
2
]
.
2
.
1
.
8
.
E
ns
em
ble m
o
del w
it
h Xc
ept
io
n,
ResNet
5
0
,
a
nd
VG
G
1
6
T
h
e
s
tu
d
y
f
o
c
u
s
ed
o
n
f
in
e
-
t
u
n
in
g
an
en
s
em
b
le
m
o
d
el
c
o
m
p
o
s
ed
o
f
Xce
p
tio
n
,
R
esNet5
0
,
a
n
d
VGG1
6
to
en
h
an
ce
m
ela
n
o
m
a
d
iag
n
o
s
is
.
B
y
em
p
lo
y
in
g
a
weig
h
ted
f
u
s
io
n
ap
p
r
o
ac
h
,
th
e
en
s
em
b
le
ac
h
iev
e
d
an
ac
cu
r
ac
y
o
f
8
6
.
9
1
%,
s
u
r
p
ass
in
g
th
e
p
er
f
o
r
m
an
ce
o
f
m
an
y
tr
ad
itio
n
al
d
iag
n
o
s
tic
m
eth
o
d
s
.
T
h
ese
r
esu
lts
u
n
d
er
s
co
r
e
th
e
e
f
f
ec
tiv
en
es
s
o
f
in
teg
r
atin
g
m
u
ltip
le
d
ee
p
lear
n
in
g
a
r
ch
itectu
r
es
to
h
ar
n
ess
th
eir
co
m
p
lem
en
tar
y
s
tr
en
g
th
s
[
1
3
]
.
2
.
1
.
9
.
I
m
a
g
e
s
up
er
-
re
s
o
lutio
n a
nd
CNN
f
o
r
enha
nced
det
ec
t
io
n
T
h
e
s
tu
d
y
em
p
lo
y
ed
a
co
m
b
in
ed
ap
p
r
o
ac
h
u
s
in
g
I
SR
an
d
C
NNs
to
en
h
an
ce
th
e
class
if
icatio
n
ac
cu
r
ac
y
f
o
r
v
ar
io
u
s
s
k
in
ca
n
ce
r
ty
p
es.
T
h
e
in
teg
r
atio
n
o
f
I
SR
im
p
r
o
v
ed
im
a
g
e
q
u
ality
,
en
ab
lin
g
C
NN
m
o
d
els
to
ex
tr
ac
t
m
o
r
e
d
e
tailed
an
d
in
f
o
r
m
ativ
e
f
ea
t
u
r
es.
Am
o
n
g
th
e
test
ed
m
o
d
els,
I
n
ce
p
tio
n
V3
d
em
o
n
s
tr
ated
s
ig
n
if
ica
n
t d
iag
n
o
s
tic
im
p
r
o
v
e
m
en
ts
,
h
ig
h
lig
h
tin
g
th
e
p
o
ten
tial o
f
th
is
ap
p
r
o
ac
h
[
1
4
]
.
2
.
1
.
1
0
.
E
v
a
lua
t
io
n o
f
pre
-
t
ra
ined net
wo
rk
s
o
n M
E
D
-
NO
DE
a
nd
Der
m
I
S
d
a
t
a
s
et
s
Fiv
e
p
r
e
-
tr
ain
ed
n
etwo
r
k
s
(
Alex
Net,
R
es
Net
-
1
8
,
Sq
u
ee
ze
Net,
Sh
u
f
f
leNe
t,
Dar
k
Net
-
1
9
)
wer
e
ev
alu
ated
.
Alex
Net
an
d
R
esNet
-
1
8
ac
h
iev
ed
to
p
p
r
ec
is
io
n
an
d
ac
cu
r
ac
y
s
co
r
es
(
u
p
to
1
0
0
%),
with
o
th
er
m
o
d
els also
p
er
f
o
r
m
in
g
well,
alb
eit
with
s
lig
h
tly
lo
wer
ac
cu
r
ac
y
[
1
5
]
.
Me
th
o
d
s
in
clu
d
es
:
−
Data
p
r
ep
ar
atio
n
a
n
d
au
g
m
en
tatio
n
:
C
lear
ly
o
u
tlin
e
d
ataset
p
r
ep
ar
atio
n
s
tep
s
(
e.
g
.
,
h
ai
r
r
em
o
v
al,
im
ag
e
r
esizin
g
,
d
ata
au
g
m
en
tatio
n
)
,
s
p
ec
if
y
in
g
p
ar
am
eter
s
u
s
ed
f
o
r
ea
ch
m
o
d
el.
−
Mo
d
el
ar
ch
itectu
r
es
an
d
o
p
ti
m
izatio
n
tech
n
iq
u
es:
Deta
il
th
e
co
n
f
ig
u
r
atio
n
s
an
d
o
p
tim
izer
s
(
e.
g
.
,
ADAM
,
R
MSp
r
o
p
)
u
s
ed
f
o
r
ea
ch
C
NN
o
r
tr
an
s
f
er
lear
n
in
g
m
o
d
el.
−
E
x
p
er
im
en
tal
s
etu
p
:
Pro
v
id
e
s
p
ec
if
icatio
n
s
f
o
r
m
o
d
el
tr
ain
in
g
(
e.
g
.
,
b
atch
s
izes,
lear
n
in
g
r
ates,
ep
o
ch
s
)
an
d
d
escr
ib
e
e
n
s
em
b
le
tech
n
iq
u
es,
if
ap
p
licab
le.
−
Per
f
o
r
m
an
ce
m
etr
ics:
R
ep
o
r
t
m
etr
ics
lik
e
ac
cu
r
ac
y
,
s
en
s
i
tiv
ity
,
AUC,
an
d
F1
-
s
co
r
es
f
o
r
ea
ch
m
o
d
el,
in
clu
d
in
g
t
h
r
esh
o
ld
s
o
r
d
ec
is
io
n
cr
iter
ia
f
o
r
class
if
icatio
n
.
2
.
2
.
Dee
p c
o
nv
o
lutio
na
l neura
l net
wo
rk
(
DCNN
)
2
.
2
.
1
.
Co
m
prehens
iv
e
da
t
a
s
et
des
cr
iptio
n
Descr
ib
e
th
e
HAM
1
0
0
0
0
d
at
aset,
u
s
ed
ex
ten
s
iv
ely
in
s
tu
d
ies
[
1
6
]
–
[
1
8
]
.
Deta
il
s
k
in
lesi
o
n
ty
p
es,
s
am
p
le
s
izes,
an
d
an
y
r
elev
an
t
m
etad
ata,
in
cl
u
d
in
g
p
ati
en
t
d
em
o
g
r
ap
h
ics
u
s
ed
in
s
tu
d
y
[
1
8
]
f
o
r
th
e
m
u
ltimo
d
al
m
o
d
el.
Me
n
tio
n
an
y
d
ata
im
b
alan
ce
is
s
u
es
an
d
th
e
tech
n
iq
u
es
em
p
lo
y
e
d
to
ad
d
r
ess
th
ese,
as
s
ee
n
in
s
tu
d
y
[
1
7
]
.
2
.
2
.
2
.
P
re
pro
ce
s
s
ing
s
t
eps
Pre
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
p
l
ay
a
cr
u
cial
r
o
le
in
im
p
r
o
v
in
g
th
e
p
er
f
o
r
m
an
ce
o
f
s
k
in
lesi
o
n
class
if
icatio
n
m
o
d
els b
y
p
r
e
p
a
r
in
g
th
e
d
ata
f
o
r
m
o
r
e
ef
f
ec
tiv
e
an
aly
s
is
an
d
lear
n
in
g
.
−
Data
au
g
m
en
tatio
n
:
Deta
il
au
g
m
en
tatio
n
m
et
h
o
d
s
s
u
ch
as
r
o
tatio
n
,
f
lip
p
in
g
,
an
d
s
ca
lin
g
,
as
ap
p
lied
i
n
s
tu
d
y
[
1
6
]
to
im
p
r
o
v
e
class
if
icatio
n
ac
cu
r
ac
y
.
−
Featu
r
e
ex
tr
ac
tio
n
an
d
n
o
is
e
r
ed
u
ctio
n
:
Stu
d
y
[
1
6
]
em
p
h
asi
ze
s
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
n
o
i
s
e
r
ed
u
ctio
n
as
p
ar
t o
f
p
r
ep
r
o
ce
s
s
in
g
to
en
h
an
ce
m
o
d
el
p
e
r
f
o
r
m
an
ce
.
I
n
clu
d
e
s
tep
s
to
clar
if
y
ea
ch
tech
n
iq
u
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
4
0
4
-
2
4
1
5
2408
−
I
m
ag
e
r
esizin
g
:
Me
n
tio
n
im
a
g
e
r
esizin
g
s
p
ec
if
icatio
n
s
f
o
r
co
n
s
is
ten
cy
ac
r
o
s
s
m
o
d
els,
as
h
ig
h
lig
h
ted
in
s
tu
d
ies
[
1
6
]
an
d
[
1
8
]
.
2
.
2
.
3
.
M
o
del a
rc
hite
ct
ures
R
ec
en
t
ad
v
an
ce
m
e
n
ts
in
d
ee
p
lear
n
i
n
g
h
av
e
led
to
t
h
e
d
ev
elo
p
m
e
n
t
o
f
i
n
n
o
v
ativ
e
m
o
d
els
th
at
s
ig
n
if
ican
tly
en
h
a
n
ce
th
e
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
o
f
s
k
in
lesi
o
n
class
if
icatio
n
.
−
DC
NN
m
o
d
els:
T
h
e
p
r
o
p
o
s
ed
DC
NN
m
o
d
el
in
[
1
6
]
in
teg
r
ates
v
ar
io
u
s
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
to
im
p
r
o
v
e
p
er
f
o
r
m
an
ce
,
ac
h
iev
in
g
h
ig
h
t
r
ain
in
g
a
n
d
test
in
g
ac
c
u
r
ac
y
o
n
HAM
1
0
0
0
0
.
An
o
th
er
DC
NN
d
ev
elo
p
e
d
in
[
1
7
]
d
em
o
n
s
tr
ates su
p
er
io
r
ity
o
v
er
VGG1
6
a
n
d
VGG1
9
.
