I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
E
lec
t
r
ical
an
d
Com
p
u
t
e
r
E
n
gin
e
e
r
in
g
(
I
JE
CE
)
Vol.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
,
pp.
319
~
327
I
S
S
N:
2088
-
8708
,
DO
I
:
10
.
11591/i
jec
e
.
v
15
i
1
.
pp
3
19
-
327
319
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
e
c
e
.
iaes
c
or
e
.
c
om
D
e
e
p
l
e
ar
n
in
g
f
or
sk
i
n
m
e
la
n
om
a c
la
ssi
f
ic
at
io
n
u
si
n
g
d
e
r
m
osc
o
p
ic
i
m
ag
e
s i
n
d
if
f
e
r
e
n
t
c
ol
o
r
sp
ac
e
s
S
an
k
ar
ak
u
t
t
i
P
alan
icham
y
M
an
ik
an
d
a
n
1
,
S
an
d
e
e
p
Re
d
d
y
Nar
an
i
2
,
S
ak
t
h
ivel
Kart
h
ik
e
yan
3
,
Nagar
aj
an
M
oh
an
k
u
m
ar
4
1
D
e
pa
r
tm
e
nt
of
E
le
c
tr
oni
c
s
a
nd C
omm
uni
c
a
ti
on E
ngi
ne
e
r
in
g, S
a
ve
e
th
a
E
ngi
ne
e
r
in
g C
ol
le
ge
, C
h
e
nna
i,
I
ndi
a
2
I
nde
pe
nde
nt
R
e
s
e
a
r
c
he
r
,
T
e
xa
s
, U
ni
te
d S
ta
te
s
of
A
me
r
ic
a
3
D
e
pa
r
tm
e
nt
of
E
le
c
tr
oni
c
s
a
nd C
omm
uni
c
a
ti
on E
ngi
ne
e
r
in
g, K
.S
.R
. C
ol
le
ge
of
E
ngi
ne
e
r
in
g, T
ir
uc
he
ngode
, I
ndi
a
4
D
e
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ie
nc
e
a
nd E
ngi
ne
e
r
in
g, S
ymbi
os
is
I
ns
ti
tu
te
of
T
e
c
hnol
ogy, Na
gpur
C
a
mpus
, S
ymbi
os
i
s
I
nt
e
r
na
ti
ona
l
(
D
e
e
me
d U
ni
ve
r
s
it
y)
, P
une
, I
ndi
a
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
M
a
y
14,
2024
R
e
vis
e
d
S
e
p
7,
2024
Ac
c
e
pted
Oc
t
1,
2024
Sk
i
n
can
cer
b
eg
i
n
s
i
n
t
h
e
s
k
i
n
cel
l
s
.
T
h
e
d
amag
e
t
o
t
h
e
s
k
i
n
ce
l
l
s
ca
n
cau
s
e
g
en
e
t
i
c
mu
t
at
i
o
n
s
t
h
at
l
ea
d
t
o
u
n
co
n
t
r
o
l
l
ed
g
r
o
w
t
h
a
n
d
t
h
e
fo
rmat
i
o
n
o
f
t
u
mo
rs
.
It
i
s
es
t
i
ma
t
ed
t
h
at
m
i
l
l
i
o
n
s
o
f
p
e
o
p
l
e
are
d
i
a
g
n
o
s
ed
w
i
t
h
s
k
i
n
can
cer
o
f
d
i
ffere
n
t
k
i
n
d
s
each
y
ear.
T
h
e
earl
i
er
a
s
k
i
n
can
cer
i
s
d
i
a
g
n
o
s
e
d
,
t
h
e
b
et
t
er
t
h
e
p
a
t
i
e
n
t
'
s
p
r
o
g
n
o
s
i
s
an
d
t
h
e
l
o
w
e
r
t
h
e
i
r
ch
a
n
ce
o
f
c
o
mp
l
i
ca
t
i
o
n
s
.
In
t
h
i
s
w
o
r
k
,
an
eff
i
ci
e
n
t
d
eep
l
ear
n
i
n
g
c
l
as
s
i
f
i
cat
i
o
n
(E
D
L
CS)
t
o
c
l
as
s
i
f
y
d
ermo
s
co
p
i
c
i
ma
g
es
i
s
d
ev
e
l
o
p
e
d
.
T
h
e
i
m
p
o
r
t
an
ce
o
f
c
o
l
o
r
i
n
t
h
e
d
i
a
g
n
o
s
i
s
o
f
s
k
i
n
mel
an
o
ma
h
a
s
ca
u
s
e
d
co
l
o
r
an
a
l
y
s
i
s
t
o
a
t
t
rac
t
c
o
n
s
i
d
erab
l
e
at
t
en
t
i
o
n
fro
m
res
earch
er
s
o
f
i
mag
e
-
b
as
e
d
s
k
i
n
mel
an
o
ma
an
al
y
s
i
s
.
T
h
ree
d
i
fferen
t
co
l
o
r
s
p
a
ces
s
u
ch
a
s
red
-
g
ree
n
-
b
l
u
e
(
RG
B
)
,
hue
-
s
at
u
r
at
i
o
n
-
l
i
g
h
t
n
e
s
s
(
H
IS
)
an
d
L
A
B
are
i
n
v
e
s
t
i
g
a
t
ed
i
n
t
h
i
s
s
t
u
d
y
.
T
h
e
o
b
t
a
i
n
e
d
d
ermo
s
co
p
i
c
i
ma
g
es
are
i
n
RG
B
co
l
o
r
s
p
ace.
T
h
e
RG
B
d
ermo
s
c
o
p
i
c
i
mag
e
s
are
fi
rs
t
co
n
v
ert
e
d
i
n
t
o
H
SV
an
d
L
A
B
s
p
aces
t
o
i
n
v
es
t
i
g
at
e
t
h
e
H
SV
an
d
L
A
B
co
l
o
r
s
p
ace
s
fo
r
mel
an
o
ma
cl
as
s
i
f
i
cat
i
o
n
.
T
h
en
,
t
h
e
co
l
o
r
s
p
ace
co
n
v
er
t
e
d
i
mag
e
i
s
fe
d
t
o
t
h
e
p
ro
p
o
s
ed
E
D
L
CS
t
o
ev
a
l
u
a
t
e
t
h
ei
r
p
erfo
rma
n
ces
.
Re
s
u
l
t
s
s
h
o
w
t
h
a
t
t
h
e
p
ro
p
o
s
ed
E
D
L
CS
p
ro
v
i
d
es
9
9
.
5
8
%
accu
rac
y
w
h
i
l
e
u
s
i
n
g
t
h
e
L
A
B
co
l
o
r
mo
d
e
l
t
o
c
l
as
s
i
f
y
p
rep
r
o
ces
s
ed
i
mag
e
s
w
h
i
l
e
o
t
h
er
mo
d
e
l
s
p
r
o
v
i
d
e
9
9
.
1
7
%
.
K
e
y
w
o
r
d
s
:
C
olor
s
pa
c
e
s
De
e
p
lea
r
ning
I
nc
e
pti
on
a
r
c
hit
e
c
tur
e
M
e
lanoma
c
las
s
if
ica
ti
on
S
kin
c
a
nc
e
r
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
Na
ga
r
a
jan
M
oha
nkumar
De
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ienc
e
a
nd
E
nginee
r
ing
,
S
ymbi
os
is
I
ns
ti
tut
e
of
T
e
c
hnology,
Na
gpur
C
a
mp
us
,
S
ymbi
os
is
I
nter
na
ti
ona
l
(
De
e
med
Unive
r
s
it
y)
P
une
,
I
ndia
E
mail:
nmkp
r
of
e
s
s
or
@gmail.
c
om
1.
I
NT
RODU
C
T
I
ON
S
k
in
c
a
n
c
e
r
o
c
c
u
r
s
whe
n
t
he
r
e
is
a
n
a
b
no
r
ma
l
p
r
oli
f
e
r
a
ti
on
of
s
ki
n
c
e
l
ls
,
w
hi
c
h
may
be
p
r
od
uc
e
d
b
y
e
xp
os
u
r
e
t
o
s
u
n’
s
u
lt
r
a
vi
o
let
(
UV
)
li
gh
t
a
nd
or
f
r
om
a
r
ti
f
ic
ia
l
s
o
u
r
c
e
s
.
T
he
t
a
n
ni
ng
be
ds
a
nd
s
ome
t
ype
s
o
f
las
e
r
s
a
r
e
e
x
a
m
pl
e
s
o
f
a
r
ti
f
ic
ia
l
s
o
u
r
c
e
s
o
f
UV
r
a
d
ia
ti
on
.
T
h
e
t
h
r
e
e
p
r
i
ma
r
y
f
o
r
ms
o
f
s
k
in
c
a
nc
e
r
a
r
e
a
s
f
o
ll
ows
:
a.
B
a
s
a
l
c
e
ll
c
a
r
c
inom
a
(
B
C
C
)
:
B
C
C
us
ua
ll
y
manif
e
s
ts
it
s
e
lf
on
r
e
gions
of
the
s
kin,
s
uc
h
a
s
the
f
a
c
e
a
n
d
the
ne
c
k
that
a
r
e
of
ten
e
xpos
e
d
to
the
s
un.
I
n
mos
t
c
a
s
e
s
,
it
take
s
the
f
or
m
of
a
ti
ny
,
e
leva
ted
hump
that
may
ha
ve
a
wa
xy
or
pe
a
r
les
c
e
nt
look.
b.
S
qua
mous
c
e
ll
c
a
r
c
inom
a
(
S
C
C
)
:
T
his
kind
of
s
ki
n
c
a
nc
e
r
is
a
ls
o
of
ten
s
e
e
n
on
por
ti
ons
of
the
s
kin
that
a
r
e
e
xpos
e
d
to
the
s
un.
I
t
is
a
ls
o
pos
s
ibl
e
on
other
pa
r
ts
of
the
body
a
s
we
ll
.
I
t
manif
e
s
ts
it
s
e
lf
mos
t
of
ten
a
s
a
r
a
is
e
d
lum
p
that
r
e
s
e
mbl
e
s
a
wa
r
t
or
a
s
c
a
ly
r
e
d
pa
tch.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
319
-
327
320
c.
M
e
lanoma
:
T
his
is
the
mos
t
s
e
r
ious
s
kin
c
a
nc
e
r
’
s
type.
I
t
c
a
n
e
xtend
to
other
a
r
e
a
s
o
f
the
body
i
n
it
s
unti
mely
s
tage
s
if
not
tr
e
a
ted
pr
ope
r
ly.
I
t
typ
ica
ll
y
s
e
e
ms
a
s
a
da
r
k,
une
ve
nly
de
ter
mi
ne
d
mol
e
or
s
pot
on
the
s
kin.
An
e
s
ti
mate
d
106
,
110
ne
w
c
a
s
e
s
of
mela
noma
a
nd
4
.
3
m
il
li
on
ne
w
non
-
mela
noma
ins
ta
nc
e
s
oc
c
ur
in
the
United
S
tate
s
e
ve
r
y
ye
a
r
[
1]
.
He
r
e
a
r
e
s
ome
a
ddit
ional
s
kin
c
a
nc
e
r
s
tatis
ti
c
s
:
i)
Ove
r
their
li
f
e
ti
me,
one
out
of
e
ve
r
y
f
ive
Ame
r
ica
ns
will
b
e
diagnos
e
d
with
s
kin
c
a
nc
e
r
;
ii
)
M
e
lanoma
is
t
he
5
th
f
r
e
que
nt
c
a
nc
e
r
to
oc
c
ur
in
male
s
a
nd
the
7
th
c
omm
on
type
of
c
a
nc
e
r
to
oc
c
ur
in
wome
n
;
ii
i
)
I
n
the
ye
a
r
2021,
it
is
a
nti
c
ipate
d
that
mela
noma
would
c
lai
m
the
li
ve
s
of
a
r
ound
7
,
180
pe
r
s
ons
;
iv)
T
he
f
iv
e
-
ye
a
r
s
ur
vival
pe
r
c
e
ntage
f
or
pe
r
s
ons
with
mela
noma
th
a
t
ha
s
not
gone
be
yond
the
s
kin
i
s
99%
,
pr
ovided
that
the
c
a
nc
e
r
ha
s
not
pr
ogr
e
s
s
e
d
;
v)
J
u
s
t
27%
of
pe
r
s
ons
with
mela
noma
that
ha
s
s
pr
e
a
d
to
other
r
e
gions
of
the
body
wil
l
be
a
li
ve
a
f
ter
f
ive
ye
a
r
s
;
vi)
I
f
a
pe
r
s
on
ha
s
ha
d
mo
r
e
than
f
ive
s
unbur
ns
in
their
li
f
e
ti
me,
their
c
ha
nc
e
of
a
c
quir
ing
mela
noma
is
incr
e
a
s
e
d
b
y
a
f
a
c
tor
of
two
;
a
nd
vii
)
T
a
nning
be
d
us
a
ge
be
f
or
e
the
a
ge
of
35
is
a
s
s
oc
iate
d
with
a
59
%
incr
e
a
s
e
d
r
is
k
of
a
c
quir
ing
mela
noma
in
thos
e
who
ha
ve
a
lr
e
a
dy
ha
d
the
dis
e
a
s
e
.
