T
E
L
KO
M
N
I
KA
T
e
lec
om
m
u
n
icat
ion
,
Com
p
u
t
i
n
g,
E
lec
t
r
on
ics
an
d
Cont
r
ol
Vol.
1
8
,
No.
1
,
F
e
br
ua
r
y
2020
,
pp.
4
41
~
4
49
I
S
S
N:
1693
-
6930,
a
c
c
r
e
dit
e
d
F
ir
s
t
G
r
a
de
by
Ke
me
nr
is
tekdikti
,
De
c
r
e
e
No:
21/E
/KP
T
/2018
DO
I
:
10.
12928/
T
E
L
KO
M
NI
KA
.
v1
8
i
1
.
12997
441
Jou
r
n
al
h
omepage
:
ht
tp:
//
jour
nal.
uad
.
ac
.
id/
index
.
php/T
E
L
K
OM
N
I
K
A
D
e
e
p
l
e
ar
n
in
g
m
od
e
l
f
or
t
h
o
r
ax d
is
e
ase
s
d
e
t
e
c
t
io
n
Gh
ad
a
A.
S
h
ad
e
e
d
,
M
oh
a
m
m
e
d
A
.
T
awf
e
e
q
,
S
a
ws
an
M
.
M
ah
m
ou
d
Co
mp
u
t
er
E
n
g
i
n
eer
i
n
g
D
e
p
art
me
n
t
,
Co
l
l
e
g
e
o
f
E
n
g
i
n
eer
i
n
g
,
Mu
s
t
a
n
s
i
ri
y
a
h
U
n
i
v
er
s
i
t
y
,
Iraq
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
Apr
25
,
2019
R
e
vis
e
d
J
ul
7
,
20
19
Ac
c
e
pted
J
ul
1
8
,
20
19
D
es
p
i
t
e
t
h
e
av
ai
l
ab
i
l
i
t
y
o
f
rad
i
o
l
o
g
y
d
ev
i
ces
i
n
s
o
me
h
eal
t
h
care
cen
t
ers
,
t
h
o
rax
d
i
s
ea
s
es
are
co
n
s
i
d
ere
d
as
o
n
e
o
f
t
h
e
mo
s
t
co
m
mo
n
h
ea
l
t
h
p
ro
b
l
em
s
,
es
p
ec
i
al
l
y
i
n
ru
ral
area
s
.
By
ex
p
l
o
i
t
i
n
g
t
h
e
p
o
w
er
o
f
t
h
e
In
t
er
n
et
o
f
t
h
i
n
g
s
an
d
s
p
ec
i
fi
c
p
l
a
t
fo
rm
s
t
o
an
a
l
y
ze
a
l
ar
g
e
v
o
l
u
me
o
f
med
i
cal
d
at
a,
t
h
e
h
eal
t
h
o
f
a
p
at
i
en
t
co
u
l
d
b
e
i
m
p
ro
v
ed
ear
l
i
er.
I
n
t
h
i
s
p
ap
er,
t
h
e
p
ro
p
o
s
ed
mo
d
el
i
s
b
as
e
d
o
n
p
re
-
t
ra
i
n
e
d
Res
N
et
-
5
0
fo
r
d
i
a
g
n
o
s
i
n
g
t
h
o
r
ax
d
i
s
eas
e
s
.
Ch
es
t
x
-
ray
i
mag
e
s
are
cro
p
p
e
d
t
o
ex
t
rac
t
t
h
e
ri
b
cag
e
p
art
fro
m
t
h
e
ch
es
t
rad
i
o
g
ra
p
h
s
.
Res
N
e
t
-
5
0
w
as
re
-
t
ra
i
n
o
n
Ch
es
t
x
-
ray
1
4
d
at
a
s
et
w
h
er
e
a
ch
es
t
rad
i
o
g
ra
p
h
i
mag
e
s
are
i
n
s
er
t
ed
i
n
t
o
t
h
e
mo
d
el
t
o
d
e
t
ermi
n
e
i
f
t
h
e
p
ers
o
n
i
s
h
eal
t
h
y
o
r
n
o
t
.
In
t
h
e
cas
e
o
f
an
u
n
h
eal
t
h
y
p
at
i
en
t
,
t
h
e
mo
d
e
l
can
cl
as
s
i
fy
t
h
e
d
i
s
eas
e
i
n
t
o
o
n
e
o
f
t
h
e
fo
u
rt
ee
n
c
h
es
t
d
i
s
ea
s
es
.
T
h
e
re
s
u
l
t
s
s
h
o
w
t
h
e
a
b
i
l
i
t
y
o
f
Res
N
et
-
5
0
i
n
ach
i
ev
i
n
g
i
m
p
res
s
i
v
e
p
erfo
rma
n
ce
i
n
cl
a
s
s
i
fy
i
n
g
t
h
o
rax
d
i
s
ea
s
es
.
K
e
y
w
o
r
d
s
:
C
he
s
t
r
a
diogr
a
phy
De
e
p
l
e
a
r
ning
I
nter
ne
t
of
t
hings
R
e
s
Ne
t
-
50
T
hor
a
x
dis
e
a
s
e
s
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
:
Gha
da
A.
S
ha
de
e
d,
C
omput
e
r
E
nginee
r
ing
De
p
a
r
tm
e
nt,
C
oll
e
ge
of
E
n
ginee
r
ing
,
M
us
tans
ir
iyah
Unive
r
s
it
y
,
B
a
ghda
d,
I
r
a
q
.
E
mail:
gha
da
.
s
ha
de
e
d@gmail.
c
om
1.
I
NT
RODU
C
T
I
ON
C
he
s
t
dis
e
a
s
e
s
a
r
e
one
of
the
mos
t
im
por
tant
he
a
lt
h
pr
oblems
pe
ople
e
xpe
r
ienc
e
.
M
or
e
than
1
mi
ll
ion
a
dult
s
with
pne
umoni
a
a
r
e
hos
pit
a
li
z
e
d
,
with
a
bout
50,
000
dying
e
a
c
h
ye
a
r
in
the
Unite
d
S
tate
s
a
lone
[
1,
2]
.
C
he
s
t
x
-
r
a
y
im
a
ge
s
a
r
e
the
mos
t
c
omm
on
tool
u
s
e
d
to
diagnos
e
c
he
s
t
dis
e
a
s
e
s
,
s
i
nc
e
their
de
vice
s
,
in
a
ddit
ion
to
making
the
pa
ti
e
nt
e
xpos
e
d
to
li
tt
le
r
a
diation,
a
r
e
a
ls
o
f
a
ir
ly
c
he
a
p
[
3
]
.
De
pe
n
ding
on
wor
ld
he
a
lt
h
or
ga
niza
ti
on
e
s
ti
mate
s
,
a
bout
two
-
thi
r
ds
of
the
plane
t's
population
s
uf
f
e
r
s
f
r
om
a
lac
k
o
f
a
c
c
e
s
s
to
r
a
diation
diagnos
is
[
4
]
.
E
ve
n
wi
th
the
a
v
a
il
a
bil
it
y
of
the
ne
c
e
s
s
a
r
y
e
quipm
e
nt
f
or
r
a
di
ogr
a
phy,
the
e
xpe
r
ts
who
a
r
e
a
ble
to
int
e
r
pr
e
t
the
x
-
r
a
ys
a
r
e
f
e
w,
e
s
pe
c
ially
in
r
ur
a
l
a
r
e
a
s
,
lea
ding
to
a
n
in
c
r
e
a
s
e
in
the
mor
talit
y
r
a
te
of
tr
e
a
tab
le
dis
e
a
s
e
s
in
many
c
ountr
ies
[
5
]
.
S
o
e
a
r
ly
diagnos
is
a
nd
tr
e
a
tm
e
nt
s
hould
be
a
va
il
a
ble
to
pr
e
ve
nt
c
ompl
ica
ti
ons
o
f
pne
umoni
a
t
ha
t
may
lea
d
to
de
a
th.
I
n
r
e
c
e
nt
ye
a
r
s
,
de
e
p
lea
r
ning
models
ha
ve
ma
de
s
igni
f
ica
nt
a
dva
nc
e
s
in
many
dig
it
a
l
im
a
ge
a
ppli
c
a
ti
ons
[
6
-
9
]
,
whic
h
ha
ve
include
d
f
a
s
ter
a
nd
e
a
r
li
e
r
de
tec
ti
on
of
a
ny
dis
e
a
s
e
s
with
the
he
lp
of
medic
a
l
im
a
ge
c
las
s
if
ica
ti
on
a
nd
de
tec
ti
on.