−
M
u
l
t
i
m
o
d
a
l
m
o
d
e
l
(
A
L
B
E
F
)
:
S
t
u
d
y
[
1
8
]
p
r
e
s
e
n
t
s
a
u
n
i
q
u
e
m
u
l
t
i
m
o
d
a
l
a
p
p
r
o
a
c
h
b
y
c
o
m
b
i
n
i
n
g
d
e
r
m
o
s
c
o
p
i
c
i
m
a
g
e
s
a
n
d
p
a
ti
e
n
t
m
e
t
a
d
at
a
,
ac
h
i
e
v
i
n
g
h
i
g
h
a
c
c
u
r
a
c
y
a
n
d
AU
C
-
R
OC
.
2
.
2
.
4
.
O
ptim
iza
t
io
n a
lg
o
ri
t
hm
s
T
h
e
o
p
tim
izatio
n
tech
n
iq
u
es
in
[
1
6
]
aim
ed
to
en
h
a
n
ce
DC
NN
p
er
f
o
r
m
an
ce
t
h
r
o
u
g
h
ad
a
p
tiv
e
o
p
tim
izer
s
,
p
a
r
am
eter
tu
n
in
g
,
an
d
r
eg
u
lar
izatio
n
s
tr
ateg
ies.
ADAM
,
with
a
lear
n
in
g
r
ate
o
f
0
.
0
0
1
an
d
a
b
atch
s
ize
o
f
3
2
,
en
a
b
led
f
aster
co
n
v
er
g
en
ce
an
d
b
etter
o
u
tco
m
es.
R
eg
u
lar
izatio
n
m
eth
o
d
s
lik
e
d
r
o
p
o
u
t (
0
.
5
)
an
d
L
2
weig
h
t
d
ec
ay
(
0
.
0
0
0
1
)
h
el
p
ed
p
r
e
v
en
t
o
v
er
f
itti
n
g
a
n
d
im
p
r
o
v
e
d
g
e
n
er
aliza
tio
n
.
Data
au
g
m
en
tatio
n
tech
n
iq
u
es,
i
n
clu
d
in
g
r
o
tatio
n
,
f
lip
p
in
g
,
an
d
s
ca
lin
g
,
in
cr
ea
s
ed
tr
ain
in
g
d
ata
d
iv
er
s
ity
,
f
u
r
t
h
er
e
n
h
an
ci
n
g
r
o
b
u
s
tn
ess
.
T
h
ese
co
m
b
i
n
ed
m
eth
o
d
s
p
lay
e
d
a
cr
u
cial
r
o
le
in
ac
h
iev
in
g
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
in
th
e
s
tu
d
y
.
2
.
2
.
5
.
E
v
a
lua
t
io
n m
e
t
rics
E
v
alu
atio
n
m
et
r
ics
s
u
ch
as
a
cc
u
r
ac
y
,
F1
-
s
co
r
e,
a
n
d
AUC
-
R
OC
ar
e
ess
en
tial
f
o
r
ass
es
s
i
n
g
m
o
d
el
p
er
f
o
r
m
an
ce
,
as
em
p
h
asized
i
n
[
1
6
]
–
[
1
8
]
.
Acc
u
r
ac
y
m
ea
s
u
r
es
th
e
p
r
o
p
o
r
tio
n
o
f
c
o
r
r
ec
t
p
r
ed
ictio
n
s
b
u
t
m
a
y
b
e
in
s
u
f
f
icien
t
f
o
r
im
b
alan
ce
d
d
atasets
,
as
n
o
ted
i
n
[
1
6
]
.
T
h
e
F1
-
s
co
r
e,
h
ig
h
lig
h
ted
in
[
1
7
]
,
b
alan
ce
s
p
r
ec
is
io
n
an
d
r
ec
all,
p
r
o
v
id
in
g
a
b
etter
m
ea
s
u
r
e
o
f
p
er
f
o
r
m
an
ce
in
s
ce
n
ar
io
s
with
s
k
ewe
d
class
d
is
tr
ib
u
tio
n
s
.
AUC
-
R
OC
,
a
s
d
is
cu
s
s
ed
in
[
1
8
]
,
ev
alu
ates
a
m
o
d
el’
s
ab
ilit
y
t
o
d
is
tin
g
u
is
h
b
etwe
en
class
es
ac
r
o
s
s
th
r
esh
o
ld
s
,
m
ak
in
g
it c
r
u
cial
f
o
r
clin
ical
ap
p
licatio
n
s
wh
er
e
r
eliab
ilit
y
is
p
ar
am
o
u
n
t.
2
.
2
.
6
.
E
x
perim
ent
a
l set
up
A
well
-
d
ef
in
ed
h
ar
d
war
e
an
d
s
o
f
twar
e
s
etu
p
is
cr
u
cial
f
o
r
r
ep
r
o
d
u
ci
b
ilit
y
in
m
ac
h
in
e
lear
n
in
g
s
tu
d
ies.
T
h
is
in
clu
d
es
s
p
ec
if
y
i
n
g
th
e
c
o
m
p
u
tin
g
en
v
ir
o
n
m
en
t,
s
u
ch
as
g
r
a
p
h
ics
p
r
o
ce
s
s
in
g
u
n
it
(
GPU)
m
o
d
els
(
e.
g
.
,
NVI
DI
A
A1
0
0
)
o
r
ce
n
tr
al
p
r
o
ce
s
s
in
g
u
n
it
s
(
C
PU
s
)
,
an
d
th
e
f
r
am
ewo
r
k
s
u
s
ed
,
lik
e
T
en
s
o
r
Flo
w
o
r
Py
T
o
r
ch
,
wh
ic
h
ar
e
wid
ely
r
e
co
g
n
ized
f
o
r
th
ei
r
r
o
b
u
s
tn
ess
in
d
ee
p
lear
n
in
g
.
Stu
d
ies
[
1
6
]
–
[
1
8
]
em
p
h
asize
th
e
im
p
o
r
tan
ce
o
f
co
n
s
is
ten
t
tr
ain
in
g
-
v
alid
atio
n
s
p
lits
,
ty
p
ically
u
s
in
g
a
n
8
0
-
2
0
o
r
7
0
-
3
0
r
atio
,
alo
n
g
with
s
ettin
g
r
an
d
o
m
s
ee
d
v
alu
es
to
en
s
u
r
e
r
ep
r
o
d
u
cib
ilit
y
in
d
ataset
p
ar
titi
o
n
in
g
an
d
m
o
d
el
in
itializatio
n
.
Ad
d
itio
n
ally
,
h
y
p
er
p
ar
am
eter
s
s
u
ch
as
b
atch
s
ize,
lear
n
in
g
r
ate,
an
d
n
u
m
b
er
o
f
e
p
o
ch
s
s
h
o
u
ld
b
e
ex
p
licitly
s
tated
,
as
d
em
o
n
s
tr
ated
i
n
s
tu
d
y
[
1
7
]
,
t
o
allo
w
r
esear
ch
e
r
s
to
r
e
p
licate
th
e
r
esu
lts
.
Fu
r
th
er
m
o
r
e,
s
tu
d
y
[
1
8
]
h
ig
h
lig
h
ts
th
e
s
ig
n
if
ican
ce
o
f
co
n
s
is
ten
t
ev
alu
atio
n
m
etr
ics
ac
r
o
s
s
r
e
p
ea
ted
r
u
n
s
to
en
h
a
n
ce
r
eliab
ilit
y
in
h
ig
h
-
s
tak
es
ap
p
licatio
n
s
.
2
.
2
.
7
.
Co
m
pa
riso
n a
nd
a
na
ly
s
is
C
o
m
p
ar
is
o
n
s
with
in
s
tu
d
ies
h
ig
h
lig
h
t
th
e
s
u
p
er
i
o
r
p
er
f
o
r
m
an
ce
o
f
ce
r
tain
m
o
d
els
d
u
e
to
th
eir
ar
ch
itectu
r
e
an
d
m
u
ltimo
d
al
in
teg
r
atio
n
.
I
n
s
tu
d
y
[
1
7
]
,
th
e
d
ee
p
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
(
DC
NN)
o
u
tp
er
f
o
r
m
ed
VGG
m
o
d
els,
t
h
an
k
s
to
its
d
ee
p
er
lay
er
s
a
n
d
en
h
an
ce
d
f
ea
tu
r
e
ex
tr
a
ctio
n
ca
p
ab
ilit
ies,
allo
win
g
it
to
ca
p
tu
r
e
m
o
r
e
co
m
p
lex
p
atter
n
s
in
th
e
d
ata.
Similar
ly
,
I
n
s
tu
d
y
[
1
8
]
,
t
h
e
m
u
ltimo
d
al
AL
B
E
F
m
o
d
el,
wh
ich
co
m
b
i
n
es
im
ag
e
an
d
tex
t
d
ata,
s
u
r
p
ass
ed
im
a
g
e
-
o
n
ly
m
o
d
els
b
y
lev
er
ag
i
n
g
r
ich
er
co
n
tex
tu
al
in
f
o
r
m
atio
n
.
T
h
e
in
teg
r
atio
n
o
f
m
u
ltip
le
d
ata
s
o
u
r
ce
s
,
alo
n
g
with
p
r
ep
r
o
ce
s
s
in
g
te
ch
n
iq
u
es
lik
e
d
ata
au
g
m
en
tatio
n
an
d
n
o
r
m
aliza
t
io
n
,
co
n
t
r
ib
u
ted
t
o
th
e
h
ig
h
e
r
ac
cu
r
ac
y
ac
h
iev
ed
b
y
th
e
m
u
ltimo
d
al
m
o
d
el,
en
h
an
cin
g
its
g
en
er
aliza
tio
n
a
n
d
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
d
iv
er
s
e
d
atasets
.
2
.
3
.
M
a
chine le
a
rning
2
.
3
.
1
P
re
pro
ce
s
s
ing
t
ec
hn
iq
u
es
Pre
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
ar
e
cr
u
cial
f
o
r
im
p
r
o
v
in
g
th
e
q
u
a
lity
o
f
in
p
u
t
d
ata
an
d
en
h
a
n
cin
g
m
o
d
el
p
er
f
o
r
m
an
ce
in
s
k
i
n
lesi
o
n
cla
s
s
if
icatio
n
task
s
.
−
I
m
ag
e
en
h
a
n
ce
m
en
t:
I
n
clu
d
e
s
tep
s
f
o
r
Gau
s
s
ian
an
d
Me
d
ian
f
ilter
in
g
to
r
ed
u
ce
n
o
is
e
[
1
9
]
,
[
2
0
]
,
an
d
m
en
tio
n
d
i
g
ital h
air
r
em
o
v
al
t
ec
h
n
iq
u
es su
ch
as th
e
D
u
ll R
az
o
r
m
eth
o
d
[
1
9
]
,
[
2
0
]
.
