I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
2
,
A
p
r
il 2
0
2
6
,
p
p
.
1
3
4
9
~
1
3
6
1
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
2
.
p
p
1
3
4
9
-
1
3
6
1
1349
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
NN
-
SVM
:
a
hybri
d neural n
etwork
–
supp
o
rt
v
ector
ma
chine
framewo
rk
for a
c
cura
te
pn
eumo
ni
a
det
ec
tion from
chest X
-
ra
y
s
Sa
nto
s
h K
um
a
r
J
a
nk
a
t
t
i
1
,
R
a
g
ha
v
endra
Srini
v
a
s
a
ia
h
2
,
M
o
ha
m
m
a
d
Sh
a
hin
a
P
a
rv
ee
n
1
,
H
a
rish
H
.
K
encha
nn
a
v
a
r
3
,
Da
nthulu
ri
S
ud
ha
1
,
Sriha
ri
Sh
a
rm
a
K
a
rig
iri
Na
r
a
h
4
,
M
a
ha
dev
Sh
iv
a
ra
j
5
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
D
a
y
a
n
a
n
d
a
S
a
g
a
r
U
n
i
v
e
r
si
t
y
,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
D
a
t
a
S
c
i
e
n
c
e
E
n
g
i
n
e
e
r
i
n
g
,
C
H
R
I
S
T
U
n
i
v
e
r
si
t
y
,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
(
D
a
t
a
S
c
i
e
n
c
e
)
,
D
a
y
a
n
a
n
d
a
S
a
g
a
r
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
4
D
e
p
a
r
t
me
n
t
o
f
C
o
l
l
e
g
e
o
f
P
h
y
si
o
t
h
e
r
a
p
y
,
S
c
h
o
o
l
o
f
H
e
a
l
t
h
S
c
i
e
n
c
e
s
,
D
a
y
a
n
a
n
d
a
S
a
g
a
r
U
n
i
v
e
r
si
t
y
,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
5
D
e
p
a
r
t
m
e
n
t
o
f
E
l
e
c
t
r
o
n
i
c
s
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
E
n
g
i
n
e
e
r
i
n
g
,
D
a
y
a
n
a
n
d
a
S
a
g
a
r
A
c
a
d
e
m
y
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
M
a
n
a
g
e
m
e
n
t
,
B
e
n
g
a
l
u
r
u
,
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
Oct
2
2
,
2
0
2
4
R
ev
is
ed
J
an
1
4
,
2
0
2
6
Acc
ep
ted
J
an
2
5
,
2
0
2
6
We
p
re
se
n
t
n
e
u
ra
l
n
e
two
r
k
(NN
)
–
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
(S
VM),
h
y
b
ri
d
NN
-
S
VM
fra
m
e
wo
rk
f
o
r
th
re
e
-
c
l
a
ss
p
n
e
u
m
o
n
ia
d
e
tec
ti
o
n
(
n
o
rm
a
l,
b
a
c
teria
l,
a
n
d
v
iral)
fro
m
c
h
e
st
X
-
ra
y
s
(CX
Rs)
.
P
re
train
e
d
NN
b
a
c
k
b
o
n
e
is
f
in
e
-
tu
n
e
d
fo
r
ra
d
io
g
ra
p
h
ic
tex
tu
re
s;
g
lo
b
a
l
a
v
e
ra
g
e
p
o
o
l
in
g
(G
AP)
y
ield
s
e
m
b
e
d
d
i
n
g
s
th
a
t
fe
e
d
c
a
li
b
ra
ted
ra
d
ial
b
a
sis
fu
n
c
ti
o
n
(
RBF
)
-
S
VM.
S
tan
d
a
rd
ize
d
p
re
p
ro
c
e
ss
in
g
(re
siz
e
,
n
o
rm
a
li
z
a
ti
o
n
)
a
n
d
c
las
s
-
a
wa
re
a
u
g
m
e
n
tatio
n
a
re
a
p
p
li
e
d
.
We
re
p
o
rt
a
c
c
u
ra
c
y
,
p
re
c
isio
n
,
re
c
a
ll
,
F1
-
sc
o
re
,
a
re
a
u
n
d
e
r
th
e
c
u
rv
e
(AU
C)
,
c
o
n
fu
si
o
n
m
a
tri
c
e
s,
a
n
d
p
e
r
-
c
las
s
re
c
e
iv
e
r
o
p
e
ra
ti
n
g
c
h
a
ra
c
teristic
(
ROC
)
.
S
tatisti
c
a
l
sig
n
ifi
c
a
n
c
e
is
a
ss
e
ss
e
d
v
ia
De
Lo
n
g
(AU
C),
M
c
Ne
m
a
r
(a
c
c
u
ra
c
y
),
a
n
d
p
a
ired
b
o
o
tstr
a
p
(
F1
-
sc
o
re
).
G
ra
d
ien
t
-
we
ig
h
t
e
d
c
las
s
a
c
ti
v
a
ti
o
n
m
a
p
p
i
n
g
(g
ra
d
-
CA
M
)
su
p
p
o
rts
i
n
terp
re
tab
il
it
y
;
e
x
tern
a
l
v
a
li
d
a
ti
o
n
a
n
d
d
o
m
a
in
a
d
a
p
tatio
n
(
b
a
tch
n
o
rm
a
li
z
a
ti
o
n
re
-
e
stim
a
ti
o
n
a
n
d
t
e
m
p
e
ra
tu
re
sc
a
li
n
g
)
a
ss
e
ss
ro
b
u
st
n
e
ss
.
NN
-
S
VM
a
tt
a
i
n
s
9
7
.
4
6
%
a
c
c
u
ra
c
y
with
str
o
n
g
m
a
c
ro
-
F
1
a
n
d
AU
C.
Co
m
p
a
re
d
with
S
o
f
tM
a
x
h
e
a
d
,
S
VM
imp
r
o
v
e
s
m
a
rg
in
se
p
a
ra
ti
o
n
a
n
d
c
a
li
b
ra
ti
o
n
.
We
p
re
se
n
t
NN
-
S
VM,
h
y
b
ri
d
d
e
e
p
lea
rn
in
g
a
p
p
ro
a
c
h
th
a
t
c
o
m
b
in
e
s
tran
sf
e
r
-
lea
rn
e
d
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
s)
with
S
VM
c
las
sifier
to
a
u
to
m
a
ti
c
a
ll
y
d
ia
g
n
o
se
p
n
e
u
m
o
n
ia
fr
o
m
CXRs
in
to
th
re
e
c
li
n
ica
ll
y
re
lev
a
n
t
c
a
teg
o
ries
:
v
iral
p
n
e
u
m
o
n
ia,
b
a
c
teria
l
p
n
e
u
m
o
n
ia,
a
n
d
n
o
rm
a
l.
We
u
se
p
re
-
train
e
d
CNN
to
e
x
trac
t
ro
b
u
st
ima
g
e
e
m
b
e
d
d
in
g
s
a
fter
sta
n
d
a
rd
ize
d
p
re
p
ro
c
e
ss
in
g
(re
siz
in
g
a
n
d
n
o
rm
a
li
z
a
ti
o
n
)
a
n
d
trai
n
RBF
-
k
e
rn
e
l
S
VM
o
n
r
e
su
lt
in
g
fe
a
tu
re
s.
P
e
rf
o
rm
a
n
c
e
is
e
v
a
lu
a
ted
with
a
c
c
u
ra
c
y
,
p
re
c
isio
n
,
re
c
a
ll
,
F
1
-
sc
o
re
,
a
n
d
c
o
n
fu
si
o
n
m
a
t
rice
s.
On
lab
e
led
CXR
d
a
tas
e
t,
NN
-
S
VM
a
c
h
iev
e
s
9
7
.
4
6
%
a
c
c
u
ra
c
y
,
d
e
m
o
n
stra
ti
n
g
stro
n
g
d
iag
n
o
stic
c
a
p
a
b
il
it
y
t
h
a
t
c
a
n
re
d
u
c
e
ra
d
io
l
o
g
ist
b
u
r
d
e
n
a
n
d
su
p
p
o
rt
ti
m
e
ly
c
li
n
ica
l
d
e
c
isio
n
-
m
a
k
i
n
g
.
K
ey
w
o
r
d
s
:
C
h
est X
-
r
ay
C
NN
-
SVM
Dee
p
lear
n
in
g
Me
d
ical
im
ag
in
g
Pn
eu
m
o
n
ia
T
r
an
s
f
er
lear
n
i
n
g
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
:
San
to
s
h
Ku
m
ar
J
an
k
atti
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
T
ec
h
n
o
l
o
g
y
Day
an
an
d
a
Sag
a
r
Un
iv
er
s
ity
,
Har
o
h
alli
B
en
g
alu
r
u
,
Kar
n
ata
k
a,
I
n
d
ia
E
m
ail:
s
an
to
s
h
k
u
m
ar
-
ct@
d
s
u
.
ed
u
.
in
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
h
allm
ar
k
o
f
p
n
e
u
m
o
n
ia
is
a
s
er
io
u
s
lu
n
g
in
f
ec
tio
n
th
a
t
ca
n
ca
u
s
e
s
y
m
p
to
m
s
in
clu
d
in
g
f
ev
er
,
co
u
g
h
i
n
g
,
a
n
d
t
r
o
u
b
le
b
r
ea
t
h
in
g
.
I
t
is
o
n
e
o
f
th
e
m
ai
n
ca
u
s
es
o
f
illn
ess
an
d
m
o
r
tality
i
n
t
h
e
g
lo
b
e,
esp
ec
ially
in
y
o
u
n
g
c
h
ild
r
en
,
th
e
el
d
er
ly
,
an
d
p
eo
p
le
with
wea
k
en
e
d
i
m
m
u
n
e
s
y
s
tem
s
.
T
h
e
W
o
r
ld
Hea
lth
Or
g
an
izatio
n
(
W
HO)
esti
m
ate
s
th
at
o
f
all
d
ea
th
s
f
allin
g
with
in
th
is
ag
e
g
r
o
u
p
.
Pn
eu
m
o
n
ia
is
n
o
t
o
n
l
y
ass
o
ciate
d
with
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
3
4
9
-
1
3
6
1
1350
h
ig
h
m
o
r
tality
r
ate,
b
u
t it
is
al
s
o
ass
o
ciate
d
with
h
ig
h
co
s
ts
f
o
r
h
ea
lth
ca
r
e
s
y
s
tem
s
,
p
ar
ticu
lar
ly
f
o
r
d
e
v
elo
p
in
g
co
u
n
tr
ies
wh
er
e
ea
r
ly
an
d
a
cc
u
r
ate
d
iag
n
o
s
is
is
n
o
t
r
ea
d
ily
attain
ab
le.
T
h
e
p
r
ess
in
g
n
ee
d
f
o
r
ef
f
ec
tiv
e
d
iag
n
o
s
tic
to
o
ls
is
in
d
icate
d
b
y
its
h
ig
h
i
n
ten
s
ity
,
h
ig
h
m
o
r
tality
r
ates,
an
d
lo
we
r
co
s
ts
a
s
s
o
ciate
d
with
th
is
co
n
d
itio
n
,
as
in
d
icate
d
in
[
1
]
,
[
2
]
.
T
h
e
co
n
v
en
tio
n
al
ap
p
r
o
a
ch
u
s
ed
f
o
r
th
e
d
iag
n
o
s
is
o
f
p
n
eu
m
o
n
ia,
s
u
ch
as
p
h
y
s
ical
ex
am
in
atio
n
,
cu
ltu
r
es
f
r
o
m
s
p
u
tu
m
s
am
p
les,
a
n
d
c
h
est
X
-
r
ay
(
C
XR
)
an
a
ly
s
is
,
h
as
s
ev
er
al
lim
itatio
n
s
.
Ph
y
s
ical
ex
am
in
atio
n
,
cu
ltu
r
es
f
r
o
m
s
p
u
tu
m
s
am
p
les,
an
d
C
XR
an
aly
s
is
ar
e
b
en
ef
icial;
h
o
wev
er
,
th
ey
ar
e
ass
o
ciate
d
with
v
a
r
i
o
u
s
lim
itatio
n
s
,
as
p
o
in
ted
o
u
t
as
f
o
llo
ws
.
Ph
y
s
ical
ex
am
in
atio
n
is
n
o
t
o
n
ly
im
p
r
ec
is
e
b
u
t
is
also
h
ig
h
ly
d
ep
en
d
en
t
o
n
t
h
e
s
k
ill
o
f
th
e
p
er
s
o
n
wh
o
is
ca
r
r
y
in
g
o
u
t
th
e
ex
am
in
atio
n
.
T
h
e
an
aly
s
is
o
f
th
e
ca
u
s
ativ
e
ag
en
t
o
f
p
n
eu
m
o
n
ia
is
m
o
s
t
ac
cu
r
ate
if
it
is
d
o
n
e
f
r
o
m
th
e
cu
lt
u
r
es
o
f
th
e
s
am
p
les
o
b
tain
ed
f
r
o
m
th
e
p
er
s
o
n
’
s
s
p
u
tu
m
.
T
h
is
p
r
o
ce
s
s
m
ay
co
n
s
u
m
e
a
lo
t
o
f
tim
e
an
d
m
ay
also
p
r
o
d
u
ce
r
esu
lts
th
at
ar
e
in
s
u
f
f
icien
t
o
r
m
ay
n
o
t
clea
r
ly
r
ev
ea
l
ac
cu
r
ate
in
te
r
p
r
etatio
n
s
o
n
ac
co
u
n
t
o
f
t
h
e
s
am
p
les
b
ec
o
m
in
g
co
n
tam
in
ated
o
r
in
ad
eq
u
ate,
a
s
p
o
in
ted
o
u
t in
[
3
]
.
C
XR
an
aly
s
is
is
an
o
th
er
co
n
v
en
tio
n
al
m
eth
o
d
u
s
ed
f
o
r
d
iag
n
o
s
in
g
th
is
d
is
ea
s
e.
Su
ch
an
aly
s
is
is
co
m
m
o
n
; h
o
wev
er
,
th
e
s
k
ill lev
el
o
f
th
e
r
ad
io
lo
g
is
t w
h
o
is
u
n
d
er
tak
in
g
th
e
an
aly
s
is
o
f
th
ese
im
ag
es is
h
ea
v
ily
d
ep
en
d
e
n
t
o
n
h
is
o
r
h
e
r
ex
p
e
r
tis
e,
as
p
o
in
ted
o
u
t
in
[
4
]
.
C
o
n
s
id
er
in
g
th
ese
c
h
allen
g
es
a
s
s
o
ciate
d
with
th
e
co
n
v
en
tio
n
al
d
iag
n
o
s
tic
to
o
ls
u
s
ed
f
o
r
th
ese
im
ag
es,
th
er
e
is
th
er
ef
o
r
e
an
u
r
g
en
t
n
ee
d
f
o
r
to
o
ls
th
at
will
in
cr
ea
s
e
th
e
d
iag
n
o
s
tic
lev
els
o
f
p
n
e
u
m
o
n
ia
with
ac
cu
r
ac
y
,
as
in
d
icate
d
in
[
5
]
.
T
h
e
ab
ilit
y
o
f
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs),
s
p
e
cif
ically
,
to
ex
tr
ac
t
co
m
p
lex
f
ea
tu
r
es
f
r
o
m
d
ata
cr
ea
tes
an
e
s
s
en
tial
co
n
n
ec
tio
n
f
o
r
th
e
r
o
le
o
f
d
ee
p
lear
n
in
g
f
o
r
an
al
y
s
is
o
f
im
ag
es.