T
he
s
e
s
tatis
ti
c
s
highl
ight
the
im
por
tanc
e
of
pr
ote
c
ti
ng
your
s
kin
f
r
om
the
s
un's
ha
r
mf
ul
r
a
ys
a
nd
us
ua
ll
y
ve
r
if
ying
your
s
kin
f
or
a
ny
e
xc
ha
nge
s
or
a
bnor
malit
ies
.
E
a
r
ly
de
tec
ti
on
a
nd
tr
e
a
tm
e
nt
c
a
n
gr
e
a
tl
y
im
pr
ove
outcome
s
f
or
pe
ople
with
s
kin
c
a
nc
e
r
.
I
n
I
ndia
,
s
kin
c
a
nc
e
r
is
not
a
s
c
omm
on
c
ompar
e
d
to
other
pa
r
ts
of
the
wor
ld,
pa
r
ti
c
ula
r
ly
in
r
e
gions
with
li
g
hter
-
s
kinned
populations
that
a
r
e
mor
e
s
u
s
c
e
pti
bl
e
to
UV
r
a
diation
da
mage
.
How
e
ve
r
,
s
kin
c
a
nc
e
r
is
s
ti
ll
a
c
onc
e
r
n
in
I
ndia
due
to
the
high
leve
ls
o
f
UV
r
a
diation
e
xpos
ur
e
in
many
r
e
gions
,
pa
r
ti
c
ular
ly
in
a
r
e
a
s
c
los
e
r
to
the
e
qua
tor
.
S
kin
c
a
nc
e
r
a
c
c
ounts
f
o
r
1
%
to
2
%
o
f
a
l
l
c
a
nc
e
r
s
i
n
I
ndia
[
2]
.
T
he
incide
nc
e
of
s
kin
c
a
nc
e
r
is
higher
in
nor
ther
n
I
ndia,
whe
r
e
UV
r
a
diation
e
xpos
ur
e
is
higher
,
c
ompar
e
d
to
s
outher
n
I
ndia.
T
he
types
of
s
kin
c
a
nc
e
r
O
c
c
ur
s
in
I
ndia
a
r
e
B
C
C
a
nd
S
C
C
,
whi
c
h
a
r
e
pr
im
a
r
il
y
c
a
us
e
d
by
s
un
e
xpos
ur
e
.
M
e
lanoma
is
r
e
latively
r
a
r
e
in
I
ndia
.
R
is
k
f
a
c
tor
s
f
o
r
s
kin
c
a
nc
e
r
in
I
ndia
include
s
pe
nding
pr
olonged
pe
r
iods
of
ti
me
in
the
s
un,
ha
ving
f
a
ir
s
kin
,
a
f
a
mi
ly
his
tor
y,
a
nd
a
his
tor
y
of
s
unbur
ns
.
P
r
otec
ti
ng
one
's
s
kin
f
r
o
m
the
ha
r
mf
ul
e
f
f
e
c
ts
of
UV
f
r
om
the
s
un
is
the
mos
t
e
f
f
e
c
ti
ve
s
tr
a
tegy
to
s
tave
a
ga
ins
t
s
kin
c
a
nc
e
r
by
we
a
r
ing
pr
otec
ti
ve
c
lot
hing,
us
ing
s
uns
c
r
e
e
n
with
a
t
lea
s
t
S
P
F
30,
a
nd
a
voidi
ng
pr
olonged
e
xpos
ur
e
to
the
s
un
dur
ing
pe
a
k
hour
s
.
I
t
is
a
ls
o
im
por
tant
to
pe
r
f
or
m
r
e
gular
s
ki
n
s
e
lf
-
e
xa
mi
na
ti
ons
a
nd
to
ha
ve
a
ny
unus
ua
l
or
c
ha
nging
mol
e
s
or
s
pots
on
the
s
kin
e
va
luate
d
by
a
doc
to
r
.
2.
RE
L
AT
E
D
WORKS
T
he
r
e
ha
ve
be
e
n
s
e
ve
r
a
l
r
e
late
d
wor
ks
in
the
f
ield
of
s
kin
c
a
nc
e
r
c
a
tegor
iza
ti
on
uti
li
z
ing
de
e
p
lea
r
ning
(
DL
)
models
.
A
DL
model
f
or
s
kin
c
a
nc
e
r
c
las
s
if
ica
ti
on
is
de
s
c
r
ibed
in
[
3]
.
I
t
ha
s
7
c
onvolut
ion
laye
r
s
a
nd
3
ne
ur
a
l
laye
r
s
f
or
the
pur
pos
e
of
c
las
s
if
ying
de
r
mos
c
opic
im
a
ge
s
.
An
e
f
f
icie
nt
m
e
thod
is
de
s
c
r
ibed
in
[
4]
f
o
r
the
e
a
r
ly
identi
f
ica
ti
on
of
s
kin
c
a
nc
e
r
.
DL
a
r
c
hit
e
c
tur
e
s
s
uc
h
a
s
I
nc
e
pti
on
-
v3
a
nd
R
e
s
Ne
t
-
101
a
r
e
be
ing
us
e
d
f
or
the
c
las
s
if
ica
ti
on
c
ha
ll
e
nge
.
A
s
kin
c
a
nc
e
r
c
la
s
s
if
ica
ti
on
a
pp
r
oa
c
h
is
de
mons
tr
a
ted
in
[
5]
.
T
he
d
e
e
p
c
onvolut
ion
ne
ur
a
l
ne
twor
k
(
D
C
NN
)
,
VG
G16,
a
nd
VG
G19
models
a
r
e
tr
a
ined
a
nd
a
s
s
e
s
s
e
d
f
or
s
kin
c
a
nc
e
r
diagnos
i
s
.
An
im
pr
ove
d
im
a
ge
c
las
s
if
ica
ti
on
model
is
buil
t
in
[
6]
t
o
a
s
s
is
t
de
r
matologi
s
ts
in
the
pr
oc
e
s
s
of
making
diagnos
e
s
.
T
his
model
is
int
e
nde
d
to
s
e
r
ve
a
s
a
pr
e
li
mi
na
r
y
c
he
c
k
to
a
void
a
mor
e
e
xpe
ns
ive
biops
y.
Dur
ing
the
c
las
s
if
ica
ti
on
pr
oc
e
s
s
,
tr
a
ns
f
e
r
lea
r
ning
is
e
mpl
oye
d
by
c
ombi
ning
it
with
da
ta
a
ugmenta
ti
on
a
nd
c
las
s
-
we
ight
e
d
los
s
a
ppr
oa
c
he
s
.
A
method
that
make
s
e
f
f
e
c
ti
ve
us
e
of
DL
to
ident
if
y
s
kin
c
a
nc
e
r
is
s
ugge
s
ted
in
[
7]
.
A
f
ter
making
a
djus
tm
e
nts
to
the
pr
e
-
tr
a
ined
M
obil
e
Ne
t
c
onvolut
ion
ne
ur
a
l
ne
twor
k
(
C
NN
)
,
c
las
s
if
ica
ti
on
of
de
r
mos
c
opic
im
a
ge
s
is
a
c
hieve
d.
T
his
a
ppr
oa
c
h
of
t
r
a
ns
f
e
r
lea
r
ning
ha
s
s
hown
outs
tanding
a
c
c
ur
a
c
y
ove
r
a
br
oa
d
s
pe
c
tr
um.
T
he
s
kin
les
ions
a
r
e
c
las
s
if
ied
in
[
8]
us
i
ng
thr
e
e
c
utt
ing
-
e
dge
DL
pr
e
-
tr
a
ined
models
.
T
he
s
e
models
we
r
e
R
e
s
Ne
t,
Xc
e
pti
on,
a
nd
De
ns
e
Ne
t.
A
pr
e
s
c
r
e
e
ning
a
ppr
oa
c
h
to
diagnos
e
c
a
nc
e
r
s
oone
r
in
r
ur
a
l
l
oc
a
ti
ons
is
de
s
c
r
ibed
in
[
9
]
.
A
p
r
otot
ype
of
a
ga
dge
t
c
a
pa
b
le
of
s
e
gmenting
the
a
f
f
e
c
ted
r
e
gion
of
the
s
kin
in
to
s
e
ve
n
pr
im
a
r
y
c
a
tegor
ies
,
a
s
we
ll
a
s
c
las
s
if
ying
the
s
ki
n
a
bnor
malit
y
it
s
e
lf
is
p
r
ovided.
A
R
a
s
pbe
r
r
y
P
i
3B
+
,
a
magnif
ying
c
a
mer
a
a
tt
a
c
hment,
a
C
NN
that
po
we
r
s
s
kin
c
a
nc
e
r
r
e
c
ognit
ion.
S
kin
c
a
nc
e
r
boun
da
r
ies
a
r
e
s
e
gmente
d
us
ing
a
nother
model
,
a
nd
a
n
int
e
r
a
c
ti
ve
touchs
c
r
e
e
n
us
e
r
int
e
r
f
a
c
e
is
a
ll
include
d
in
the
p
r
otot
ype
.
T
he
c
ha
ll
e
nge
of
mer
ging
im
a
ge
s
a
nd
meta
da
ta
c
h
a
r
a
c
ter
is
ti
c
s
to
the
c
las
s
if
ica
ti
on
o
f
s
kin
c
a
nc
e
r
is
d
e
s
c
r
ibed
in
[
10
]
.
Dur
ing
the
whole
pr
oc
e
s
s
of
da
ta
c
las
s
if
ic
a
ti
on,
meta
da
ta
pr
oc
e
s
s
ing
block
pr
oc
e
s
s
is
s
ugge
s
ted.
T
his
a
lgor
it
hm
make
s
us
e
of
meta
da
ta
to
he
lp
da
ta
c
la
s
s
if
ica
ti
on
a
long
with
the
im
por
tant
c
ha
r
a
c
ter
is
ti
c
s
de
r
ived
f
r
om
i
mage
s
.
An
e
ns
e
mbl
e
of
de
e
p
lea
r
ne
r
s
c
a
pa
ble
of
de
tec
ti
ng
s
kin
c
a
nc
e
r
ha
s
be
e
n
c
ons
tr
uc
ted
in
[
11]
by
c
ombi
ning
the
lea
r
ne
r
s
of
the
VG
G
,
C
a
ps
Ne
t,
a
nd
R
e
s
Ne
t
models
.
T
wo
dif
f
e
r
e
nt
a
pp
r
oa
c
he
s
f
or
c
r
os
s
-
domain
s
kin
c
a
nc
e
r
identif
ica
ti
on
is
inves
ti
ga
ted
i
n
[
12]
.
A
two
-
s
tep
pr
ogr
e
s
s
ive
tr
a
ns
f
e
r
lea
r
n
ing
a
p
pr
oa
c
h
is
e
mpl
oye
d
thr
ough
the
us
e
o
f
two
dif
f
e
r
e
nt
s
kin
dis
e
a
s
e
da
tas
e
ts
that
he
lps
f
ine
tuni
ng
the
ne
twor
k
.
A
t
f
ir
s
t,
a
de
e
p
C
NN
c
las
s
if
ier
is
pr
e
-
tr
a
ined
on
I
mage
Ne
t.
T
he
n,
a
dve
r
s
a
r
ial
lea
r
ning
is
de
s
igned
to
c
on
duc
t
a
n
invar
iant
a
tt
r
ibut
e
tr
a
ns
lation
to
pr
ovide
good
r
e
s
ul
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
De
e
p
lear
ning
for
s
k
in
me
lanoma
c
las
s
if
ication
us
ing
.
.
.