Ac
c
or
ding
to
t
he
s
uc
c
e
s
s
of
de
e
p
lea
r
ning,
many
r
e
s
e
a
r
c
he
r
s
ha
ve
s
ought
to
be
ne
f
it
f
r
o
m
de
e
p
ne
ur
a
l
ne
twor
ks
(
DN
Ns
)
f
o
r
diag
nos
ing
many
dis
e
a
s
e
s
,
including
thor
a
x
dis
e
a
s
e
s
on
c
he
s
t
r
a
diogr
a
phy,
whe
r
e
numer
ous
r
e
po
r
ts
ha
ve
be
e
n
publi
s
he
d
c
onf
ir
mi
ng
h
igh
a
c
c
ur
a
c
y
of
de
e
p
lea
r
ning
in
dis
e
a
s
e
s
diagnos
is
.
M
uc
h
r
e
s
e
a
r
c
h
ha
s
be
e
n
do
ne
us
ing
de
e
p
lea
r
ning
methods
to
de
tec
t
a
bnor
ma
l
obj
e
c
ts
in
medic
a
l
im
a
ge
s
[1
0
-
1
5
].
F
or
ins
tanc
e
,
in
[
1
6
]
,
d
igi
tal
im
a
ge
pr
oc
e
s
s
ing
tec
hniques
a
r
e
us
e
d
to
de
ve
lop
a
s
im
ple
pr
e
pr
oc
e
s
s
ing
pipeline
a
nd
e
xpe
r
t
r
a
diol
ogis
t
a
dvice
.
T
hr
e
e
ne
u
r
a
l
ne
twor
k
a
r
c
hit
e
c
tur
e
s
:
GoogL
e
Ne
t,
I
nc
e
pti
on
Ne
t,
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
1
8
,
No
.
1
,
F
e
br
ua
r
y
2020
:
4
41
-
4
49
442
R
e
s
Ne
t
a
r
e
c
r
e
a
te
d.
T
he
s
e
models
ha
ve
e
f
f
icie
ntl
y
pr
ove
n
in
c
las
s
if
ica
ti
on
tas
ks
whic
h
a
r
e
a
ppli
e
d
on
c
he
s
t
x
-
r
a
y
im
a
ge
s
.
I
n
[
1
7
]
a
unif
ied
c
onvolut
ional
ne
ur
a
l
ne
twor
k
(
C
NN
)
f
r
a
mew
or
k
wa
s
pr
opos
e
d
us
ing
we
a
kly
-
s
up
e
r
vis
e
d
mul
ti
-
labe
l
c
la
s
s
if
ica
ti
on,
taking
int
o
a
c
c
ount
t
ha
t
pooli
ng
s
tr
a
tegie
s
a
r
e
dif
f
e
r
e
nt
a
s
we
ll
a
s
va
r
ious
C
NN
's
mul
ti
-
labe
l
los
s
e
s
.
Als
o,
in
[
1
8
]
the
C
he
XN
e
t
model,
whic
h
is
a
model
of
de
e
p
l
e
a
r
ning,
ha
s
be
e
n
pr
opos
e
d
f
or
the
de
tec
ti
on
of
pne
umoni
a
whe
r
e
the
a
r
e
a
wa
s
diagno
s
e
d
with
the
dis
e
a
s
e
is
i
de
nti
f
i
e
d
in
the
im
a
ge
a
nd
us
e
d
de
ns
e
c
onne
c
ti
ons
[
19
]
a
nd
ba
tch
nor
maliza
ti
on
[
2
0
]
f
or
making
the
opti
m
iza
ti
on
mor
e
pos
s
ibi
li
ty
to
e
xe
c
uti
on
f
o
r
s
uc
h
a
model.
Ac
c
or
di
ng
to
the
r
e
s
ult
s
they
r
e
por
ted,
C
he
XN
e
t
ha
s
the
c
a
pa
bil
it
y
to
de
tec
t
pne
umoni
a
a
t
a
leve
l
e
qua
l
to
or
g
r
e
a
ter
than
that
of
r
a
diol
ogis
ts
.
I
n
[
2
1
]
ba
c
kpr
opa
ga
ti
o
n
ne
ur
a
l
ne
twor
k
(
B
P
NN
)
,
C
NN
plus
c
ompetit
ive
ne
ur
a
l
ne
twor
k
(
C
P
NN
)
we
r
e
tes
ted
f
or
the
mos
t
c
omm
on
dis
e
a
s
e
c
las
s
if
ica
ti
on
in
C
he
s
t
x
-
r
a
y.
T
he
r
e
c
ognit
ion
r
a
t
e
s
we
r
e
high
a
nd
pe
r
f
or
manc
e
wa
s
g
ood
whe
r
e
t
he
input
im
a
ge
ha
s
a
s
ize
of
32×
32
pixels
a
c
c
or
ding
to
the
r
e
s
ult
s
that
they
pr
e
s
e
nted.
C
P
NN
a
nd
B
P
NN
a
c
hieve
d
les
s
ge
ne
r
a
li
z
a
ti
on
powe
r
than
that
a
c
hieve
d
by
C
NN
.
I
n
[
2
2
]
C
he
s
tNe
t
is
pr
opos
e
d
to
a
ddr
e
s
s
the
diag
nos
is
of
thor
a
x
dis
e
a
s
e
s
on
c
h
e
s
t
r
a
diogr
a
phy
a
nd
wa
s
c
o
mpar
e
d
with
thr
e
e
de
e
p
lea
r
ning
models
on
C
he
s
t
x
-
r
a
y14
da
tas
e
t
[
1
7
]
us
ing
the
of
f
icia
l
pa
ti
e
nt
-
wis
e
s
pli
t.
Ac
c
or
ding
to
the
r
e
s
ult
s
pr
e
s
e
nted
the
r
e
s
ult
s
we
r
e
higher
than
thos
e
a
c
hieve
d
by
pr
e
vious
methods
.
I
n
a
ll
of
the
a
bove
,
a
nu
mber
o
f
methods
we
r
e
of
f
e
r
e
d
to
diagnos
e
c
he
s
t
dis
e
a
s
e
s
with
the
he
lp
of
c
omput
e
r
-
a
ided
diagnos
is
.
How
e
ve
r
,
the
pr
oble
m
of
incr
e
a
s
ing
the
s
uc
c
e
s
s
r
a
te
of
diagnos
is
of
dis
e
a
s
e
s
r
e
mains
one
o
f
the
mos
t
im
po
r
tant
tas
ks
to
c
omp
lete
the
diagnos
is
pr
oc
e
s
s
a
nd
ma
ke
it
mo
r
e
e
f
f
ic
ient.
S
o
the
model
wa
s
pr
opos
e
d
to
diagnos
e
the
c
he
s
t
c
ondit
ion
ba
s
e
d
on
r
a
diogr
a
phy.
T
he
diagnos
is
de
ter
mi
ne
s
whe
ther
the
pe
r
s
on
is
nor
mal
(
no
f
indi
ng
)
or
a
bnor
mal.
I
n
c
a
s
e
of
a
bnor
mal,
the
model
c
a
n
de
tec
t
f
our
tee
n
types
of
c
he
s
t
dis
e
a
s
e
s
.
One
of
the
de
e
p
lea
r
ning
models
,
R
e
s
tNe
t
-
50,
ha
s
be
e
n
s
ugge
s
ted
f
or
it
s
hi
gh
a
bil
it
y
to
diagnos
e
c
he
s
t
dis
e
a
s
e
s
on
X
-
r
a
y
s
,
a
s
we
ll
a
s
high
potential
to
a
void
many
of
the
pr
obl
e
ms
that
the
ne
twor
k
may
e
nc
ounter
whe
n
they
be
c
ome
de
e
pe
r
.
T
he
pr
opos
e
d
R
e
s
Ne
t
-
50
model
wa
s
e
va
luate
d
a
ga
ins
t
f
our
de
e
p
lea
r
ning
models
on
C
he
s
t
x
-
r
a
y14
da
tas
e
t.