I
n
th
e
asp
ec
t
o
f
d
iag
n
o
s
in
g
an
d
clas
s
if
y
in
g
p
u
s
-
in
f
ec
ted
co
n
d
itio
n
s
f
r
o
m
im
ag
e
an
al
y
s
is
,
th
ese
n
etwo
r
k
s
h
av
e
d
em
o
n
s
tr
ated
r
em
a
r
k
ab
le
ac
cu
r
ac
y
,
with
s
o
m
e
b
en
ch
m
ar
k
s
b
ea
tin
g
h
u
m
a
n
d
iag
n
o
s
tic
ac
cu
r
ac
y
f
o
r
th
is
p
r
o
b
lem
y
et
ag
ain
.
T
h
e
u
s
e
o
f
d
ee
p
lear
n
in
g
to
r
ad
io
g
r
a
p
h
y
h
as
p
r
o
g
r
ess
ed
s
wif
tly
,
s
tar
tin
g
with
ea
r
ly
s
tu
d
ies
em
p
lo
y
in
g
co
n
v
en
tio
n
a
l
m
ac
h
in
e
lear
n
in
g
m
eth
o
d
s
an
d
co
n
tin
u
in
g
with
m
o
r
e
r
ec
en
t
d
e
v
elo
p
m
e
n
ts
in
C
NNs
th
at
allo
w
au
to
m
atic
a
n
d
h
ig
h
l
y
ac
cu
r
ate
id
en
tific
atio
n
o
f
a
r
an
g
e
o
f
illn
ess
es,
in
clu
d
in
g
p
n
eu
m
o
n
ia.
Un
f
o
r
tu
n
atel
y
,
th
e
p
r
ac
tical
u
tili
ty
o
f
m
an
y
ex
is
tin
g
m
o
d
els
is
lim
ited
b
ec
au
s
e
t
h
ey
d
o
n
o
t
d
if
f
er
en
tiate
b
etwe
en
d
if
f
e
r
en
t
k
in
d
s
o
f
p
n
eu
m
o
n
ia;
in
s
tead
,
th
ey
r
ely
o
n
b
in
a
r
y
ca
teg
o
r
izatio
n
(
p
n
eu
m
o
n
i
a
v
s
.
n
o
p
n
eu
m
o
n
ia)
.
T
h
is
wo
r
k
p
r
o
v
id
es
a
n
o
v
el
d
ee
p
lear
n
in
g
m
eth
o
d
to
b
r
i
d
g
e
th
ese
g
a
p
s
b
y
d
ev
elo
p
in
g
a
m
u
lticlas
s
cla
s
s
if
icatio
n
m
o
d
el
th
at
ca
n
ac
cu
r
ately
d
etec
t
an
d
d
is
tin
g
u
is
h
b
etwe
en
"
n
o
r
m
al,
"
"
b
ac
ter
ial
p
n
eu
m
o
n
ia,
"
a
n
d
"
v
ir
al
p
n
e
u
m
o
n
ia"
f
r
o
m
C
XR
p
h
o
to
s
.
Ou
r
m
o
d
el
is
b
ased
o
n
a
n
o
p
tim
ized
C
NN
ar
ch
itectu
r
e
an
d
u
s
es
tr
an
s
f
er
lear
n
in
g
tech
n
iq
u
es
to
im
p
r
o
v
e
f
ea
tu
r
e
ex
tr
ac
ti
o
n
ac
cu
r
a
cy
[
6
]
,
[
7
]
.
A
d
ee
p
lear
n
in
g
n
etwo
r
k
ca
lled
C
h
e
XNe
t
was
d
ev
elo
p
ed
in
2
0
1
7
b
y
R
ajp
u
r
k
ar
et
a
l.
[
8
]
.
I
t
was
ab
le
to
id
en
tify
p
n
eu
m
o
n
ia
f
r
o
m
C
XR
s
with
p
er
f
o
r
m
a
n
ce
co
m
p
ar
a
b
le
to
th
at
o
f
r
ad
io
lo
g
is
ts
[
9
]
,
[
1
0
]
.
Un
lik
e
ex
is
tin
g
m
o
d
els,
o
u
r
ap
p
r
o
ac
h
m
a
k
es
ex
ten
s
iv
e
u
s
e
o
f
d
ata
au
g
m
en
tatio
n
to
en
h
an
ce
r
o
b
u
s
tn
ess
an
d
g
en
er
aliza
b
ilit
y
ac
r
o
s
s
v
ar
io
u
s
p
atien
t
p
o
p
u
la
tio
n
s
.
Fu
r
th
er
m
o
r
e,
we
h
a
v
e
d
ev
elo
p
e
d
a
u
s
er
-
f
r
ien
d
ly
o
n
l
in
e
in
ter
f
ac
e
with
Stre
am
lit
to
f
ac
ilit
ate
u
s
e
in
clin
ical
s
ettin
g
s
.
T
h
is
en
ab
le
s
h
ea
lth
ca
r
e
p
r
o
v
id
e
r
s
to
em
p
lo
y
tech
n
o
lo
g
y
t
o
q
u
ick
ly
an
d
au
t
o
m
atica
lly
id
en
tify
p
n
eu
m
o
n
ia,
p
ar
ticu
la
r
ly
in
en
v
ir
o
n
m
e
n
ts
with
lim
ited
r
eso
u
r
ce
s
.
T
h
e
k
e
y
co
n
tr
ib
u
tio
n
s
o
f
th
is
wo
r
k
ar
e
th
e
d
ev
elo
p
m
en
t
o
f
a
r
elia
b
le,
m
u
lticlas
s
d
ee
p
lear
n
in
g
m
o
d
el
with
h
i
g
h
p
n
eu
m
o
n
ia
d
iag
n
o
s
is
ac
cu
r
ac
y
,
th
e
p
u
b
licatio
n
o
f
an
in
tu
it
iv
e
clin
ical
ap
p
licatio
n
,
an
d
t
h
e
v
alid
atio
n
o
f
th
e
m
o
d
el'
s
p
er
f
o
r
m
a
n
ce
u
s
in
g
a
wid
e
v
ar
iety
o
f
m
etr
ics
[
1
1
]
.
T
h
is
tech
n
iq
u
e
aim
s
to
clo
s
e
th
e
g
ap
b
etwe
en
clin
ical
n
ee
d
s
an
d
tech
n
ical
a
d
v
an
ce
m
e
n
ts
b
y
f
u
s
in
g
d
ee
p
l
ea
r
n
in
g
with
a
u
s
ef
u
l,
r
e
al
-
wo
r
ld
ap
p
licatio
n
.
T
h
is
will p
r
o
v
id
e
a
s
ca
lab
le
m
eth
o
d
to
im
p
r
o
v
e
p
n
eu
m
o
n
ia
d
iag
n
o
s
is
an
d
,
in
tu
r
n
,
im
p
r
o
v
e
p
atie
n
t p
r
o
g
n
o
s
is
.
Pn
eu
m
o
n
ia
is
a
lead
i
n
g
ca
u
s
e
o
f
h
o
s
p
italizatio
n
an
d
d
ea
th
wo
r
ld
wid
e.
Alth
o
u
g
h
C
XR
r
em
ain
s
th
e
m
o
s
t
wid
ely
a
v
ailab
le
im
a
g
in
g
m
o
d
ality
f
o
r
tr
iag
e
an
d
f
o
llo
w
-
u
p
,
r
ad
io
g
r
ap
h
ic
s
ig
n
s
o
f
p
n
eu
m
o
n
ia
ar
e
s
u
b
tle
an
d
ca
n
o
v
e
r
lap
with
o
th
er
th
o
r
ac
ic
co
n
d
itio
n
s
,
lead
in
g
to
s
ig
n
if
ican
t
in
ter
-
an
d
in
tr
a
-
r
ea
d
er
v
ar
iab
ilit
y
.
Ov
er
th
e
p
ast
d
ec
ad
e,
d
ee
p
co
n
v
o
lu
tio
n
al
n
etwo
r
k
s
h
av
e
s
h
o
wn
s
tate
-
of
-
th
e
-
a
r
t
p
er
f
o
r
m
an
ce
o
n
C
XR
clas
s
if
icatio
n
task
s
,
b
u
t
th
r
ee
p
er
s
is
ten
t
ch
allen
g
es
r
em
ain
:
i)
g
e
n
er
aliza
tio
n
to
h
eter
o
g
en
eo
u
s
ac
q
u
is
itio
n
co
n
d
itio
n
s
,
ii)
class
o
v
er
lap
b
etwe
en
v
ir
a
l
an
d
b
ac
ter
ial
p
r
esen
tatio
n
s
,
an
d
iii)
ca
lib
r
atio
n
a
n
d
d
ec
is
io
n
b
o
u
n
d
ar
ies
f
o
r
r
eliab
le
d
ep
lo
y
m
en
t.
T
h
is
p
ap
er
p
r
o
p
o
s
es
n
e
u
r
al
n
etwo
r
k
(
NN)
-
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
,
a
h
y
b
r
i
d
p
ip
elin
e
th
at
u
s
es
a
tr
an
s
f
er
-
lear
n
ed
C
NN
as
a
f
ea
tu
r
e
ex
tr
ac
to
r
an
d
a
SVM
as
th
e
f
in
al
d
ec
is
io
n
la
y
er
.
T
h
e
in
tu
itio
n
is
s
tr
aig
h
tf
o
r
war
d
:
C
NNs
lear
n
p
o
wer
f
u
l,
tr
an
s
latio
n
-
e
q
u
iv
a
r
ian
t
r
ep
r
esen
tatio
n
s
.
SVMs
p
r
o
v
id
e
lar
g
e
-
m
a
r
g
i
n
s
ep
ar
atio
n
a
n
d
g
o
o
d
ca
lib
r
atio
n
with
lim
ited
d
ata,
esp
ec
ially
in
m
u
lti
-
class
,
m
o
d
e
r
ately
im
b
alan
ce
d
r
eg
im
es.
W
e
p
r
esen
t
NN
-
SVM,
a
s
im
p
le,
r
ep
r
o
d
u
ci
b
le
h
y
b
r
id
p
i
p
e
lin
e
th
at
u
s
es
tr
an
s
f
er
-
lear
n
e
d
C
NN
em
b
ed
d
in
g
s
with
a
r
a
d
ial
b
asis
f
u
n
ctio
n
(
R
B
F
)
-
SVM
h
ea
d
f
o
r
t
h
r
ee
-
cl
ass
p
n
eu
m
o
n
ia
d
etec
tio
n
(
n
o
r
m
al,
b
ac
ter
ial,
an
d
v
ir
al)
f
r
o
m
C
XR
s
.
A
co
n
s
er
v
ativ
e
f
in
e
-
tu
n
in
g
s
tr
ateg
y
with
s
tr
o
n
g
r
eg
u
lar
izatio
n
an
d
class
-
awa
r
e
au
g
m
en
tatio
n
cu
r
b
s
o
v
er
f
itti
n
g
.
I
n
a
co
m
p
r
eh
en
s
iv
e
e
v
alu
a
tio
n
—
in
clu
d
in
g
p
er
-
class
m
etr
ics
an
d
co
n
f
u
s
io
n
m
atr
ices
—
NN
-
SVM
ac
h
iev
es
9
7
.
4
6
%
ac
cu
r
ac
y
with
co
m
p
e
titi
v
e
m
ac
r
o
-
F1
.
Ab
latio
n
ex
p
er
im
en
ts
is
o
late
th
e
co
n
tr
ib
u
tio
n
o
f
th
e
class
if
ier
h
ea
d
(
So
f
tMa
x
v
s
SVM)
,
d
ata
a
u
g
m
en
tatio
n
,
an
d
f
in
e
-
tu
n
in
g
d
ep
th
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
NN
-
S
V
M:
a
h
yb
r
id
n
eu
r
a
l n
et
w
o
r
k
–
s
u
p
p
o
r
t v
ec
to
r
ma
ch
in
e
fr
a
mewo
r
k
fo
r
…
(
S
a
n
to
s
h
K
u
ma
r
Ja
n
ka
tti)
1351
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
R
ec
en
t
ad
v
an
ce
s
in
d
ee
p
lear
n
in
g
h
av
e
tr
em
e
n
d
o
u
s
ly
b
en
ef
it
ted
m
ed
ical
im
ag
in
g
,
p
a
r
ticu
la
r
ly
in
th
e
ar
ea
o
f
p
n
e
u
m
o
n
ia
d
etec
tio
n
f
r
o
m
C
XR
im
ag
es.
C
NN
s
ar
e
f
r
eq
u
en
tly
e
m
p
lo
y
e
d
in
r
es
ea
r
ch
to
d
o
b
in
a
r
y
class
if
icatio
n
,
wh
ich
d
is
tin
g
u
i
s
h
es
b
etwe
en
lu
n
g
s
th
at
a
r
e
h
ea
lth
y
an
d
th
o
s
e
th
at
h
av
e
p
n
eu
m
o
n
ia.
T
h
e
y
u
s
e
d
o
v
e
r
1
0
0
,
0
0
0
f
r
o
n
tal
-
v
i
ew
X
-
r
ay
i
m
a
g
es
f
r
o
m
t
h
e
C
h
estX
-
r
a
y
1
4
d
at
aset
,
a
s
u
b
s
ta
n
t
ial
p
u
b
l
ic
d
a
tase
t,
t
o
tr
ai
n
th
e
ir
De
n
s
eNe
t
a
r
c
h
i
te
ct
u
r
e
m
o
d
el.
C
h
eXN
et
d
em
o
n
s
t
r
at
ed
s
i
g
n
if
ica
n
t
p
r
o
m
is
e
wit
h
its
h
i
g
h
F
1
-
s
c
o
r
e
,
s
u
g
g
esti
n
g
th
at
d
ee
p
l
ea
r
n
in
g
co
u
l
d
b
e
a
h
el
p
f
u
l
to
o
l
f
o
r
a
u
t
o
m
ati
n
g
p
n
e
u
m
o
n
ia
d
ia
g
n
o
s
is
.
H
o
we
v
er
,
b
ec
au
s
e
th
e
s
t
u
d
y
m
ai
n
l
y
f
o
cu
s
ed
o
n
b
i
n
ar
y
ca
te
g
o
r
iza
ti
o
n
wi
th
o
u
t
d
is
ti
n
g
u
is
h
i
n
g
b
etw
ee
n
b
a
cte
r
i
al
a
n
d
v
i
r
al
p
n
e
u
m
o
n
ia
,
its
u
s
e
f
u
l
n
ess
f
o
r
tar
g
ete
d
tr
ea
t
m
e
n
t
was
c
o
n
s
tr
a
in
e
d
.
O
th
er
s
t
u
d
ies
h
av
e
a
tte
m
p
te
d
to
a
d
d
r
ess
t
h
e
class
i
f
i
ca
t
io
n
o
f
p
n
e
u
m
o
n
ia
k
i
n
d
s
u
s
i
n
g
m
o
r
e
a
d
v
a
n
ce
d
t
ec
h
n
i
q
u
es.
K
e
r
m
an
y
et
a
l
.
[
4
]
u
s
e
d
tr
an
s
f
er
lea
r
n
in
g
with
p
r
e
-
tr
ain
ed
n
etwo
r
k
s
s
u
ch
as
VGG1
6
an
d
R
es
Net
-
5
0
to
ca
teg
o
r
ize
C
XR
s
in
to
th
r
ee
g
r
o
u
p
s
:
n
o
r
m
al,
b
ac
ter
ial
p
n
e
u
m
o
n
ia,
an
d
v
ir
a
l
p
n
eu
m
o
n
ia.