(
Sank
ar
ak
utt
i
P
alani
c
hamy
M
anikandan
)
321
A
s
tr
a
tegy
is
of
f
e
r
e
d
in
[
13
]
to
f
us
e
the
c
ha
r
a
c
ter
is
ti
c
s
of
DL
with
the
c
ha
r
a
c
ter
is
ti
c
of
tr
a
dit
ional
im
a
ge
pr
oc
e
s
s
ing.
A
hypothes
is
whic
h
ha
s
dis
ti
nc
t
e
r
r
or
pr
of
il
e
s
is
de
ve
loped
a
nd
they
a
r
e
c
ompl
e
m
e
ntar
y
to
one
a
nother
.
T
he
t
r
a
dit
ional
im
a
ge
pr
oc
e
s
s
ing
a
r
m
c
ons
is
ts
of
a
c
li
nica
l
modul
e
a
nd
thr
e
e
im
a
ge
pr
oc
e
s
s
ing
modul
e
s
.
T
he
im
a
ge
pr
oc
e
s
s
ing
modul
e
s
a
r
e
a
bl
e
to
identif
y
les
ion
c
ha
r
a
c
ter
is
ti
c
s
that
a
r
e
a
na
lo
gous
to
c
li
nica
l
de
r
mos
c
opy
da
ta,
s
uc
h
a
s
a
n
a
bnor
mal
pigm
e
nt
ne
twor
k
,
c
olor
dis
tr
ibut
ion,
a
nd
bloo
d
ve
s
s
e
l
dis
tr
ibut
ion.
A
f
ull
y
a
utom
a
ted
DL
e
ns
e
mbl
e
s
a
r
e
pr
e
s
e
nted
in
[
14]
f
or
s
kin
c
a
nc
e
r
c
las
s
if
ica
t
ion.
T
he
e
ns
e
mbl
e
tec
hniques
us
ing
M
a
s
k
R
-
C
NN
a
nd
De
e
plabV
3+
a
ppr
oa
c
he
s
a
r
e
de
ve
loped.
A
de
e
p
s
upe
r
vis
e
d
mul
ti
-
s
c
a
le
ne
twor
k
(
DSM
-
Ne
twor
k)
is
de
s
c
r
ibed
in
[
15
]
.
T
o
ha
ndle
di
f
f
e
r
e
nt
s
ize
s
of
s
kin
les
ions
,
a
mul
ti
-
s
c
a
le
c
onne
c
ti
on
bloc
k
is
planne
d
a
nd
a
ggr
e
ga
tes
inf
or
mation
f
r
om
s
ha
l
low
a
nd
de
e
p
laye
r
s
.
I
n
a
ddit
ion,
a
c
ondit
ional
r
a
ndom
f
ie
ld
model
us
e
s
pos
t
-
pr
oc
e
s
s
ing
to
r
e
f
ine
the
s
kin
c
ontour
.
A
li
ghtwe
ight
model
f
or
the
de
tec
ti
on
of
s
kin
c
a
nc
e
r
with
f
e
a
tur
e
dis
c
r
im
ination
is
s
ugge
s
ted
in
[
16]
.
I
t
is
ba
s
e
d
on
the
noti
on
of
f
ine
-
gr
a
ined
c
a
tegor
iza
ti
on
a
nd
ha
s
two
f
e
a
tur
e
e
xtr
a
c
ti
on
modul
e
s
that
a
r
e
s
ha
r
e
d
a
c
r
os
s
them:
f
e
a
tur
e
dis
c
r
im
ination
a
nd
a
les
ion
c
las
s
if
ica
ti
on
ne
twor
k.
A
hybr
id
da
ta
mi
ning
s
tr
a
tegy
is
s
ugge
s
ted
in
[
17]
.
I
t
in
tegr
a
tes
k
-
ne
a
r
e
s
t
ne
ighbor
(
KN
N)
a
nd
s
uppor
t
ve
c
tor
mac
hine
(
S
VM
)
to
a
s
s
e
mbl
e
up
an
a
c
c
ur
a
te
pr
e
pa
r
a
ti
on
f
or
br
e
a
s
t
c
a
nc
e
r
ous
de
ve
l
opment
e
s
ti
mate
.
T
he
pr
e
diction
of
Alz
he
im
e
r
's
di
s
e
a
s
e
h
a
s
r
e
c
e
ived
a
20
pe
r
c
e
nt
e
nha
nc
e
ment
in
c
ha
r
a
c
ter
iza
ti
on
a
c
c
ur
a
c
y.
T
he
s
c
ope
of
th
is
model
is
a
ppli
e
d
t
o
hybr
id
a
r
ti
f
icia
l
int
e
ll
igenc
e
(
AI
)
c
a
lcula
ti
ons
that
or
ga
nize
S
VM
with
C
NN
to
e
xpe
c
t
Alz
he
im
e
r
's
s
ickn
e
s
s
a
nd
make
a
he
lpf
ul
model
[
18]
.
T
h
e
r
a
d
i
o
f
r
e
q
u
e
n
c
y
(
RF
)
m
o
d
u
l
e
b
u
i
l
d
s
w
i
r
e
l
e
s
s
d
a
t
a
t
r
a
n
s
f
e
r
a
n
d
t
r
a
n
s
m
i
s
s
i
o
n
b
e
t
w
e
e
n
t
h
e
w
e
a
r
a
b
l
e
d
e
v
i
c
e
s
.
T
h
e
s
e
n
s
o
r
d
a
t
a
a
n
d
a
l
e
r
t
s
a
n
n
o
u
n
c
e
m
e
n
t
r
a
p
i
d
l
y
s
e
n
d
t
o
t
h
e
c
l
o
u
d
[
1
9
]
.
3.
M
E
T
HO
DS
AN
D
M
AT
E
RI
AL
S
T
he
p
r
opos
e
d
e
f
f
icie
nt
de
e
p
lea
r
ning
c
las
s
if
ica
ti
on
(
E
DL
C
S
)
is
de
s
igned
to
c
las
s
if
y
the
de
r
mos
c
opic
im
a
ge
s
f
or
s
kin
c
a
nc
e
r
diagnos
is
.
An
im
a
ge
c
las
s
if
ica
ti
on
s
ys
tem
is
a
type
of
AI
s
ys
tem
that
is
de
s
igned
to
a
utom
a
ti
c
a
ll
y
c
las
s
if
y
im
a
ge
s
ba
s
e
d
on
their
c
onte
nt.
I
mage
c
las
s
if
ica
ti
on
s
ys
tems
us
e
DL
a
lgor
it
hms
,
s
uc
h
a
s
C
NN
s
,
to
a
na
lyze
the
f
e
a
tu
r
e
s
a
nd
pa
tt
e
r
ns
with
in
a
n
im
a
ge
a
nd
c
las
s
if
y
it
int
o
one
or
mor
e
pr
e
de
f
ined
c
a
tegor
ies
.
T
he
pr
oc
e
s
s
of
c
r
e
a
ti
ng
a
n
im
a
ge
c
las
s
if
ica
ti
on
s
ys
tem
typi
c
a
ll
y
invol
ve
s
tr
a
ini
ng
the
s
ys
tem
with
a
labe
led
im
a
ge
.
Dur
ing
the
tr
a
ini
ng
p
r
oc
e
s
s
,
the
s
ys
tem
lea
r
ns
to
identi
f
y
the
unique
f
e
a
tur
e
s
a
nd
pa
tt
e
r
ns
withi
n
e
a
c
h
c
a
tegor
y
of
im
a
ge
s
.
Onc
e
the
s
ys
tem
is
tr
a
ined,
it
c
a
n
then
be
us
e
d
to
tes
t
ne
w
im
a
ge
s
int
o
the
pr
e
de
f
ined
c
a
tegor
ies
.
T
he
wor
king
f
low
of
the
pr
opos
e
d
E
DL
C
S
f
or
s
kin
mela
noma
c
las
s
if
ica
ti
on
us
ing
de
r
mos
c
opic
im
a
ge
s
in
dif
f
e
r
e
nt
c
olor
s
pa
c
e
s
is
s
h
own
in
F
igu
r
e
1.
F
igur
e
1.
S
kin
mela
noma
c
las
s
if
ica
ti
on
pr
oc
e
s
s
3.
1.
P
r
e
p
r
oc
e
s
s
in
g
I
mage
pr
e
pr
oc
e
s
s
ing
is
a
s
e
r
ies
of
ope
r
a
ti
ons
pe
r
f
or
med
on
a
n
im
a
ge
be
f
or
e
it
is
a
na
lyze
d
or
us
e
d
f
or
f
ur
the
r
pr
oc
e
s
s
ing.
I
n
im
a
ge
a
na
lys
is
a
nd
c
om
puter
vis
ion
tas
ks
,
p
r
e
pr
oc
e
s
s
ing
is
a
n
im
po
r
tant
s
tep,
a
s
it
c
a
n
im
pr
ove
the
a
c
c
ur
a
c
y
a
nd
r
e
li
a
bil
it
y
of
the
a
na
lys
is
by
r
e
movi
ng
nois
e
,
e
nha
nc
ing
c
ontr
a
s
t,
a
nd
r
e
duc
ing
the
im
pa
c
t
of
a
r
ti
f
a
c
ts
in
the
im
a
ge
.
S
ome
c
omm
on
im
a
ge
pr
e
pr
oc
e
s
s
ing
tec
hniques
include
im
a
ge
r
e
s
izing,
im
a
ge
c
r
opping,
im
a
ge
f
il
ter
ing
,
im
a
ge
no
r
maliza
ti
on
a
nd
th
r
e
s
holdi
ng.
I
n
thi
s
wor
k
,
media
n
f
i
lt
e
r
in
g
[
20]
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
319
-
327
322
e
mpl
oye
d
f
or
nois
e
r
e
moval
a
nd
im
a
ge
r
e
s
izing
is
pe
r
f
or
med
be
f
or
e
c
las
s
if
ica
ti
on.
F
il
ter
ing
a
n
im
a
ge
invol
ve
s
a
pplyi
ng
a
mathe
matica
l
ope
r
a
ti
on
to
the
pixels
in
the
im
a
ge
to
im
pr
ove
or
e
li
mi
na
te
de
f
ini
te
f
e
a
tur
e
s
.
S
ome
c
omm
on
f
i
lt
e
r
s
include
Ga
us
s
ian
f
il
ter
s
,
media
n
f
il
ter
s
,
a
nd
e
dge
de
tec
ti
on
f
il
ter
s
.
M
e
dian
f
il
ter
is
de
f
ined
in
(
1)
.
=
{
+
,
+
:
,
=
−
,
.
.
.
.
}
,
=
(
+
1
)
,
.
.
.
.
.
(
−
)
(
1)
He
r
e
is
the
f
il
ter
s
ize
.
F
igur
e
2
s
hows
the
pr
e
pr
oc
e
s
s
ing
r
e
s
ult
s
of
the
E
DL
C
S
s
ys
tem.
F
igur
e
2(
a
)
i
ll
us
tr
a
tes
the
input
de
r
mos
c
opic
im
a
ge
s
a
nd
F
igur
e
2(
b)
dis
plays
the
media
n
f
il
ter
e
d
de
r
mos
c
opic
im
a
ge
s
us
ing
a
window
of
s
ize
(
2s
+
1,
2s
+
1)
f
il
ter
.
(
a
)
(
b)
F
igur
e
2.
P
r
e
p
r
oc
e
s
s
ing
r
e
s
ult
s
of
the
E
DL
C
S
s
ys
tem
is
(
a
)
i
nput
de
r
mos
c
opic
im
a
ge
a
nd
(
b
)
media
n
f
il
ter
e
d
im
a
ge
3.
2.
Color
d
o
m
ain
A
c
olor
s
pa
c
e
de
s
c
r
ibes
how
c
olor
s
c
a
n
b
e
r
e
pr
e
s
e
nted
us
ing
number
s
.
C
olor
s
pa
c
e
s
a
r
e
us
e
d
to
s
pe
c
if
y
a
nd
c
omm
unica
te
c
olor
inf
o
r
mation
in
va
r
ious
a
ppli
c
a
ti
ons
,
including
digi
tal
im
a
ging
,
p
r
int
ing,
a
nd
video
pr
oduc
ti
on.
T
he
opti
on
o
f
c
olo
r
s
pa
c
e
c
a
lcula
tes
on
the
de
tailed
a
ppli
c
a
ti
on
a
nd
the
type
of
c
olor
inf
or
mation
that
ne
e
ds
to
be
r
e
pr
e
s
e
nted.