T
he
b
lock
diag
r
a
m
o
f
the
p
r
opos
e
d
model
is
s
hown
in
F
igur
e
1.
T
he
r
e
maining
of
thi
s
pa
pe
r
is
s
tr
uc
tur
e
d
a
s
f
oll
ows
:
i
n
s
e
c
ti
on
2
,
the
pr
opos
e
d
model
R
e
s
Ne
t
-
50
a
nd
it
s
a
r
c
hit
e
c
tur
e
s
a
r
e
de
s
c
r
ibed.
T
he
da
tas
e
t
u
s
e
d
in
the
e
xpe
r
im
e
ntation
is
de
s
c
r
ibed
in
s
e
c
ti
on
3
a
long
with
it
s
p
r
e
pr
oc
e
s
s
ing.
F
inally
,
the
r
e
s
ult
s
a
r
e
pr
e
s
e
nted
in
s
e
c
ti
on
4
with
c
or
r
e
s
ponding
d
is
c
us
s
ions
f
oll
owe
d
by
the
c
onc
lus
ions
in
s
e
c
ti
on
5
.
F
igur
e
1.
I
l
lus
tr
a
ti
on
of
c
he
s
t
im
a
ge
a
na
lys
is
with
R
e
s
Ne
t
-
50
model
2.
P
ROP
OS
E
D
M
ODE
L
I
n
thi
s
wor
k,
the
pr
opos
e
d
model
is
divi
de
d
int
o
thr
e
e
s
teps
:
pr
e
pr
oc
e
s
s
ing,
R
e
s
Ne
t
-
50,
dis
e
a
s
e
diagnos
is
a
s
s
hown
in
F
igur
e
2.
I
n
the
pr
e
pr
oc
e
s
s
ing
s
t
e
p,
ther
e
a
r
e
thr
e
e
blocks
:
f
ir
s
t,
to
obt
a
in
mor
e
a
c
c
ur
a
te
a
nd
r
e
li
a
ble
da
ta
f
r
om
C
he
s
t
x
-
r
a
y
im
a
ge
,
the
r
ib
c
a
ge
a
r
e
a
is
c
ut
to
r
e
tain
on
it
a
n
d
lea
ve
the
r
e
maining
a
r
e
a
s
in
the
im
a
ge
.
T
his
c
r
oppin
g
pr
oc
e
s
s
make
s
the
tr
a
ini
ng
da
ta
mo
r
e
us
e
f
ul
a
nd
thus
incr
e
a
s
e
s
the
a
c
c
ur
a
c
y
of
the
r
e
s
ult
s
a
nd
r
e
duc
e
s
t
he
ti
me
o
f
tr
a
ini
ng
.
S
e
c
ondly,
the
im
a
ge
s
a
r
e
c
on
ve
r
ted
to
R
GB
.
T
hir
dly,
the
im
a
ge
s
c
ome
with
dif
f
e
r
e
nt
s
iz
e
s
.
T
he
r
e
f
or
e
,
the
im
a
ge
s
mus
t
be
unif
ied
withi
n
a
c
e
r
tain
s
ize
a
s
r
e
quir
e
d
by
the
p
r
opos
e
d
ne
twor
k.
Dur
in
g
the
tr
a
ini
ng
pr
oc
e
s
s
,
s
ome
de
tails
may
be
f
or
g
ott
e
n
in
the
im
a
ge
s
,
s
o
the
im
a
ge
s
a
r
e
r
e
pe
a
ted
to
make
the
ne
twor
k
r
e
membe
r
the
mos
t
de
tails
.
I
n
a
ddit
ion,
thi
s
pr
oc
e
s
s
incr
e
a
s
e
s
ne
twor
k
r
e
s
olut
ion.
I
n
R
e
s
Ne
t
-
5
0,
the
e
a
r
li
e
r
laye
r
s
a
r
e
f
r
e
e
z
e
,
while
the
f
ull
y
c
on
ne
c
ted
is
r
e
plac
e
d
a
c
c
or
ding
to
r
e
quir
e
ments
of
the
wor
k.
T
he
las
t
de
c
is
ion
a
pp
e
a
r
s
to
de
ter
mi
ne
the
c
a
s
e
of
the
c
he
s
t
is
take
n
in
the
las
t
s
tep.
2.
1.
R
es
N
et
-
50
T
he
a
r
c
hit
e
c
tur
e
of
the
R
e
s
Ne
t
-
50
de
e
p
lea
r
nin
g
models
is
s
hown
in
F
igur
e
3
.
I
t
c
ons
is
ts
of
50
laye
r
s
.
Unlike
othe
r
DN
N,
R
e
s
Ne
t
-
50
model
is
c
ha
r
a
c
ter
ize
d
by
it
s
a
bil
it
y
to
a
void
s
ome
of
the
p
r
oblems
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
De
e
p
lear
ning
mode
l
for
thor
ax
dis
e
as
e
s
de
tec
ti
on
(
Ghada
A
.
Shade
e
d)
443
c
onf
r
ont
the
ne
twor
k
whe
n
incr
e
a
s
ing
i
ts
laye
r
s
.
One
of
thes
e
pr
oblems
is
whe
n
the
numbe
r
o
f
laye
r
s
incr
e
a
s
e
s
to
mor
e
than
25
laye
r
s
s
tar
ti
ng
with
th
e
pr
oblem
o
f
va
nis
hing
gr
a
dients
,
whic
h
is
f
or
m
e
d
whe
n
the
gr
a
dient
is
ve
r
y
s
mall
then
the
we
ight
s
will
not
be
c
ha
nge
e
f
f
e
c
ti
ve
ly
a
nd
it
may
c
a
u
s
e
the
ne
ur
ona
l
ne
twor
k
to
s
top
c
ompl
e
tely
f
or
f
utu
r
e
tr
a
ini
ng
[
23
]
.
S
o,
thi
s
model
ha
s
be
e
n
uti
l
ize
d
in
thi
s
wo
r
k
due
to
it
s
high
a
bil
it
y
to
a
void
many
pr
oblems
a
s
we
ll
a
s
i
ts
e
f
f
icie
nt
pe
r
f
o
r
manc
e
in
the
diagnos
is
of
c
he
s
t
dis
e
a
s
e
s
.
F
igur
e
2
.
P
r
opos
e
d
model
block
diag
r
a
m
F
igur
e
3
.
Ar
c
hit
e
c
tur
e
of
R
e
s
Ne
t
-
50
model
I
n
thi
s
pa
pe
r
,
two
im
po
r
tant
s
tr
a
tegie
s
f
or
R
e
s
Ne
t
-
50
is
inves
ti
ga
ted.
F
ir
s
t,
the
model
pa
r
a
mete
r
s
with
r
a
ndom
va
lues
we
r
e
in
it
ialize
d,
s
o
the
mod
e
l
is
tr
a
ined
f
r
om
the
be
ginni
ng.
I
n
the
s
e
c
ond
s
tr
a
tegy,
the
we
ight
s
of
the
model
a
r
e
ini
t
ialize
d
f
r
om
p
r
e
-
tr
a
ined.
I
n
the
pr
opos
e
d
model,
the
f
ir
s
t
laye
r
r
e
qui
r
e
s
input
im
a
ge
s
of
s
ize
224×
224×
3,
whe
r
e
3
is
the
number
of
c
olor
s
(
R
e
s
Ne
t
-
50
we
r
e
de
s
igned
in
or
de
r
to
pr
oc
e
s
s
the
R
GB
im
a
ge
s
de
pe
nding
on
the
I
mage
Ne
t
[
24]
da
tas
e
t)
,
s
o
gr
a
y
im
a
ge
s
c
onve
r
ted
to
c
olor
im
a
g
e
s
.
T
he
n
c
olor
im
a
ge
s
a
r
e
pa
s
s
e
d
to
the
pr
opos
e
d
model
a
s
the
ini
ti
a
l
laye
r
s
whe
r
e
their
we
ight
s
a
r
e
f
r
oz
e
n
by
making
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
1
8
,
No
.
1
,
F
e
br
ua
r
y
2020
:
4
41
-
4
49
444
lea
r
ning
r
a
tes
e
qua
l
to
z
e
r
o
.