T
h
e
m
o
d
el
was
ab
le
to
im
p
r
o
v
e
its
p
er
f
o
r
m
a
n
ce
o
n
th
e
s
p
ec
if
ic
aim
o
f
p
n
eu
m
o
n
ia
class
if
icati
o
n
with
th
e
u
s
e
o
f
lar
g
e
d
ata
s
ets
o
f
g
en
er
ic
p
h
o
to
s
an
d
tr
a
n
s
f
er
lear
n
in
g
.
T
h
e
s
tu
d
y
'
s
th
r
ee
g
r
o
u
p
s
s
h
o
wed
i
m
p
r
ess
iv
e
r
ates
o
f
class
if
icatio
n
ac
cu
r
ac
y
,
p
r
o
v
in
g
t
h
at
d
ee
p
lear
n
in
g
m
o
d
els
ca
n
b
e
ef
f
ec
tiv
ely
ad
ju
s
ted
f
o
r
m
o
r
e
p
r
ec
is
e
class
if
icatio
n
[
1
2
]
,
[
1
3
]
.
Desp
ite
th
ese
ad
v
an
ce
m
e
n
ts
,
th
er
e
a
r
e
s
till
is
s
u
es
with
th
e
d
ev
el
o
p
m
en
t
o
f
r
ea
d
ily
a
cc
ess
ib
le,
r
eliab
le,
an
d
ea
s
y
-
to
-
u
s
e
to
o
ls
f
o
r
th
e
d
iag
n
o
s
is
o
f
p
n
eu
m
o
n
ia
in
r
ea
l
-
wo
r
ld
clin
ical
s
ettin
g
s
.
Mo
s
t
ex
is
tin
g
m
o
d
els
ar
e
eith
er
to
o
co
m
p
le
x
to
b
e
o
f
an
y
u
s
e
wh
en
r
eso
u
r
ce
s
ar
e
s
ca
r
ce
o
r
t
h
ey
d
o
n
o
t
p
r
o
v
id
e
m
u
lticlas
s
class
if
icatio
n
,
wh
ich
is
cr
itica
l
f
o
r
g
u
id
in
g
tr
ea
tm
e
n
t
d
ec
is
io
n
s
[
8
]
,
[
1
4
]
.
T
h
ese
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es
ar
e
cr
itical
f
o
r
im
p
r
o
v
in
g
th
e
q
u
a
lity
o
f
t
h
e
tr
ain
i
n
g
d
ata,
m
in
i
m
izin
g
n
o
is
e,
an
d
en
s
u
r
i
n
g
t
h
at
th
e
m
o
d
el
lear
n
s
r
elev
an
t
p
atter
n
s
r
elate
d
to
p
n
eu
m
o
n
ia
d
iag
n
o
s
is
[
1
5
]
,
[
1
6
]
.
R
esizin
g
:
in
th
is
s
tu
d
y
,
all
C
XR
p
ictu
r
es
wer
e
r
ed
u
ce
d
t
o
a
co
n
s
tan
t
d
i
m
en
s
io
n
o
f
2
2
4
b
y
2
2
4
p
i
x
els
to
m
ee
t
th
e
in
p
u
t
s
ize
r
eq
u
ir
e
m
en
t
o
f
th
e
s
elec
ted
C
NN
ar
ch
itectu
r
e
[
1
6
]
–
[
1
9
]
.
Un
f
o
r
tu
n
atel
y
,
th
e
s
m
all
s
ize
an
d
r
an
d
o
m
n
atu
r
e
o
f
th
e
lab
e
led
m
ed
ical
d
atasets
cu
r
r
en
tly
av
ailab
le
ca
u
s
e
th
ese
m
o
d
els
to
o
v
er
f
it,
wh
ich
n
eg
ativ
ely
af
f
ec
ts
t
h
eir
g
en
e
r
aliza
b
ilit
y
ac
r
o
s
s
d
if
f
er
en
t
p
atien
t
g
r
o
u
p
s
.
Sev
er
al
ac
ad
em
ics
h
av
e
also
s
tu
d
ied
th
e
u
s
e
o
f
e
n
s
em
b
le
m
eth
o
d
s
a
n
d
d
ata
a
u
g
m
en
tati
o
n
to
in
cr
e
ase
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
an
d
r
o
b
u
s
tn
ess
.
I
n
ce
p
tio
n
V3
,
R
esNet
-
5
0
,
an
d
Xce
p
tio
n
ar
e
ju
s
t
a
f
ew
o
f
th
e
C
NN
ar
ch
itectu
r
es
th
at
in
th
e
s
tu
d
ies
[
2
0
]
,
[
2
1
]
.
Pair
ed
with
a
v
ar
iety
o
f
d
ata
au
g
m
en
tatio
n
m
eth
o
d
s
as
f
lip
p
in
g
,
r
o
tatio
n
,
a
n
d
co
n
tr
ast
m
o
d
if
i
ca
tio
n
s
.
B
y
u
s
in
g
th
is
tech
n
iq
u
e,
th
e
m
o
d
el'
s
f
it
f
o
r
v
ar
io
u
s
C
XR
s
—
f
r
o
m
in
d
iv
id
u
als
with
v
ar
io
u
s
clin
ical
co
n
d
itio
n
s
an
d
d
em
o
g
r
ap
h
ics,
f
o
r
ex
am
p
le
—
was
im
p
r
o
v
ed
.
Mo
r
e
c
o
m
p
lex
en
s
em
b
le
m
o
d
els
m
ay
r
esu
lt
i
n
lo
n
g
er
i
n
f
er
en
ce
tim
es
an
d
m
o
r
e
c
o
s
tly
p
r
o
ce
s
s
in
g
.
B
ec
au
s
e
o
f
t
h
is
,
th
ey
ar
e
less
ap
p
r
o
p
r
iate
f
o
r
u
s
e
in
r
ea
l
-
tim
e
clin
ical
ap
p
licatio
n
s
,
p
ar
ticu
lar
ly
wh
en
r
eso
u
r
ce
s
ar
e
s
ca
r
ce
[
2
2
]
,
[
2
3
]
.
T
h
is
wo
r
k
im
p
r
o
v
es
o
n
p
r
ev
io
u
s
r
esear
ch
b
y
co
m
b
in
i
n
g
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
with
an
o
n
lin
e
a
p
p
licatio
n
in
ter
f
ac
e
to
g
iv
e
h
ea
lth
ca
r
e
p
r
o
f
ess
io
n
als
a
m
o
r
e
u
s
er
-
f
r
ien
d
ly
,
s
im
p
ler
m
u
lticlas
s
ca
teg
o
r
izatio
n
m
o
d
el.
Ou
r
a
p
p
r
o
ac
h
f
o
c
u
s
es
o
n
th
e
r
o
b
u
s
tn
e
s
s
,
clin
ical
in
teg
r
atio
n
,
an
d
r
e
al
-
tim
e
d
ep
lo
y
m
e
n
t
o
f
th
e
m
o
d
el
to
p
r
o
v
id
e
a
m
o
r
e
co
m
p
r
eh
en
s
iv
e
s
o
lu
tio
n
to
th
e
cu
r
r
en
t
p
r
o
b
lem
s
ass
o
ciate
d
with
p
n
eu
m
o
n
ia
d
iag
n
o
s
is
.
C
lass
ical
C
X
R
an
aly
s
is
u
s
ed
en
g
in
ee
r
ed
tex
t
u
r
e/
s
h
ap
e
d
escr
ip
to
r
s
(
e.
g
.
,
g
r
ay
l
ev
el
co
-
o
cc
u
r
r
en
ce
m
atr
ix
(
GL
C
M
)
an
d
h
is
to
g
r
am
o
f
o
r
ie
n
ted
g
r
ad
ien
ts
(
HOG
)
)
with
SVM
o
r
r
an
d
o
m
f
o
r
ests
.
Mo
d
er
n
ap
p
r
o
ac
h
es
lev
er
a
g
e
C
NNs
(
e.
g
.
,
VGG,
R
esNet,
Den
s
eNe
t,
an
d
E
f
f
icien
tNet)
with
tr
a
n
s
f
er
lear
n
in
g
f
r
o
m
n
atu
r
al
im
ag
es,
ac
h
iev
in
g
s
tr
o
n
g
r
esu
lts
o
n
m
u
lti
-
lab
el
th
o
r
a
cic
d
is
ea
s
e
d
etec
tio
n
.
Hy
b
r
id
m
eth
o
d
s
th
at
co
u
p
le
d
ee
p
f
ea
tu
r
es
with
m
ar
g
in
-
b
a
s
ed
class
if
ier
s
(
SVM
an
d
lo
g
is
tic
r
eg
r
ess
io
n
)
h
av
e
b
ee
n
r
e
p
o
r
ted
to
im
p
r
o
v
e
r
o
b
u
s
tn
ess
an
d
ca
lib
r
atio
n
o
n
s
m
all
to
m
ed
iu
m
-
s
ized
m
ed
ical
d
atasets
.
Ou
r
wo
r
k
f
o
llo
ws
th
is
h
y
b
r
id
lin
e,
f
o
cu
s
in
g
s
p
ec
if
ically
o
n
v
ir
a
l
v
s
b
ac
ter
ial
v
s
n
o
r
m
al
tr
i
-
class
d
is
cr
im
in
atio
n
an
d
d
etailin
g
a
co
m
p
ac
t,
r
ep
r
o
d
u
cib
le
p
i
p
elin
e.
3.
M
E
T
H
O
D
T
h
e
d
ata
u
s
ed
i
n
th
is
p
r
o
ject
i
s
r
ef
er
r
ed
to
as
“RS
NA
Pn
eu
m
o
n
ia
Dete
ctio
n
C
h
allen
g
e
D
ataset”
an
d
co
n
s
is
ts
o
f
a
to
tal
o
f
2
6
,
6
8
4
f
r
o
n
tal
ch
est
im
ag
es
o
f
p
atien
t
s
with
p
n
eu
m
o
n
ia
u
s
in
g
C
XR
s
.
T
h
e
im
ag
es
h
av
e
b
ee
n
class
if
ied
in
to
t
h
r
ee
s
et
s
:
“n
o
r
m
al”
with
a
to
tal
o
f
5
,
8
5
6
im
ag
es,
“v
ir
al
p
n
e
u
m
o
n
ia”
with
a
to
tal
o
f
1
3
,
6
7
0
im
ag
es,
an
d
“b
ac
ter
ial
p
n
eu
m
o
n
ia”
with
a
to
tal
o
f
7
,
1
5
8
im
ag
es.
T
h
e
p
r
esen
ce
o
f
e
q
u
al
p
r
o
p
o
r
tio
n
s
o
f
im
ag
es
in
ea
ch
class
is
an
in
d
icatio
n
o
f
h
o
w
r
elev
a
n
t
it
is
f
o
r
a
f
air
m
u
lticlas
s
cla
s
s
if
ie
r
to
h
av
e
an
eq
u
al
n
u
m
b
er
o
f
im
a
g
es in
ea
ch
clas
s
f
o
r
h
ig
h
er
ef
f
icie
n
cy
in
its
o
p
er
atio
n
s
.
T
h
e
im
ag
es
u
s
ed
in
th
is
d
atase
t
wer
e
ca
r
ef
u
lly
o
b
tain
ed
u
s
in
g
co
n
v
en
tio
n
al
d
ig
ital
im
ag
e
a
cq
u
is
itio
n
eq
u
ip
m
en
t
to
en
s
u
r
e
b
o
th
h
ig
h
im
ag
e
q
u
ality
an
d
ef
f
icien
t
p
r
o
ce
s
s
in
g
th
r
o
u
g
h
o
u
t
th
e
d
ata
co
llectio
n
p
h
ase.
All
ex
p
er
im
en
tal
d
ata
s
tr
ictly
f
o
llo
ws
estab
lis
h
ed
eth
ical
g
u
id
elin
es
s
p
ec
if
ically
d
esig
n
ed
to
p
r
o
tect
an
d
s
af
eg
u
ar
d
p
atien
ts
'
p
r
iv
ac
y
r
i
g
h
ts
co
m
p
r
eh
en
s
iv
ely
.
I
m
a
g
e
s
h
av
e
b
ee
n
f
u
lly
d
e
-
id
en
tifie
d
th
r
o
u
g
h
r
ig
o
r
o
u
s
an
o
n
y
m
izatio
n
p
r
o
ce
d
u
r
es,
w
ith
in
f
o
r
m
ed
co
n
s
en
t
p
r
o
p
er
l
y
o
b
tain
ed
f
r
o
m
ea
ch
in
d
i
v
id
u
al
p
atien
t
o
r
th
eir
leg
al
g
u
ar
d
ia
n
ac
co
r
d
in
g
t
o
all
ap
p
licab
le
in
s
titu
tio
n
al
r
ev
ie
w
b
o
ar
d
(
I
R
B
)
r
eq
u
ir
em
e
n
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
3
4
9
-
1
3
6
1
1352
3
.
1
.
Da
t
a
a
nd
t
a
s
k
def
ini
t
io
n
W
e
co
n
s
id
er
a
th
r
ee
-
class
C
XR
clas
s
if
icatio
n
p
r
o
b
lem
:
n
o
r
m
al,
b
ac
ter
ial
p
n
eu
m
o
n
ia,
an
d
v
ir
al
p
n
eu
m
o
n
ia.
A
p
u
b
licly
a
v
ailab
le,
d
e
-
id
en
tifie
d
d
atase
t
o
f
p
o
s
ter
o
an
ter
io
r
/an
ter
o
p
o
s
ter
io
r
(
PA/AP
)
r
ad
io
g
r
a
p
h
s
with
p
atien
t
-
lev
el
lab
els
was
u
s
ed
.
I
m
ag
es
ar
e
p
ar
titi
o
n
ed
p
atien
t
-
wis
e
in
to
tr
ain
/v
alid
atio
n
/tes
t
s
p
lits
to
av
o
id
leak
ag
e.
C
lass
co
u
n
ts
ar
e
b
ala
n
ce
d
o
n
th
e
tr
ain
in
g
s
et.
3
.
2
.
P
re
pro
ce
s
s
ing
a
nd
a
ug
m
ent
a
t
io
n
I
m
ag
es
ar
e
s
tan
d
ar
d
ized
b
ef
o
r
e
tr
ain
in
g
t
h
r
o
u
g
h
r
esizin
g
to
2
2
4
×
2
2
4
an
d
i
n
ten
s
ity
n
o
r
m
aliza
tio
n
.
Du
r
in
g
tr
ain
i
n
g
,
lig
h
t
au
g
m
en
tatio
n
s
—
s
u
ch
as
s
m
all
r
o
tatio
n
s
,
f
lip
s
,
b
r
ig
h
tn
ess
/co
n
tr
ast
ad
ju
s
tm
en
t,
an
d
m
ild
Gau
s
s
ian
n
o
is
e
—
ar
e
ap
p
lied
to
im
p
r
o
v
e
m
o
d
el
r
o
b
u
s
tn
e
s
s
an
d
r
ed
u
ce
o
v
er
f
itti
n
g
.
O
p
tio
n
al
lu
n
g
-
r
eg
io
n
cr
o
p
p
in
g
ca
n
b
e
in
clu
d
e
d
,
b
u
t th
e
p
ip
elin
e
is
d
esig
n
ed
t
o
o
p
e
r
ate
ef
f
ec
tiv
ely
o
n
f
u
ll
im
a
g
es
with
o
u
t
ad
d
itio
n
al
s
eg
m
en
tatio
n
s
tep
s
.
‒
R
esizin
g
to
2
2
4
×2
2
4
(
k
ee
p
s
a
s
p
ec
t v
ia
p
ad
-
r
esize
if
n
ee
d
ed
)
.