T
he
r
e
a
r
e
s
e
ve
r
a
l
types
of
c
olor
s
pa
c
e
s
,
but
the
mos
t
c
om
mon
a
r
e
r
e
d,
g
r
e
e
n,
blue
(
R
GB
)
,
L
AB
a
nd
hue
,
s
a
tur
a
ti
on,
va
lue
(
HSV
)
.
−
R
e
d,
gr
e
e
n,
blue
(
R
GB
)
c
olor
s
pa
c
e
:
E
a
c
h
c
olor
c
ha
nne
l
in
the
R
GB
c
olou
r
s
pa
c
e
s
uc
h
a
s
r
e
d,
g
r
e
e
n,
a
nd
blue
li
ght
r
a
nging
f
r
om
0
to
255
.
B
y
a
gg
r
e
ga
ti
ng
va
r
ious
leve
ls
of
r
e
d,
gr
e
e
n
,
a
nd
blue
,
a
wide
r
a
nge
of
c
olor
s
c
a
n
be
r
e
pr
e
s
e
nted
[
21
]
.
−
L
AB
c
olor
s
pa
c
e
:
I
n
the
L
AB
c
olor
s
pa
c
e
,
the
L
p
a
r
a
mete
r
r
e
pr
e
s
e
nts
li
ghtnes
s
a
nd
r
a
nge
s
f
r
om
0
(
b
lac
k)
to
100
(
whi
te)
.
T
h
e
a
pa
r
a
mete
r
r
e
pr
e
s
e
nts
the
de
gr
e
e
of
r
e
dne
s
s
or
g
r
e
e
nne
s
s
,
with
pos
it
ive
v
a
lues
s
howing
r
e
d
a
nd
ne
ga
ti
ve
va
lues
s
howing
gr
e
e
n.
T
he
b
pa
r
a
mete
r
r
e
pr
e
s
e
nts
the
de
gr
e
e
of
bluene
s
s
or
ye
ll
owne
s
s
,
with
pos
it
ive
va
lues
s
howing
ye
ll
ow
with
ne
ga
ti
ve
va
lues
s
howing
blue.
T
he
a
a
nd
b
pa
r
a
mete
r
s
r
a
nge
f
r
om
-
128
to
127
.
M
or
e
inf
or
mation
c
a
n
be
f
ound
in
[
22]
.
−
Hue
,
s
a
tur
a
ti
on,
li
gh
tnes
s
/value
(
HSL
/HS
V
)
c
ol
or
s
pa
c
e
:
I
t
p
lays
c
olor
s
a
s
a
a
r
r
a
nge
ment
o
f
hue
,
s
a
tur
a
ti
on,
a
nd
li
ghtnes
s
or
va
lue.
L
ight
ne
s
s
or
va
lue
r
e
pr
e
s
e
nts
the
br
ight
ne
s
s
of
the
c
olor
,
s
a
tur
a
ti
on
indi
c
a
tes
the
pur
it
y
other
wis
e
the
c
olo
r
int
e
ns
it
y,
a
nd
h
ue
indi
c
a
tes
the
c
olo
r
i
ts
e
lf
.
M
or
e
in
f
or
mation
c
a
n
be
f
ound
in
[
23]
.
3.
3.
D
e
e
p
lear
n
in
g
a
r
c
h
it
e
c
t
u
r
e
DL
ha
s
be
e
n
highl
y
s
uc
c
e
s
s
f
ul
in
r
e
c
e
nt
ye
a
r
s
[
24]
,
[
25]
a
nd
many
of
the
mos
t
s
igni
f
ica
nt
br
e
a
kthr
oughs
in
a
r
ti
f
icia
l
int
e
ll
igenc
e
ha
ve
be
e
n
made
pos
s
ibl
e
by
a
dva
nc
e
s
in
DL
a
lgor
it
h
ms
a
nd
tec
hniques
.
DL
ha
s
a
gr
e
a
t
de
a
l
of
potential
to
t
r
a
ns
f
or
m
va
r
ious
a
s
pe
c
ts
of
he
a
lt
hc
a
r
e
that
include
s
i
mpr
ove
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
De
e
p
lear
ning
for
s
k
in
me
lanoma
c
las
s
if
ication
us
ing
.
.
.
(
Sank
ar
ak
utt
i
P
alani
c
hamy
M
anikandan
)
323
diagnos
ti
c
s
,
opti
mi
z
e
d
dr
ug
dis
c
ove
r
y
p
r
oc
e
s
s
,
a
nd
pe
r
s
ona
li
z
e
d
tr
e
a
tm
e
nt.
X
-
r
a
ys
,
magne
ti
c
r
e
s
ona
nc
e
im
a
ging,
a
nd
pa
thol
ogy
s
li
de
s
a
r
e
a
ll
e
xa
mpl
e
s
o
f
medic
a
l
im
a
ge
s
,
whic
h
a
r
e
s
uc
c
e
s
s
f
ull
y
a
na
lyze
d
by
DL
models
.
T
his
he
lps
in
the
e
a
r
ly
identif
ica
ti
on
a
nd
diagnos
is
of
dis
e
a
s
e
s
s
uc
h
a
s
c
a
nc
e
r
,
Alz
he
im
e
r
's
dis
e
a
s
e
,
a
nd
diabe
ti
c
r
e
ti
nopa
thy.
F
igur
e
3
s
hows
the
p
r
opo
s
e
d
s
tr
uc
tur
e
of
the
E
DL
C
S
f
o
r
s
kin
c
a
nc
e
r
diagno
s
is
.
F
igur
e
3.
P
r
opos
e
d
E
DL
C
S
de
s
ign
3.
3.
1.
Convol
u
t
ion
layer
I
t
is
a
p
r
im
a
r
y
bui
ldi
ng
block
in
C
NN
s
uti
li
z
e
d
f
or
im
a
ge
p
r
oc
e
s
s
ing
a
nd
c
omput
e
r
vis
ion
tas
ks
.
T
o
de
r
ive
f
e
a
tur
e
maps
f
r
om
a
n
input
im
a
ge
,
a
c
o
nvolut
ional
laye
r
e
mpl
oys
a
s
e
t
of
lea
r
na
ble
f
il
ter
s
on
the
input
,
a
nd
then
c
onvolves
the
two.
T
he
input
im
a
ge
is
pa
s
s
e
d
thr
ough
e
a
c
h
f
il
ter
,
whic
h
is
a
ti
ny
matr
ix
of
we
ight
s
that
pe
r
f
or
ms
a
dot
p
r
oduc
t
ope
r
a
ti
on
a
t
e
a
c
h
s
pa
ti
a
l
poin
t.
T
h
is
ope
r
a
ti
on
will
r
e
s
ult
in
the
pr
oduc
ti
on
o
f
a
f
e
a
tur
e
map
a
s
it
s
output
.
T
his
f
e
a
tur
e
map
will
r
e
f
lec
t
the
a
c
ti
va
ti
on
of
that
f
il
te
r
a
t
e
a
c
h
pos
it
ion
in
the
input
.
T
he
main
a
dva
ntage
s
of
c
o
nvolut
ional
laye
r
s
a
r
e
their
a
bil
it
y
to
e
xtr
a
c
t
loca
l
f
e
a
tur
e
s
a
nd
their
pa
r
a
mete
r
s
ha
r
ing
pr
ope
r
ty
.
B
y
lea
r
nin
g
s
ha
r
e
d
f
il
ter
s
,
a
c
onvolut
ional
laye
r
c
a
n
c
a
ptur
e
s
pa
ti
a
l
pa
tt
e
r
ns
a
c
r
os
s
the
e
nti
r
e
input
im
a
ge
,
making
it
we
ll
-
s
uit
e
d
f
or
tas
ks
li
ke
objec
t
de
tec
ti
on
a
nd
r
e
c
ognit
ion.
C
onvolut
ional
laye
r
s
typi
c
a
ll
y
include
s
e
ve
r
a
l
hy
pe
r
-
pa
r
a
mete
r
s
,
s
uc
h
a
s
the
numbe
r
o
f
f
il
ter
s
,
f
il
t
e
r
’
s
s
ize
,
the
s
tr
ide
of
the
c
onvolut
ion
ope
r
a
ti
on
,
a
nd
the
pa
dding
a
ppli
e
d
to
the
input
im
a
ge
.
T
he
s
e
hype
r
-
pa
r
a
mete
r
s
P
r
edi
ct
ed
C
l
as
s
L
a
ye
r
-
1
&2
(
3x3,
32)
M
a
x
P
ooli
ng
L
a
ye
r
(
2x2)
L
a
ye
r
-
3&4
(
3x3,
64)
M
a
x
P
ooli
ng
L
a
ye
r
(
2x2)
L
a
ye
r
-
5&6
(
3x3
,
128
)
M
a
x
P
ooli
ng
L
a
ye
r
(
2x2)
L
a
ye
r
-
7&8
(
3x3,
256)
M
a
x
P
ooli
ng
L
a
ye
r
(
2x2)
Dens
e
L
ayer
S
of
tM
a
x
L
a
ye
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
319
-
327
324
c
a
n
be
tuned
to
opt
im
ize
the
f
unc
ti
oning
o
f
the
m
ode
l
on
a
s
pe
c
if
ic
tas
k.
I
n
thi
s
mec
ha
nis
m,
the
c
o
nvolut
ion
laye
r
s
a
r
e
a
r
r
a
nge
d
in
a
n
e
f
f
icie
nt
manne
r
to
a
c
hie
ve
mor
e
a
c
c
ur
a
c
y
than
othe
r
a
r
c
hit
e
c
tur
e
s
.
3.
3.
2.
M
ax
p
oo
li
n
g
layer
I
t
r
e
duc
e
s
the
input
’
s
s
pa
ti
a
l
dim
e
ns
ions
by
taki
ng
the
maximum
va
lue
of
a
f
ixed
-
s
ize
window
(
us
ua
ll
y
2
×
2
or
3
×
3)
a
nd
s
li
ding
it
ove
r
the
input
f
e
a
tur
e
map.
Du
r
ing
max
pooli
ng,
the
input
is
divi
de
d
int
o
s
ub
-
r
e
gions
,
a
nd
e
a
c
h
s
ub
-
r
e
gion’
s
maximum
va
lue
is
take
n
to
pr
oduc
e
a
s
maller
output
f
e
a
tur
e
m
a
p.
T
his
ope
r
a
ti
on
he
lps
r
e
duc
ing
the
s
pa
ti
a
l
dim
e
ns
ions
,
while
r
e
taining
the
mos
t
s
igni
f
ica
nt
inf
o
r
mation
p
r
e
s
e
nt
in
the
input
.
M
a
x
pooli
ng
laye
r
s
ha
ve
s
e
ve
r
a
l
a
dva
ntage
s
,
including
the
a
bil
it
y
to
e
xtr
a
c
t
domi
na
nt
f
e
a
tu
r
e
s
f
r
om
the
input
f
e
a
tur
e
map
while
r
e
duc
ing
the
e
f
f
e
c
t
o
f
loca
l
va
r
iations
a
nd
nois
e
.
T
he
y
a
ls
o
he
lp
in
r
e
du
c
ing
the
c
omput
a
ti
ona
l
c
os
t
a
nd
the
number
of
pa
r
a
mete
r
s
r
e
quir
e
d
in
the
ne
twor
k,
by
down
-
s
a
mpl
ing
the
f
e
a
tur
e
maps
.
M
a
x
pooli
ng
laye
r
s
in
c
ombi
na
ti
on
wi
th
c
onvolut
ional
laye
r
s
c
a
n
be
us
e
d
to
e
xtr
a
c
t
hier
a
r
c
hica
l
r
e
pr
e
s
e
ntations
of
the
input
im
a
ge
.
B
y
a
lt
e
r
na
ti
n
g
be
twe
e
n
c
onvolut
ional
laye
r
s
a
nd
max
poo
li
n
g
laye
r
s
,
C
NN
s
c
a
n
lea
r
n
to
de
tec
t
f
e
a
tur
e
s
o
f
incr
e
a
s
ing
c
o
mpl
e
xit
y
in
the
input
im
a
ge
.
3.
3.
3
.
De
n
s
e
l
aye
r
A
de
ns
e
laye
r
is
a
type
o
f
ne
u
r
a
l
ne
twor
k
.