V
a
r
ious
leve
ls
of
th
e
f
e
a
tur
e
s
of
C
he
s
t
x
-
r
a
y
im
a
ge
s
a
r
e
e
xtr
a
c
ted
in
the
f
ir
s
t
c
onvolut
ional
laye
r
a
nd
th
is
is
s
hown
in
F
igur
e
4
whe
r
e
the
lea
r
ne
d
f
i
lt
e
r
s
a
r
e
s
hown
a
t
c
onvolut
ion
laye
r
.
T
o
r
e
tr
a
in
the
model
f
or
ne
w
c
las
s
if
ica
ti
on
tas
ks
,
whe
r
e
the
las
t
f
ull
c
onne
c
ti
on
ha
s
be
e
n
r
e
moved,
whic
h
c
ontent
1000
c
las
s
e
s
a
nd
r
e
plac
e
d
by
f
ull
y
c
onne
c
ted
w
it
h
f
if
tee
n
c
las
s
e
s
.
Dur
ing
t
r
a
ini
ng,
a
t
the
f
r
oz
e
n
laye
r
s
,
the
pa
r
a
mete
r
s
do
not
upda
te.
T
he
s
pe
e
d
of
model
tr
a
ini
ng
is
gr
e
a
tl
y
incr
e
a
s
e
d
whe
n
the
we
ight
s
of
t
he
ini
ti
a
l
laye
r
s
a
r
e
f
r
oz
e
n
a
s
a
r
e
s
ult
of
c
ounti
ng
the
gr
a
dien
ts
of
the
laye
r
s
that
ha
ve
be
e
n
f
r
oz
e
n.
R
e
s
Ne
t
-
50
c
ontains
two
blocks
,
C
onvolut
ional
block
a
nd
I
de
nti
ty
block
de
pe
nding
on
the
dim
e
ns
ions
of
the
input
/ou
tput
,
whe
ther
s
im
il
a
r
or
di
f
f
e
r
e
nt
.
I
de
nti
ty
block
is
us
e
d
whe
n
the
di
mens
ions
a
r
e
s
im
il
a
r
,
while
the
c
onvolut
ional
b
lock
is
u
s
e
d
whe
n
the
di
mens
ions
a
r
e
di
f
f
e
r
e
nt
.
F
igur
e
5
s
hows
the
a
r
c
hit
e
c
tur
e
of
e
a
c
h
of
c
onvolut
ional
block
a
n
d
identit
y
block.
I
n
the
ne
xt
s
e
c
ti
on,
the
da
tas
e
t
us
e
d
in
thi
s
wor
k
is
de
s
c
r
ibed
.
F
igur
e
4
.
T
r
a
ined
c
onvolut
ional
f
il
ter
s
in
the
f
ir
s
t
l
a
ye
r
(
a
)
(
b)
F
igur
e
5
.
T
ype
s
o
f
blocks
in
R
e
s
Ne
t
model
:
(
a
)
i
de
nti
ty
b
lock
,
(
b)
c
onvolut
ional
b
lock
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
De
e
p
lear
ning
mode
l
for
thor
ax
dis
e
as
e
s
de
tec
ti
on
(
Ghada
A
.
Shade
e
d)
445
3.
DA
T
ASE
T
I
n
thi
s
wor
k
,
the
pub
li
c
ly
a
va
il
a
ble
r
a
diog
r
a
phic
da
tas
e
t,
C
he
s
t
X
-
r
a
y
14
r
e
lea
s
e
d
by
W
a
ng
e
t
a
l.
[
17
]
wa
s
us
e
d.
T
he
da
tas
e
t
include
s
3182
x
-
r
a
y
im
a
ge
s
,
s
ome
e
xa
mpl
e
s
of
thi
s
da
tas
e
t
a
r
e
s
hown
in
F
igur
e
6
.
T
his
da
tas
e
t
ha
s
f
if
tee
n
labe
ls
c
ons
is
ti
ng
of
Nor
mal
labe
l
a
nd
14
dis
e
a
s
e
labe
ls
include
:
e
f
f
us
ion,
c
ons
oli
da
ti
on,
e
de
ma,
c
a
r
diom
e
ga
ly,
a
tele
c
tas
is
,
e
m
phys
e
ma,
f
ibr
os
is
,
nodule,
he
r
nia
mas
s
,
inf
il
tr
a
ti
on,
pne
umot
hor
a
x,
pleur
a
l
thi
c
ke
ning
,
a
nd
pne
u
mo
nia.
T
he
labe
ls
of
our
da
tas
e
t
a
r
e
c
lea
r
ly
il
lus
tr
a
ted
in
F
igur
e
7
,
whic
h
dis
plays
the
tot
a
l
number
o
f
im
a
g
e
s
.
F
or
the
de
tec
ti
on
tas
k,
the
da
tas
e
t
is
r
a
ndoml
y
s
pli
t
int
o
tes
ti
ng
15%
(
476
im
a
ge
s
)
,
va
li
da
ti
on
15
%
(
480
i
mage
s
)
,
a
nd
tr
a
ini
ng
70%
(
2226
im
a
ge
s
)
.
T
he
im
a
ge
s
we
r
e
s
a
ve
d
in
P
NG
f
or
mat.
T
he
digi
ti
z
e
d
im
a
ge
s
we
r
e
c
r
opping
a
nd
dupli
c
a
ti
ng
.
F
igur
e
6
.
E
ight
e
xa
mpl
e
s
f
r
om
c
he
s
t
x
-
r
a
y
14
da
tas
e
t,
whe
r
e
the
c
he
s
t
x
-
r
a
y
14
include
s
112,
120
im
a
ge
s
f
r
om
30,
805
pa
ti
e
nts
F
igur
e
7
.
Numbe
r
of
im
a
ge
s
in
e
a
c
h
labe
l
4.
RE
S
NE
T
-
50
M
ODE
L
RE
S
UL
T
S
T
he
r
e
s
ult
s
obtaine
d
f
r
om
the
pr
opos
e
d
model
we
r
e
p
r
e
s
e
nted
in
thi
s
s
e
c
ti
on.
T
he
model
is
im
pleme
nted
in
M
a
tl
a
b
2018a
us
ing
De
e
p
L
e
a
r
ning
T
oolbox
f
o
r
R
e
s
Ne
t
-
50
Ne
twor
k
[
6
]
,
wor
k
ing
on
a
c
omput
e
r
with
8
GB
memor
y
a
nd
I
ntel
(
R
)
C
or
e
(
T
M
)
i7
-
8550U
C
P
U
@
1
.
80GH
z
.
T
he
t
r
a
ini
ng
pa
r
a
mete
r
s
a
r
e
s
e
t
a
s
f
oll
ows
:
the
mi
ni
-
ba
tch
s
tocha
s
ti
c
gr
a
di
e
nt
de
s
c
e
nt
a
lgor
it
h
m
ha
s
be
e
n
a
dopted
whe
r
e
the
lea
r
ning
r
a
te
to
0.
0001
,
a
nd
the
ba
tch
s
ize
is
10,
maxim
um
it
e
r
a
ti
on
number
is
13320
a
nd
a
t
the
laye
r
of
f
ull
y
c
onne
c
ted
that
the
f
a
c
tor
of
bias
lea
r
n
r
a
te
is
10
a
n
d
the
we
ight
o
f
f
a
c
tor
of
lea
r
ning
r
a
te
is
10
.
T
he
pr
o
pos
e
d
model
a
c
hieve
d
ve
r
y
high
r
e
s
ult
s
i
n
r
e
a
ding
r
a
diogr
a
phy
a
nd
c
he
s
t
diagnos
is
whe
r
e
the
a
r
e
a
unde
r
c
ur
ve
(
AU
C
)
r
e
a
c
he
d
to
0
.
9261.
I
n
F
igu
r
e
8,
the
e
xa
mpl
e
s
hows
the
e
nd
r
e
s
ult
of
how
the
diagnos
is
is
made
thr
ough
the
pr
opos
e
d
model.
T
he
pa
ti
e
nt
wa
s
di
a
gnos
e
d
with
the
E
de
ma.
T
a
ble
1
s
hows
the
r
e
c
ognit
ion
r
a
tes
obtaine
d
f
or
the
p
r
o
pos
e
d
ne
twor
k
whe
r
e
it
s
howe
d
that
the
a
c
c
ur
a
c
y
us
ing
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
1
8
,
No
.