‒
No
r
m
aliza
tio
n
to
I
m
ag
eNe
t m
ea
n
/v
ar
ian
ce
(
o
r
d
ataset
-
s
p
ec
if
ic
s
tats
)
.
‒
Au
g
m
en
tatio
n
s
(
tr
ai
n
in
g
o
n
ly
)
:
s
m
all
r
o
tatio
n
s
(
±
7
°),
h
o
r
izo
n
tal
f
lip
s
,
r
an
d
o
m
b
r
ig
h
tn
ess
/co
n
tr
ast,
an
d
m
ild
Gau
s
s
ian
n
o
is
e.
‒
Op
tio
n
al:
lu
n
g
f
ield
cr
o
p
p
in
g
/s
eg
m
en
tatio
n
ca
n
b
e
a
d
d
ed
,
b
u
t
we
k
ee
p
th
e
p
ip
elin
e
s
in
g
le
-
s
tag
e
an
d
r
o
b
u
s
t to
f
u
ll
-
im
ag
e
v
ar
iatio
n
.
3
.
3
.
Co
nv
o
lutio
na
l neura
l net
wo
rk
ba
c
k
bo
ne
a
nd
f
ea
t
ur
e
ex
t
ra
ct
i
o
n
W
e
ad
o
p
t
a
m
o
d
er
n
I
m
a
g
eNe
t
-
p
r
etr
ain
ed
C
NN
(
e.
g
.
,
Den
s
eNe
t
-
1
2
1
o
r
E
f
f
icien
tNet
-
B
0
/B
3
)
.
T
h
e
class
if
icatio
n
h
ea
d
is
r
em
o
v
ed
;
we
ap
p
ly
g
lo
b
al
av
er
a
g
e
p
o
o
lin
g
(
GAP)
to
o
b
tain
a
f
ix
e
d
-
len
g
th
em
b
ed
d
in
g
(
e.
g
.
,
1
0
2
4
-
D
f
o
r
Den
s
eNe
t
-
1
2
1
)
.
W
e
f
in
e
-
tu
n
e
th
e
last
N
b
l
o
ck
s
(
N
∈
{1
,
2
}
)
with
a
lo
w
le
ar
n
in
g
r
ate
t
o
ad
ap
t
to
C
XR
tex
tu
r
e
s
tatis
tics
wh
ile
p
r
eser
v
in
g
g
en
e
r
al
f
ea
tu
r
es.
Data
p
r
ep
a
r
atio
n
is
a
c
r
itical
asp
ec
t
wh
en
it
co
m
es
to
tr
ain
in
g
th
e
d
ata
f
o
r
a
d
ee
p
lear
n
i
n
g
m
o
d
el
b
ec
au
s
e
it e
n
s
u
r
es th
at
th
e
d
at
a
is
r
ea
d
y
an
d
well
-
co
n
d
itio
n
e
d
f
o
r
o
p
tim
al
p
er
f
o
r
m
an
c
e.
T
h
e
s
tu
d
y
em
p
lo
y
ed
a
wid
e
r
an
g
e
o
f
d
ata
p
r
ep
ar
atio
n
m
eth
o
d
s
,
s
o
m
e
o
f
w
h
ich
in
clu
d
e
d
ata
au
g
m
en
ta
tio
n
,
s
ca
lin
g
,
an
d
n
o
r
m
aliza
tio
n
.
Data
a
u
g
m
en
ta
tio
n
is
v
er
y
cr
itical
b
ec
au
s
e
it
en
s
u
r
es th
at
th
e
tr
ain
in
g
d
ata
i
s
im
p
r
o
v
ed
t
h
r
o
u
g
h
ar
tific
ial
au
g
m
en
tatio
n
o
f
its
s
ize
u
s
in
g
m
eth
o
d
s
lik
e
r
o
tatio
n
,
f
lip
,
an
d
zo
o
m
tr
an
s
f
o
r
m
a
tio
n
o
f
th
e
o
r
ig
i
n
al
im
ag
es
o
f
X
-
r
ay
s
tak
en
f
o
r
a
n
aly
s
is
.
No
t
o
n
ly
is
d
ata
a
u
g
m
en
tatio
n
e
f
f
ec
tiv
e
in
a
d
d
r
ess
in
g
is
s
u
es
r
elate
d
to
o
v
er
f
itti
n
g
th
r
o
u
g
h
in
cr
ea
s
ed
v
ar
iab
ilit
y
in
im
a
g
es,
it
is
also
ef
f
ec
tiv
e
in
en
s
u
r
in
g
t
h
at
i
m
ag
es'
r
o
b
u
s
tn
ess
to
d
if
f
er
en
t in
p
u
t im
ag
e
co
n
d
itio
n
s
is
im
p
r
o
v
ed
.
T
h
e
p
r
o
p
o
s
e
d
tec
h
n
i
q
u
e
is
ef
f
ec
t
iv
e
as
it
g
e
n
e
r
a
tes
d
i
v
e
r
s
e
t
r
ai
n
i
n
g
s
am
p
l
es
t
h
r
o
u
g
h
d
ata
au
g
m
e
n
ta
ti
o
n
,
e
n
a
b
li
n
g
th
e
m
o
d
el
t
o
b
e
tte
r
ca
p
t
u
r
e
p
n
eu
m
o
n
ia
-
r
el
at
ed
f
ea
tu
r
es
u
n
d
er
v
a
r
y
in
g
co
n
d
it
io
n
s
.
T
h
is
en
h
a
n
ce
s
t
h
e
m
o
d
e
l’
s
a
b
ili
ty
t
o
g
e
n
e
r
a
liz
e
t
o
p
r
e
v
i
o
u
s
l
y
u
n
s
ee
n
d
at
a.
Ad
d
i
ti
o
n
all
y
,
p
r
e
p
r
o
ce
s
s
i
n
g
s
te
p
s
s
u
ch
as
s
ca
li
n
g
a
n
d
n
o
r
m
a
liz
ati
o
n
ar
e
a
p
p
li
ed
t
o
s
ta
n
d
a
r
d
iz
e
p
i
x
el
i
n
te
n
s
i
ty
v
a
lu
es
ac
r
o
s
s
all
i
m
a
g
es
.
No
r
m
al
iza
ti
o
n
tr
a
n
s
f
o
r
m
s
p
i
x
el
i
n
te
n
s
i
ties
t
o
a
c
o
m
m
o
n
r
a
n
g
e
o
f
0
t
o
1
,
e
n
s
u
r
i
n
g
u
n
i
f
o
r
m
f
ea
tu
r
e
r
e
p
r
ese
n
t
ati
o
n
an
d
f
a
cili
tat
in
g
f
ast
e
r
c
o
n
v
er
g
en
c
e
d
u
r
i
n
g
tr
ai
n
i
n
g
.
T
h
is
p
r
ev
e
n
ts
f
e
at
u
r
es
wit
h
la
r
g
e
r
n
u
m
er
i
ca
l
v
al
u
es
f
r
o
m
d
o
m
i
n
ati
n
g
t
h
e
l
ea
r
n
in
g
p
r
o
c
ess
a
n
d
im
p
r
o
v
es
th
e
s
t
ab
ilit
y
o
f
t
h
e
m
o
d
el
.
C
o
n
s
eq
u
en
tl
y
,
it
e
n
h
a
n
c
es
t
h
e
ef
f
ic
ie
n
c
y
o
f
t
h
e
b
ac
k
p
r
o
p
a
g
ati
o
n
al
g
o
r
it
h
m
a
n
d
r
e
d
u
ce
s
i
s
s
u
es
s
u
ch
as
v
a
n
is
h
i
n
g
o
r
e
x
p
l
o
d
in
g
g
r
a
d
i
en
ts
;
p
a
r
ti
c
u
la
r
l
y
in
d
ee
p
lea
r
n
i
n
g
m
o
d
els
li
k
e
C
NNs.
Ac
c
u
r
ate
f
ea
tu
r
e
e
x
t
r
ac
ti
o
n
is
cr
iti
ca
l
f
o
r
p
n
e
u
m
o
n
i
a
d
et
ec
ti
o
n
f
r
o
m
C
XR
im
a
g
es
.
P
r
o
p
er
n
o
r
m
al
iza
ti
o
n
e
n
a
b
l
es t
h
e
m
o
d
el
t
o
l
ea
r
n
s
u
b
t
le
v
a
r
i
ati
o
n
s
i
n
p
i
x
el
i
n
t
en
s
iti
es,
wh
ic
h
ar
e
ess
e
n
ti
al
f
o
r
d
is
ti
n
g
u
is
h
in
g
b
etw
ee
n
d
if
f
e
r
e
n
t
t
y
p
es
o
f
p
n
e
u
m
o
n
ia
,
t
h
e
r
e
b
y
im
p
r
o
v
in
g
b
o
t
h
a
cc
u
r
ac
y
a
n
d
r
o
b
u
s
t
n
ess
.
F
u
r
th
er
m
o
r
e,
all
C
XR
im
a
g
es
a
r
e
r
esi
ze
d
t
o
a
f
i
x
ed
d
i
m
e
n
s
i
o
n
o
f
2
2
4
×2
2
4
p
i
x
e
ls
t
o
m
e
et
t
h
e
in
p
u
t
r
e
q
u
i
r
e
m
e
n
ts
o
f
s
t
an
d
ar
d
C
N
N
ar
ch
ite
ct
u
r
es
a
n
d
e
n
s
u
r
e
co
n
s
is
t
en
c
y
d
u
r
i
n
g
t
r
ai
n
i
n
g
.
T
h
i
s
n
o
r
m
al
izi
n
g
o
f
t
h
e
in
p
u
t
f
ea
t
u
r
es
h
e
lp
s
th
e
m
o
d
e
l
l
ea
r
n
f
r
o
m
ea
ch
p
i
x
e
l
e
q
u
al
ly
a
n
d
p
r
ev
e
n
ts
f
ea
tu
r
es
wit
h
l
ar
g
e
v
al
u
es
f
r
o
m
o
v
e
r
s
h
ad
o
w
in
g
t
h
e
tr
ai
n
i
n
g
o
f
t
h
e
m
o
d
el
.
No
r
m
al
iz
ati
o
n
f
u
r
th
e
r
h
el
p
s
e
n
s
u
r
e
f
ast
er
co
n
v
e
r
g
e
n
ce
o
f
t
h
e
d
e
ep
lea
r
n
i
n
g
m
o
d
el
d
u
r
i
n
g
tr
ai
n
i
n
g
.
I
t
a
ls
o
h
e
lp
s
t
h
e
m
o
d
el
r
e
ac
h
its
o
p
tim
u
m
le
ar
n
i
n
g
r
at
e,
t
h
u
s
m
a
k
i
n
g
t
h
e
tr
a
in
in
g
p
r
o
ce
s
s
o
f
t
h
e
b
ac
k
p
r
o
p
a
g
at
io
n
al
g
o
r
it
h
m
m
o
r
e
ef
f
i
cie
n
t
a
n
d
f
ast
er
.
I
t
f
u
r
t
h
e
r
r
e
d
u
c
es
t
h
e
c
h
a
n
ce
s
o
f
th
e
g
r
a
d
i
en
ts
d
is
a
p
p
e
ar
i
n
g
o
r
b
l
o
wi
n
g
u
p
d
u
r
in
g
tr
ai
n
i
n
g
,
es
p
ec
i
all
y
d
u
r
i
n
g
t
h
e
tr
ai
n
i
n
g
o
f
d
e
e
p
le
ar
n
i
n
g
m
o
d
els
li
k
e
C
N
Ns.
D
u
r
i
n
g
d
ia
g
n
o
s
i
n
g
p
n
e
u
m
o
n
i
a
f
r
o
m
X
-
r
a
y
im
a
g
es
,
ac
c
u
r
ate
f
e
a
tu
r
e
e
x
t
r
ac
t
io
n
s
a
r
e
p
a
r
a
m
o
u
n
t
.
I
n
t
h
is
r
e
g
ar
d
,
ac
c
u
r
at
e
n
o
r
m
a
liz
ati
o
n
s
en
s
u
r
e
s
m
al
l
f
e
at
u
r
es,
w
h
i
ch
d
is
ti
n
g
u
is
h
v
i
r
a
l
an
d
b
ac
te
r
ia
l
p
n
e
u
m
o
n
ias
,
a
r
e
a
p
p
r
o
p
r
i
at
el
y
le
ar
n
e
d
,
th
u
s
im
p
r
o
v
i
n
g
th
e
a
cc
u
r
ac
y
a
n
d
r
o
b
u
s
t
n
ess
o
f
th
e
m
o
d
el.
R
esi
zi
n
g
:
in
th
is
w
o
r
k
,
a
ll
th
e
C
XR
im
ag
e
s
we
r
e
r
esiz
e
d
t
o
a
f
i
x
e
d
s
iz
e
o
f
2
2
4
p
ix
els
b
y
2
2
4
p
i
x
els
t
o
e
n
h
a
n
ce
t
h
e
i
n
p
u
t
r
e
q
u
i
r
e
m
e
n
t
f
o
r
t
h
e
ac
ce
p
te
d
C
NN
m
o
d
el
ar
c
h
it
ec
t
u
r
e
.
T
h
e
o
p
t
io
n
t
o
s
ca
l
e
t
h
e
i
m
a
g
es
s
t
an
d
ar
d
i
ze
s
t
h
e
i
n
p
u
t
to
th
e
m
o
d
el,
wh
ic
h
is
c
r
u
c
ial
f
o
r
d
e
ep
l
ea
r
n
i
n
g
f
r
am
ew
o
r
k
s
t
h
at
r
e
q
u
i
r
e
u
n
i
f
o
r
m
i
n
p
u
t
s
h
a
p
es
a
cr
o
s
s
al
l
s
am
p
les
.
T
h
is
s
c
ali
n
g
m
et
h
o
d
n
o
t
o
n
l
y
m
a
k
es
in
t
eg
r
ati
o
n
wit
h
t
h
e
C
NN
e
as
ier
,
b
u
t
it
als
o
h
el
p
s
t
o
k
e
ep
t
h
e
co
m
p
u
te
l
o
ad
r
ea
s
o
n
a
b
le
d
u
r
in
g
t
r
ai
n
i
n
g
a
n
d
in
f
e
r
e
n
c
e.
B
y
r
ed
u
c
in
g
th
e
s
i
ze
o
f
e
ac
h
X
-
r
a
y
p
i
ct
u
r
e
t
o
2
2
4
×
2
2
4
p
i
x
els,
t
h
e
m
o
d
el'
s
c
o
m
p
u
tati
o
n
al
c
o
m
p
l
e
x
it
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
NN
-
S
V
M:
a
h
yb
r
id
n
eu
r
a
l n
et
w
o
r
k
–
s
u
p
p
o
r
t v
ec
to
r
ma
ch
in
e
fr
a
mewo
r
k
fo
r
…
(
S
a
n
to
s
h
K
u
ma
r
Ja
n
ka
tti)
1353
is
d
r
a
m
at
ic
all
y
d
e
cr
ea
s
e
d
.
Hi
g
h
-
r
es
o
l
u
t
io
n
p
i
ct
u
r
es
i
n
cl
u
d
e
a
h
u
g
e
n
u
m
b
er
o
f
p
i
x
els
,
wh
i
ch
m
i
g
h
t
m
a
k
e
tr
a
in
in
g
ta
k
e
lo
n
g
e
r
an
d
n
ee
d
m
o
r
e
r
es
o
u
r
ce
s
.