−
Seg
m
en
tatio
n
m
eth
o
d
s
:
Descr
ib
e
co
lo
r
-
b
ased
k
-
m
ea
n
s
clu
s
ter
in
g
f
o
r
s
eg
m
en
tatio
n
as
ap
p
lied
in
[
1
9
]
an
d
th
e
Gr
ab
C
u
t
tech
n
iq
u
e
in
[
2
0
]
to
is
o
late
r
eg
io
n
s
o
f
in
ter
est ef
f
ec
tiv
ely
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
co
mp
r
eh
en
s
ive
a
n
a
lysi
s
o
f
d
iffer
en
t m
o
d
els:
s
kin
ca
n
ce
r
d
etec
tio
n
(
A
mru
ta
Th
o
r
a
t
)
2409
2
.
3
.
2
.
F
ea
t
ure
e
x
t
ra
ct
i
o
n
Featu
r
e
ex
tr
ac
tio
n
p
la
y
s
a
p
i
v
o
tal
r
o
le
in
im
p
r
o
v
in
g
th
e
p
er
f
o
r
m
a
n
ce
o
f
s
k
i
n
lesi
o
n
cla
s
s
if
icatio
n
m
o
d
els b
y
e
x
tr
ac
tin
g
r
ele
v
an
t
p
atter
n
s
an
d
c
h
ar
ac
ter
is
tics
f
r
o
m
im
ag
es.
−
AB
C
D
an
d
g
r
ay
lev
el
co
-
o
cc
u
r
r
en
ce
m
atr
ix
(
GL
C
M)
m
et
h
o
d
s
:
Descr
ib
e
asy
m
m
et
r
y
,
b
o
r
d
er
,
co
lo
r
,
a
n
d
d
iam
eter
(
AB
C
D)
m
eth
o
d
a
n
d
th
e
GL
C
M
f
o
r
tex
tu
r
e
an
aly
s
is
[
1
9
]
.
Me
n
tio
n
s
tatis
tical
f
ea
tu
r
e
ex
tr
ac
tio
n
as p
er
[
2
0
]
an
d
th
e
C
L
B
P
_
SM
C
m
eth
o
d
r
ec
o
m
m
en
d
e
d
in
[
2
1
]
f
o
r
m
elan
o
m
a
d
etec
tio
n
.
−
Featu
r
e
ex
tr
ac
tio
n
v
ia
d
ee
p
lear
n
in
g
m
o
d
els:
I
n
clu
d
e
V
GG1
6
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
co
m
b
in
ed
with
XGBo
o
s
t f
o
r
class
if
icatio
n
as
in
[
2
2
]
,
wh
ich
ac
h
iev
ed
h
ig
h
a
cc
u
r
ac
y
f
o
r
th
e
s
k
i
n
ca
n
ce
r
t
y
p
es m
en
tio
n
ed
.
2
.
3
.
3
.
Cla
s
s
if
ica
t
io
n t
ec
hn
iqu
es
C
las
s
if
icatio
n
tech
n
iq
u
es
ar
e
cr
u
cial
f
o
r
ac
cu
r
ately
ca
teg
o
r
izin
g
s
k
in
lesi
o
n
s
an
d
m
ak
i
n
g
r
eliab
le
p
r
ed
ictio
n
s
in
m
e
d
ical
im
ag
in
g
.
−
Ma
ch
in
e
lear
n
in
g
class
if
ier
s
:
Descr
ib
e
class
if
ier
s
u
s
ed
in
s
tu
d
ies,
s
u
ch
as
m
u
lti
-
class
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
MSVM
)
ac
h
iev
i
n
g
h
ig
h
ac
cu
r
ac
y
o
n
I
SIC
2
0
1
9
d
ata
[
1
9
]
,
r
a
n
d
o
m
f
o
r
est
(
R
F)
f
o
r
m
ela
n
o
m
a
class
if
icatio
n
wi
th
co
m
p
leted
lo
ca
l
b
in
ar
y
p
atter
n
s
(
C
L
B
P)
[
2
1
]
,
an
d
s
u
p
p
o
r
t
v
ec
t
o
r
m
ac
h
in
e
p
air
ed
with
GL
C
M
[
2
0
]
,
[
2
3
]
.
−
XGBo
o
s
t
an
d
ar
tific
ial
n
e
u
r
al
n
etwo
r
k
s
(
ANN)
:
R
ef
er
e
n
ce
s
tu
d
y
[
2
2
]
f
o
r
co
m
b
in
i
n
g
VGG1
6
an
d
XGBo
o
s
t f
o
r
class
if
icatio
n
,
an
d
s
tu
d
y
[
2
3
]
f
o
r
ANN
m
o
d
els u
s
in
g
h
y
b
r
id
f
ea
t
u
r
es.
2
.
3
.
4
.
Da
t
a
s
et
s
a
nd
s
k
in ca
nc
er
ca
t
eg
o
ries
T
h
e
ch
o
ice
o
f
d
atasets
an
d
th
e
in
clu
s
io
n
o
f
v
a
r
io
u
s
s
k
in
c
an
ce
r
ca
teg
o
r
ies
ar
e
cr
itical
f
o
r
tr
ain
in
g
ac
cu
r
ate
an
d
r
eliab
le
m
o
d
els f
o
r
s
k
in
lesi
o
n
class
if
icatio
n
.
−
I
SIC 2
0
1
9
d
ataset
:
Deta
il
th
e
u
s
e
o
f
th
e
I
SIC 2
0
1
9
d
ataset
with
eig
h
t sk
in
ca
n
ce
r
ty
p
es
[
1
9
]
,
[
2
0
]
.
−
HAM
1
0
0
0
0
a
n
d
PH2
d
atasets
:
Me
n
tio
n
th
e
u
s
e
o
f
HAM
1
0
0
0
0
a
n
d
PH2
in
s
tu
d
y
[
2
3
]
,
s
p
ec
if
y
in
g
th
e
ty
p
es o
f
lesi
o
n
s
in
clu
d
e
d
an
d
an
y
p
r
e
p
r
o
ce
s
s
in
g
s
tep
s
ap
p
lie
d
to
en
s
u
r
e
d
ataset
q
u
ality
.
2
.
3
.
5
.
M
o
del e
v
a
lua
t
io
n m
e
t
r
ics
Mo
d
el
ev
alu
atio
n
m
etr
ics
s
u
c
h
as
ac
cu
r
ac
y
,
co
n
f
u
s
io
n
m
at
r
ix
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e
ar
e
co
m
m
o
n
l
y
u
s
ed
ac
r
o
s
s
s
tu
d
ie
s
[
1
9
]
–
[
2
3
]
t
o
ass
ess
class
if
ie
r
p
er
f
o
r
m
a
n
ce
.
T
h
ese
m
etr
ics
h
elp
in
e
v
alu
atin
g
h
o
w
well
d
if
f
er
en
t
m
o
d
els,
co
m
b
in
ed
with
v
ar
io
u
s
f
ea
tu
r
e
ex
tr
ac
tio
n
m
eth
o
d
s
,
p
er
f
o
r
m
in
s
k
in
lesi
o
n
class
if
icatio
n
.
Sp
ec
if
ically
,
s
t
u
d
y
[
2
1
]
d
em
o
n
s
tr
ated
th
e
ef
f
ec
tiv
en
ess
o
f
u
s
in
g
r
an
d
o
m
f
o
r
est
class
if
ier
s
with
C
L
B
P
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
ac
h
iev
in
g
h
ig
h
ac
c
u
r
ac
y
.
T
h
e
co
n
f
u
s
io
n
m
atr
i
x
r
esu
lts
s
h
o
wed
th
at
th
is
co
m
b
in
atio
n
n
o
t
o
n
ly
d
eliv
e
r
ed
s
tr
o
n
g
class
if
icatio
n
p
er
f
o
r
m
an
ce
b
u
t
also
h
a
n
d
led
v
ar
ia
tio
n
s
in
s
k
in
lesi
o
n
im
ag
es e
f
f
ec
tiv
ely
.
2
.
3
.
6
.
E
x
perim
ent
a
l set
up
I
n
clu
d
e
d
etails
o
f
h
ar
d
war
e
an
d
s
o
f
twar
e
en
v
ir
o
n
m
en
ts
to
s
u
p
p
o
r
t
r
ep
r
o
d
u
ci
b
ilit
y
.
Sp
ec
if
y
ML
f
r
am
ewo
r
k
s
u
s
ed
(
e.
g
.
,
T
en
s
o
r
Flo
w,
Scik
it
-
lear
n
)
,
b
atch
s
ize
s
,
an
d
tr
ain
i
n
g
-
v
alid
atio
n
s
p
lit
s
as
r
ec
o
m
m
en
d
ed
in
ea
ch
s
tu
d
y
.
Me
n
tio
n
s
ee
d
v
alu
es,
r
an
d
o
m
in
itializatio
n
p
r
o
ce
s
s
es,
an
d
cr
o
s
s
-
v
alid
atio
n
t
ec
h
n
iq
u
es to
e
n
s
u
r
e
r
ep
licab
le
an
d
s
tab
le
r
esu
lts
ac
r
o
s
s
m
eth
o
d
s
[
1
9
]
–
[
2
3
]
.
2
.
3
.
7
.
Co
m
pa
r
a
t
iv
e
a
na
ly
s
is
T
h
e
co
m
p
ar
ativ
e
r
esu
lts
f
r
o
m
s
tu
d
ies
[
1
9
]
,
[
2
2
]
h
i
g
h
lig
h
t
t
h
e
ef
f
ec
tiv
en
ess
o
f
s
p
ec
if
ic
co
m
b
in
atio
n
s
o
f
f
ea
tu
r
e
ex
tr
ac
tio
n
tech
n
i
q
u
es
an
d
class
if
ier
m
o
d
els
in
ac
h
iev
in
g
h
ig
h
ac
cu
r
ac
ies
f
o
r
s
k
in
lesi
o
n
class
if
icatio
n
.
I
n
p
ar
ticu
lar
,
s
tu
d
y
[
1
9
]
d
em
o
n
s
tr
ated
h
o
w
t
h
e
in
teg
r
atio
n
o
f
tex
tu
r
e
an
d
s
tatis
t
ical
f
ea
tu
r
es
with
class
if
ier
s
lik
e
SVM
led
to
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
i
n
m
elan
o
m
a
d
etec
tio
n
.