E
a
c
h
n
e
ur
on
in
the
f
e
e
df
or
wa
r
d
ne
twor
k
is
a
s
s
oc
iate
d
to
e
a
c
h
ne
ur
on
in
the
pr
e
vious
laye
r
.
E
a
c
h
ne
ur
on
c
omput
e
s
a
we
ight
e
d
s
um
of
the
pr
e
vious
laye
r
’
s
input
s
.
T
he
n,
a
n
a
c
ti
va
ti
on
f
unc
ti
on
is
e
mpl
oye
d
to
pr
odu
c
e
a
n
output
.
De
ns
e
laye
r
s
a
r
e
c
omm
only
us
e
d
f
or
va
r
ious
tas
ks
s
u
c
h
a
s
na
tur
a
l
langua
ge
pr
oc
e
s
s
ing
(
NL
P
)
,
im
a
ge
c
las
s
if
ica
ti
on
a
nd
the
pr
e
diction
of
ti
me
s
e
r
ies
.
T
he
y
a
r
e
pa
r
ti
c
ular
ly
us
e
f
ul
f
or
tas
ks
that
r
e
quir
e
lea
r
ni
ng
c
ompl
e
x
a
nd
nonli
ne
a
r
r
e
lations
hips
be
twe
e
n
i
nputs
a
nd
output
s
.
T
he
c
ompl
e
xit
y
of
model
de
pe
nds
on
the
number
of
ne
ur
ons
a
nd
it
s
a
bil
it
y
to
r
e
pr
e
s
e
nt
the
unde
r
lyi
ng
f
unc
ti
on.
T
he
a
c
ti
va
ti
on
f
unc
ti
on
he
l
ps
int
r
oduc
e
non
li
ne
a
r
it
y
in
the
model
a
nd
is
ty
pica
ll
y
a
r
e
c
ti
f
ied
li
ne
a
r
unit
(
R
e
L
U
)
o
r
s
igm
oid
.
De
ns
e
lay
e
r
s
c
a
n
be
us
e
d
a
s
ne
u
r
a
l
ne
twor
k
’
s
output
laye
r
,
whe
r
e
the
number
of
ne
ur
ons
in
the
laye
r
r
e
pr
e
s
e
nts
the
num
be
r
of
c
a
tegor
ies
in
a
c
a
tegor
iza
ti
on
is
s
ue
or
the
n
umber
of
output
s
.
3.
3.
4
.
S
o
f
t
M
ax
layer
A
S
of
tM
a
x
laye
r
is
a
c
omm
on
ly
us
e
d
laye
r
in
ne
ur
a
l
ne
twor
ks
,
pa
r
ti
c
ular
ly
in
the
c
ontext
of
c
las
s
if
ica
ti
on
pr
oblems
.
I
t
is
typi
c
a
ll
y
the
las
t
laye
r
a
nd
is
us
e
d
to
tr
a
ns
f
or
m
the
output
s
of
the
pr
e
vious
laye
r
int
o
a
p
r
oba
bil
it
y
dis
tr
ibut
ion
ove
r
the
dif
f
e
r
e
nt
c
l
a
s
s
e
s
.
T
he
S
o
f
tM
a
x
f
unc
ti
on
r
e
tur
ns
a
ve
c
tor
o
f
t
he
s
a
me
s
ize
a
s
the
input
ve
c
tor
(
r
e
a
l
number
s
)
,
whe
r
e
e
a
c
h
e
leme
nt
is
a
non
-
ne
ga
ti
ve
number
be
twe
e
n
0
a
n
d
1,
a
nd
the
s
um
is
1.
T
o
do
thi
s
,
e
xpone
nti
a
te
e
ve
r
y
e
leme
nt
in
the
input
ve
c
tor
,
a
nd
then
divi
de
the
output
ve
c
tor
's
e
leme
nts
by
the
tot
a
l
o
f
a
ll
the
e
xpone
nti
a
ted
va
l
ue
s
to
nor
malize
it
.
I
t
is
c
omm
on
pr
a
c
ti
c
e
to
c
om
bine
the
S
of
tM
a
x
laye
r
with
a
los
s
f
unc
ti
on
s
uc
h
c
r
os
s
e
ntr
opy
los
s
.
T
he
dif
f
e
r
e
nc
e
be
twe
e
n
the
a
nti
c
ipate
d
a
nd
a
c
tual
pr
oba
bil
it
y
d
is
tr
ibut
ions
—
that
is
,
the
one
-
hot
e
nc
oding
of
the
tr
ue
c
las
s
labe
l
—
is
c
a
lcula
ted
us
ing
the
c
r
os
s
e
ntr
opy
los
s
.
Ove
r
a
ll
,
the
S
of
tM
a
x
laye
r
p
lays
a
c
r
uc
ial
r
ole
in
c
onve
r
t
ing
the
output
s
o
f
a
ne
ur
a
l
ne
twor
k
int
o
a
pr
oba
bil
it
y
dis
tr
ibut
ion
ove
r
the
c
las
s
e
s
,
whic
h
c
a
n
then
be
us
e
d
to
make
pr
e
dictions
.
4.
RE
S
UL
T
AN
D
DI
S
CU
S
S
I
ON
T
he
P
H2
da
taba
s
e
[
26]
,
[
27
]
is
a
publi
c
ly
a
va
il
a
b
le
da
taba
s
e
of
de
r
mos
c
opic
im
a
ge
s
of
pigm
e
nted
s
kin
les
ions
,
whic
h
is
de
s
igned
to
a
id
in
the
diagno
s
is
of
mela
noma.
T
he
P
H2
da
taba
s
e
wa
s
c
r
e
a
ted
b
y
a
tea
m
of
r
e
s
e
a
r
c
he
r
s
f
r
om
the
Unive
r
s
it
y
of
P
or
to
in
P
o
r
tugal
a
nd
c
ontains
a
tot
a
l
of
200
im
a
ge
s
of
s
kin
les
ions
.
T
his
da
taba
s
e
is
notable
f
or
it
s
high
-
qua
li
ty
de
r
mo
s
c
opic
im
a
ge
s
,
whic
h
a
r
e
c
a
ptur
e
d
us
ing
a
high
-
r
e
s
olut
ion
c
a
mer
a
a
nd
a
de
r
matos
c
ope
.
T
he
im
a
ge
s
a
r
e
c
a
ptur
e
d
unde
r
s
tanda
r
dize
d
c
ondit
ions
,
with
c
ons
is
tent
li
ghti
ng
a
nd
c
a
mer
a
s
e
tt
ings
,
to
e
ns
ur
e
that
the
im
a
ge
s
a
r
e
of
high
qua
li
ty
a
nd
c
ompar
a
ble
to
e
a
c
h
other
.
E
a
c
h
im
a
ge
in
the
da
taba
s
e
is
a
c
c
ompanie
d
by
a
s
e
t
o
f
gr
ound
tr
u
th
a
nnotations
,
including
the
diagnos
is
,
the
type
of
les
ion,
a
nd
the
loca
ti
on
o
f
th
e
les
ion
on
the
body
[
28]
,
[
29]
.
T
he
P
H2
da
taba
s
e
ha
s
be
e
n
us
e
d
in
s
e
ve
r
a
l
r
e
s
e
a
r
c
h
s
tudi
e
s
to
de
v
e
lop
a
nd
e
va
luate
the
de
ve
loped
a
lgor
it
hms
.
F
igur
e
4
s
hows
the
s
a
mpl
e
de
r
mos
c
opic
im
a
ge
s
in
the
da
taba
s
e
.
F
igur
e
4
(
a
)
s
hows
the
nor
mal
im
a
ge
s
f
r
om
the
P
H
2
da
taba
s
e
,
a
nd
the
a
typi
c
a
l
ne
vus
a
nd
mela
noma
im
a
ge
s
a
r
e
s
hown
in
F
igur
e
s
4(
b
)
a
nd
4(
c
)
c
or
r
e
s
pondingl
y.
Hype
r
-
pa
r
a
mete
r
tuni
ng
is
a
n
im
por
tant
pa
r
t
o
f
t
he
mac
hine
lea
r
ning
wor
kf
low,
a
s
the
c
hoice
of
hype
r
-
pa
r
a
mete
r
s
c
a
n
gr
e
a
tl
y
a
f
f
e
c
t
the
pe
r
f
or
man
c
e
of
the
model.
T
he
y
a
r
e
s
e
t
pr
ior
to
the
tr
a
ini
ng
pr
oc
e
s
s
,
a
nd
a
ls
o
gove
r
n
the
be
ha
vior
of
the
tr
a
ini
ng
pr
oc
e
s
s
of
a
mac
hine
lea
r
ning
model.
T
he
s
e
pa
r
a
mete
r
s
c
a
nnot
be
lea
r
ne
d
f
r
om
the
da
ta
a
nd
mus
t
be
s
e
t
manua
ll
y
or
us
ing
s
ome
a
utom
a
ted
s
e
a
r
c
h
a
lgor
it
hm
T
a
ble
1
s
hows
the
hype
r
-
pa
r
a
mete
r
s
us
e
d
in
thi
s
wor
k
.
T
he
pe
r
f
or
manc
e
of
the
p
r
opos
e
d
E
DL
C
S
de
s
ign
is
a
na
ly
z
e
d
us
ing
thr
e
e
mea
s
ur
e
ments
:
f
or
e
xa
mpl
e
,
a
c
c
ur
a
c
y,
s
e
ns
it
ivi
ty
a
nd
s
pe
c
if
icity.
T
a
ble
2
s
hows
the
obtaine
d
pe
r
f
or
manc
e
metr
ics
f
or
d
if
f
e
r
e
nt
c
olor
models
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
De
e
p
lear
ning
for
s
k
in
me
lanoma
c
las
s
if
ication
us
ing
.
.
.
(
Sank
ar
ak
utt
i
P
alani
c
hamy
M
anikandan
)
325
(
a
)
(
b)
(
c
)
F
igur
e
4.
Da
taba
s
e
im
a
ge
s
(
a
)
n
or
mal
,
(
b)
Atypica
l
Ne
vus
,
a
nd
(
c
)
M
a
li
gna
nt
or
M
e
lanoma
T
a
ble
1.
Hype
r
pa
r
a
mete
r
s
us
e
d
in
the
p
r
opos
e
d
E
DL
C
S
de
s
ign
H
ype
r
pa
r
a
me
te
r
s
V
a
lu
e
s
C
onvolut
io
n
l
a
ye
r
f
il
te
r
s
iz
e
3
×
3
N
umbe
r
of
c
onvolut
io
n f
il
te
r
s
32, 64, 128
a
nd 256
S
tr
id
e
i
n c
onvolut
io
n
1
A
c
ti
va
ti
on f
unc
ti
on (
in
put
)
R
e
L
U
M
a
x pooli
ng l
a
ye
r
f
il
te
r
s
iz
e
2
×
2
S
tr
id
e
i
n ma
x pooli
ng l
a
ye
r
2
A
c
ti
va
ti
on f
unc
ti
on (
out
put
)
s
of
tm
a
x
L
os
s
f
unc
ti
on
C
r
os
s
e
nt
r
opy los
s
O
pt
im
iz
a
ti
on
M
in
i
-
ba
tc
h gr
a
di
e
nt
de
s
c
e
nt
a
lg
or
it
hm
T
a
ble
2.
P
e
r
f
o
r
manc
e
of
the
pr
opos
e
d
E
DL
C
S
de
s
ign
unde
r
di
f
f
e
r
e
nt
c
olor
models
C
ondi
ti
on of
i
nput
de
r
mos
c
opi
c
i
ma
ge
s
R
G
B
L
A
B
H
S
V
A
c
c
ur
a
c
y (
%
)
N
o pr
e
pr
oc
e
s
s
in
g
96.67
98.54
97.29
P
r
e
pr
oc
e
s
s
in
g by me
di
a
n f
il
te
r
99.17
99.58
99.17
S
e
ns
it
iv
it
y (
%
)
N
o pr
e
pr
oc
e
s
s
in
g
97.57
98.96
98.26
P
r
e
pr
oc
e
s
s
in
g by me
di
a
n f
il
te
r
99.31
99.65
99.31
S
pe
c
if
ic
it
y (
%
)
N
o pr
e
pr
oc
e
s
s
in
g
95.31
97.92
95.83
P
r
e
pr
oc
e
s
s
in
g by me
di
a
n f
il
te
r
98.96
99.48
98.96
I
t
is
ob
s
e
r
v
e
d
f
r
om
T
a
bl
e
2
th
a
t
the
L
A
B
c
o
lor
model
pr
o
vid
e
s
b
e
tt
e
r
r
e
s
u
lt
s
t
ha
n
o
ther
tw
o
-
c
ol
o
r
model
s
u
s
e
d
in
th
is
s
tud
y
f
or
the
c
l
a
s
s
if
i
c
a
ti
on
of
de
r
mo
s
c
o
pic
im
a
ge
s
.