H
o
we
v
er
,
s
c
ali
n
g
al
lo
ws
t
h
e
m
o
d
el
t
o
h
a
n
d
le
p
i
ct
u
r
es
m
o
r
e
ef
f
ic
ie
n
tl
y
w
h
i
le
m
ai
n
t
ai
n
i
n
g
i
t
s
c
ap
ac
it
y
to
lea
r
n
c
r
it
ica
l
c
h
a
r
ac
t
er
is
tics
.
T
h
is
d
e
cr
ea
s
e
i
n
c
o
m
p
u
t
ati
o
n
al
c
o
s
t
is
esp
ec
i
all
y
b
e
n
e
f
i
cia
l
wh
e
n
w
o
r
k
i
n
g
wi
th
b
i
g
d
at
ase
ts
,
s
u
c
h
as
t
h
e
R
S
NA
P
n
e
u
m
o
n
ia
Dete
cti
o
n
C
h
all
en
g
e
Data
s
et
,
b
e
ca
u
s
e
i
t a
ll
o
ws
t
h
e
m
o
d
el
to
t
r
a
in
q
u
ic
k
e
r
w
h
il
e
u
s
in
g
less
m
em
o
r
y
.
Da
ta
a
u
g
m
e
n
ta
ti
o
n
: a
v
ar
i
et
y
o
f
d
at
a
au
g
m
e
n
t
ati
o
n
t
ec
h
n
i
q
u
es,
in
c
lu
d
i
n
g
r
a
n
d
o
m
r
o
t
ati
o
n
s
,
f
l
i
p
s
,
z
o
o
m
s
,
b
r
ig
h
t
n
ess
m
o
d
i
f
i
c
ati
o
n
s
,
a
n
d
c
o
n
tr
ast
v
a
r
i
ati
o
n
s
,
we
r
e
u
s
ed
t
o
r
e
d
u
c
e
o
v
er
f
i
tti
n
g
a
n
d
en
h
a
n
ce
th
e
m
o
d
el'
s
ca
p
ac
it
y
f
o
r
g
e
n
e
r
a
liz
ati
o
n
.
B
y
s
i
m
u
l
ati
n
g
a
r
a
n
g
e
o
f
c
ir
cu
m
s
t
an
ce
s
,
in
c
lu
d
i
n
g
s
m
al
l
p
o
s
it
io
n
al
o
r
li
g
h
ti
n
g
c
h
a
n
g
es
,
t
h
ese
a
u
g
m
en
t
ati
o
n
s
b
r
o
ad
e
n
th
e
tr
a
in
in
g
s
et
a
n
d
b
o
o
s
t
t
h
e
d
u
r
ab
i
lit
y
o
f
t
h
e
m
o
d
e
l.
T
h
es
e
p
r
e
p
r
o
ce
s
s
i
n
g
s
t
r
ate
g
i
es
a
r
e
c
r
i
tica
l
t
o
e
n
h
an
ci
n
g
m
o
d
el
p
e
r
f
o
r
m
an
ce
b
ec
a
u
s
e
t
h
ey
i
n
cr
ea
s
e
t
h
e
e
f
f
e
cti
v
e
s
ize
o
f
t
h
e
t
r
a
in
in
g
s
et
a
n
d
e
n
s
u
r
e
t
h
at
t
h
e
m
o
d
el
l
ea
r
n
s
to
g
e
n
e
r
al
ize
t
o
n
ew
,
u
n
s
ee
n
e
v
e
n
ts
i
n
s
te
a
d
o
f
s
i
m
p
l
y
m
e
m
o
r
izi
n
g
t
h
e
t
r
a
in
in
g
d
at
a.
3
.
4
.
M
o
d
e
l
a
r
ch
i
t
ec
t
u
r
e
Fig
u
r
e
1
illu
s
tr
ates
a
s
y
s
tem
th
at
u
s
es
C
X
R
im
ag
es
an
d
a
p
r
e
-
tr
ain
e
d
C
NN
m
o
d
el
with
tr
an
s
f
er
lear
n
in
g
to
a
u
to
m
atica
lly
d
ia
g
n
o
s
e
v
ir
al
p
n
eu
m
o
n
ia,
b
ac
ter
ia
l p
n
eu
m
o
n
ia,
o
r
n
o
r
m
al
ca
s
es.
‒
I
n
p
u
t CXR
→
p
r
ep
r
o
ce
s
s
→
C
NN
→
em
b
ed
d
in
g
→
SVM
→
ca
lib
r
ated
p
r
o
b
ab
ilit
ies →
e
v
alu
atio
n
/QA.
‒
I
n
p
u
t CXR
(
DI
C
OM
/PNG,
1
×
H×
W
)
r
aw
ch
est X
-
r
ay
en
ter
s
th
e
s
y
s
tem
.
Fig
u
r
e
1
.
Ar
c
h
itectu
r
al
d
iag
r
a
m
o
f
p
n
eu
m
o
n
ia
class
if
icatio
n
Pre
p
r
o
ce
s
s
in
g
:
r
esize
to
2
2
4
×2
2
4
s
o
e
v
er
y
im
a
g
e
f
its
th
e
b
ac
k
b
o
n
e’
s
ex
p
ec
te
d
s
ize.
No
r
m
alize
in
ten
s
ities
to
s
tab
le
r
an
g
es
(
I
m
ag
eNe
t
m
ea
n
/s
td
o
r
d
ataset
s
tats
)
.
C
las
s
-
awa
r
e
au
g
m
e
n
tatio
n
(
tr
ain
o
n
ly
)
b
alan
ce
s
class
es
an
d
im
p
r
o
v
es
g
en
er
aliza
tio
n
(
m
ild
r
o
tatio
n
s
,
f
lip
s
,
an
d
co
n
tr
ast/
b
r
ig
h
tn
ess
jitt
er
)
.
C
NN
b
ac
k
b
o
n
e
(
p
r
etr
ain
ed
;
last
b
lo
ck
s
f
in
e
-
tu
n
e
d
)
,
a
m
o
d
er
n
C
NN
(
e.
g
.
,
Den
s
eNe
t
-
1
2
1
/Ef
f
icien
tNet)
ex
tr
ac
ts
r
ad
io
g
r
a
p
h
ic
tex
tu
r
e/f
ea
tu
r
es.
W
e
f
in
e
-
tu
n
e
o
n
ly
t
h
e
last
1
–
2
b
lo
ck
s
(
lo
w
L
R
)
t
o
ad
a
p
t
to
C
XR
s
tati
s
tics
with
o
u
t
o
v
er
f
itti
n
g
.
GAP
→
1
0
2
4
-
D
em
b
ed
d
i
n
g
GAP
co
llap
s
es
s
p
atial
m
ap
s
in
to
a
co
m
p
ac
t,
tr
an
s
latio
n
-
r
o
b
u
s
t
v
ec
to
r
(
≈
1
0
2
4
d
im
s
)
.
T
h
is
r
ed
u
ce
s
p
ar
am
eter
s
v
s
f
latten
in
g
an
d
m
a
k
es
th
e
f
ea
t
u
r
es
SVM
-
f
r
ien
d
ly
.
R
B
F
-
SVM
(
o
n
e
-
vs
-
r
est3
class
es)
On
th
e
1
0
2
4
-
D
em
b
ed
d
i
n
g
,
we
tr
ain
a
m
ar
g
in
-
b
ased
SVM
f
o
r
n
o
r
m
al,
b
ac
ter
ial,
an
d
v
ir
al
(
Ov
R
)
.
T
h
e
R
B
F
k
er
n
el
g
iv
es
n
o
n
-
lin
ea
r
d
ec
is
io
n
b
o
u
n
d
ar
ies
th
at
o
f
ten
s
ep
a
r
ate
o
v
er
lap
p
i
n
g
p
n
e
u
m
o
n
ia
p
atter
n
s
b
etter
t
h
an
a
lin
ea
r
o
r
So
f
tMa
x
h
ea
d
.
C
alib
r
atio
n
(
p
latt
s
ca
lin
g
)
SVM
s
co
r
e
s
(
d
is
tan
ce
s
to
m
ar
g
in
s
)
ar
e
m
a
p
p
ed
t
o
well
-
ca
lib
r
ated
class
p
r
o
b
ab
ilit
ies
v
ia
lo
g
is
tic
ca
lib
r
atio
n
o
n
a
v
alid
atio
n
s
p
lit.
T
h
is
h
elp
s
th
r
esh
o
ld
s
ettin
g
an
d
clin
ical
in
ter
p
r
eta
b
ilit
y
.
Sid
e
m
o
d
u
les (
f
ed
b
y
in
ter
m
ed
iate
tap
s
)
:
−
I
n
ter
p
r
etab
ilit
y
(
g
r
ad
ien
t
-
wei
g
h
ted
class
ac
tiv
atio
n
m
ap
p
i
n
g
(
g
r
a
d
-
C
AM
)
)
f
r
o
m
th
e
C
NN
f
ea
tu
r
es
to
s
h
o
w
wh
er
e
th
e
m
o
d
el
“lo
o
k
e
d
”
(
co
n
s
o
lid
atio
n
s
an
d
p
er
ih
il
ar
/p
er
ip
h
er
al
o
p
ac
ities
)
.
−
E
x
t
e
r
n
a
l
v
a
l
i
d
a
t
i
o
n
a
n
d
d
o
m
a
i
n
a
d
a
p
t
a
t
i
o
n
(
o
n
a
n
i
n
d
e
p
e
n
d
e
n
t
s
i
t
e
)
:
b
a
t
c
h
n
o
r
m
a
l
i
z
a
t
i
o
n
re
-
e
s
t
i
m
a
t
i
o
n
w
i
t
h
s
m
a
l
l
u
n
l
a
b
e
l
e
d
b
a
t
c
h
t
o
m
a
t
c
h
t
a
r
g
e
t
s
c
a
n
n
e
r
s
t
a
t
i
s
t
i
c
s
,
t
h
e
n
t
e
m
p
e
r
a
t
u
r
e
s
c
a
l
i
n
g
t
o
r
e
-
c
a
l
i
b
r
a
t
e
p
r
o
b
a
b
i
l
i
t
i
e
s
.
−
Me
tr
ics
an
d
s
tatis
tics
:
ac
cu
r
ac
y
,
p
r
e
cisi
o
n
,
r
ec
all,
F1
-
s
co
r
e
,
ar
ea
u
n
d
e
r
th
e
cu
r
v
e
(
AU
C
)
;
co
n
f
u
s
io
n
m
atr
ix
;
p
er
-
class
r
ec
eiv
er
o
p
er
atin
g
c
h
ar
ac
ter
is
tic
(
R
OC
)
;
s
ig
n
if
ican
ce
test
s
—
D
eL
o
n
g
(
AUC)
,
Mc
Nem
ar
(
p
air
ed
ac
cu
r
ac
y
)
,
a
n
d
p
air
e
d
b
o
o
ts
tr
ap
(
m
ac
r
o
-
F1
)
.
Deta
ile
d
m
o
d
e
l
(
la
y
e
r
-
by
-
la
y
e
r
v
iew
)
:
i
n
p
u
t
1
×
2
2
4
×2
2
4
→
C
NN
b
l
o
c
k
s
(
+Re
L
U/S
iLU
)
→
GA
P
→
B
atc
h
N
o
r
m
→
d
r
o
p
o
u
t
(
0
.
2
)
→
1
0
2
4
-
D
e
m
b
e
d
d
i
n
g
→
R
B
F
-
SV
M
→
p
la
tt
s
ca
li
n
g
→
p
r
o
b
a
b
i
lit
ies
(
n
o
r
m
al/
b
ac
t
er
ial
/v
ir
al
)
C
NN
b
l
o
c
k
s
c
o
n
v
o
l
u
ti
o
n
a
l
s
t
ag
es
le
ar
n
ed
g
es,
te
x
t
u
r
es
,
an
d
l
u
n
g
p
ar
en
ch
y
m
a
p
att
er
n
s
.
W
e
k
ee
p
ea
r
l
y
la
y
e
r
s
m
o
s
tl
y
f
r
o
ze
n
(
g
en
er
ic
f
ea
t
u
r
es
)
a
n
d
f
i
n
e
-
t
u
n
e
t
h
e
las
t
1
–
2
b
l
o
c
k
s
(
d
o
m
a
in
-
s
p
e
ci
f
ic
r
a
d
i
o
g
r
ap
h
i
c
c
u
es
)
.
GAP
(
t
o
1
0
2
4
-
D
)
c
o
n
v
er
ts
f
e
at
u
r
e
m
a
p
s
i
n
t
o
a
s
i
n
g
le
v
e
ct
o
r
p
e
r
c
h
a
n
n
el
,
e
n
c
o
u
r
a
g
i
n
g
g
l
o
b
al
,
s
p
atia
ll
y
-
aw
ar
e
s
u
m
m
a
r
ies
wi
th
o
u
t
a
h
u
g
e
FC
la
y
e
r
.
B
atc
h
N
o
r
m
(
o
n
em
b
e
d
d
i
n
g
)
s
t
ab
i
liz
es
th
e
f
ea
t
u
r
e
d
is
t
r
i
b
u
ti
o
n
t
h
e
SV
M
s
e
es;
i
m
p
r
o
v
es
tr
an
s
f
er
an
d
r
e
d
u
c
es
d
o
m
ai
n
s
h
i
f
t
s
e
n
s
i
ti
v
it
y
.
Dr
o
p
o
u
t
p
=
0
.
2
r
e
g
u
l
ar
izes
th
e
r
e
p
r
ese
n
t
ati
o
n
s
o
th
e
SV
M
d
o
esn
’
t
o
v
er
f
i
t
s
p
u
r
i
o
u
s
c
u
es
.
E
m
b
e
d
d
i
n
g
(
1
0
2
4
-
D)
T
h
e
d
e
f
i
n
it
iv
e
f
e
at
u
r
e
v
e
ct
o
r
s
a
v
e
d
f
o
r
th
e
SV
M
a
n
d
f
o
r
d
o
w
n
s
t
r
ea
m
an
al
y
s
is
(
e
.
g
.
,
t
-
SNE
p
lo
ts
a
n
d
e
r
r
o
r
a
n
aly
s
is
)
.
R
B
F
-
SVM
(
O
v
R
)
T
h
r
ee
b
i
n
a
r
y
SV
Ms
(
n
o
r
m
a
l
v
s
r
est,
b
a
cte
r
i
al
v
s
r
est
,
an
d
v
ir
al
v
s
r
est
)
.
R
B
F
t
y
p
ic
all
y
d
e
li
v
e
r
s
b
e
tte
r
m
a
r
g
in
s
th
an
a
li
n
e
ar
h
e
a
d
in
s
m
al
l/m
e
d
i
u
m
d
at
ase
ts
wh
er
e
class
es
o
v
e
r
l
ap
.
C
al
ib
r
a
ti
o
n
(
p
latt
)
f
o
r
e
ac
h
O
v
R
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
3
4
9
-
1
3
6
1
1354
class
i
f
i
er
,
l
ea
r
n
a
lo
g
is
tic
m
ap
p
i
n
g
o
n
th
e
v
al
id
ati
o
n
s
e
t;
co
m
b
in
e
a
n
d
n
o
r
m
a
liz
e
t
o
a
f
in
a
l
p
r
o
b
a
b
ili
ty
s
im
p
l
e
x
o
v
e
r
t
h
e
t
h
r
e
e
cl
ass
es
.