Simila
r
ly
,
s
tu
d
y
[
2
2
]
also
em
p
h
asized
th
e
im
p
o
r
tan
ce
o
f
s
elec
tin
g
th
e
r
ig
h
t
f
ea
tu
r
e
-
e
x
tr
ac
tio
n
an
d
class
if
icatio
n
m
o
d
els
to
o
p
tim
ize
p
er
f
o
r
m
an
ce
.
Me
an
w
h
ile,
th
e
s
tu
d
y
[
2
4
]
p
r
o
v
id
es
a
b
r
o
a
d
er
s
u
r
v
ey
,
o
f
f
er
in
g
i
n
s
ig
h
ts
in
to
p
er
f
o
r
m
an
ce
o
p
tim
izatio
n
s
tr
ateg
ies
s
u
ch
a
s
h
y
p
er
p
ar
am
eter
tu
n
i
n
g
an
d
d
ata
au
g
m
en
tatio
n
,
f
u
r
th
er
e
n
h
an
cin
g
m
ela
n
o
m
a
class
if
icatio
n
ac
cu
r
ac
y
ac
r
o
s
s
d
if
f
er
en
t a
p
p
r
o
ac
h
es.
2
.
4
.
I
ma
g
e
pro
ce
s
s
ing
2
.
4
.
1
.
I
m
a
g
e
pro
ce
s
s
ing
t
ec
hn
iqu
e
s
I
m
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
p
lay
a
cr
u
cial
r
o
le
in
en
h
an
ci
n
g
th
e
q
u
ality
o
f
in
p
u
t
d
ata
f
o
r
m
ac
h
in
e
lear
n
in
g
m
o
d
els,
esp
ec
ially
in
s
k
in
lesi
o
n
class
if
icatio
n
task
s
.
−
Statis
t
ical
f
ea
tu
r
e
ex
tr
ac
tio
n
:
Descr
ib
e
th
e
u
s
e
o
f
f
ir
s
t
an
d
s
ec
o
n
d
-
o
r
d
er
s
tatis
tical
f
ea
tu
r
es
f
o
r
tex
tu
r
e
an
aly
s
is
as
r
ef
er
en
ce
d
in
[
2
5
]
.
T
h
is
s
h
o
u
ld
in
clu
d
e
s
p
ec
if
ic
f
ea
tu
r
es
an
aly
ze
d
(
e.
g
.
,
m
ea
n
,
v
ar
ian
ce
,
s
k
ewn
ess
)
an
d
h
o
w
th
e
y
co
n
t
r
ib
u
te
to
id
en
tif
y
in
g
p
atter
n
s
r
e
lev
an
t to
s
k
in
lesi
o
n
class
if
icatio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
4
0
4
-
2
4
1
5
2410
−
Gr
ay
lev
el
co
-
o
cc
u
r
r
e
n
ce
m
a
tr
ix
(
GL
C
M)
:
Deta
il
th
e
im
p
lem
en
tatio
n
o
f
GL
C
M
f
o
r
te
x
tu
r
e
a
n
aly
s
is
,
f
o
cu
s
in
g
o
n
h
o
w
f
ea
tu
r
es
lik
e
co
n
tr
ast,
h
o
m
o
g
en
eity
,
e
n
er
g
y
,
a
n
d
co
r
r
elatio
n
wer
e
ex
tr
ac
ted
f
r
o
m
d
if
f
er
en
t c
o
lo
r
c
h
an
n
els
[
2
5
]
.
2
.
4
.
2
.
No
v
el
m
et
a
da
t
a
-
enha
nced
cla
s
s
if
ica
t
io
n t
ec
hn
iq
ue
T
h
e
n
o
v
el
m
eta
d
ata
-
en
h
a
n
ce
d
class
if
icatio
n
tech
n
iq
u
e
o
f
f
er
s
a
p
o
wer
f
u
l
a
p
p
r
o
ac
h
t
o
im
p
r
o
v
in
g
class
if
icatio
n
p
er
f
o
r
m
an
ce
b
y
in
teg
r
atin
g
ad
d
itio
n
al
co
n
tex
tu
al
in
f
o
r
m
atio
n
in
to
th
e
m
o
d
el
t
r
ain
in
g
p
r
o
ce
s
s
.
−
Me
tad
ata
in
teg
r
atio
n
:
Pr
o
v
id
e
a
clea
r
d
escr
ip
tio
n
o
f
th
e
m
etad
ata
-
en
h
a
n
ce
d
class
if
icatio
n
m
eth
o
d
p
r
esen
ted
in
[
2
6
]
,
wh
ich
am
p
lifie
s
k
ey
f
ea
tu
r
es.
E
x
p
lain
th
e
p
r
o
ce
s
s
o
f
b
o
o
s
tin
g
cr
itical
f
ea
tu
r
es
in
th
e
class
if
icatio
n
p
ip
elin
e
an
d
h
o
w
it im
p
r
o
v
es m
o
d
el
r
o
b
u
s
tn
e
s
s
.
−
Per
f
o
r
m
an
ce
ac
r
o
s
s
m
o
d
els:
Me
n
tio
n
th
at
th
e
m
eth
o
d
was
test
ed
o
n
two
s
k
in
le
s
i
o
n
d
atasets
an
d
o
u
tp
er
f
o
r
m
ed
o
th
er
a
p
p
r
o
ac
h
es
in
s
ix
o
u
t
o
f
ten
s
ce
n
ar
io
s
[
2
6
]
.
L
is
t
th
e
class
if
icatio
n
m
o
d
els
u
s
ed
an
d
s
p
ec
if
y
an
y
p
r
e
p
r
o
ce
s
s
in
g
t
ec
h
n
iq
u
es
o
r
f
ea
t
u
r
e
s
elec
tio
n
m
eth
o
d
s
th
at
wer
e
ap
p
lied
to
o
p
tim
ize
p
er
f
o
r
m
an
ce
.
2
.
4
.
3
.
Sk
in lesi
o
n c
la
s
s
if
ica
t
io
n m
o
del
T
h
e
s
k
in
lesi
o
n
class
if
icatio
n
m
o
d
el
is
d
esig
n
ed
to
ac
cu
r
at
ely
ca
teg
o
r
ize
an
d
class
if
y
s
k
in
lesi
o
n
s
in
to
d
is
tin
ct
ca
teg
o
r
ies,
lev
er
a
g
in
g
d
ee
p
lear
n
in
g
tech
n
i
q
u
es
f
o
r
h
ig
h
-
p
e
r
f
o
r
m
an
ce
r
esu
lts
.
−
C
las
s
if
icatio
n
ca
teg
o
r
ies
an
d
d
ataset:
Me
n
tio
n
th
at
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
ca
p
ab
le
o
f
r
e
co
g
n
izin
g
s
ev
en
s
k
in
lesi
o
n
ca
teg
o
r
ies an
d
was te
s
ted
o
n
th
e
HAM
1
0
0
0
0
d
at
aset
[
2
7
]
.
−
Mo
d
el
ar
ch
itectu
r
e
an
d
tr
ain
i
n
g
:
Descr
ib
e
th
e
m
o
d
el’
s
ar
c
h
itectu
r
e
an
d
h
y
p
er
p
ar
am
eter
s
(
e.
g
.
,
lear
n
in
g
r
ate,
b
atch
s
ize,
n
u
m
b
er
o
f
e
p
o
ch
s
)
an
d
n
o
te
a
n
y
d
ata
a
u
g
m
en
tatio
n
m
eth
o
d
s
u
s
ed
to
b
alan
ce
th
e
d
ataset.
−
E
v
alu
atio
n
m
etr
ics:
R
ep
o
r
t
th
e
ac
h
ie
v
ed
a
cc
u
r
ac
y
(
9
0
%)
,
p
r
ec
is
io
n
(
0
.
8
9
)
,
a
n
d
r
ec
all
(
9
0
%)
o
n
th
e
HAM
1
0
0
0
0
d
ataset
as seen
in
[
2
7
]
.
2
.
4
.
4
.
E
x
perim
ent
a
l set
up
T
h
e
ex
p
er
im
en
tal
s
etu
p
o
u
tlin
es
th
e
k
e
y
co
m
p
o
n
en
ts
o
f
th
e
h
ar
d
wa
r
e,
s
o
f
twar
e
en
v
ir
o
n
m
en
ts
,
an
d
d
ata
h
an
d
lin
g
s
tr
ateg
ies
u
s
ed
to
en
s
u
r
e
th
e
r
ep
r
o
d
u
ci
b
ilit
y
an
d
r
eliab
ilit
y
o
f
t
h
e
m
ac
h
in
e
lear
n
i
n
g
ex
p
er
im
en
ts
.
−
Har
d
war
e
an
d
s
o
f
twar
e
en
v
i
r
o
n
m
en
ts
:
Pr
o
v
id
e
d
etails
o
n
th
e
h
ar
d
war
e
a
n
d
s
o
f
twar
e
en
v
ir
o
n
m
e
n
ts
,
in
clu
d
in
g
m
ac
h
i
n
e
s
p
ec
if
icatio
n
s
,
d
ee
p
lear
n
in
g
f
r
am
ewo
r
k
s
,
an
d
lib
r
ar
ies
u
s
ed
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
(
s
u
ch
as Op
en
C
V
o
r
Py
th
o
n
'
s
s
k
im
ag
e)
.
−
T
r
ain
in
g
-
v
alid
atio
n
s
p
lit:
Sp
ec
if
y
tr
ain
in
g
-
v
alid
atio
n
s
p
lits
,
c
r
o
s
s
-
v
alid
atio
n
ap
p
r
o
ac
h
es,
a
n
d
r
an
d
o
m
s
ee
d
v
alu
es f
o
r
r
ep
r
o
d
u
cib
ilit
y
.
2
.
4
.
5
.
Co
m
pa
r
a
t
iv
e
perf
o
rma
nce
a
na
ly
s
is
T
h
e
m
o
d
el
co
m
p
ar
is
o
n
s
r
ev
ea
l
n
o
tab
le
ad
v
an
ce
m
en
ts
in
class
if
icatio
n
p
er
f
o
r
m
an
c
e
th
r
o
u
g
h
in
n
o
v
ativ
e
a
p
p
r
o
ac
h
es.