1
,
F
e
br
ua
r
y
2020
:
4
41
-
4
49
446
the
pr
opos
e
d
ne
twor
k
ha
s
93.
03%
f
o
r
t
r
a
ini
ng
a
nd
94.
49%
f
o
r
tes
ti
ng
.
Als
o,
the
ove
r
a
ll
pe
r
f
or
m
a
nc
e
f
or
tr
a
ini
ng
ti
me,
a
mo
unt
o
f
da
ta
,
a
nd
the
r
e
c
ognit
ion
r
a
te
is
de
s
c
r
ibed
in
T
a
ble
2.
T
he
a
ve
r
a
ge
d
ti
me
of
c
las
s
if
ica
ti
on
a
n
im
a
ge
us
ing
the
tr
a
ined
ne
twor
k
wa
s
0.
3
s
e
c
ond
pe
r
im
a
ge
.
T
hus
p
r
ovidi
ng
les
s
t
im
e
a
nd
les
s
e
f
f
or
t
to
obtain
a
d
iagnos
is
.
T
he
tot
a
l
tr
a
ini
ng
t
im
e
to
the
p
r
opos
e
d
ne
twor
k
took
a
bout
29
-
30
hou
r
s
.
F
igur
e
8
.
C
he
s
t
x
-
r
a
ys
im
a
ge
with
it
s
diagnos
is
a
n
d
the
c
las
s
e
s
pr
oba
bil
it
ies
T
a
ble
1.
P
e
r
f
o
r
manc
e
r
a
tes
f
o
r
R
e
s
Ne
t
-
50
on
tr
a
ini
ng,
va
li
da
ti
on,
tes
ti
ng,
a
nd
ove
r
a
ll
da
ta
T
a
ble
2
.
P
e
r
f
o
r
manc
e
r
a
tes
f
o
r
R
e
s
Ne
t
-
50
of
the
pr
opos
e
d
ne
twor
k
N
e
twor
k
mode
l
D
a
ta
f
or
tr
a
in
in
g
(
70%
)
D
a
ta
f
or
va
li
da
ti
on
(
15%
)
D
a
ta
f
or
te
s
ti
ng
(
15%
)
O
ve
r
a
ll
da
ta
(
100%
)
R
e
s
N
e
t
-
50
93.03%
93.42%
94.49%
93.18%
N
e
twor
k
mode
l
T
r
a
in
in
g
time
R
e
c
ogni
ti
on
r
a
te
O
ve
r
a
ll
da
ta
M
a
xi
mum
numbe
r
of
it
e
r
a
ti
ons
R
e
s
N
e
t
-
50
1055
min
93.18%
3182
im
a
ge
s
13320
T
he
model
a
c
hieve
d
good
diagnos
ti
c
r
e
s
ult
s
f
or
a
ll
gr
oups
,
with
the
highes
t
a
c
c
ur
a
c
y
be
ing
96.
5%
o
f
the
e
de
ma
dis
e
a
s
e
c
a
tegor
y,
whic
h
86
im
a
ge
s
a
nd
les
s
a
c
c
ur
a
te
be
ing
85.
71
%
of
the
he
r
nia
dis
e
a
s
e
c
a
tegor
y,
whic
h
70
im
a
ge
s
a
s
s
hown
in
F
igur
e
9
.
I
n
T
a
bl
e
3,
the
p
r
opos
e
d
model
wa
s
c
ompar
e
d
with
thr
e
e
de
e
p
lea
r
ning
models
whe
r
e
the
pe
r
-
c
las
s
AU
C
obtaine
d
by
a
pplyi
ng
the
tes
t
da
tas
e
t
to
thos
e
models
.
T
he
pr
opos
e
d
model
a
c
hieve
d
the
highes
t
r
a
te
f
or
mos
t
o
f
the
c
las
s
e
s
a
s
s
hown
by
the
r
e
s
ult
s
.
Among
the
models
that
we
r
e
tr
a
ined
on
the
C
he
s
t
x
-
r
a
y14
da
tas
e
t,
our
model
a
c
hieve
d
the
highes
t
a
c
c
ur
a
c
y
r
a
ti
os
f
or
mos
t
c
a
s
e
s
.
T
he
r
e
s
ult
s
s
howe
d
that
the
pr
opos
e
d
model
with
the
de
e
pe
r
ne
t
ha
d
a
lowe
r
e
r
r
o
r
r
a
te
t
ha
n
thos
e
with
lowe
r
laye
r
's
de
pth.
T
he
r
e
s
ult
s
s
how
that
incr
e
a
s
ing
ne
twor
k
de
pth
incr
e
a
s
e
s
the
ne
twor
k's
a
bil
it
y
to
c
las
s
if
y.
T
he
pr
opos
e
d
model
a
c
hieve
d
a
r
a
ti
ng
a
c
c
ur
a
c
y
of
92
.
71%
,
while
the
r
e
s
ult
s
o
f
p
r
e
vious
ne
twor
ks
we
r
e
84.
1378%
,
80
.
2714%
a
nd
73
.
8142%
.
De
s
pit
e
the
de
pth
of
the
model,
the
c
ompl
e
xit
y
is
s
t
il
l
low.
An
im
a
ge
wa
s
take
n
f
or
e
a
c
h
c
a
s
e
(
15
dif
f
e
r
e
nt
i
mage
s
)
a
nd
then
wa
s
pr
e
s
e
nted
to
the
pr
opos
e
d
models
to
de
ter
mi
ne
the
a
ppr
op
r
iate
diagnos
is
whe
r
e
the
r
e
s
ult
s
a
r
e
a
s
s
hown
in
F
igu
r
e
10
.
T
a
ble
3
.
C
ompar
is
on
AU
C
on
c
he
s
t
x
-
r
a
y
14
C
he
s
t
c
a
s
e
w
a
ng e
t
a
l.
[
18]
Y
e
a
e
t
a
l.
[
25]
R
a
jp
ur
ka
r
e
t
a
l.
[
19]
P
r
opos
e
d ne
twor
k
A
te
l
e
c
t
a
s
is
71.6%
77.2%
80.94%
93.5%
C
a
r
di
ome
ga
ly
80.7%
90.4%
92.48%
94%
C
on
s
ol
i
da
ti
on
70.8%
78.8%
79.01%
91.33%
E
f
f
us
io
n
78.4%
85.9%
86.38%
91%
E
mphys
e
m
a
81.5%
82.9%
93.71%
94.5%
F
ib
r
os
i
s
76.9%
76.7%
80.47%
96.5%
I
nf
il
tr
a
ti
o
n
60.9%
69.5%
73.45%
88.5%
M
a
s
s
70.6%
79.2%
86.76%
94.06%
N
odul
e
67.1%
71. 7%
78.02%
91.58%
P
ne
umo
ni
a
63.3%
71.3%
76.8%
92.16%
P
ne
umo
th
or
a
x
80.6%
84.1%
88.87%
93.5%
E
de
ma
83.5%
88.2%
88.78%
96.51%
P
le
ur
a
l
T
hi
c
ke
n
in
g
70.8%
76.5%
80.62%
91.5%
H
e
r
ni
a
76.7%
91.4%
91.64%
85.71%
N
or
ma
l
-
-
-
96.3%
A
ve
r
a
ge
73.8142%
80.2714%
84.1378%
92.71%
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
De
e
p
lear
ning
mode
l
for
thor
ax
dis
e
as
e
s
de
tec
ti
on
(
Ghada
A
.
Shade
e
d)
447
F
igur
e
9
.
P
e
r
f
or
manc
e
r
a
tes
f
or
f
if
tee
n
c
a
s
e
s
in
c
la
s
s
if
ica
ti
on
F
igur
e
10
.
S
ome
r
e
s
ult
s
of
diagnos
ti
c
de
tec
ti
on
in
r
a
diogr
a
phy
im
a
ge
s
.
An
e
xa
mpl
e
of
e
a
c
h
c
las
s
if
ica
ti
on
wa
s
take
n
5.
CONC
L
USI
ON
I
n
thi
s
pa
pe
r
,
a
pr
opos
e
d
c
he
s
t
diagnos
e
model
wa
s
a
ppli
e
d
to
c
he
s
t
r
a
diogr
a
phs
to
diagno
s
e
15
c
a
s
e
s
(
14
c
he
s
t
dis
e
a
s
e
s
a
nd
1
nor
mal
c
ond
it
ion)
,
ba
s
e
d
on
R
e
s
Ne
t
-
50
r
e
-
tr
a
ini
ng.