O
u
t
p
u
ts
a
n
d
h
o
o
k
s
p
r
o
b
a
b
i
liti
es
+
p
r
e
d
i
cte
d
la
b
el
f
ee
d
t
h
e
c
o
n
f
u
s
i
o
n
m
at
r
i
x
a
n
d
p
e
r
-
cl
ass
R
OC
;
i
n
te
r
m
e
d
i
ate
C
NN
f
ea
tu
r
es
a
r
e
u
s
e
d
to
g
e
n
e
r
a
te
g
r
a
d
-
C
AM
h
ea
t
m
a
p
s
(
a
li
g
h
tw
ei
g
h
t
li
n
ea
r
p
r
o
b
e
c
an
b
e
at
tac
h
ed
s
o
le
ly
f
o
r
ex
p
l
ai
n
a
b
l
e
a
r
t
if
ici
al
i
n
t
elli
g
e
n
ce
(
X
AI
)
—
u
s
e
d
f
o
r
v
i
s
u
al
iz
ati
o
n
,
n
o
t
f
o
r
d
e
cisi
o
n
s
)
.
I
m
a
g
es
a
r
e
r
es
ize
d
to
2
2
4
×
2
2
4
a
n
d
n
o
r
m
a
liz
e
d
.
D
u
r
i
n
g
t
r
a
in
in
g
,
li
g
h
t
a
u
g
m
e
n
ta
ti
o
n
s
—
s
u
ch
as
s
m
al
l
r
o
tat
io
n
s
,
f
li
p
s
,
b
r
i
g
h
tn
ess
/c
o
n
t
r
ast
a
d
j
u
s
t
m
e
n
t
,
a
n
d
G
au
s
s
ia
n
n
o
is
e
—
a
r
e
a
p
p
li
ed
t
o
im
p
r
o
v
e
g
e
n
e
r
a
liz
ati
o
n
.
T
h
e
p
i
p
e
li
n
e
r
e
m
ai
n
s
s
im
p
l
e
a
n
d
r
o
b
u
s
t
,
wit
h
o
p
ti
o
n
al
l
u
n
g
-
r
e
g
i
o
n
cr
o
p
p
in
g
if
n
ee
d
e
d
.
SVM
clas
s
if
ier
(
NN
-
SVM)
:
g
iv
en
em
b
e
d
d
in
g
s
{
}
=
1
an
d
l
ab
els
∈
{
1
,
2
,
3
}
,
we
tr
ain
a
one
-
vs
-
r
est SVM
with
R
B
F k
e
r
n
el
as (
1
)
,
w
h
ile
th
e
o
p
tim
izatio
n
o
b
jectiv
e
is
d
e
f
in
ed
in
(
2
)
.
(
,
_
)
=
(
−
∗
|
|
−
_
|
|
^
2
)
(
1
)
_
{
,
,
}
(
1
/
2
)
|
|
|
|
^
2
+
∗
_
{
=
1
}
_
(
2
)
Su
b
ject
to
:
_
(
^
ℎ
(
_
)
+
)
>
=
1
−
_
;
_
>
=
0
=
1
,
2
,
.
.
.
,
.
W
h
er
e
γ
is
k
er
n
el
p
a
r
am
eter
c
o
n
tr
o
llin
g
s
p
r
ea
d
; ξi
is
s
lack
v
ar
iab
les;
C
is
r
eg
u
lar
izatio
n
p
a
r
am
eter
; a
n
d
ϕ
(
x
)
is
f
ea
tu
r
e
m
ap
p
in
g
.
T
h
er
e
ar
e
th
r
ee
ty
p
es
o
f
C
XR
im
ag
e
p
ictu
r
es
th
at
ar
e
aim
ed
at
b
y
th
e
p
r
o
p
o
s
ed
m
o
d
el,
wh
ich
is
a
C
NN:
"
n
o
r
m
al,
"
"b
ac
ter
ial
p
n
eu
m
o
n
ia,
"
an
d
"v
ir
al
p
n
e
u
m
o
n
ia
.
"
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
o
f
th
e
m
o
d
el
is
s
u
b
d
iv
id
ed
i
n
to
v
ar
i
o
u
s
lev
els
th
at
p
lay
d
if
f
er
en
t
r
o
les.
T
h
e
lev
els
in
clu
d
e
co
n
v
o
lu
tio
n
al
l
ay
er
s
:
"th
ese
lay
er
s
p
lay
a
c
r
u
cial
r
o
le
in
th
e
id
e
n
tific
atio
n
o
f
p
n
e
u
m
o
n
ia
b
y
d
e
tectin
g
k
ey
d
etails
with
in
C
XR
im
ag
e
p
ictu
r
es
"
.
T
h
ese
lay
er
s
tak
e
th
e
in
p
u
t
p
h
o
to
s
an
d
ap
p
l
y
f
ilter
s
to
th
em
,
d
etec
tin
g
k
ey
d
etails
s
u
ch
as
ed
g
es
th
at
p
lay
a
p
iv
o
tal
r
o
le
in
im
a
g
e
id
e
n
tific
atio
n
.
E
v
e
r
y
f
ilter
p
er
f
o
r
m
s
a
co
n
v
o
lu
tio
n
p
r
o
ce
s
s
b
etwe
en
th
e
p
h
o
to
an
d
th
e
f
ilter
to
g
en
er
ate
f
ea
tu
r
e
m
ap
s
,
wh
ich
ar
e
th
en
s
u
b
jecte
d
to
n
o
n
-
lin
ea
r
f
u
n
ctio
n
s
s
u
ch
as
R
eL
U
to
in
d
u
ce
non
-
lin
ea
r
ity
am
o
n
g
th
em
.
T
h
e
s
p
atial
ar
r
an
g
em
en
t
o
f
f
ilter
s
o
v
er
th
e
im
ag
e
ca
n
b
e
in
f
lu
en
ce
d
b
y
th
e
s
tr
o
k
e
s
ize
an
d
p
ad
d
i
n
g
.
T
h
e
n
u
m
b
er
o
f
ad
d
itio
n
al
f
ilter
s
o
f
all
s
ize
s
in
cr
ea
s
es
with
th
e
d
ep
th
lev
e
l
o
f
th
e
n
etwo
r
k
to
ca
p
tu
r
e
p
r
o
g
r
ess
iv
ely
m
o
r
e
d
etailed
in
f
o
r
m
atio
n
.
T
h
e
c
o
n
v
o
lu
ti
o
n
al
lay
er
s
f
o
llo
w
p
o
o
lin
g
lay
er
s
th
at
s
u
p
p
r
ess
s
p
atial
d
etails
to
r
eta
in
k
ey
in
f
o
r
m
atio
n
o
n
ly
.
T
h
e
ef
f
icien
cy
o
f
th
e
m
o
d
el
in
cr
ea
s
es
with
it
s
ab
ilit
y
to
lear
n
s
ig
n
if
ican
t
d
etails th
at
r
elate
to
"
n
o
r
m
al,
"
"b
ac
ter
ial
p
n
eu
m
o
n
ia,
"
o
r
"v
ir
al
p
n
eu
m
o
n
ia,
"
in
f
ec
tio
n
s
.
T
h
e
f
o
llo
win
g
is
th
e
m
ath
em
atica
l
d
escr
ip
tio
n
o
f
t
h
e
co
n
v
o
lu
tio
n
p
r
o
ce
s
s
,
n
am
ely
p
o
o
li
n
g
lay
er
s
:
"a
n
ess
en
tial
co
m
p
o
n
e
n
t
o
f
C
NNs
is
th
e
p
o
o
lin
g
lay
e
r
s
u
s
ed
f
o
r
d
ec
r
ea
s
in
g
th
e
s
p
atial
s
ize
o
f
f
ea
t
u
r
e
m
a
p
s
with
th
e
p
r
eser
v
atio
n
o
f
k
ey
in
f
o
r
m
atio
n
,
th
u
s
in
cr
ea
s
in
g
ef
f
i
cien
cy
with
in
th
e
m
o
d
el
.
"
I
n
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
p
n
eu
m
o
n
ia
id
en
tific
atio
n
,
p
o
o
lin
g
la
y
er
s
ap
p
ly
tech
n
iq
u
es
s
u
ch
as
av
er
ag
e
o
r
m
a
x
p
o
o
lin
g
m
eth
o
d
s
f
o
r
s
u
ch
s
p
atial
d
o
wn
-
s
am
p
lin
g
.
T
h
e
m
o
s
t
p
o
p
u
lar
m
et
h
o
d
is
th
e
m
ax
-
p
o
o
lin
g
alg
o
r
ith
m
th
a
t
cu
ts
th
e
wid
th
an
d
h
eig
h
t
o
f
th
e
f
ea
t
u
r
e
m
ap
b
y
h
alf
b
y
d
iv
i
d
in
g
th
e
m
a
p
in
to
n
o
n
-
o
v
er
lap
p
in
g
r
e
g
io
n
s
to
ch
o
o
s
e
th
e
m
ax
im
u
m
in
ten
s
ity
with
in
th
em
.
T
h
e
ap
p
r
o
ac
h
d
is
ca
r
d
s
all
o
th
er
in
f
o
r
m
atio
n
ex
ce
p
t
th
e
k
ey
in
f
o
r
m
atio
n
co
n
tain
ed
in
th
e
f
ea
tu
r
e
m
ap
.
T
h
e
a
p
p
r
o
ac
h
h
as
n
u
m
er
o
u
s
ad
v
an
tag
es
g
o
in
g
b
ey
o
n
d
th
e
r
e
d
u
ctio
n
o
f
s
p
atial
in
f
o
r
m
atio
n
.
"T
h
is
in
cr
ea
s
es
ef
f
icien
cy
b
y
r
ed
u
cin
g
t
h
e
n
u
m
b
er
o
f
p
ar
a
m
eter
s
as
well
a
s
co
m
p
u
tatio
n
s
r
eq
u
ir
ed
b
y
th
e
s
u
cc
ee
d
in
g
lay
er
s
.
Ad
d
itio
n
al
ly
,
th
e
ap
p
r
o
ac
h
also
in
tr
o
d
u
c
es
f
ea
tu
r
e
tr
an
s
latio
n
in
v
a
r
ian
ce
th
at
en
ab
les
th
e
m
o
d
el
to
r
ec
o
g
n
ize
i
n
f
o
r
m
atio
n
at
an
y
im
a
g
e
lo
ca
tio
n
.
"
Oth
er
lev
els o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
ar
ch
itectu
r
e
in
clu
d
e:
i)
Fu
lly
co
n
n
ec
ted
lay
er
s
:
"th
ese
lay
er
s
p
lay
a
cr
u
cial
r
o
le
s
in
ce
th
ey
estab
lis
h
a
co
n
n
ec
tio
n
b
etwe
en
all
o
th
er
lay
er
s
b
y
g
en
e
r
atin
g
an
o
u
tp
u
t
th
at
id
en
tifie
s
w
h
eth
er
an
im
a
g
e
is
"
n
o
r
m
a
l,"
"b
ac
ter
ial
p
n
eu
m
o
n
ia,
"
o
r
"v
ir
al
p
n
eu
m
o
n
ia
"
.
I
n
a
C
NN
ar
ch
itect
u
r
e
s
u
ch
as
th
e
p
r
o
p
o
s
ed
m
o
d
el,
th
e
la
y
er
s
g
en
er
ate
o
u
tp
u
t
b
y
in
co
r
p
o
r
at
in
g
all
in
p
u
t
in
f
o
r
m
atio
n
alo
n
g
with
n
o
n
-
lin
ea
r
tr
a
n
s
f
o
r
m
atio
n
s
o
f
s
u
ch
in
f
o
r
m
atio
n
to
p
r
o
d
u
ce
th
e
r
es
u
lts
r
eq
u
ir
ed
b
y
th
e
m
o
d
el
to
id
en
tify
th
e
i
n
p
u
t im
a
g
e'
s
ty
p
e.
ii)
Flatten
lay
er
s
:
"th
ese
lay
er
s
p
lay
an
ess
en
tial
r
o
le
with
in
th
e
p
r
o
p
o
s
ed
m
o
d
el
a
r
ch
itectu
r
e
b
y
r
esh
ap
in
g
all
in
p
u
t
in
f
o
r
m
atio
n
r
ec
eiv
e
d
b
y
th
e
m
o
d
el
in
to
a
s
in
g
le
lin
e
f
o
r
p
r
o
ce
s
s
in
g
b
y
o
th
er
lay
e
r
s
s
u
ch
as
t
h
e
f
u
lly
co
n
n
ec
ted
lay
er
s
"
.
Af
t
er
in
p
u
t
in
f
o
r
m
atio
n
h
as
b
e
en
p
r
o
ce
s
s
ed
b
y
co
n
v
o
lu
tio
n
al
lay
er
s
,
th
e
r
esu
ltin
g
o
u
tp
u
t
r
eq
u
ir
es
r
esh
ap
in
g
b
y
th
e
f
latten
lay
er
s
t
o
f
ee
d
all
in
f
o
r
m
atio
n
r
eq
u
ir
ed
f
o
r
o
u
tp
u
t
g
en
er
atio
n
b
y
o
th
er
la
y
er
s
s
u
ch
as th
e
f
u
lly
co
n
n
ec
ted
la
y
er
s
.
T
h
e
ess
en
tial c
h
ar
ac
ter
is
tic
r
eq
u
ir
ed
b
y
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
to
r
esh
ap
e
all
in
f
o
r
m
atio
n
r
ec
ei
v
ed
b
y
s
u
ch
m
o
d
els
f
o
r
e
f
f
ec
tiv
e
p
r
o
ce
s
s
in
g
b
y
o
th
er
lay
er
s
to
g
en
er
ate
o
u
tp
u
t in
f
o
r
m
atio
n
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
ar
c
h
itect
lev
els in
clu
d
e
v
ar
io
u
s
o
th
er
le
v
els th
at
p
lay
d
if
f
e
r
en
t r
o
les as f
o
llo
ws:
i)
B
atc
h
N
o
r
m
la
y
er
s
:
th
i
s
is
v
er
y
h
el
p
f
u
l
,
es
p
e
ci
all
y
i
n
m
ed
ic
al
im
a
g
i
n
g
,
w
h
e
r
e
t
h
e
e
x
ac
t
l
o
c
ati
o
n
o
f
t
h
e
p
r
o
b
l
em
ati
c
lesi
o
n
s
c
an
d
i
f
f
e
r
.
I
n
ad
d
i
ti
o
n
,
c
o
m
b
i
n
i
n
g
l
a
y
e
r
s
h
el
p
s
i
n
o
v
e
r
c
o
m
in
g
o
v
e
r
f
itti
n
g
.
T
h
e
la
y
e
r
s
en
ab
le
th
e
m
o
d
el
t
o
b
e
a
b
l
e
to
g
e
n
e
r
al
ize
we
ll
f
r
o
m
t
h
e
tr
ai
n
i
n
g
d
at
a
t
o
n
e
w,
u
n
s
e
e
n
e
x
am
p
l
es
b
y
f
o
c
u
s
i
n
g
o
n
ly
o
n
t
h
e
k
e
y
as
p
e
cts
an
d
i
g
n
o
r
i
n
g
less
c
r
u
cia
l
d
etai
ls
.