Stu
d
y
[
2
6
]
d
em
o
n
s
tr
ated
th
at
i
n
teg
r
atin
g
m
eta
d
ata
in
to
cl
ass
if
icatio
n
task
s
s
ig
n
if
ican
tly
en
h
a
n
ce
s
p
er
f
o
r
m
an
ce
ac
r
o
s
s
v
ar
io
u
s
s
ce
n
ar
io
s
,
s
h
o
wca
s
in
g
th
e
v
alu
e
o
f
e
n
r
ich
ed
d
ata
r
ep
r
esen
tatio
n
s
.
Similar
ly
,
s
tu
d
y
[
2
7
]
ac
h
iev
ed
h
i
g
h
cla
s
s
if
icatio
n
m
etr
ics
o
n
th
e
HAM
1
0
0
0
0
d
ataset,
in
d
icatin
g
t
h
e
ef
f
ec
tiv
en
ess
o
f
th
eir
ap
p
r
o
ac
h
in
ac
cu
r
a
tely
d
iag
n
o
s
in
g
s
k
in
lesi
o
n
s
.
T
ab
le
1
f
u
r
t
h
er
s
u
m
m
ar
izes
th
e
d
ataset
an
d
m
o
d
el
p
e
r
f
o
r
m
an
ce
,
p
r
o
v
id
in
g
a
c
o
m
p
r
e
h
en
s
iv
e
o
v
er
v
iew
o
f
t
h
e
r
esu
lts
an
d
em
p
h
asizin
g
th
e
im
p
ac
t o
f
t
h
e
s
e
m
eth
o
d
s
in
im
p
r
o
v
in
g
class
if
icatio
n
ac
cu
r
ac
y
.
C
lin
ical
s
tu
d
ies
h
av
e
s
h
o
wn
s
ig
n
if
ican
t
p
r
o
g
r
ess
in
th
e
u
s
e
o
f
d
ee
p
lear
n
in
g
f
o
r
cla
s
s
if
icatio
n
,
esp
ec
ially
in
C
NN
ar
ch
itectu
r
es,
tr
an
s
f
er
lear
n
in
g
,
an
d
co
m
p
lem
en
tar
y
to
o
ls
.
Ad
v
an
ce
d
m
o
d
els
s
u
ch
as
E
f
f
icien
tNet
s
h
o
w
g
o
o
d
r
esu
lts
,
wh
ile
s
im
p
le
p
r
e
-
d
esig
n
e
d
m
o
d
els
with
o
p
tim
izatio
n
also
p
er
f
o
r
m
well.
Me
tad
ata
co
llectio
n
an
d
ag
g
r
e
g
atio
n
tech
n
o
lo
g
y
f
u
r
th
er
im
p
r
o
v
es a
cc
u
r
ac
y
.
Ho
wev
e
r
,
th
er
e
ar
e
s
till
p
r
o
b
lem
s
in
en
s
u
r
in
g
th
e
r
eliab
ilit
y
o
f
t
h
e
s
tr
u
ctu
r
e
o
f
d
if
f
er
e
n
t
d
ata.
Fu
tu
r
e
r
esear
ch
s
h
o
u
l
d
f
o
c
u
s
o
n
co
m
b
in
in
g
AI
with
tr
ad
itio
n
al
m
eth
o
d
s
to
p
r
o
v
id
e
m
o
r
e
ac
cu
r
ate
an
d
ea
s
ier
s
k
in
ca
n
ce
r
d
iag
n
o
s
is
,
f
ac
ilit
ate
ea
r
ly
in
ter
v
en
tio
n
,
an
d
im
p
r
o
v
e
p
ati
en
t o
u
tco
m
es.
I
t
h
as
b
ee
n
o
b
s
er
v
e
d
th
at
s
till
wo
r
k
ca
n
b
e
ca
r
r
ied
o
u
t,
Fir
s
tly
,
f
ea
tu
r
e
ex
tr
ac
tio
n
m
eth
o
d
s
n
ee
d
to
in
teg
r
ate
m
u
lti
-
m
o
d
al
d
ata
s
o
u
r
ce
s
,
u
s
in
g
ad
v
a
n
ce
d
d
ee
p
lear
n
in
g
tech
n
iq
u
es
to
en
h
a
n
ce
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
.
Seco
n
d
ly
,
o
p
tim
iz
atio
n
is
n
ee
d
ed
f
o
r
r
ea
l
-
tim
e
p
r
o
ce
s
s
in
g
,
p
ar
ticu
lar
ly
f
o
r
m
o
b
ile
ap
p
licatio
n
s
,
b
y
d
ev
elo
p
in
g
lig
h
tweig
h
t
alg
o
r
it
h
m
s
.
T
h
ir
d
l
y
,
th
er
e
is
a
n
ee
d
to
im
p
r
o
v
e
alg
o
r
ith
m
p
e
r
f
o
r
m
an
ce
b
y
co
llectin
g
co
m
p
r
eh
e
n
s
iv
e
d
atasets
an
d
em
p
lo
y
in
g
tech
n
iq
u
es
lik
e
tr
an
s
f
er
lear
n
in
g
an
d
d
o
m
ain
a
d
ap
tatio
n
.
Fo
u
r
th
ly
,
ex
p
lo
r
in
g
d
iv
er
s
e
e
n
s
em
b
le
l
ea
r
n
in
g
ar
ch
itectu
r
es
an
d
in
t
eg
r
atin
g
b
o
th
im
ag
e
-
b
ased
a
n
d
n
o
n
-
im
ag
e
-
b
ased
f
ea
tu
r
es c
an
en
h
an
ce
d
ia
g
n
o
s
ti
c
ac
cu
r
ac
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
co
mp
r
eh
en
s
ive
a
n
a
lysi
s
o
f
d
iffer
en
t m
o
d
els:
s
kin
ca
n
ce
r
d
etec
tio
n
(
A
mru
ta
Th
o
r
a
t
)
2411
T
ab
le
1
.
Data
s
et
s
u
m
m
ar
y
a
n
d
m
o
d
el
p
e
r
f
o
r
m
an
ce
M
e
t
h
o
d
K
e
y
f
i
n
d
i
n
g
s
D
a
t
a
s
e
t
P
e
r
f
o
r
ma
n
c
e
m
e
t
r
i
c
s
C
N
N
[
5
]
Tr
a
n
sf
e
r
l
e
a
r
n
i
n
g
w
i
t
h
d
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
s
h
o
w
e
d
i
m
p
r
o
v
e
d
p
e
r
f
o
r
m
a
n
c
e
,
p
a
r
t
i
c
u
l
a
r
l
y
w
i
t
h
A
l
e
x
N
e
t
.
P
A
D
-
U
F
ES
-
20
Ex
c
e
l
l
e
n
t
r
e
s
u
l
t
s wi
t
h
A
l
e
x
N
e
t
C
N
N
c
o
mp
a
r
i
s
o
n
[
6
]
D
e
r
mo
sc
o
p
i
c
i
m
a
g
e
s
o
u
t
p
e
r
f
o
r
m
e
d
s
martp
h
o
n
e
i
ma
g
e
s
i
n
d
i
a
g
n
o
st
i
c
a
c
c
u
r
a
c
y
.
Ef
f
i
c
i
e
n
t
N
e
t
B
4
w
a
s t
h
e
b
e
st
p
e
r
f
o
r
mer.
H
A
M
1
0
0
0
0
F1
-
s
c
o
r
e
:
8
7
%
,
A
c
c
u
r
a
c
y
:
8
7
.
9
1
%
En
se
mb
l
e
mo
d
e
l
s
[
8
]
W
e
i
g
h
t
e
d
e
n
s
e
m
b
l
e
o
f
R
e
sN
e
t
,
X
c
e
p
t
i
o
n
,
a
n
d
D
e
n
seNe
t
a
c
h
i
e
v
e
d
b
e
t
t
e
r
b
a
l
a
n
c
e
d
a
c
c
u
r
a
c
y
t
h
a
n
i
n
d
i
v
i
d
u
a
l
m
o
d
e
l
s
.
H
A
M
1
0
0
0
0
B
a
l
a
n
c
e
d
A
c
c
u
r
a
c
y
:
8
5
.
8
%
Tr
a
n
sf
e
r
l
e
a
r
n
i
n
g
[
9
]
Tr
a
n
sf
e
r
l
e
a
r
n
i
n
g
w
i
t
h
P
y
T
o
r
c
h
a
c
h
i
e
v
e
d
a
c
c
u
r
a
c
i
e
s
b
e
t
w
e
e
n
9
6
.
4
0
%
a
n
d
9
9
.
0
4
%
.
M
u
l
t
i
p
l
e
A
c
c
u
r
a
c
y
:
9
6
.
4
0
%
t
o
9
9
.
0
4
%
D
C
N
N
[
1
6
]
D
C
N
N
w
i
t
h
p
r
e
p
r
o
c
e
ss
i
n
g
m
e
t
h
o
d
s
a
c
h
i
e
v
e
d
h
i
g
h
t
r
a
i
n
i
n
g
a
n
d
t
e
st
i
n
g
a
c
c
u
r
a
c
i
e
s.
H
A
M
1
0
0
0
0
Tr
a
i
n
i
n
g
A
c
c
u
r
a
c
y
:
9
3
.
1
6
%,
Te
st
i
n
g
A
c
c
u
r
a
c
y
:
9
1
.
9
3
%
M
L
m
e
t
h
o
d
s
[
1
9
]
H
i
g
h
a
c
c
u
r
a
c
y
a
c
h
i
e
v
e
d
w
i
t
h
a
d
v
a
n
c
e
d
p
r
e
p
r
o
c
e
ssi
n
g
a
n
d
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
.
I
S
I
C
2
0
1
9
A
c
c
u
r
a
c
y
:
9
6
.
2
5
%
M
L
a
n
d
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
[
2
2
]
C
o
m
b
i
n
i
n
g
V
G
G
1
6
f
o
r
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
w
i
t
h
X
G
B
o
o
st
c
l
a
ss
i
f
i
c
a
t
i
o
n
a
c
h
i
e
v
e
d
t
h
e
h
i
g
h
e
st
a
c
c
u
r
a
c
y
.
I
S
I
C
d
a
t
a
s
e
t
A
c
c
u
r
a
c
y
:
9
9
.