I
t
a
c
hie
ve
d
high
e
f
f
icie
nc
y
in
the
diagnos
is
of
c
he
s
t
r
a
diogr
a
phy
us
ing
the
de
e
p
lea
r
n
ing
model
with
a
n
AU
C
r
a
t
e
f
or
a
ll
c
las
s
e
s
of
0
.
9261.
T
his
model
wa
s
then
c
ompar
e
d
to
th
r
e
e
models
of
de
e
p
lea
r
ning
us
e
d
C
he
s
t
X
-
r
a
y
14
da
ta
s
e
t
whe
r
e
it
wa
s
s
up
e
r
ior
.
I
t
is
hope
d
that
thi
s
model
will
im
pr
ove
the
pr
ogr
e
s
s
of
he
a
lt
h
c
a
r
e
a
nd
incr
e
a
s
e
a
c
c
e
s
s
to
the
medic
a
l
e
xpe
r
ienc
e
th
r
oughout
the
wor
ld
whe
n
a
c
c
e
s
s
to
s
kil
led
r
a
diol
ogis
ts
is
li
mi
te
d.
I
n
ou
r
f
utur
e
wor
k
,
wor
k
s
hould
be
done
to
incr
e
a
s
e
c
ont
r
ol
of
R
e
s
Ne
t
-
50
us
ing
a
c
tual
x
-
r
a
y
da
ta,
whe
r
e
R
e
s
Ne
t
-
50
c
a
n
be
s
igni
f
ica
ntl
y
incr
e
a
s
e
d
by
thi
s
c
onf
igur
a
ti
on.
T
he
r
e
s
ult
s
s
howe
d
that
the
us
e
of
de
e
p
lea
r
ning
methods
is
us
e
f
ul
f
or
de
tec
ti
ng
x
-
r
a
y
dis
e
a
s
e
s
on
the
c
he
s
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
1
8
,
No
.
1
,
F
e
br
ua
r
y
2020
:
4
41
-
4
49
448
AC
KNOWL
E
DGM
E
N
T
W
e
c
omm
e
nd
the
e
f
f
o
r
ts
made
to
make
the
C
he
s
t
X
-
r
a
y
14
da
tas
e
t
a
va
il
a
ble,
making
it
e
a
s
ier
to
c
ompar
e
the
diag
nos
is
of
14
thor
a
x
dis
e
a
s
e
s
on
c
he
s
t
r
a
diogr
a
phs
.
Als
o,
the
a
utho
r
s
would
li
ke
t
o
thank
M
us
tans
ir
iyah
Unive
r
s
it
y
(
ww
w.
uomus
tans
ir
iyah.
e
du.
iq)
B
a
ghda
d
–
I
r
a
q
f
or
it
s
s
uppor
t
o
f
th
is
wor
k.
RE
F
E
RE
NC
E
S
[1
]
Cen
t
er
s
fo
r
D
i
s
eas
e
C
o
n
t
ro
l
an
d
Prev
en
t
i
o
n
(
C
D
C
)
,
[O
n
l
i
n
e
],
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
c
d
c.
g
o
v
/
p
n
e
u
mo
n
i
a
/
p
rev
e
n
t
i
o
n
.
h
t
m
l
,
O
ct
2
0
1
8
.
[2
]
Ch
eat
S
h
eet
,
[O
n
l
i
n
e],
A
v
ai
l
ab
l
e:
h
t
t
p
s
:
/
/
w
w
w
.
c
h
eat
s
h
e
et
.
co
m/
h
eal
t
h
-
f
i
t
n
es
s
/
t
h
e
s
e
-
are
-
the
-
l
ead
i
n
g
-
cau
s
es
-
of
-
d
e
at
h
-
in
-
the
-
u
-
s
.
h
t
ml
,
2
0
1
8
.
[3
]
R.
G
ru
et
zemach
er
a
n
d
A
.
G
u
p
t
a,
“U
s
i
n
g
d
ee
p
l
e
arn
i
n
g
f
o
r
p
u
l
m
o
n
ar
y
n
o
d
u
l
e
d
e
t
ect
i
o
n
&
d
i
a
g
n
o
s
i
s
,
”
Twen
t
y
-
s
eco
n
d
A
m
er
i
ca
s
Co
n
f
e
r
en
ce
o
n
In
f
o
r
m
a
t
i
o
n
S
y
s
t
e
m
s
,
p
p
.
1
-
9
,
2
0
1
6
.
[4
]
D
.
J
.
Mo
l
l
u
ra,
et
a
l
.
,
“W
h
i
t
e
p
a
p
er
rep
o
rt
o
f
t
h
e
rad
-
ai
d
co
n
feren
ce
o
n
i
n
t
ern
a
t
i
o
n
a
l
rad
i
o
l
o
g
y
f
o
r
d
ev
e
l
o
p
i
n
g
co
u
n
t
r
i
es
:
i
d
en
t
i
f
y
i
n
g
ch
a
l
l
e
n
g
e
s
,
o
p
p
o
rt
u
n
i
t
i
e
s
,
an
d
s
t
rat
e
g
i
e
s
fo
r
i
mag
i
n
g
s
er
v
i
ce
s
i
n
t
h
e
d
e
v
el
o
p
i
n
g
w
o
r
l
d
,
”
Jo
u
r
n
a
l
o
f
t
h
e
A
m
e
r
i
c
a
n
C
o
l
l
eg
e
o
f
R
a
d
i
o
l
o
g
y,
J
A
m
Co
l
l
R
a
d
i
o
l
,
v
o
l
.
7
,
n
o
.
7
,
p
p
.
4
9
5
-
5
0
0
,
2
0
1
0
.
[5
]
A
.
K
es
s
el
ma
n
,
et
al
.
,
“2
0
1
5
rad
-
a
i
d
c
o
n
fere
n
ce
o
n
i
n
t
e
rn
at
i
o
n
al
rad
i
o
l
o
g
y
f
o
r
d
ev
e
l
o
p
i
n
g
c
o
u
n
t
r
i
es
:
T
h
e
e
v
o
l
v
i
n
g
g
l
o
b
a
l
rad
i
o
l
o
g
y
l
an
d
s
ca
p
e,
”
Jo
u
r
n
a
l
o
f
t
h
e
A
m
e
r
i
c
a
n
Co
l
l
e
g
e
o
f
R
a
d
i
o
l
o
g
y
,
v
o
l
.
13
,
n
o
.
9
,
p
p
.
1
1
3
9
-
1
1
4
4
,
Sep
2
0
1
6
.
[6
]
K
.
H
e,
et
al
.
,
“D
eep
res
i
d
u
al
l
earn
i
n
g
fo
r
i
mag
e
reco
g
n
i
t
i
o
n
,
”
P
r
o
ceed
i
n
g
s
o
f
t
h
e
IE
E
E
co
n
f
e
r
en
ce
o
n
co
m
p
u
t
e
r
vi
s
i
o
n
a
n
d
p
a
t
t
er
n
r
eco
g
n
i
t
i
o
n
,
p
p
.
7
7
0
-
7
7
8
,
2
0
1
6
.
[7
]
S
.
Y
an
,
et
al
.
,
“
D
ri
v
er
Beh
av
i
o
r
Reco
g
n
i
t
i
o
n
Bas
e
d
o
n
D
eep
Co
n
v
o
l
u
t
i
o
n
a
l
N
eu
ra
l
N
et
w
o
r
k
s
,
”
12
th
In
t
er
n
a
t
i
o
n
a
l
Co
n
f
er
e
n
ce
o
n
N
a
t
u
r
a
l
Co
m
p
u
t
a
t
i
o
n
,
F
u
z
z
y
S
ys
t
em
s
a
n
d
Kn
o
wl
e
d
g
e
D
i
s
co
ve
r
y
(ICNC
-
F
S
K
D
)
,
2
0
1
6
.
[8
]
J
.
L
i
u
,
et
al
.