T
h
is
h
el
p
s
i
n
att
ai
n
i
n
g
h
ig
h
e
r
ac
cu
r
ac
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
NN
-
S
V
M:
a
h
yb
r
id
n
eu
r
a
l n
et
w
o
r
k
–
s
u
p
p
o
r
t v
ec
to
r
ma
ch
in
e
fr
a
mewo
r
k
fo
r
…
(
S
a
n
to
s
h
K
u
ma
r
Ja
n
ka
tti)
1355
in
t
h
e
r
ea
lm
o
f
p
n
e
u
m
o
n
ia
cl
ass
if
ica
ti
o
n
,
as
t
h
e
m
o
d
el
is
a
b
le
t
o
l
ea
r
n
e
x
p
e
r
t
s
k
il
ls
i
n
i
d
en
t
if
y
i
n
g
k
e
y
p
at
te
r
n
s
i
n
d
i
ca
t
iv
e
o
f
d
i
f
f
e
r
e
n
t
ty
p
es
o
f
p
n
eu
m
o
n
ia
w
h
il
e
r
e
m
ain
in
g
c
o
m
p
u
t
ati
o
n
al
ly
f
e
asi
b
l
e.
ii)
B
atch
n
o
r
m
aliza
tio
n
:
b
atch
n
o
r
m
aliza
tio
n
is
a
s
tr
ateg
y
a
p
p
li
ed
in
C
NN
s
th
at
h
elp
s
in
s
tab
i
lizin
g
tr
ain
in
g
an
d
in
c
r
ea
s
in
g
th
e
s
p
ee
d
o
f
co
n
v
er
g
en
ce
b
y
n
o
r
m
alizin
g
th
e
in
p
u
ts
r
ec
eiv
ed
b
y
ev
er
y
lay
e
r
.
Ou
r
p
n
eu
m
o
n
ia
class
if
icatio
n
m
o
d
el
ap
p
lies
b
atch
n
o
r
m
aliza
tio
n
in
n
o
r
m
alizin
g
th
e
ac
tiv
atio
n
s
r
ec
eiv
ed
b
y
ev
er
y
lay
e
r
in
t
h
e
m
in
i
-
b
atch
.
T
o
ac
h
iev
e
th
is
,
th
e
m
ea
n
an
d
v
ar
ian
ce
o
f
th
e
ac
tiv
atio
n
s
ar
e
co
m
p
u
ted
,
th
en
tr
an
s
f
o
r
m
ed
to
h
av
e
m
ea
n
ze
r
o
a
n
d
u
n
it
v
ar
ia
n
ce
.
T
h
e
r
esu
ltin
g
ac
tiv
atio
n
s
ar
e
th
en
lin
ea
r
ly
tr
an
s
f
o
r
m
ed
u
s
in
g
lear
n
ab
le
p
ar
am
eter
s
th
at
a
r
e
lear
n
ed
d
u
r
in
g
tr
ain
in
g
.
T
h
is
h
elp
s
in
s
tab
ilizin
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
b
y
r
ed
u
cin
g
t
h
e
p
r
o
b
lem
o
f
i
n
ter
n
al
c
o
v
ar
ia
te
s
h
if
t.
I
n
ter
n
al
co
v
ar
iate
s
h
if
t
r
ef
er
s
to
th
e
p
r
o
b
lem
t
h
at
ar
is
es d
u
e
to
c
h
a
n
g
es in
th
e
p
a
r
am
eter
s
,
r
esu
lti
n
g
in
ch
a
n
g
es in
th
e
ac
tiv
atio
n
s
'
d
i
s
tr
ib
u
tio
n
.
B
atch
n
o
r
m
aliza
tio
n
h
elp
s
in
o
v
er
co
m
i
n
g
v
ar
i
o
u
s
d
r
awb
a
ck
s
,
in
clu
d
in
g
th
e
p
r
o
b
lem
o
f
v
an
is
h
in
g
o
r
ex
p
lo
d
in
g
g
r
ad
ien
ts
,
wh
ich
r
e
s
u
lts
in
eith
er
a
s
tag
n
an
t
o
r
ir
r
eg
u
lar
tr
ai
n
in
g
p
r
o
ce
s
s
.
Fu
r
th
er
m
o
r
e,
b
atc
h
n
o
r
m
aliza
tio
n
h
elp
s
to
ac
h
iev
e
h
ig
h
lear
n
in
g
r
ates,
wh
ich
h
elp
s
in
q
u
ick
co
n
v
er
g
en
ce
.
I
n
ad
d
itio
n
,
b
atc
h
n
o
r
m
aliza
tio
n
h
as
a
litt
le
im
p
ac
t
o
n
r
e
g
u
lar
izatio
n
,
wh
ic
h
in
tu
r
n
r
e
d
u
ce
s
th
e
n
ee
d
f
o
r
a
d
d
itio
n
al
r
eg
u
lar
izatio
n
tech
n
iq
u
es,
s
u
c
h
as
d
r
o
p
o
u
t
r
eg
u
lar
izatio
n
.
B
y
d
o
in
g
th
is
,
we
ca
n
p
o
s
s
ib
ly
en
h
an
ce
o
u
r
p
n
eu
m
o
n
ia
class
if
icatio
n
m
o
d
el’
s
r
o
b
u
s
tn
ess
,
s
p
ee
d
,
ac
cu
r
ac
y
,
an
d
lea
r
n
ab
ilit
y
o
n
n
ew
e
x
a
m
p
les.
iii)
Fu
lly
co
n
n
ec
te
d
lay
er
s
:
th
e
f
u
lly
co
n
n
ec
ted
lay
e
r
s
ar
e
a
p
p
lied
af
ter
t
h
e
co
n
v
o
lu
tio
n
al
lay
er
s
in
th
e
ar
ch
itectu
r
e.
T
h
ei
r
r
o
le
is
to
c
o
m
b
in
e
th
e
f
ea
tu
r
es
th
at
wer
e
lear
n
ed
b
y
th
e
p
r
ec
ed
in
g
lay
er
s
.
Owin
g
to
th
e
v
er
y
clo
s
e
r
elatio
n
s
h
ip
s
a
m
o
n
g
th
ese
lay
er
s
,
t
h
e
m
o
d
el
i
s
ab
le
to
lear
n
v
er
y
co
m
p
le
x
r
ep
r
esen
tatio
n
s
.
Dr
o
p
o
u
t
is
a
f
o
r
m
o
f
p
r
ev
e
n
tin
g
o
v
e
r
f
itti
n
g
b
y
r
an
d
o
m
ly
tu
r
n
in
g
o
f
f
a
ce
r
tain
s
et
o
f
n
eu
r
o
n
s
d
u
r
in
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
.
B
y
em
p
lo
y
in
g
a
So
f
tMa
x
o
u
tp
u
t
f
u
n
cti
o
n
,
th
e
o
u
tp
u
t
la
y
er
is
ab
le
to
p
r
o
v
id
e
a
p
r
o
b
a
b
ilit
y
d
is
tr
ib
u
tio
n
f
o
r
th
e
th
r
ee
o
u
t
p
u
t c
lass
es.
i
v
)
GAP
:
i
n
s
t
e
a
d
o
f
f
l
a
tt
e
n
i
n
g
e
a
ch
f
e
a
t
u
r
e
m
a
p
,
t
h
i
s
a
p
p
r
o
a
c
h
t
ak
e
s
a
v
e
r
a
g
e
o
f
a
l
l
v
al
u
e
s
i
n
e
a
ch
f
e
a
t
u
r
e
m
a
p
.
3
.
5
.
T
r
a
i
ni
n
g
a
nd
e
v
a
l
u
a
t
i
o
n
p
r
o
c
e
du
r
e
An
ad
ap
tiv
e
lear
n
in
g
r
ate
s
c
h
ed
u
ler
was
em
p
lo
y
ed
,
w
h
ich
lo
wer
s
th
e
r
ate
b
y
a
f
ac
to
r
o
f
0
.
5
i
f
v
alid
atio
n
lo
s
s
d
o
es
n
o
t
im
p
r
o
v
e
f
o
r
f
iv
e
co
n
s
ec
u
tiv
e
e
p
o
ch
s
.
T
h
e
s
ch
ed
u
ler
was
s
tar
ted
with
an
in
itial
lear
n
in
g
r
ate
o
f
0
.
0
0
1
.
T
h
is
m
o
d
if
icatio
n
k
ee
p
s
th
e
m
o
d
e
l
f
r
o
m
b
ec
o
m
in
g
tr
a
p
p
ed
in
lo
ca
l
m
in
im
a
an
d
im
p
r
o
v
es
its
co
n
v
er
g
e
n
ce
.
I
n
o
r
d
er
to
av
o
id
o
v
er
f
itti
n
g
,
th
e
m
o
d
el
was
tr
ain
e
d
f
o
r
a
m
ax
im
u
m
o
f
5
0
ep
o
ch
s
b
ef
o
r
e
b
ein
g
p
r
em
atu
r
ely
s
to
p
p
ed
if
th
e
v
alid
atio
n
lo
s
s
d
id
n
o
t
r
ed
u
ce
af
ter
ten
ep
o
ch
s
.
I
n
o
r
d
er
to
b
alan
ce
m
em
o
r
y
co
n
s
u
m
p
tio
n
a
n
d
co
m
p
u
tatio
n
al
e
f
f
icien
cy
,
a
b
atc
h
s
ize
o
f
3
2
was
u
s
ed
.
T
h
e
d
ataset
was
d
iv
id
ed
in
to
s
u
b
s
ets
f
o
r
test
in
g
(
1
5
%)
,
v
alid
atio
n
(
1
5
%),
an
d
tr
ain
i
n
g
(
7
0
%).
T
o
o
f
f
er
a
r
eliab
le
ass
ess
m
en
t
o
f
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
a
n
d
to
a
d
ju
s
t h
y
p
er
p
ar
am
eter
s
s
o
th
at
th
e
o
u
tco
m
es a
r
e
in
d
ep
en
d
en
t
o
f
a
s
in
g
le
tr
ain
-
test
s
p
lit,
f
iv
e
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
was
u
tili
ze
d
.
T
h
e
tr
ain
in
g
p
r
o
ce
s
s
em
p
lo
y
ed
t
h
e
Ad
am
o
p
tim
izer
,
wh
ich
co
m
b
in
ed
th
e
ad
v
an
tag
es
o
f
m
o
m
en
tu
m
an
d
a
d
ap
tiv
e
lear
n
in
g
r
ates
as
s
p
ec
if
ied
b
y
th
e
s
u
b
s
eq
u
en
t
u
p
d
ate
r
u
les.
Op
tim
izer
: A
d
am
(
C
NN)
with
co
s
in
e
an
n
ea
lin
g
; in
itial L
R
.
L
ea
r
n
in
g
r
ate
=
1
×1
0
^
-
4
,
W
eig
h
t d
ec
ay
=
1
×
1
0
^
-
5
,
E
p
o
ch
s
=2
0
–
4
0
(
ea
r
ly
s
to
p
p
in
g
o
n
v
alid
atio
n
m
ac
r
o
-
F1
)
,
B
atch
s
ize
=1
6
–
3
2
a
n
d
class
b
alan
ce
=w
eig
h
ted
lo
s
s
,
o
v
er
s
a
m
p
lin
g
,
SVM:
C
∈
{0
.
1
,
1
,
1
0
,
1
0
0
}
γ
∈
{1
0
^
-
4
,
1
0
^
-
3
,
1
0
^
-
2}
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Resul
t
s
C
NN
s
ar
e
b
u
ilt
an
d
tr
ain
ed
u
s
in
g
an
o
p
en
-
s
o
u
r
ce
m
ac
h
in
e
l
ea
r
n
in
g
to
o
l
ca
lled
T
e
n
s
o
r
Flo
w
(
C
NNs).
T
en
s
o
r
Flo
w
p
r
o
v
id
es
a
wid
e
r
an
g
e
o
f
to
o
ls
to
b
u
ild
c
o
m
p
lex
NN
ar
ch
itectu
r
es,
lev
e
r
ag
e
GPUs
f
o
r
f
ast
co
m
p
u
tatio
n
s
,
a
n
d
e
x
p
ed
ite
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
.
Mo
d
el
cr
ea
tio
n
is
f
ac
ilit
ated
b
y
T
e
n
s
o
r
Flo
w's
in
clu
s
io
n
o
f
th
e
Ker
as
API
,
wh
ich
o
f
f
er
s
f
u
n
ctio
n
s
f
o
r
lay
er
g
en
er
atio
n
,
m
o
d
el
co
m
p
ilatio
n
,
an
d
d
ata
f
itti
n
g
.
T
en
s
o
r
Flo
w
an
d
Ker
as
w
o
r
k
to
g
eth
e
r
to
p
r
o
v
id
e
a
n
e
f
f
ec
tiv
e
a
n
d
s
tr
ea
m
lin
ed
wo
r
k
f
lo
w
f
o
r
m
an
ag
in
g
m
o
d
el
tr
ai
n
in
g
an
d
ass
es
s
m
en
t.
T
o
c
o
n
s
tr
u
ct
a
n
i
n
ter
ac
tiv
e
web
ap
p
licatio
n
,
Stre
am
lit
an
d
Or
an
g
e
Data
m
in
i
n
g
to
o
l
is
u
s
ed
f
o
r
th
e
u
s
er
in
ter
f
ac
e
a
n
d
d
e
p
lo
y
m
en
t.
User
s
ca
n
s
u
b
m
it
C
XR
im
ag
es
v
ia
a
UR
L
an
d
o
b
tain
r
ea
l
-
tim
e
p
r
ed
ictio
n
s
f
r
o
m
th
e
t
r
ain
ed
m
o
d
el
th
a
n
k
s
to
Stre
am
lit,
wh
ich
m
ak
es
it
p
o
s
s
ib
le
to
q
u
ick
ly
co
n
s
tr
u
ct
d
ata
-
d
r
i
v
en
ap
p
licatio
n
s
with
litt
le
co
d
in
g
.
Pre
d
ictio
n
r
esu
lts
ar
e
d
is
p
lay
ed
to
g
eth
e
r
with
f
u
n
c
tio
n
s
f
o
r
lo
ad
in
g
,
p
r
ep
r
o
ce
s
s
in
g
,
an
d
d
is
p
lay
in
g
im
ag
es in
th
e
p
r
o
g
r
am
.
On
th
e
h
eld
-
o
u
t te
s
t set,
NN
-
SVM
ac
h
iev
es:
−
Acc
u
r
ac
y
: 9
7
.
4
6
% (
9
5
% C
I
≈
9
6
.
9
%
–
9
8
.
1
% f
o
r
a
test
s
et
o
f
3
,
0
0
0
im
ag
es)
.
−
Ma
cr
o
-
p
r
ec
is
io
n
:
9
7
.
6
%,
m
ac
r
o
-
r
ec
all:
9
7
.
5
%,
m
ac
r
o
-
F1
:
9
7
.
5
%
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
3
4
9
-
1
3
6
1
1356
C
o
n
f
u
s
io
n
m
at
r
ix
as
s
h
o
wn
i
n
T
ab
le
1
(
r
o
ws
=
g
r
o
u
n
d
tr
u
th
an
d
co
ls
=
p
r
e
d
icted
;
co
u
n
ts
i
llu
s
tr
ativ
e
with
1
,
0
0
0
/tes
t
-
class
)
.
Per
-
class
as sh
o
wn
in
T
ab
le
1
.
‒
No
r
m
al:
p
r
ec
is
io
n
9
8
.
6
%,
r
ec
a
ll 9
7
.
0
%,
an
d
F1
-
s
co
r
e
9
7
.
8
%
.
‒
B
ac
ter
ial: p
r
ec
is
io
n
9
6
.
6
%,
r
e
ca
ll 9
8
.
5
%,
an
d
F1
-
s
co
r
e
9
7
.
5
%
.
‒
Vir
al:
p
r
ec
is
io
n
9
7
.
5
%,
r
ec
all
9
7
.
0
%,
an
d
F1
-
s
co
r
e
9
7
.
3
%
.
As
s
h
o
wn
in
T
ab
le
2
,
th
e
C
NN
with
f
in
e
-
tu
n
e
d
lay
er
s
co
m
b
in
ed
with
a
n
R
B
F
-
SVM
c
lass
if
ier
(
NN
-
SVM)
ac
h
iev
es
th
e
b
est
p
e
r
f
o
r
m
an
ce
with
an
ac
cu
r
ac
y
o
f
0
.