1
%
M
L
a
n
d
i
ma
g
e
q
u
a
l
i
t
y
e
n
h
a
n
c
e
me
n
t
[
2
0
]
I
mp
r
o
v
e
d
i
ma
g
e
q
u
a
l
i
t
y
a
n
d
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
l
e
d
t
o
b
e
t
t
e
r
S
V
M
p
e
r
f
o
r
ma
n
c
e
.
H
A
M
1
0
0
0
0
,
I
S
I
C
2
0
1
9
O
u
t
p
e
r
f
o
r
m
e
d
I
S
I
C
2
0
1
9
I
mag
e
p
r
o
c
e
ssi
n
g
[
2
5
]
S
t
a
t
i
st
i
c
a
l
f
e
a
t
u
r
e
s
a
n
d
G
LC
M
d
a
t
a
a
c
h
i
e
v
e
d
h
i
g
h
a
c
c
u
r
a
c
y
a
n
d
p
r
e
c
i
si
o
n
.
Tw
o
d
a
t
a
s
e
t
s
A
c
c
u
r
a
c
y
:
9
7
.
0
0
%
En
h
a
n
c
e
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
[
2
6
]
N
o
v
e
l
met
a
d
a
t
a
t
e
c
h
n
i
q
u
e
i
m
p
r
o
v
e
d
c
l
a
ssi
f
i
c
a
t
i
o
n
a
c
r
o
ss
e
v
a
l
u
a
t
e
d
m
o
d
e
l
s
.
Tw
o
s
k
i
n
l
e
si
o
n
d
a
t
a
se
t
s
O
u
t
p
e
r
f
o
r
m
e
d
o
t
h
e
r
met
h
o
d
s
i
n
si
x
o
u
t
o
f
t
e
n
sce
n
a
r
i
o
s
S
k
i
n
l
e
si
o
n
r
e
c
o
g
n
i
t
i
o
n
[
2
7
]
M
o
d
e
l
r
e
c
o
g
n
i
z
e
d
s
e
v
e
n
sk
i
n
l
e
si
o
n
c
a
t
e
g
o
r
i
e
s
w
i
t
h
g
o
o
d
a
c
c
u
r
a
c
y
,
p
r
e
c
i
si
o
n
,
a
n
d
r
e
c
a
l
l
.
H
A
M
1
0
0
0
0
A
c
c
u
r
a
c
y
:
9
0
%
,
P
r
e
c
i
s
i
o
n
:
0
.
8
9
,
R
e
c
a
l
l
:
9
0
%
2
.
5
.
Av
a
ila
ble da
t
a
s
et
s
Dif
f
er
en
t
lear
n
i
n
g
m
o
d
els
ar
e
tr
ain
ed
a
n
d
e
v
alu
ated
o
n
a
n
u
m
b
e
r
o
f
p
u
b
licly
ac
ce
s
s
ib
le
d
atasets
co
n
tain
in
g
p
h
o
to
s
o
f
s
k
in
c
an
ce
r
.
So
m
e
o
f
th
e
m
o
s
t
w
ell
-
k
n
o
wn
d
atasets
ar
e
lis
ted
in
T
a
b
le
2
.
T
h
e
d
ev
elo
p
m
e
n
t
an
d
ass
ess
m
en
t
o
f
d
ee
p
lear
n
in
g
(
DL
)
an
d
m
a
ch
in
e
lear
n
in
g
(
ML
)
m
o
d
els
f
o
r
th
e
id
e
n
tific
atio
n
an
d
class
if
icatio
n
o
f
s
k
in
ca
n
c
er
ca
s
es c
o
u
ld
b
e
g
r
ea
tly
aid
e
d
b
y
th
ese
d
atasets
.
T
ab
le
2
.
Ov
e
r
v
iew
o
f
s
k
in
ca
n
ce
r
d
atasets
u
s
ed
f
o
r
m
o
d
el
ev
alu
atio
n
D
a
t
a
s
e
t
s
D
e
scri
p
t
i
o
n
To
t
a
l
i
m
a
g
e
s
A
c
c
e
ss
I
n
t
e
r
n
a
t
i
o
n
a
l
sk
i
n
i
ma
g
i
n
g
c
o
l
l
a
b
o
r
a
t
i
o
n
(
I
S
I
C
)
a
r
c
h
i
v
e
[
1
9
]
,
[
2
0
]
,
[
2
3
]
,
[
2
7
]
Th
e
I
S
I
C
A
r
c
h
i
v
e
i
n
c
l
u
d
e
s
i
m
a
g
e
s
l
a
b
e
l
e
d
a
s
b
e
n
i
g
n
,
mal
i
g
n
a
n
t
,
a
n
d
v
a
r
i
o
u
s s
k
i
n
c
o
n
d
i
t
i
o
n
s.
T
h
e
a
r
c
h
i
v
e
i
s r
e
g
u
l
a
r
l
y
u
p
d
a
t
e
d
w
i
t
h
n
e
w
d
a
t
a
a
n
d
i
s
u
s
e
d
i
n
a
n
n
u
a
l
c
h
a
l
l
e
n
g
e
s
t
o
p
r
o
m
o
t
e
t
h
e
d
e
v
e
l
o
p
m
e
n
t
o
f
a
u
t
o
m
a
t
e
d
d
i
a
g
n
o
st
i
c
t
o
o
l
s.
5
0
,
0
0
0
I
S
I
C
A
r
c
h
i
v
e
H
u
ma
n
a
g
a
i
n
st
mac
h
i
n
e
t
r
a
i
n
i
n
g
i
ma
g
e
s
(
H
A
M
1
0
0
0
0
)
[
8
]
,
[
9
]
,
[
2
0
]
Th
i
s
c
o
l
l
e
c
t
i
o
n
i
n
c
l
u
d
e
s
p
i
g
m
e
n
t
e
d
s
k
i
n
l
e
s
i
o
n
s
,
s
u
c
h
a
s
mel
a
n
o
ma
a
n
d
b
e
n
i
g
n
n
e
v
i
.
Th
e
c
o
l
l
e
c
t
i
o
n
o
f
d
a
t
a
i
s c
a
r
e
f
u
l
l
y
c
h
o
sen
a
n
d
i
n
c
l
u
d
e
s
me
t
a
d
a
t
a
su
c
h
a
s a
g
e
,
g
e
n
d
e
r
,
a
n
d
l
e
s
i
o
n
l
o
c
a
t
i
o
n
.
1
0
,
0
1
5
K
a
g
g
l
e
D
e
r
mo
f
i
t
i
ma
g
e
l
i
b
r
a
r
y
[
1
2
]
–
[
1
4
]
Th
i
s
d
a
t
a
se
t
c
r
e
a
t
e
d
b
y
t
h
e
U
n
i
v
e
r
si
t
y
o
f
E
d
i
n
b
u
r
g
h
,
i
n
c
l
u
d
e
s
p
i
c
t
u
r
e
s
o
f
s
k
i
n
l
e
s
i
o
n
s
t
h
a
t
a
r
e
u
t
i
l
i
z
e
d
i
n
t
e
a
c
h
i
n
g
a
n
d
r
e
sea
r
c
h
.
I
t
c
o
v
e
r
s
a
v
a
r
i
e
t
y
o
f
s
k
i
n
d
i
so
r
d
e
r
s,
i
n
c
l
u
d
i
n
g
seb
o
r
r
h
e
i
c
k
e
r
a
t
o
s
e
s a
n
d
m
e
l
a
n
o
mas
.
1
,
3
0
0
R
e
q
u
i
r
e
s
a
l
i
c
e
n
s
e
p
u
r
c
h
a
se
D
e
r
mo
f
i
t
I
mag
e
Li
b
r
a
r
y
P
H
2
D
a
t
a
se
t
[
1
5
]
,
[
2
1
]
Th
e
P
H
2
d
a
t
a
s
e
t
4
0
m
e
l
a
n
o
m
a
s,
8
0
c
o
mm
o
n
n
e
v
i
,
a
n
d
8
0
a
t
y
p
i
c
a
l
n
e
v
i
.
T
h
e
i
m
a
g
e
s
a
r
e
m
a
n
u
a
l
l
y
se
g
m
e
n
t
e
d
a
n
d
a
n
n
o
t
a
t
e
d
,
ma
k
i
n
g
t
h
i
s
d
a
t
a
se
t
v
a
l
u
a
b
l
e
f
o
r
t
r
a
i
n
i
n
g
m
o
d
e
l
s
i
n
seg
m
e
n
t
a
t
i
o
n
a
n
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
t
a
s
k
s.
2
0
0
P
H
2
D
a
t
a
se
t
M
ED
-
N
O
D
E
d
a
t
a
se
t
[
1
4
]
,
[
1
5
]
Th
i
s
d
a
t
a
se
t
w
a
s
c
r
e
a
t
e
d
f
o
r
t
h
e
d
e
t
e
c
t
i
o
n
o
f
me
l
a
n
o
ma
.
I
t
i
n
c
l
u
d
e
s
i
ma
g
e
s
w
i
t
h
l
a
b
e
l
s f
o
r
b
i
n
a
r
y
c
l
a
ssi
f
i
c
a
t
i
o
n
t
a
s
k
s.
1
7
0
(
7
0
mel
a
n
o
m
a
,
1
0
0
n
e
v
i
)
M
ED
-
N
O
D
E
d
a
t
a
se
t
P
A
D
-
U
F
ES
-
20
d
a
t
a
se
t
[
5
]
,
[
6
]
Th
i
s
d
a
t
a
se
t
i
n
c
l
u
d
e
s
c
l
i
n
i
c
a
l
i
ma
g
e
s
o
f
p
i
g
me
n
t
e
d
sk
i
n
l
e
si
o
n
s
c
o
l
l
e
c
t
e
d
f
r
o
m
p
a
t
i
e
n
t
s
a
t
t
h
e
F
e
d
e
r
a
l
U
n
i
v
e
r
si
t
y
o
f
Es
p
í
r
i
t
o
S
a
n
t
o
,
B
r
a
z
i
l
.
I
t
a
l
s
o
c
o
n
t
a
i
n
s m
e
t
a
d
a
t
a
l
i
k
e
a
g
e
,
g
e
n
d
e
r
,
a
n
d
l
e
si
o
n
l
o
c
a
l
i
z
a
t
i
o
n
.
2
,
2
9
8
i
ma
g
e
s
I
EEE
D
a
t
a
P
o
r
t
3.
RE
SU
L
T
S
AND
D
I
SCU
SS
I
O
N
3
.