T
his
i
s
du
e
to
the
f
a
c
t
tha
t
the
L
A
B
c
olor
s
pa
c
e
de
f
in
e
s
c
ol
or
dif
f
e
r
e
nc
e
s
,
b
uil
di
ng
it
f
unc
ti
o
na
l
f
or
c
ol
or
a
n
a
ly
s
i
s
.
I
t
is
a
thr
e
e
-
dim
e
n
s
io
na
l
c
olor
s
pa
c
e
w
hich
in
dic
a
te
s
th
e
c
olor
s
e
s
ta
bli
s
he
d
on
th
e
thr
e
e
p
a
r
a
m
e
ter
s
of
li
ghtn
e
s
s
,
a
(
r
e
d
-
gr
e
e
n)
,
a
nd
b
(
blue
-
ye
ll
o
w)
.
5.
CONC
L
USI
ON
S
kin
c
a
nc
e
r
de
ve
lops
us
ua
ll
y
by
e
xpos
ur
e
to
UV
r
a
diation
f
r
om
the
s
un
or
lamps
or
tanning
be
ds
.
T
he
unc
ontr
oll
e
d
g
r
owth
of
da
mage
d
s
kin
c
e
ll
s
l
e
a
ds
to
the
f
or
mation
o
f
s
kin
c
a
nc
e
r
.
T
he
e
a
r
ly
de
tec
ti
on
he
lps
to
r
e
duc
e
the
r
is
k
of
c
ompl
ica
ti
ons
a
nd
be
tt
e
r
tr
e
a
tm
e
nt.
I
n
thi
s
pa
pe
r
,
E
DL
C
S
is
de
ve
loped
f
or
s
kin
c
a
nc
e
r
diagnos
is
.
I
t
us
e
s
thr
e
e
c
olor
s
p
a
c
e
model
s
s
uc
h
a
s
,
R
GB
,
L
AB
a
nd
HSV
a
r
e
e
mpl
oye
d,
a
nd
their
indi
vidual
c
olor
c
omponents
a
r
e
f
e
d
to
the
pr
opos
e
d
E
DL
C
S
.
B
e
f
or
e
f
e
d
to
the
s
ys
tem,
the
input
im
a
ge
s
a
r
e
f
il
ter
e
d
to
r
e
move
the
nois
e
s
a
nd
ha
ir
s
by
media
n
f
il
ter
a
nd
then
they
a
r
e
r
e
pr
e
s
e
nted
by
dif
f
e
r
e
nt
c
olor
models
.
T
he
f
inal
pr
e
diction
r
e
s
ult
s
f
r
o
m
the
E
D
L
C
S
is
e
it
he
r
nor
mal
o
r
a
bnor
mal
R
e
s
ult
s
s
how
that
the
pr
opos
e
d
E
DC
L
S
pr
ovides
pr
omi
s
ing
r
e
s
ult
s
f
or
a
ll
c
olor
models
a
nd
the
L
AB
c
olor
model
pr
ovid
e
s
be
tt
e
r
r
e
s
ult
s
than
other
models
with
a
n
ove
r
a
ll
a
c
c
ur
a
c
y
of
99.
58
%
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
15
,
No.
1
,
F
e
br
ua
r
y
20
25
:
319
-
327
326
RE
F
E
RE
NC
E
S
[
1]
R
.
R
a
me
r
e
t
al
.
,
“
I
mpa
c
t
of
c
a
nna
bi
noi
d
c
ompounds
o
n
s
ki
n
c
a
nc
e
r
,”
C
an
c
e
r
s
,
vol
.
14,
no.
7,
M
a
r
.
2022,
doi
:
10.3390/c
a
nc
e
r
s
14071769.
[
2]
P
.
M
a
th
ur
e
t
al
.
,
“
I
C
M
R
-
N
C
D
I
R
-
N
C
R
P
in
ve
s
ti
ga
to
r
gr
ou
p.
c
a
nc
e
r
s
ta
ti
s
ti
c
s
,
2020:
r
e
por
t
f
r
om
na
ti
ona
l
c
a
nc
e
r
r
e
g
is
tr
y
pr
ogr
a
mm
e
, I
ndi
a
,”
J
C
O
G
lo
b O
nc
ol
, vol
. 6, no. 6, pp. 1063
–
1075, 2020.
[
3]
M
.
U
c
kune
r
a
nd
H
.
E
r
ol
,
“
A
ne
w
de
e
p
le
a
r
ni
ng
mode
l
f
or
s
ki
n
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on,”
in
2021
6t
h
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
Sc
ie
n
c
e
and E
ngi
ne
e
r
in
g (
U
B
M
K
)
, S
e
p. 2021, pp. 27
–
31, doi:
10.1109/UB
M
K
52708.2021.9558936.
[
4]
A
.
D
e
mi
r
,
F
.
Y
il
ma
z
,
a
nd
O
.
K
os
e
,
“
E
a
r
ly
de
te
c
ti
on
of
s
ki
n
c
a
nc
e
r
us
in
g
de
e
p
le
a
r
ni
ng
a
r
c
hi
te
c
tu
r
e
s
:
r
e
s
ne
t
-
101
a
nd
in
c
e
pt
i
on
-
v3,”
i
n
2019 M
e
di
c
al
T
e
c
hnol
ogi
e
s
C
ong
r
e
s
s
(
T
I
P
T
E
K
N
O
)
, O
c
t.
2019, pp. 1
–
4,
doi
:
10.1109/T
I
P
T
E
K
N
O
47231.2019.8972045.
[
5]
N
.
A
bur
a
e
d,
A
.
P
a
nt
ha
kka
n,
M
.
A
l
-
S
a
a
d,
S
.
A
.
A
mi
n,
a
nd
W
.
M
a
ns
oor
,
“
D
e
e
p
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
D
C
N
N
)
f
or
s
ki
n
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
on,”
in
2020
27t
h
I
E
E
E
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
E
le
c
tr
oni
c
s
,
C
ir
c
ui
ts
and
Sy
s
te
m
s
(
I
C
E
C
S)
,
N
ov.
20
20,
pp. 1
–
4, doi:
10.1109/I
C
E
C
S
49266.2020.9294814.
[
6]
H
.
K
.
K
onda
ve
e
ti
a
nd
P
.
E
dupuga
nt
i,
“
S
ki
n
c
a
nc
e
r
c
la
s
s
if
ic
a
ti
o
n
us
in
g
tr
a
ns
f
e
r
le
a
r
ni
ng,”
in
2020
I
E
E
E
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
A
dv
e
nt
T
r
e
nds
in
M
ul
ti
di
s
c
ip
li
nar
y
R
e
s
e
a
r
c
h
a
nd
I
nnov
at
io
n
(
I
C
A
T
M
R
I
)
,
D
e
c
.
2020,
pp.
1
–
4,
doi
:
10.1109/I
C
A
T
M
R
I
51801.2020.9398388.
[
7]
H
.
Y
ouni
s
,
M
.
H
.
B
ha
tt
i,
a
nd M
.
A
z
e
e
m,
“
C
la
s
s
if
ic
a
ti
on
of
s
ki
n
c
a
nc
e
r
de
r
mos
c
opy
im
a
ge
s
u
s
in
g
tr
a
ns
f
e
r
le
a
r
ni
ng,”
in
2019
15t
h
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on E
m
e
r
gi
ng T
e
c
hnol
ogi
e
s
(
I
C
E
T
)
, D
e
c
. 2019, pp. 1
–
4,
doi
:
10.1109/I
C
E
T
48972.2019.8994508.
[
8]
Z
. R
a
hma
n a
nd A
. M
. A
mi
, “
A
t
r
a
n
s
f
e
r
le
a
r
ni
ng
-
ba
s
e
d
a
ppr
oa
c
h f
or
s
ki
n l
e
s
io
n c
la
s
s
if
ic
a
ti
on f
r
om i
mba
la
nc
e
d da
ta
,
”
i
n
2020 11th
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
E
le
c
t
r
ic
al
and
C
om
put
e
r
E
ngi
ne
e
r
in
g
(
I
C
E
C
E
)
,
D
e
c
.
2020,
pp.
65
–
68,
doi
:
10.1109/I
C
E
C
E
51571.2020.9393155.
[
9]
C
.
J
e
n
N
ge
h,
C
.
M
a
,
T
.
K
ua
n
-
W
e
i
H
o,
Y
.
W
a
ng,
a
nd
J
.
R
a
i
ti
,
“
D
e
e
p
le
a
r
ni
ng
on
e
dge
de
vi
c
e
f
or
e
a
r
ly
pr
e
s
c
r
e
e
ni
ng
of
s
ki
n
c
a
nc
e
r
s
in
r
ur
a
l
c
omm
uni
ti
e
s
,”
in
2020
I
E
E
E
G
lo
bal
H
um
ani
ta
r
ia
n
T
e
c
hnol
ogy
C
onf
e
r
e
nc
e
(
G
H
T
C
)
,
O
c
t.
2020,
pp.
1
–
4,
doi
:
10.1109/G
H
T
C
46280.2020.9342911.
[
10]
A
.
G
.
C
.
P
a
c
he
c
o
a
nd
R
.
A
.
K
r
ohl
in
g,
“
A
n
a
tt
e
nt
io
n
-
ba
s
e
d
m
e
c
ha
ni
s
m
to
c
ombi
ne
im
a
g
e
s
a
nd
me
ta
da
ta
in
de
e
p
le
a
r
ni
ng
mode
ls
a
ppl
ie
d
to
s
ki
n
c
a
n
c
e
r
c
la
s
s
if
ic
a
ti
on,”
I
E
E
E
J
our
nal
of
B
io
m
e
di
c
al
and
H
e
al
th
I
nf
or
m
at
ic
s
,
vol
.
25,
no.
9,
pp.
3554
–
3563,
S
e
p.
2021, doi:
10.1109/J
B
H
I
.2021.3062002.
[
11]
A
.
I
mr
a
n,
A
.
N
a
s
ir
,
M
.
B
il
a
l,
G
.
S
un,
A
.
A
lz
a
hr
a
ni
,
a
nd
A
.
A
lm
uha
im
e
e
d,
“
S
ki
n
c
a
nc
e
r
de
te
c
ti
on
us
in
g
c
ombi
ne
d
de
c
is
io
n
of
de
e
p l
e
a
r
ne
r
s
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 10, pp. 118198
–
118212, 202
2, doi:
10.1109/AC
C
E
S
S
.2022.3220329.
[
12]
Y
.
G
u,
Z
.
G
e
,
C
.
P
.
B
onni
ngt
on,
a
nd
J
.
Z
hou,
“
P
r
ogr
e
s
s
iv
e
tr
a
ns
f
e
r
le
a
r
ni
ng
a
nd
a
dve
r
s
a
r
ia
l
doma
in
a
da
pt
a
ti
on
f
or
c
r
os
s
-
do
ma
in
s
ki
n
di
s
e
a
s
e
c
la
s
s
if
ic
a
ti
on,”
I
E
E
E
J
our
nal
of
B
io
m
e
di
c
al
and
H
e
al
th
I
nf
or
m
at
ic
s
,
vol
.
24,
no.
5,
pp.
1379
–
1393,
M
a
y
2020,
doi
:
10.1109/J
B
H
I
.2019.2942429.
[
13]
J
.
R
.
H
a
ge
r
ty
e
t
al
.
,
“
D
e
e
p
le
a
r
ni
ng
a
nd
ha
ndc
r
a
f
te
d
me
th
od
f
us
io
n:
hi
ghe
r
di
a
gnos
ti
c
a
c
c
ur
a
c
y
f
or
me
la
noma
de
r
mos
c
opy
im
a
ge
s
,”
I
E
E
E
J
our
nal
of
B
io
m
e
di
c
al
and
H
e
al
th
I
nf
or
m
at
ic
s
,
vol
.
23,
no.
4,
pp.
1385
–
1391,
J
ul
.
2019,
doi
:
10.1109/J
B
H
I
.2019.2891049.
[
14]
M
.
G
oya
l,
A
.