,
“
Sk
el
et
o
n
Bas
e
d
H
u
ma
n
A
ct
i
o
n
Reco
g
n
i
t
i
o
n
w
i
t
h
G
l
o
b
a
l
Co
n
t
ex
t
-
A
w
are
A
t
t
en
t
i
o
n
L
ST
M
N
et
w
o
r
k
s
,
”
IE
E
E
Tr
a
n
s
a
ct
i
o
n
s
o
n
Im
a
g
e
P
r
o
ces
s
i
n
g
,
A
p
r
2
0
1
8
.
[9
]
M
.
Ch
en
,
et
al
.
,
“
D
ee
p
Feat
u
re
L
earn
i
n
g
fo
r
Me
d
i
ca
l
Imag
e
A
n
a
l
y
s
i
s
w
i
t
h
Co
n
v
o
l
u
t
i
o
n
al
A
u
t
o
e
n
co
d
er
N
e
u
r
al
,
”
IE
E
E
Tr
a
n
s
a
ct
i
o
n
s
o
n
B
i
g
D
a
t
a
,
p
p
.
1
-
1
,
2
0
1
7
.
[1
0
]
H
.
A
.
Al
Mu
b
arak
,
et
al
.
,
“
A
H
y
b
r
i
d
D
eep
L
earn
i
n
g
an
d
H
an
d
craft
e
d
Feat
u
re
A
p
p
r
o
ac
h
fo
r
Cer
v
i
ca
l
Can
cer
D
i
g
i
t
a
l
H
i
s
t
o
l
o
g
y
Ima
g
e
Cl
as
s
i
f
i
cat
i
o
n
,
”
In
t
er
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
H
e
a
l
t
h
c
a
r
e
In
f
o
r
m
a
t
i
o
n
S
ys
t
em
s
a
n
d
In
f
o
r
m
a
t
i
cs
,
v
o
l
.
1
4
,
n
o
.
2
,
p
p
.
6
6
-
8
7
,
2
0
1
9
.
[1
1
]
T
.
K
o
o
i
,
et
al
.,
“
L
arg
e
s
cal
e
d
eep
l
ear
n
i
n
g
f
o
r
co
mp
u
t
er
ai
d
ed
d
e
t
ect
i
o
n
o
f
mammo
g
rap
h
i
c
l
e
s
i
o
n
s
,”
M
ed
Im
a
g
e
A
n
a
l
,
v
o
l
.
3
5
,
pp.
3
0
3
-
3
1
2
,
2
0
1
6
.
[1
2
]
M.
G
h
afo
o
ri
an
,
et
al
.,
“
N
o
n
-
u
n
i
f
o
rm
p
at
c
h
s
amp
l
i
n
g
w
i
t
h
d
ee
p
co
n
v
o
l
u
t
i
o
n
a
l
n
eu
ra
l
n
et
w
o
r
k
s
fo
r
w
h
i
t
e
mat
t
e
r
h
y
p
eri
n
t
e
n
s
i
t
y
s
eg
me
n
t
a
t
i
o
n
,”
IE
E
E
I
n
t
S
y
m
p
B
i
o
m
e
d
i
c
a
l
Im
a
g
i
n
g
,
p
p
.
1
4
1
4
-
1
4
1
7
,
2
0
1
6
.
[1
3
]
J
.
Ch
arb
o
n
n
i
er,
et
al
.,
“
Imp
ro
v
i
n
g
ai
rw
a
y
s
eg
me
n
t
a
t
i
o
n
i
n
co
m
p
u
t
ed
t
o
mo
g
rap
h
y
u
s
i
n
g
l
ea
k
d
et
ec
t
i
o
n
w
i
t
h
co
n
v
o
l
u
t
i
o
n
a
l
n
et
w
o
r
k
s
,”
M
e
d
Im
a
g
e
A
n
a
l
,
v
o
l
.
36,
p
p
.
52
-
60
,
2
0
1
7
.
[1
4
]
M.
J
.
J
.
P.
v
an
G
ri
n
s
v
en
,
e
t
al
.,
“
Fas
t
co
n
v
o
l
u
t
i
o
n
al
n
eu
ra
l
n
e
t
w
o
rk
t
rai
n
i
n
g
u
s
i
n
g
s
e
l
ect
i
v
e
d
a
t
a
s
am
p
l
i
n
g
:
A
p
p
l
i
cat
i
o
n
t
o
h
em
o
rrh
a
g
e
d
et
ect
i
o
n
i
n
co
l
o
r
fu
n
d
u
s
i
mag
e
s
,”
IE
E
E
Tr
a
n
s
M
e
d
Im
a
g
i
n
g
,
v
o
l
.
3
5
,
n
o
.
5
,
pp.
1
2
7
3
-
1
2
8
4
,
2
0
1
6
.
[1
5
]
E
.
K
es
i
m,
et
al
.
,
“X
-
Ray
Ch
e
s
t
Ima
g
e
Cl
as
s
i
f
i
cat
i
o
n
b
y
A
Smal
l
-
Si
z
e
d
Co
n
v
o
l
u
t
i
o
n
al
N
eu
ra
l
N
et
w
o
r
k
,
”
S
ci
e
n
t
i
f
i
c
M
eet
i
n
g
o
n
E
l
ec
t
r
i
ca
l
-
E
l
ect
r
o
n
i
c
&
B
i
o
m
ed
i
ca
l
E
n
g
i
n
ee
r
i
n
g
a
n
d
Co
m
p
u
t
e
r
S
c
i
en
ce
(
E
B
B
T)
,
2
0
1
9
.
[1
6
]
C.
T
at
aru
,
et
al
.
,
“D
eep
L
earn
i
n
g
fo
r
ab
n
o
rmal
i
t
y
d
et
ec
t
i
o
n
i
n
Ch
es
t
X
-
Ray
i
ma
g
es
,
”
T
e
s
i
s
St
an
f
o
r
d
U
n
i
v
er
s
i
t
y
,
2
0
1
7
.
[1
7
]
X
.
W
an
g
,
et
al
.
,
“Ch
es
t
X
-
Ra
y
8
:
H
o
s
p
i
t
al
-
Sca
l
e
Ch
es
t
X
-
Ra
y
D
at
a
b
as
e
an
d
Ben
c
h
mark
s
o
n
W
ea
k
l
y
-
Su
p
erv
i
s
e
d
Cl
as
s
i
fi
cat
i
o
n
an
d
L
o
cal
i
zat
i
o
n
o
f
Co
mmo
n
T
h
o
rax
D
i
s
eas
e
s
,
”
2
0
1
7
IE
E
E
Co
n
f
er
e
n
ce
o
n
Co
m
p
u
t
er
V
i
s
i
o
n
a
n
d
P
a
t
t
e
r
n
R
ec
o
g
n
i
t
i
o
n
(C
V
P
R
)
,
p
p
.
3
4
6
2
-
3
4
7
1
,
2
0
1
7
.
[1
8
]
P.
Raj
p
u
r
k
ar,
et
a
l
.
,
“
Ch
eX
N
et
:
Rad
i
o
l
o
g
i
s
t
-
L
ev
e
l
Pn
e
u
mo
n
i
a
D
e
t
ect
i
o
n
o
n
Ch
e
s
t
X
-
Ray
s
w
i
t
h
D
ee
p
L
earn
i
n
g
,
”
i
n
P
r
o
ceed
i
n
g
s
o
f
t
h
e
IE
E
E
co
n
f
e
r
en
ce
o
n
co
m
p
u
t
e
r
vi
s
i
o
n
a
n
d
p
a
t
t
e
r
n
r
eco
g
n
i
t
i
o
n
,
D
ec
2
0
1
7
.
[1
9
]
G
.
H
u
an
g
,
e
t
al
.
,
“D
e
n
s
e
l
y
c
o
n
n
ect
e
d
co
n
v
o
l
u
t
i
o
n
al
n
et
w
o
r
k
s
,
”
P
r
o
cee
d
i
n
g
s
o
f
t
h
e
IE
E
E
c
o
n
f
er
en
ce
o
n
c
o
m
p
u
t
e
r
vi
s
i
o
n
a
n
d
p
a
t
t
er
n
r
eco
g
n
i
t
i
o
n
,
A
u
g
2
0
1
7
.
[2
0
]
S.
Io
ffe
an
d
C.