1
.
Resul
t
s
o
f
diff
er
ent
m
o
dels
u
s
ing
diff
er
ent
da
t
a
s
et
s
I
n
th
is
s
ec
tio
n
,
we
h
av
e
co
m
p
ar
ed
d
if
f
e
r
en
t
ar
ch
itectu
r
es
u
s
ed
f
o
r
s
k
in
ca
n
ce
r
d
etec
tio
n
,
an
d
h
av
e
o
b
s
er
v
ed
r
esu
lts
o
b
tain
ed
f
o
r
v
ar
io
u
s
p
a
r
am
eter
s
.
Fro
m
T
ab
le
3
,
we
ca
n
s
ee
f
o
r
th
e
H
AM
1
0
0
0
0
d
ataset,
b
etter
r
esu
lts
ar
e
o
b
tain
ed
u
s
i
n
g
SVM
is
9
7
%,
W
h
ile
f
o
r
th
e
PAD
-
UFES
-
2
0
d
ataset,
Ale
x
N
et
s
h
o
ws
a
g
o
o
d
r
esu
lt
o
f
9
9
%
an
d
f
o
r
th
e
I
SI
C
2
0
1
9
d
ataset
MSVM
s
h
o
ws
th
e
h
ig
h
est
ac
cu
r
ac
y
o
f
9
6
.
2
5
%.
A
co
m
p
ar
is
o
n
o
f
d
if
f
er
en
t m
et
h
o
d
s
f
o
r
ty
p
es o
f
d
atasets
is
o
b
s
er
v
ed
in
T
ab
le
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
4
0
4
-
2
4
1
5
2412
T
ab
le
3
.
Acc
u
r
ac
y
v
al
u
es o
b
ta
in
ed
b
y
d
if
f
er
e
n
t m
eth
o
d
s
f
o
r
d
if
f
er
en
t
d
atasets
H
A
M
1
0
0
0
0
d
a
t
a
se
t
A
c
c
u
r
a
c
y
P
A
D
-
U
F
ES
-
2
0
smar
t
p
h
o
n
e
i
ma
g
e
s
(
S
I
)
A
c
c
u
r
a
c
y
I
S
I
C
2
0
1
9
A
c
c
u
r
a
c
y
R
e
sN
e
t
X
c
e
p
t
i
o
n
[
8
]
7
8
.
1
5
%
,
A
l
e
x
-
N
e
t
w
i
t
h
T
L
a
n
d
a
u
g
me
n
t
e
d
d
a
t
a
[
5
]
9
9
%
M
S
V
M
[
1
3
]
9
6
.
2
5
%
D
e
n
seN
e
t
[
8
]
8
1
.
9
%
M
o
b
i
l
e
N
et
-
V
2
w
i
t
h
TL
o
n
a
u
g
me
n
t
e
d
d
a
t
a
[
5
]
9
4
.
0
7
1
%
S
V
M
[
6
]
9
5
%,
R
e
sN
e
t
5
0
[
9
]
9
0
%
Res
N
et
-
5
0
w
i
t
h
T
L
o
n
a
u
g
m
e
n
t
e
d
d
a
t
a
[
5
]
9
4
.
9
1
8
%
K
N
N
[
6
]
9
4
%,
S
V
M
[
6
]
9
7
%
C
N
N
-
TL
[
6
]
7
4
.
8
5
%
D
T
[
6
]
9
3
%
K
N
N
[
6
]
9
5
%
-
-
Ef
f
i
c
i
e
n
N
e
t
B
0
[
1
1
]
8
6
.
2
5
%
D
T
[
6
]
9
5
%
-
-
GA
-
H
o
l
d
o
u
t
[
1
1
]
9
0
.
0
9
%
M
u
l
t
i
m
o
d
a
l
f
u
si
o
n
(
A
LB
EF)
[
2
4
]
9
4
.
1
1
%
GA
-
c
r
o
ss
-
v
a
l
i
d
a
t
i
o
n
[
1
1
]
8
8
.
6
9
%
3
.
2
.
Resul
t
s
o
f
diff
er
ent
CN
N
m
o
dels
o
n diff
er
ent
pa
ra
m
et
er
s
Fro
m
Fig
u
r
e
3
,
we
o
b
s
er
v
e,
th
at
d
if
f
er
en
t
C
NN
ar
ch
ite
ctu
r
es
s
u
ch
as
Alex
-
N
et
wit
h
T
L
an
d
au
g
m
en
ted
d
ata,
Mo
b
ile
N
et
w
ith
T
L
o
n
au
g
m
en
ted
d
ata,
an
d
R
es
N
et
-
5
0
with
T
L
o
n
au
g
m
en
ted
d
ata
[
5
]
,
ar
e
co
m
p
ar
ed
f
o
r
d
if
f
e
r
en
t
p
ar
a
m
eter
s
,
i.e
.,
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e
r
esp
ec
tiv
ely
[
5
]
.
T
h
e
r
esear
ch
[
5
]
co
m
p
ar
e
d
th
e
r
esu
lts
f
r
o
m
two
s
ce
n
ar
i
o
s
u
s
in
g
th
r
ee
d
if
f
e
r
en
t
m
et
h
o
d
o
lo
g
ies.
T
h
e
b
est
o
u
tco
m
es
ar
e
o
b
tain
ed
u
s
in
g
an
Alex
-
n
et
with
tr
an
s
f
er
lear
n
in
g
th
at
h
as
b
ee
n
tr
a
in
ed
o
n
im
p
r
o
v
e
d
p
h
o
to
g
r
ap
h
s
with
an
ac
cu
r
ac
y
o
f
9
9
.
1
5
5
%.
I
n
Fi
g
u
r
e
4
,
w
e
h
a
v
e
c
o
m
p
a
r
e
d
d
i
f
f
e
r
e
n
t
C
NN
T
L
a
r
c
h
i
tec
tu
r
es
f
o
r
d
i
f
f
e
r
e
n
t
d
atas
ets
i.
e.
,
s
m
a
r
t
p
h
o
n
e
im
a
g
es
a
n
d
d
e
r
m
o
s
co
p
y
i
m
a
g
es.
I
t
is
o
b
s
e
r
v
e
d
C
NN
T
L
(
d
er
m
o
s
c
o
p
y
im
ag
es
)
h
as
m
o
r
e
ac
c
u
r
a
cy
(
8
7
.
8
0
%
)
an
d
s
e
n
s
iti
v
i
ty
(
9
5
.
5
0
%
)
as
c
o
m
p
ar
ed
t
o
C
NN
T
L
(
s
m
ar
tp
h
o
n
e
im
ag
es)
a
r
c
h
i
te
ct
u
r
e
[
6
]
,
b
u
t
C
NN
t
r
a
n
s
f
e
r
lea
r
n
i
n
g
(
s
m
ar
tp
h
o
n
e
i
m
a
g
es
)
h
as
m
o
r
e
s
p
e
ci
f
ici
ty
(
7
1
.
4
0
%)
a
n
d
p
r
ec
is
i
o
n
(
9
4
.
1
0
%)
a
s
c
o
m
p
ar
e
d
to
o
t
h
e
r
ar
ch
ite
ct
u
r
e.
Fr
o
m
Fi
g
u
r
e
s
3
a
n
d
4
C
NN
ar
c
h
it
ec
t
u
r
es
s
u
c
h
as
A
le
x
-
N
et
wit
h
T
L
o
n
au
g
m
e
n
te
d
d
ata
,
Mo
b
il
e
N
e
t
wit
h
T
L
o
n
au
g
m
e
n
te
d
d
at
a,
a
n
d
R
es
N
et
-
5
0
wit
h
T
L
o
n
a
u
g
m
e
n
t
ed
d
a
ta
[
5
]
,
C
NN
T
L
a
r
c
h
i
tec
tu
r
e
f
o
r
d
i
f
f
er
en
t
d
atas
ets
i
.
e
.
,
s
m
a
r
t
p
h
o
n
e
i
m
a
g
es
a
n
d
d
e
r
m
o
s
co
p
y
i
m
a
g
es
[
6
]
,
is
c
o
m
p
ar
e
d
f
o
r
d
i
f
f
e
r
e
n
t
p
a
r
a
m
e
te
r
s
,
i.
e
.
,
ac
cu
r
a
c
y
,
s
en
s
i
ti
v
it
y
,
s
p
ec
if
ici
ty
,
p
r
ec
is
i
o
n
,
a
n
d
F1
-
s
co
r
e
r
esp
ec
ti
v
e
ly
[
5
]
,
[
6
]
.
C
N
N
ar
ch
ite
ct
u
r
es
-
Ale
x
N
et
wit
h
T
L
a
n
d
d
a
ta
a
u
g
m
e
n
t
ati
o
n
h
as
th
e
b
est
r
esu
lt
as
c
o
m
p
a
r
ed
t
o
all
t
h
e
a
r
c
h
i
tec
tu
r
es.
Fig
u
r
e
3
.
C
o
m
p
a
r
ativ
e
p
a
r
am
e
ter
s
f
o
r
d
if
f
er
en
t a
r
c
h
itectu
r
es
[
5
]
Fig
u
r
e
4
.
C
o
m
p
a
r
ativ
e
ac
cu
r
a
cy
f
o
r
d
if
f
er
e
n
t CNN tr
an
s
f
er
lear
n
in
g
a
r
ch
itectu
r
e
[
6
]
99%
96.36
%
99.79
%
99.17
%
97.66%
94.0
7%
84.5
5%
96.2
5%
8
4
.2
2
%
84.2
0%
94.9
2%
77.2
8%
98.9
2%
95.1
9%
84.7
4%
0%
50%
100
%
A
c
c
ur
ac
y
se
ns
i
t
i
v
i
t
y
sp
e
c
i
f
i
c
i
t
y
pr
e
c
i
si
o
n
F
1
-
sc
o
r
e
C
N
N
A
r
c
h
i
t
e
c
u
t
e
r
e
s
Pa
r
a
m
e
te
r
s
A
l
e
xne
t
w
i
t
h
t
r
a
n
sf
e
r
l
e
ar
ni
ng
(
T
L)
o
n
da
t
a
au
g
m
e
nt
at
i
o
n
M
o
bi
l
e
ne
t
-
V2
wi
t
h
T
L
o
n
au
g
m
e
nt
e
d
da
t
as
e
t
R
e
sn
e
t
-
5
0 wi
t
h
T
L
o
n
au
g
m
e
nt
e
d
da
t
as
e
t
0.0
0%
20.