O
a
kl
e
y,
P
.
B
a
ns
a
l,
D
.
D
a
nc
e
y,
a
nd
M
.
H
.
Y
a
p,
“
S
ki
n
le
s
io
n
s
e
gme
nt
a
ti
on
in
de
r
mos
c
opi
c
im
a
ge
s
w
it
h
e
n
s
e
m
bl
e
de
e
p l
e
a
r
ni
ng me
th
ods
,”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 4171
–
4181, 2
020, doi:
10.1109/AC
C
E
S
S
.2019.2960504.
[
15]
G
.
Z
ha
ng
e
t
al
.
,
“
D
S
M
:
a
de
e
p
s
upe
r
vi
s
e
d
mul
ti
-
s
c
a
le
ne
twor
k
le
a
r
ni
ng
f
o
r
s
ki
n
c
a
nc
e
r
s
e
gme
nt
a
ti
on,”
I
E
E
E
A
c
c
e
s
s
,
vo
l.
7,
pp. 140936
–
140945, 2019, doi:
10.1109/AC
C
E
S
S
.2019.29436
28.
[
16]
L
.
W
e
i,
K
.
D
in
g,
a
nd
H
.
H
u,
“
A
ut
oma
ti
c
s
ki
n
c
a
nc
e
r
de
te
c
ti
on
in
de
r
mos
c
opy
im
a
ge
s
ba
s
e
d
on
e
ns
e
mbl
e
li
ght
w
e
ig
ht
de
e
p
le
a
r
ni
ng ne
twor
k,”
I
E
E
E
A
c
c
e
s
s
, vol
. 8, pp. 99633
–
99647, 202
0, doi:
10.1109/AC
C
E
S
S
.2020.2997710.
[
17]
K
.
P
a
dma
na
ba
n,
A
.
M
.
S
e
nt
hi
l
K
uma
r
,
H
.
A
z
a
th
,
A
.
K
.
V
e
l
mur
uga
n,
a
nd
M
.
S
ubbi
a
h,
“
H
ybr
id
da
ta
mi
ni
ng
te
c
hni
que
b
a
s
e
d
br
e
a
s
t
c
a
nc
e
r
pr
e
di
c
ti
on,”
i
n
A
I
P
C
onf
e
r
e
nc
e
P
r
oc
e
e
di
ng
s
, 202
3, vol. 2523, doi
:
10.1063/5.0110216.
[
18]
M
.
S
e
nt
hi
l
K
uma
r
,
H
.
A
z
a
th
,
A
.
K
.
V
e
lm
ur
uga
n,
K
.
P
a
dma
n
a
ba
n,
a
nd
M
.
S
ubbi
a
h,
“
P
r
e
di
c
ti
on
of
A
lz
he
im
e
r
’
s
di
s
e
a
s
e
us
in
g
hybr
id
ma
c
hi
ne
l
e
a
r
ni
ng t
e
c
hni
que
,”
i
n
A
I
P
C
onf
e
r
e
nc
e
P
r
oc
e
e
di
ngs
, 2023, vol. 2523, do
i:
10.1063/5.0110283
.
[
19]
A
. D
e
e
pa
, R
. L
a
th
a
, T
. S
. K
uma
r
, N
. K
. M
a
ni
ka
nd
a
n, J
. P
r
e
e
th
a
, a
nd S
. M
ur
uga
n, “
I
oT
-
ba
s
e
d w
e
a
r
a
bl
e
de
vi
c
e
s
f
or
pe
r
s
ona
l
s
a
f
e
ty
a
nd a
c
c
id
e
nt
pr
e
ve
nt
io
n s
ys
te
ms
,”
i
n
2023 2nd I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on Smar
t
T
e
c
hnol
ogi
e
s
f
or
Sm
ar
t
N
at
io
n, Smar
tT
e
c
h
C
on
2023
, A
ug. 2023, pp. 1510
–
1514, doi:
10.1109/S
ma
r
tT
e
c
hC
on
57526.2023.10391691.
[
20]
S
.
P
.
M
a
ni
r
a
j
a
nd
P
.
S
a
r
da
r
ma
r
a
n,
“
C
la
s
s
if
ic
a
ti
on
of
de
r
mos
c
opi
c
im
a
ge
s
us
in
g
s
of
t
c
omput
in
g
te
c
hni
que
s
,”
N
e
ur
al
C
om
pu
ti
ng
and A
ppl
ic
at
io
ns
, vol
. 33, no. 19, pp. 13015
–
13026, 2021, doi:
10.1007/s
00521
-
021
-
05998
-
5.
[
21]
M
.
T
k
a
lc
ic
a
nd
J
.
F
.
T
a
s
ic
,
“
C
ol
our
s
p
a
c
e
s
:
pe
r
c
e
pt
ua
l,
hi
s
to
r
ic
a
l
a
nd
a
ppl
ic
a
ti
ona
l
ba
c
kgr
ound,”
in
T
he
I
E
E
E
R
e
gi
on
8
E
U
R
O
C
O
N
2003. C
om
put
e
r
a
s
a T
ool
.
, vol
. 1, pp. 304
–
308, doi:
10.1109/E
U
R
C
O
N
.2003.1248032.
[
22]
L
.
G
e
or
gi
e
va
,
T
.
D
im
it
r
ova
,
a
nd
N
.
A
nge
lo
v,
“
R
G
B
a
nd
H
S
V
c
ol
our
mode
ls
in
c
ol
our
id
e
nt
if
ic
a
ti
on
of
di
gi
ta
l
tr
a
uma
s
im
a
g
e
s
,
”
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on C
om
put
e
r
Sy
s
te
m
s
and T
e
c
hnol
ogi
e
s
, pp. 1
–
6, 2005.
[
23]
A
.
S
.
A
.
H
a
ns
a
nd
S
.
R
a
o,
“
A
C
N
N
-
L
S
T
M
ba
s
e
d
de
e
p
n
e
ur
a
l
ne
twor
ks
f
or
f
a
c
ia
l
e
mot
io
n
de
te
c
ti
on
in
vi
de
o
s
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
A
dv
anc
e
s
i
n Si
gnal
and I
m
age
Sc
ie
n
c
e
s
, vol
. 7, no.
1
, pp. 11
–
20, M
a
r
. 2021,
doi
:
10.29284/I
J
A
S
I
S
.7.1.2021.11
-
20.
[
24]
A
.
H
us
s
a
in
de
e
n,
S
.
I
qba
l,
a
nd
T
.
D
.
A
mbe
goda
,
“
M
ul
ti
-
la
b
e
l
pr
ot
ot
ype
ba
s
e
d
in
te
r
pr
e
ta
bl
e
ma
c
hi
ne
le
a
r
ni
ng
f
or
me
la
noma
de
te
c
ti
on,”
I
nt
e
r
nat
io
nal
J
ou
r
nal
of
A
dv
anc
e
s
in
Si
gnal
a
nd
I
m
age
Sc
ie
nc
e
s
,
vol
.
8,
no.
1,
pp.
40
–
53,
J
a
n.
2022,
doi
:
10.29284/I
J
A
S
I
S
.8.1.2022.40
-
53.
[
25]
T
.
M
e
ndonc
a
,
P
.
M
.
F
e
r
r
e
ir
a
,
J
.
S
.
M
a
r
que
s
,
A
.
R
.
S
.
M
a
r
c
a
l,
a
nd
J
.
R
oz
e
ir
a
,
“
P
H
2
-
a
de
r
mos
c
opi
c
im
a
ge
d
a
ta
ba
s
e
f
or
r
e
s
e
a
r
c
h
a
nd
be
nc
hma
r
ki
ng,”
in
2013
35t
h
A
nnual
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
of
th
e
I
E
E
E
E
ngi
ne
e
r
in
g
in
M
e
di
c
in
e
and
B
io
lo
gy
Soc
ie
ty
(
E
M
B
C
)
, J
ul
. 2013, pp. 5437
–
5440, doi:
10.1109/E
M
B
C
.2013.6610779.
[
26]
P
.
M
.
F
e
r
r
e
i
r
a
,
“
P
H
2
D
a
ta
ba
s
e
,
”
A
D
D
I
P
r
oj
e
c
t
,
2012,
A
c
c
e
s
s
e
d:
N
ov.
20,
2022.
[
O
nl
in
e
]
.
A
va
i
la
bl
e
:
ht
tp
s
:/
/ww
w
.f
c
.up.pt/
a
ddi
/p
h2%
20da
ta
ba
s
e
.ht
ml
[
27]
S
.
J
us
ti
n
a
nd
M
.
P
a
tt
na
ik
,
“
S
ki
n
le
s
io
n
s
e
gme
nt
a
ti
on
by
pi
xe
l
by
pi
xe
l
a
ppr
oa
c
h
us
in
g
d
e
e
p
le
a
r
ni
ng,”
I
nt
e
r
nat
io
nal
J
ou
r
na
l
of
A
dv
anc
e
s
i
n Si
gnal
and I
m
age
Sc
ie
nc
e
s
, vol
. 6, no. 1, J
un. 2020
, doi
:
10.29284/I
J
A
S
I
S
.6.1.2020.12
-
20.
[
28]
S
.
S
e
lv
a
r
a
s
u,
K
.
B
a
s
hka
r
a
n,
K
.
R
a
dhi
k
a
,
S
.
V
a
la
r
ma
th
y,
a
nd
S
.
M
ur
uga
n,
“
I
oT
-
e
na
bl
e
d
me
di
c
a
ti
on
s
a
f
e
ty
:
r
e
a
l
-
ti
me
te
mpe
r
a
tu
r
e
a
nd
s
to
r
a
ge
moni
to
r
in
g
f
or
e
nha
nc
e
d
m
e
di
c
a
ti
on
qua
li
ty
in
h
os
pi
ta
ls
,”
in
2023
2nd
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
A
ut
om
at
i
on,
C
om
put
in
g and R
e
ne
w
abl
e
Sy
s
t
e
m
s
(
I
C
A
C
R
S)
, D
e
c
. 2023, pp. 256
–
261, doi:
10.1109/I
C
A
C
R
S
58579.2023.10405212.
[
29]
G
.
T
ha
hni
ya
th
e
t
al
.
,
“
C
lo
ud
ba
s
e
d
pr
e
di
c
ti
on
of
e
pi
le
pt
ic
s
e
iz
u
r
e
s
us
in
g
r
e
a
l
-
ti
me
e
le
c
tr
oe
nc
e
ph
a
lo
gr
a
ms
a
na
ly
s
is
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
E
le
c
tr
ic
al
and
C
om
put
e
r
E
ngi
ne
e
r
in
g
,
vol
.
14,
no.
5,
pp.
6047
–
6056,
O
c
t.
2024,
doi
:
10.11591
/i
je
c
e
.v14i5.pp6047
-
6056.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
De
e
p
lear
ning
for
s
k
in
me
lanoma
c
las
s
if
ication
us
ing
.
.
.
(
Sank
ar
ak
utt
i
P
alani
c
hamy
M
anikandan
)
327
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Sa
nka
ra
kutt
i
P
a
l
a
n
i
cha
m
y
M
a
n
i
ka
nd
a
n
h
a
s
o
b
t
a
i
n
e
d
h
i
s
B.
E
d
e
g
ree
i
n
el
ect
r
o
n
i
cs
an
d
co
mm
u
n
i
cat
i
o
n
en
g
i
n
eer
i
n
g
fro
m
A
n
n
a
U
n
i
v
er
s
i
t
y
.
A
ft
er
acq
u
i
r
i
n
g
h
i
s
B.
E
.,
h
e
co
mp
l
et
e
d
h
i
s
M.
E
.
d
eg
ree
i
n
emb
ed
d
ed
s
y
s
t
ems
fr
o
m
A
n
n
a
U
n
i
v
ers
i
t
y
an
d
c
u
rren
t
l
y
h
e
i
s
p
u
r
s
u
i
n
g
h
i
s
Ph
.
D
.
at
D
r.
MG
R
U
n
i
v
er
s
i
t
y
.
H
e
h
as
s
p
eci
al
i
zed
i
n
s
u
b
j
ec
t
s
l
i
k
e
d
i
g
i
t
al
el
ect
r
o
n
i
cs
,
ci
rcu
i
t
an
d
d
e
v
i
ce
s
,
an
d
l
i
n
ear
i
n
t
e
g
rat
e
d
ci
rcu
i
t
s
.