Szeg
ed
y
,
“Bat
ch
n
o
rmal
i
zat
i
o
n
:
A
cce
l
er
at
i
n
g
d
ee
p
n
et
w
o
r
k
t
rai
n
i
n
g
b
y
red
u
c
i
n
g
i
n
t
ern
a
l
co
v
ari
at
e
s
h
i
f
t
,
”
In
t
er
n
a
t
i
o
n
a
l
co
n
f
e
r
en
ce
o
n
m
a
ch
i
n
e
l
ea
r
n
i
n
g
,
2
0
1
5
.
[2
1
]
R.
H
.
A
b
i
y
e
v
an
d
M
.
K
.
S.
Ma’ai
t
a
h
,
“D
ee
p
Co
n
v
o
l
u
t
i
o
n
a
l
N
e
u
ral
N
et
w
o
r
k
s
f
o
r
Ch
e
s
t
D
i
s
eas
e
s
D
e
t
ect
i
o
n
,
”
H
i
n
d
a
w
i
,
Jo
u
r
n
a
l
o
f
H
ea
l
t
h
ca
r
e
E
n
g
i
n
eer
i
n
g
,
v
o
l
.
2
0
1
8
,
p
p
.
1
-
1
1
,
2
0
1
8
.
[2
2
]
H
.
W
an
g
an
d
Y
.
X
i
a,
“
Ch
es
t
N
e
t
:
A
D
ee
p
N
eu
ra
l
N
et
w
o
rk
fo
r
Cl
as
s
i
f
i
ca
t
i
o
n
o
f
T
h
o
rac
i
c
D
i
s
e
as
e
s
o
n
Ch
es
t
Rad
i
o
g
ra
p
h
y
,
”
P
r
o
ceed
i
n
g
s
o
f
t
h
e
IE
E
E
Co
n
f
e
r
en
ce
o
n
Co
m
p
u
t
er
V
i
s
i
o
n
a
n
d
P
a
t
t
er
n
R
ec
o
g
n
i
t
i
o
n
,
2
0
1
8
.
[2
3
]
T
o
w
ard
s
D
at
a
Sci
e
n
ce,
[O
n
l
i
n
e],
A
v
ai
l
ab
l
e:
h
t
t
p
s
:
/
/
t
o
w
ar
d
s
d
at
a
s
ci
e
n
ce.
co
m/
u
n
d
ers
t
an
d
i
n
g
-
res
i
d
u
al
-
n
et
w
o
r
k
s
-
9
ad
d
4
b
6
6
4
b
0
3
,
2
0
1
8
.
[2
4
]
Imag
eN
e
t
,
[O
n
l
i
n
e],
A
v
a
i
l
a
b
l
e
:
h
t
t
p
:
/
/
w
w
w
.
i
mag
e
-
n
et
.
o
r
g
.
[2
5
]
Y
.
L
i
,
et
al
.
,
“L
earn
i
n
g
t
o
d
i
a
g
n
o
s
e
fr
o
m
s
cra
t
ch
b
y
e
x
p
l
o
i
t
i
n
g
d
ep
en
d
en
c
i
es
am
o
n
g
l
a
b
el
s
,
”
P
r
o
ceed
i
n
g
s
o
f
t
h
e
IE
E
E
co
n
f
er
e
n
ce
o
n
co
m
p
u
t
er
v
i
s
i
o
n
a
n
d
p
a
t
t
er
n
r
ec
o
g
n
i
t
i
o
n
,
2
0
1
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
De
e
p
lear
ning
mode
l
for
thor
ax
dis
e
as
e
s
de
tec
ti
on
(
Ghada
A
.
Shade
e
d)
449
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
G
h
ad
a
A
.
Sh
ad
eed
recei
v
ed
h
er
B.
Sc.
i
n
Co
mp
u
t
er
E
n
g
i
n
eer
i
n
g
fro
m
Mu
s
t
an
s
i
r
i
y
a
h
U
n
i
v
er
s
i
t
y
,
Bag
h
d
ad
,
Iraq
,
i
n
2
0
1
6
,
s
h
e
i
s
c
u
rren
t
l
y
an
M.
Sc.
s
t
u
d
e
n
t
i
n
t
h
e
D
e
p
art
me
n
t
o
f
Co
m
p
u
t
er
E
n
g
i
n
eeri
n
g
,
Mu
s
t
a
n
s
i
ri
y
ah
U
n
i
v
ers
i
t
y
/
Ba
g
h
d
ad
,
Iraq
.
Mo
h
amme
d
A
l
i
T
a
w
feeq
i
s
an
as
s
o
c
i
at
e
d
Pro
fes
s
o
r
at
Co
mp
u
t
er
E
n
g
i
n
eer
i
n
g
D
ep
ar
t
men
t
,
Co
l
l
eg
e
o
f
E
n
g
i
n
eeri
n
g
,
Mu
s
t
an
s
i
r
i
y
ah
U
n
i
v
ers
i
t
y
.
H
e
recei
v
e
d
h
i
s
B.
Sc.
i
n
E
l
ec
t
ri
ca
l
E
n
g
i
n
eeri
n
g
fro
m
U
n
i
v
er
s
i
t
y
o
f
T
ec
h
n
o
l
o
g
y
/
Ba
g
h
d
ad
,
Iraq
i
n
1
9
7
9
a
n
d
h
i
s
M.
Sc.
d
e
g
ree
i
n
Co
mp
u
t
er
E
n
g
i
n
eeri
n
g
fro
m
U
n
i
v
ers
i
t
y
o
f
Bag
h
d
ad
i
n
1
9
8
9
.
H
e
o
b
t
a
i
n
e
d
h
i
s
Ph
.
D
.
i
n
Co
mp
u
t
er
E
n
g
i
n
eeri
n
g
fro
m
U
n
i
v
er
s
i
t
y
o
f
T
ec
h
n
o
l
o
g
y
,
Bag
h
d
a
d
,
Iraq
i
n
2
0
0
6
.
H
i
s
res
earc
h
i
n
t
eres
t
s
i
n
c
l
u
d
e
In
t
el
l
i
g
en
t
Sy
s
t
em
s
,
O
p
t
i
mi
za
t
i
o
n
T
ec
h
n
i
q
u
es
,
W
i
re
l
es
s
Sen
s
o
r
N
et
w
o
r
k
s
,
a
n
d
Io
T
ap
p
l
i
cat
i
o
n
s
.
Saw
s
a
n
M.
Mah
mo
u
d
recei
v
ed
h
er
B.
Sc.
i
n
Co
mp
u
t
er
Sci
e
n
ce
fro
m
U
n
i
v
er
s
i
t
y
o
f
T
ech
n
o
l
o
g
y
/
Bag
h
d
a
d
,
Iraq
i
n
1
9
9
4
.
Sh
e
o
b
t
ai
n
e
d
h
er
M.
Sc.
fro
m
U
n
i
v
er
s
i
t
y
o
f
Bag
h
d
ad
i
n
1
9
9
8
.
H
er
Ph
.
D
.
d
eg
ree
i
n
Co
m
p
u
t
at
i
o
n
a
l
In
t
e
l
l
i
g
e
n
ce
i
s
o
b
t
ai
n
ed
fr
o
m
N
o
t
t
i
n
g
h
am
T
re
n
t
U
n
i
v
er
s
i
t
y
,
N
o
t
t
i
n
g
h
am,
U
K
i
n
2
0
1
2
.
Saw
s
an
j
o
i
n
e
d
Mu
s
t
a
n
s
i
ri
y
ah
U
n
i
v
ers
i
t
y
/
E
n
g
i
n
e
er
i
n
g
Co
l
l
eg
e
i
n
1
9
9
4
a
s
a
memb
er
o
f
t
h
e
aca
d
emi
c
s
t
aff
.
H
er
res
earc
h
i
n
t
ere
s
t
s
i
n
cl
u
d
e
,
b
u
t
n
o
t
l
i
mi
t
e
d
t
o
,
Co
m
p
u
t
at
i
o
n
al
I
n
t
e
l
l
i
g
e
n
ce,
A
m
b
i
e
n
t
In
t
e
l
l
i
g
e
n
ce
(Smart
H
o
me
an
d
In
t
el
l
i
g
en
t
E
n
v
i
r
o
n
me
n
t
),
W
i
rel
e
s
s
Sen
s
o
r
N
et
w
o
r
k
,
D
at
a
Mi
n
i
n
g
,
an
d
H
ea
l
t
h
Mo
n
i
t
o
r
i
n
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.