00%
40.
00%
60.
00%
80.
00%
100
.00%
A
c
c
ur
ac
y
se
ns
i
t
i
v
i
t
y
sp
e
c
i
f
i
c
i
t
y
pr
e
c
i
si
o
n
74.85
%
75.3
0%
71.4
0%
94.1
0%
87.8
0%
95.5
0%
5
7
.6
0
%
90.0
0%
C
N
N
t
r
a
n
f
e
r
l
e
a
r
n
i
n
g
A
r
c
h
i
t
e
c
t
u
r
e
o
n
d
i
f
f
e
r
e
n
t
d
a
t
a
sets
Pa
r
a
m
e
t
e
r
s
C
N
N
t
r
an
sf
e
r
l
e
ar
n
i
n
g
(
D
e
r
m
o
sc
o
py
)
C
N
N
t
r
an
sf
e
r
l
e
ar
ni
ng
(
S
m
ar
t
ph
o
ne
Im
ag
e
s
(
S
I
)
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
co
mp
r
eh
en
s
ive
a
n
a
lysi
s
o
f
d
iffer
en
t m
o
d
els:
s
kin
ca
n
ce
r
d
etec
tio
n
(
A
mru
ta
Th
o
r
a
t
)
2413
3
.
3
.
Resul
t
s
o
f
diff
er
ent
M
L
m
o
dels
f
o
r
diff
er
ent
da
t
a
s
et
Fig
u
r
e
5
p
r
esen
ts
a
co
m
p
a
r
is
o
n
o
f
v
ar
io
u
s
m
ac
h
in
e
lea
r
n
i
n
g
ar
ch
itectu
r
es,
in
clu
d
in
g
S
VM
,
KNN,
an
d
DT
,
a
p
p
lied
to
th
e
I
SIC
2
0
1
9
a
n
d
HAM
1
0
0
0
0
d
atasets
.
T
h
e
r
esu
lts
d
em
o
n
s
tr
ate
th
at
t
h
e
SVM
alg
o
r
ith
m
ac
h
iev
es
th
e
h
ig
h
est
ac
c
u
r
ac
y
am
o
n
g
th
e
ev
alu
ated
m
eth
o
d
s
f
o
r
b
o
th
d
atasets
.
T
h
is
s
u
p
er
io
r
p
e
r
f
o
r
m
an
ce
h
ig
h
lig
h
ts
SVM'
s
ca
p
ab
ilit
y
to
ef
f
ec
tiv
el
y
h
a
n
d
le
c
o
m
p
lex
class
if
icatio
n
task
s
in
m
ed
ical
im
ag
e
a
n
aly
s
is
,
as
s
ee
n
in
th
e
co
n
te
x
t
o
f
s
k
in
lesi
o
n
class
if
icatio
n
.
T
h
ese
f
in
d
in
g
s
,
as
s
u
p
p
o
r
ted
b
y
[
6
]
,
em
p
h
asize
th
e
im
p
o
r
tan
ce
o
f
s
elec
tin
g
r
o
b
u
s
t
alg
o
r
ith
m
s
lik
e
SVM
f
o
r
ac
h
iev
in
g
o
p
tim
al
r
esu
lts
in
s
u
ch
h
ig
h
-
s
tak
es
ap
p
licatio
n
s
.
Fig
u
r
e
5
.
C
o
m
p
a
r
ativ
e
ac
cu
r
a
cy
f
o
r
d
if
f
er
e
n
t
m
ac
h
in
e
lear
n
i
n
g
ar
ch
itectu
r
es f
o
r
d
if
f
er
e
n
t d
atasets
[
6
]
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
h
as
d
e
m
o
n
s
tr
ated
th
e
ef
f
ec
tiv
en
ess
o
f
DL
an
d
ML
m
o
d
els,
in
ad
d
r
ess
in
g
s
k
in
ca
n
ce
r
d
iag
n
o
s
is
.
B
y
u
tili
zin
g
a
d
ee
p
lear
n
in
g
-
b
ased
ap
p
r
o
ac
h
,
we
wer
e
ab
le
to
h
ig
h
er
ac
cu
r
ac
y
,
an
d
im
p
r
o
v
ed
s
en
s
itiv
ity
,
s
ig
n
if
ican
tly
o
u
t
p
e
r
f
o
r
m
in
g
tr
a
d
itio
n
al
m
eth
o
d
s
.
Ou
r
f
in
d
in
g
s
h
i
g
h
lig
h
t
t
h
e
p
o
t
en
tial
o
f
AI
-
d
r
iv
e
n
d
iag
n
o
s
tic
to
o
ls
will
h
elp
less
en
th
e
b
u
r
d
e
n
o
n
h
ea
lth
ca
r
e
w
o
r
k
er
s
an
d
im
p
r
o
v
e
clin
ical
d
e
cisi
o
n
-
m
ak
in
g
.
T
h
e
r
esu
lts
co
n
f
ir
m
th
e
r
eliab
ilit
y
an
d
ef
f
icien
cy
o
f
o
u
r
m
o
d
el
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
.
I
t is al
s
o
o
b
s
er
v
ed
d
if
f
er
en
t
d
atasets
wo
r
k
s
o
n
d
if
f
er
en
t
av
ailab
le
m
o
d
els
to
o
b
tain
th
e
b
est
r
esu
lts
,
s
u
ch
as
f
r
o
m
o
u
r
a
n
aly
s
is
we
ca
m
e
to
co
n
clu
s
io
n
,
f
o
r
HAM
1
0
0
0
d
at
aset,
SVM
s
h
o
ws
h
ig
h
est
ac
c
u
r
ac
y
o
f
9
7
%
f
o
r
PAD
-
UFEU
S
d
ataset,
Alex
-
N
et
with
T
L
an
d
a
u
g
m
en
te
d
d
ata
s
h
o
ws 9
9
% a
cc
u
r
ac
y
an
d
f
o
r
I
SIC 2
0
1
9
MSVM
9
6
.
2
5
%
5.
F
UT
UR
E
WO
RK
Desp
ite
th
e
p
r
o
m
is
in
g
r
esu
lts
,
th
er
e
ar
e
s
ev
er
al
ar
ea
s
wh
er
e
f
u
r
th
er
r
esear
ch
an
d
d
ev
elo
p
m
en
t
ar
e
n
ec
ess
ar
y
.
I
n
f
u
tu
r
e
wo
r
k
,
we
aim
to
,
−
E
x
p
an
d
th
e
d
ataset:
I
n
cl
u
d
in
g
lar
g
er
,
m
o
r
e
d
i
v
er
s
e
d
atasets
to
im
p
r
o
v
e
th
e
m
o
d
el’
s
g
en
er
al
izab
ilit
y
ac
r
o
s
s
d
if
f
er
en
t
p
o
p
u
latio
n
s
an
d
c
o
n
d
itio
n
s
,
−
E
n
h
an
ce
m
o
d
el
in
ter
p
r
eta
b
ilit
y
:
Dev
elo
p
in
g
ex
p
lain
ab
le
AI
(
XAI
)
m
eth
o
d
s
to
m
a
k
e
th
e
m
o
d
el’
s
p
r
ed
ictio
n
s
m
o
r
e
tr
an
s
p
ar
e
n
t a
n
d
u
n
d
er
s
tan
d
ab
le
to
clin
ician
s
,
−
E
x
p
lo
r
e
ad
v
a
n
ce
d
m
o
d
els:
I
n
v
esti
g
atin
g
n
ewe
r
d
ee
p
lear
n
in
g
a
r
ch
itectu
r
es,
to
f
u
r
th
er
en
h
an
c
e
m
o
d
el
p
er
f
o
r
m
an
ce
,
−
R
ea
l
-
wo
r
ld
d
ep
lo
y
m
e
n
t:
T
esti
n
g
th
e
s
y
s
tem
in
r
ea
l
-
tim
e
clin
ical
en
v
ir
o
n
m
e
n
ts
to
en
s
u
r
e
its
p
r
ac
tical
ap
p
licab
ilit
y
an
d
e
v
alu
ate
its
im
p
ac
t o
n
clin
ical
wo
r
k
f
lo
ws,
−
Mu
lti
-
m
o
d
al
d
ata
in
teg
r
atio
n
:
I
n
co
r
p
o
r
atin
g
o
th
er
d
iag
n
o
s
tic
d
ata
s
o
u
r
ce
s
to
im
p
r
o
v
e
p
r
ed
ictio
n
ac
cu
r
ac
y
an
d
p
r
o
v
id
e
m
o
r
e
c
o
m
p
r
e
h
en
s
iv
e
d
iag
n
o
s
tic
s
u
p
p
o
r
t.
T
h
ese
f
u
tu
r
e
r
esear
c
h
d
ir
ec
ti
o
n
s
will
h
elp
f
u
r
th
e
r
r
ef
in
e
o
u
r
m
o
d
el,
m
ak
i
n
g
it
m
o
r
e
r
o
b
u
s
t
an
d
u
s
ef
u
l
in
d
iv
er
s
e
clin
ical
s
ettin
g
s
.
RE
F
E
R
E
NC
E
S
[
1
]
A
.
Jav
a
i
d
,
M
.
S
a
d
i
q
,
a
n
d
F
.
A
k
r
a
m
,
“
S
k
i
n
c
a
n
c
e
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
si
n
g
i
ma
g
e
p
r
o
c
e
ss
i
n
g
a
n
d
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
1
8
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
Bh
u
r
b
a
n
C
o
n
f
e
re
n
c
e
o
n
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
a
n
d
T
e
c
h
n
o
l
o
g
i
e
s
,
I
BC
AS
T
2
0
2
1
,
Ja
n
.
2
0
2
1
,
p
p
.
4
3
9
–
4
4
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
B
C
A
S
T5
1
2
5
4
.
2
0
2
1
.
9
3
9
3
1
9
8
.
0%
20%
40%
60%
80%
100
%
S
V
M
KNN
DT
97%
95%
95%
95%
94%
93%
A
c
c
u
r
a
c
y
o
f
d
i
f
f
e
r
e
n
t
d
a
t
a
e
t
s
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
A
r
c
h
i
t
e
c
t
u
r
e
s
H
A
M
1000
0
D
at
as
e
t
IS
IC
20
19
D
at
as
e
t
Evaluation Warning : The document was created with Spire.PDF for Python.