H
e
h
as
at
t
e
n
d
e
d
s
ev
era
l
N
PT
E
L
an
d
Sw
a
y
am
co
u
rs
e
s
an
d
h
a
s
t
eac
h
i
n
g
e
x
p
eri
en
ce
o
f
o
v
er
1
5
y
ear
s
.
H
e
h
a
s
p
u
b
l
i
s
h
ed
5
Pat
e
n
t
s
,
2
b
o
o
k
s
a
n
d
h
as
a
t
t
e
n
d
e
d
s
ev
eral
N
at
i
o
n
al
a
n
d
In
t
ern
a
t
i
o
n
a
l
Co
n
fere
n
ces
a
n
d
p
u
b
l
i
s
h
e
d
p
a
p
ers
i
n
Sco
p
u
s
J
o
u
r
n
al
s
.
H
e
h
a
s
g
u
i
d
ed
man
y
n
u
mb
er
s
o
f
u
n
d
erg
ra
d
u
a
t
e
an
d
p
o
s
t
g
rad
u
at
e
s
t
u
d
en
t
s
f
o
r
t
h
e
i
r
acad
e
mi
c
p
r
o
j
ec
t
s
.
H
e
ca
n
b
e
co
n
t
ac
t
ed
a
t
emai
l
:
d
rman
i
k
a
n
d
a
n
s
p
@
g
mai
l
.
co
m
an
d
man
i
k
a
n
d
a
n
s
p
1
7
@
g
ma
i
l
.
co
m.
Sa
ndeep
R
eddy
N
a
ra
n
i
an
es
t
eemed
p
r
o
fes
s
i
o
n
a
l
w
i
t
h
a
mas
t
er’
s
d
eg
ree
i
n
co
mp
u
t
er
s
c
i
en
ce
fro
m
So
u
t
h
ern
I
l
l
i
n
o
i
s
U
n
i
v
ers
i
t
y
Carb
o
n
d
al
e,
I
l
l
i
n
o
i
s
,
U
SA
s
h
i
n
e
s
b
ri
g
h
t
l
y
w
i
t
h
a
d
i
s
t
i
n
g
u
i
s
h
ed
career
s
p
an
n
i
n
g
o
v
er
1
3
y
ear
s
i
n
t
h
e
d
y
n
ami
c
rea
l
m
o
f
In
fo
rma
t
i
o
n
T
ech
n
o
l
o
g
y
.
H
i
s
v
a
s
t
s
k
i
l
l
s
e
t
en
c
o
mp
a
s
s
es
a
d
i
v
ers
e
arra
y
o
f
c
u
t
t
i
n
g
-
e
d
g
e
t
ech
n
o
l
o
g
i
e
s
,
i
n
c
l
u
d
i
n
g
O
rac
l
e,
SQ
L
Serv
er,
Po
s
t
g
reS
Q
L
,
an
d
p
o
p
u
l
a
r
cl
o
u
d
s
er
v
i
ce
s
s
u
ch
as
O
CI,
A
W
S
an
d
Sn
o
w
fl
ak
e.
W
i
t
h
a
s
t
el
l
ar
t
rac
k
reco
rd
o
f
s
u
cce
s
s
f
u
l
l
y
s
t
eer
i
n
g
co
mp
l
ex
d
at
a
m
i
g
ra
t
i
o
n
p
ro
j
ect
s
,
co
n
d
u
ct
i
n
g
i
n
-
d
e
p
t
h
d
a
t
a
an
al
y
s
i
s
,
a
n
d
craf
t
i
n
g
i
n
n
o
v
a
t
i
v
e
s
o
ft
w
are
s
o
l
u
t
i
o
n
s
i
n
c
l
o
u
d
en
v
i
r
o
n
me
n
t
s
,
Mr.
N
aran
i
al
s
o
s
h
o
w
ca
s
es
h
i
s
mas
t
ery
i
n
U
n
i
x
s
h
e
l
l
s
cri
p
t
i
n
g
an
d
Py
t
h
o
n
s
cri
p
t
i
n
g
fo
r
s
o
p
h
i
s
t
i
cat
e
d
In
t
e
g
r
at
ed
D
a
t
a
W
are
h
o
u
s
e
s
y
s
t
ems
.
H
i
s
remark
ab
l
e
p
ro
f
i
ci
en
c
y
an
d
t
ech
n
i
ca
l
fi
n
es
s
e
u
n
d
ers
c
o
re
h
i
s
remar
k
ab
l
e
v
er
s
at
i
l
i
t
y
an
d
d
eep
ex
p
er
t
i
s
e
i
n
t
h
e
e
v
er
-
ev
o
l
v
i
n
g
f
i
el
d
o
f
In
f
o
rmat
i
o
n
T
ec
h
n
o
l
o
g
y
.
Mr.
N
aran
i
i
s
t
h
e
rev
ere
d
au
t
h
o
r
b
e
h
i
n
d
t
h
e
e
n
l
i
g
h
t
e
n
i
n
g
p
l
a
t
fo
rm
w
w
w
.
d
b
a
d
eed
s
.
co
m,
w
h
ere
h
e
g
e
n
ero
u
s
l
y
s
h
are
s
p
ro
f
o
u
n
d
i
n
s
i
g
h
t
s
o
n
t
h
e
aro
u
n
d
o
f
d
a
t
ab
a
s
e
man
a
g
e
men
t
.
H
e
ca
n
b
e
co
n
t
ac
t
e
d
at
emai
l
:
s
an
d
eep
re
d
d
y
cl
o
u
d
b
a@
g
ma
i
l
.
c
o
m.
Sa
kthi
v
e
l
Ka
rthi
ke
y
a
n
recei
v
ed
t
h
e
b
ac
h
el
o
r’s
d
eg
ree
i
n
el
ec
t
ro
n
i
cs
an
d
co
mmu
n
i
ca
t
i
o
n
e
n
g
i
n
eer
i
n
g
i
n
N
a
t
i
o
n
a
l
E
n
g
i
n
eer
i
n
g
Co
l
l
eg
e
a
n
d
ma
s
t
er’
s
d
e
g
ree
fro
m
Po
n
d
i
ch
erry
E
n
g
i
n
eeri
n
g
Co
l
l
e
g
e,
Po
n
d
i
ch
err
y
an
d
Ph
.
D
.
fro
m
A
n
n
a
U
n
i
v
er
s
i
t
y
.
H
e
h
as
b
een
w
o
r
k
i
n
g
as
p
r
o
fes
s
o
r
i
n
t
h
e
D
ep
ar
t
men
t
o
f
E
l
ect
ro
n
i
c
s
an
d
Co
mm
u
n
i
cat
i
o
n
E
n
g
i
n
eeri
n
g
,
K
.
S.
R.
Co
l
l
eg
e
o
f
E
n
g
i
n
eer
i
n
g
,
N
amak
k
a
l
(D
t
),
T
ami
l
N
ad
u
,
In
d
i
a
s
i
n
ce
2
0
0
8
.
H
i
s
area
s
o
f
i
n
t
eres
t
i
n
c
l
u
d
e
d
i
g
i
t
al
i
ma
g
e
p
ro
ce
s
s
i
n
g
,
n
eu
ra
l
n
et
w
o
r
k
s
an
d
s
o
f
t
co
mp
u
t
i
n
g
.
H
e
h
as
g
u
i
d
ed
v
ar
i
o
u
s
U
G
an
d
PG
p
ro
j
ect
s
i
n
t
h
e
area
o
f
Imag
e
p
ro
ce
s
s
i
n
g
an
d
i
n
d
i
fferen
t
areas
o
f
E
l
ec
t
ro
n
i
c
s
a
n
d
C
o
mmu
n
i
ca
t
i
o
n
E
n
g
i
n
eeri
n
g
an
d
a
l
s
o
g
u
i
d
i
n
g
re
s
earc
h
w
o
rk
.
H
e
h
as
p
u
b
l
i
s
h
ed
man
y
p
ap
er
s
i
n
In
t
er
n
at
i
o
n
al
an
d
N
at
i
o
n
al
j
o
u
r
n
al
s
.
H
e
i
s
a
l
i
fe
memb
er
o
f
IST
E
an
d
IE
T
E
.
H
e
can
b
e
co
n
t
ac
t
ed
at
ema
i
l
:
s
k
k
n
0
3
@
g
mai
l
.
co
m.
N
a
g
a
r
a
ja
n
M
o
h
a
nkum
a
r
w
as
b
o
rn
i
n
I
n
d
i
a
i
n
1
9
7
8
.
H
e
rece
i
v
e
d
h
i
s
B.
E
.
D
eg
ree
fro
m
Bh
ara
t
h
i
y
ar
U
n
i
v
er
s
i
t
y
,
T
am
i
l
N
a
d
u
,
In
d
i
a
i
n
2
0
0
0
,
M.
E
.
an
d
Ph
.
D
.
d
e
g
ree
fro
m
J
ad
a
v
p
u
r
U
n
i
v
ers
i
t
y
,
K
o
l
k
a
t
a
i
n
2
0
0
4
an
d
2
0
1
0
.
H
e
j
o
i
n
ed
t
h
e
N
a
n
o
D
ev
i
ce
Si
mu
l
at
i
o
n
L
ab
o
rat
o
ry
i
n
2
0
0
7
an
d
w
o
r
k
e
d
as
a
s
en
i
o
r
res
earc
h
fel
l
o
w
u
n
d
er
CSIR
d
i
rect
Sch
eme
t
i
l
l
Sep
t
em
b
er
2
0
0
9
.
L
at
er
h
e
j
o
i
n
ed
SK
P
E
n
g
i
n
eeri
n
g
Co
l
l
eg
e
a
s
a
p
ro
fe
s
s
o
r
t
o
d
ev
e
l
o
p
res
earch
ac
t
i
v
i
t
i
e
s
i
n
t
h
e
f
i
el
d
o
f
V
L
SI
an
d
N
A
N
O
t
ech
n
o
l
o
g
y
.
H
e
i
s
cu
rre
n
t
l
y
w
o
r
k
i
n
g
as
a
res
earch
p
ro
fe
s
s
o
r
at
Sy
m
b
i
o
s
i
s
I
n
s
t
i
t
u
t
e
o
f
T
ech
n
o
l
o
g
y
,
N
a
g
p
u
r
Camp
u
s
,
Sy
m
b
i
o
s
i
s
(In
t
er
n
at
i
o
n
al
)
D
eeme
d
U
n
i
v
er
s
i
t
y
,
Pu
n
e,
In
d
i
a.
H
e
i
s
a
s
en
i
o
r
memb
e
r
o
f
IE
E
E
.
H
e
h
as
ab
o
u
t
8
5
In
t
ern
a
t
i
o
n
a
l
j
o
u
r
n
al
p
u
b
l
i
cat
i
o
n
s
i
n
re
p
u
t
ed
j
o
u
rn
a
l
s
an
d
ab
o
u
t
5
0
i
n
t
er
n
at
i
o
n
al
co
n
fere
n
ce
p
r
o
ceed
i
n
g
s
.
H
e
recei
v
ed
t
h
e
carri
er
a
w
ar
d
fo
r
y
o
u
n
g
t
eac
h
ers
(C
A
Y
T
)
fr
o
m
A
ICT
E
,
N
ew
D
el
h
i
i
n
t
h
e
y
ear
o
f
2012
-
2
0
1
4
.
H
i
s
res
e
arch
i
n
t
ere
s
t
i
n
cl
u
d
e
s
mo
d
e
l
i
n
g
an
d
s
i
mu
l
a
t
i
o
n
s
t
u
d
y
o
f
H
E
MT
s
,
o
p
t
i
mi
za
t
i
o
n
o
f
d
e
v
i
ce
s
f
o
r
rad
i
o
fre
q
u
e
n
cy
(
RF
)
a
p
p
l
i
ca
t
i
o
n
s
an
d
ch
arac
t
eri
za
t
i
o
n
o
f
ad
v
an
ce
d
H
E
M
T
arch
i
t
ect
u
re,
t
erah
ert
z
el
ec
t
ro
n
i
c
s
,
h
i
g
h
freq
u
en
c
y
i
mag
i
n
g
,
s
en
s
o
r
s
an
d
c
o
mmu
n
i
ca
t
i
o
n
.
H
e
can
b
e
co
n
t
ac
t
ed
at
emai
l
:
n
m
k
s
k
p
ec@
g
ma
i
l
.
c
o
m.
Evaluation Warning : The document was created with Spire.PDF for Python.