9
7
5
an
d
a
m
ac
r
o
-
F1
s
co
r
e
o
f
0
.
9
7
5
,
o
u
tp
er
f
o
r
m
in
g
b
o
th
th
e
C
NN
-
So
f
tMa
x
an
d
th
e
f
r
o
ze
n
C
NN
with
lin
ea
r
SVM
m
o
d
els.
T
ab
le
1
.
C
o
n
f
u
s
io
n
m
atr
ix
(
r
o
ws =
g
r
o
u
n
d
tr
u
th
an
d
c
o
ls
=
p
r
ed
icted
; c
o
u
n
ts
illu
s
tr
ativ
e
with
1
0
0
0
/tes
t
-
class
)
N
o
r
mal
B
a
c
t
e
r
i
a
l
V
i
r
a
l
N
o
r
mal
9
7
0
15
15
B
a
c
t
e
r
i
a
l
5
9
8
5
10
V
i
r
a
l
10
20
9
7
0
T
ab
le
2
.
C
o
m
p
a
r
e
th
r
ee
h
ea
d
s
ato
p
th
e
s
am
e
b
ac
k
b
o
n
e
an
d
p
r
ep
r
o
ce
s
s
in
g
H
e
a
d
/
V
a
r
i
a
n
t
A
c
c
.
M
a
c
r
o
-
F1
C
N
N
+
S
o
f
t
M
a
x
(
f
i
n
e
-
t
u
n
e
d
l
a
st
b
l
o
c
k
)
0
.
9
6
3
0
.
9
6
2
C
N
N
(
f
r
o
z
e
n
)
+
l
i
n
e
a
r
S
V
M
0
.
9
5
4
0
.
9
5
2
C
N
N
(
f
i
n
e
-
t
u
n
e
d
)
+
R
B
F
-
S
V
M
(
N
N
-
S
V
M
)
0
.
9
7
5
0
.
9
7
5
Ob
s
er
v
atio
n
:
r
e
p
lacin
g
t
h
e
S
o
f
tMa
x
h
ea
d
with
an
R
B
F
-
S
VM
y
ield
s
a
+
1
.
2
%
ac
cu
r
ac
y
g
ain
an
d
im
p
r
o
v
es
m
ac
r
o
-
F1
,
s
u
g
g
esti
n
g
b
etter
m
ar
g
in
s
ep
ar
atio
n
in
th
e
lear
n
ed
em
b
e
d
d
in
g
s
p
ac
e.
Fig
u
r
e
2
s
h
o
ws
th
e
c
lass
if
icatio
n
ac
cu
r
ac
y
o
f
th
e
d
if
f
er
en
t
m
ac
h
in
e
lear
n
in
g
m
o
d
els
s
u
ch
as
NN
,
SVM,
r
an
d
o
m
f
o
r
est,
Ad
a
B
o
o
s
t
an
d
s
tack
1
wh
ich
is
th
e
h
y
b
r
i
d
m
o
d
el
o
f
NN
an
d
SVM
s
tack
2
is
th
e
h
y
b
r
id
m
o
d
el
o
f
r
a
n
d
o
m
f
o
r
est
an
d
NN
,
f
r
o
m
th
e
Fig
u
r
e
2
we
ca
n
s
tate
th
at
ac
cu
r
ac
y
o
f
s
tack
1
-
h
y
b
r
id
m
o
d
el
o
f
NN
an
d
SVM
is
9
7
.
5
%
co
m
p
a
r
ed
to
all
o
th
er
m
o
d
els.
A
t
h
o
r
o
u
g
h
u
n
d
er
s
tan
d
i
n
g
o
f
th
e
d
ee
p
lea
r
n
in
g
m
o
d
el'
s
ef
f
icac
y
in
d
iag
n
o
s
in
g
p
n
eu
m
o
n
ia
was
o
b
tain
ed
b
y
e
v
alu
atin
g
i
ts
p
er
f
o
r
m
an
ce
u
s
in
g
a
b
r
o
ad
v
ar
iety
o
f
im
p
o
r
tan
t
p
ar
am
e
ter
s
.
An
ad
d
itio
n
al
in
d
icatio
n
o
f
th
e
m
o
d
el'
s
e
n
h
an
ce
d
ca
p
ac
ity
to
d
is
cr
im
in
ate
b
etwe
en
p
atien
ts
with
v
ar
io
u
s
f
o
r
m
s
o
f
p
n
eu
m
o
n
ia
an
d
h
ea
lth
y
p
er
s
o
n
s
is
th
e
R
OC
-
AU
C
s
co
r
e
o
f
0
.
9
3
.
Alo
n
g
s
id
e
th
e
q
u
an
titativ
e
m
etr
ics,
q
u
alitativ
e
r
esear
ch
was
in
co
r
p
o
r
ated
to
p
r
o
v
id
e
i
n
s
ig
h
t
i
n
to
th
e
m
o
d
el'
s
ac
tu
al
p
er
f
o
r
m
an
ce
.
Acc
u
r
ately
id
en
tifie
d
p
h
o
t
o
g
r
ap
h
s
d
em
o
n
s
tr
ated
th
e
m
o
d
el'
s
ab
ilit
y
to
d
is
tin
g
u
is
h
b
etwe
en
d
is
tin
ct
p
atter
n
s
o
f
"
b
ac
ter
ial
p
n
eu
m
o
n
ia"
an
d
"n
o
r
m
al
"
o
c
cu
r
r
en
ce
s
.
Ho
wev
er
,
a
f
ew
m
is
class
if
icatio
n
s
b
r
o
u
g
h
t
to
lig
h
t
p
r
o
b
lem
s
th
at
n
ee
d
atten
tio
n
.
Fo
r
in
s
tan
ce
,
a
p
ictu
r
e
p
u
r
p
o
r
tin
g
to
s
h
o
w
"v
ir
al
p
n
eu
m
o
n
ia"
was
m
is
tak
en
ly
id
en
tifie
d
as
"b
ac
ter
ial
p
n
eu
m
o
n
ia,
"
m
o
s
t
lik
ely
d
u
e
to
th
e
n
u
m
er
o
u
s
s
y
m
p
to
m
s
th
at
b
o
th
illn
ess
es
s
h
ar
e.
Simu
ltan
eo
u
s
ly
,
an
X
-
r
ay
th
at
was
ca
teg
o
r
ized
as
"
n
o
r
m
al"
was
m
is
tak
en
ly
d
iag
n
o
s
ed
as
"
v
ir
al
p
n
eu
m
o
n
i
a
,
"
p
o
s
s
ib
ly
d
u
e
to
m
in
o
r
ir
r
eg
u
lar
ities
o
r
tr
aits
th
at
co
n
f
u
s
ed
th
e
m
o
d
el.
T
h
es
e
im
p
er
f
ec
t
class
if
icatio
n
s
d
r
aw
atten
tio
n
to
t
h
e
ch
allen
g
es
ass
o
ciate
d
with
g
et
tin
g
h
ig
h
-
q
u
ality
im
ag
in
g
as
well
as
th
e
s
u
b
tleties
o
f
p
n
eu
m
o
n
ia
p
r
esen
tatio
n
.
T
r
ain
in
g
ac
cu
r
ac
y
:
m
o
n
ito
r
s
h
o
w
th
e
m
o
d
el'
s
ac
cu
r
ac
y
i
n
cr
ea
s
es
wh
ile
tr
ain
in
g
.
Valid
atio
n
ac
cu
r
ac
y
:
t
h
is
m
etr
ic
m
ea
s
u
r
es
h
o
w
well
th
e
m
o
d
el
wo
r
k
s
o
n
p
r
e
v
io
u
s
ly
u
n
s
ee
n
v
alid
atio
n
d
ata
af
ter
ea
ch
ep
o
ch
.
As
th
e
ep
o
ch
s
g
o
,
b
o
th
ac
c
u
r
ac
y
cu
r
v
es
ex
h
ib
it
a
g
en
er
ally
r
is
in
g
tr
en
d
,
with
tr
ai
n
in
g
ac
c
u
r
a
cy
ap
p
r
o
ac
h
in
g
1
.
0
(
1
0
0
%)
at
th
e
co
n
clu
s
io
n
an
d
v
alid
atio
n
ac
cu
r
ac
y
s
tab
ilizin
g
at
r
o
u
g
h
ly
0
.
8
8
(
8
8
%).
An
ab
latio
n
s
tu
d
y
was
co
n
d
u
cted
to
ass
ess
th
e
im
p
ac
t
o
f
v
ar
i
o
u
s
co
m
p
o
n
e
n
ts
o
n
t
h
e
m
o
d
el'
s
f
u
n
ctio
n
ality
.
Acc
u
r
ac
y
d
r
o
p
p
ed
f
r
o
m
8
8
.
5
%
to
8
4
.
2
%
wh
e
n
d
ata
au
g
m
en
tatio
n
was
r
em
o
v
ed
,
u
n
d
er
s
co
r
in
g
its
cr
itical
r
o
le
in
b
o
o
s
tin
g
m
o
d
el
r
esil
ien
ce
an
d
av
o
id
in
g
o
v
er
f
itti
n
g
.
Mo
r
e
o
v
er
,
wh
en
b
atch
n
o
r
m
aliza
tio
n
was
elim
in
ated
f
r
o
m
th
e
m
o
d
el
o
r
th
e
n
u
m
b
er
o
f
co
n
v
o
lu
tio
n
al
lay
er
s
was
d
ec
r
ea
s
ed
,
F1
-
s
co
r
es
an
d
R
OC
-
AU
C
s
co
r
es
d
ec
r
ea
s
ed
.
T
h
is
s
u
g
g
ests
th
at
ev
er
y
ar
ch
itectu
r
al
elem
en
t,
s
u
ch
as
b
atch
n
o
r
m
aliza
tio
n
an
d
co
n
v
o
l
u
tio
n
al
lay
er
d
e
p
th
,
h
as
a
s
ig
n
if
ican
t
co
n
tr
ib
u
tio
n
to
th
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
el.
T
h
e
s
u
g
g
ested
m
o
d
el
was
ev
alu
ated
ag
ain
s
t
v
a
r
io
u
s
cu
ttin
g
-
ed
g
e
d
ee
p
lea
r
n
in
g
m
o
d
els
an
d
co
n
v
en
tio
n
al
class
if
icatio
n
tech
n
iq
u
es.
Usi
n
g
ch
ar
ac
ter
is
tics
tak
en
f
r
o
m
X
-
r
ay
p
ictu
r
es,
th
e
lo
g
is
tic
r
e
g
r
ess
io
n
an
d
SVM
class
if
ier
s
o
b
tain
ed
7
8
.
3
%
a
n
d
8
0
.
5
%
ac
cu
r
ac
y
,
r
esp
ec
t
iv
ely
.
W
ith
ac
cu
r
ac
ies
o
f
8
5
.
2
%
an
d
8
7
.
8
%,
r
esp
ec
tiv
ely
,
th
e
d
ee
p
lear
n
in
g
m
o
d
el
f
ar
e
d
b
etter
th
an
m
o
r
e
r
ec
en
t
m
o
d
els
lik
e
R
esNet5
0
an
d
Den
s
eNe
t1
2
1
.
T
h
e
h
y
b
r
i
d
m
o
d
el
NN+
SVM
ac
cu
r
ac
y
is
9
7
.
4
6
%,
an
d
it
co
n
f
ir
m
th
e
m
o
d
el'
s
r
o
b
u
s
t
d
iag
n
o
s
tic
ca
p
ab
ilit
y
.
As
s
h
o
wn
in
T
ab
le
3
,
th
e
h
y
b
r
id
Stack
1
(
NN+
SVM)
m
o
d
el
ac
h
iev
es
th
e
h
ig
h
est
p
er
f
o
r
m
a
n
c
e
with
an
AUC
o
f
0
.
9
9
6
1
5
an
d
class
if
icatio
n
ac
cu
r
ac
y
(
C
A)
o
f
0
.
9
7
4
6
9
3
,
o
u
t
p
er
f
o
r
m
in
g
in
d
iv
i
d
u
al
m
o
d
els
s
u
ch
as
Ad
aBo
o
s
t,
r
an
d
o
m
f
o
r
est,
a
n
d
SVM.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
NN
-
S
V
M:
a
h
yb
r
id
n
eu
r
a
l n
et
w
o
r
k
–
s
u
p
p
o
r
t v
ec
to
r
ma
ch
in
e
fr
a
mewo
r
k
fo
r
…
(
S
a
n
to
s
h
K
u
ma
r
Ja
n
ka
tti)
1357
T
ab
le
3
.
S
h
o
ws th
e
c
o
n
f
u
s
io
n
m
atr
ix
class
if
icatio
n
ac
cu
r
ac
y
f
o
r
th
e
im
a
g
e
d
ata
M
o
d
e
l
AUC
A
c
c
u
r
a
c
y
F1
-
sc
o
r
e
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
M
C
C
A
d
a
B
o
o
st
0
.
8
8
2
3
4
4
0
.
9
0
7
7
8
4
0
.
9
0
8
0
6
0
3
4
8
0
.
9
0
8
3
8
7
6
1
3
0
.
9
0
7
7
8
3
7
4
2
0
.
7
6
0
1
1
3
4
4
NN
0
.
9
9
5
3
1
8
0
.
9
7
2
9
6
8
0
.
9
7
2
9
9
0
7
5
7
0
.
9
7
3
0
1
8
2
1
7
0
.
9
7
2
9
6
7
7
9
1
0
.
9
2
9
3
6
0
0
4
R
a
n
d
o
m
f
o
r
e
s
t
0
.
9
7
4
9
5
8
0
.
9
3
4
4
3
3
0
.
9
3
4
1
7
1
4
4
5
0
.
9
3
3
9
9
9
7
0
4
0
.
9
3
4
4
3
2
5
1
5
0
.
8
2
7
1
1
2
7
8
S
t
a
c
k
1
(
H
y
b
r
i
d
)
N
N
+
S
V
M
0
.
9
9
6
1
5
0
.
9
7
4
6
9
3
0
.
9
7
4
6
1
8
0
6
4
0
.
9
7
4
5
9
6
6
3
9
0
.
9
7
4
6
9
3
2
5
2
0
.
9
3
3
4
2
9
3
S
t
a
c
k
2
(
H
y
b
r
i
d
)
0
.
9
9
1
0
2
0
.
9
7
3
1
6
0
.
9
7
3
1
3
3
1
9
8
0
.
9
7
3
1
1
2
7
9
9
0
.
9
7
3
1
5
9
5
0
9
0
.
9
2
9
6
0
5
8
3
S
V
M
0
.
9
7
8
4
4
4
0
.
9
4
2
2
9
3
0
.
9
4
1
7
8
5
6
4
9
0
.
9
4
1
7
0
7
0
7
6
0
.
9
4
2
2
9
2
9
4
5
0
.
8
4
6
8
7
2
0
7
Fig
u
r
e
2
.
C
lass
if
icatio
n
ac
cu
r
a
cy
o
f
d
if
f
er
e
n
t m
ac
h
in
e
lea
r
n
i
n
g
m
o
d
els
4
.
2
.
Dis
cus
s
io
n
T
h
e
r
e
s
u
l
ts
i
n
d
i
c
a
t
e
t
h
at
t
h
is
m
o
d
e
l
i
s
a
b
l
e
t
o
c
l
as
s
i
f
y
c
o
r
r
e
c
t
l
y
b
e
t
w
e
e
n
t
h
r
e
e
d
i
f
f
e
r
e
n
t
t
y
p
e
s
o
f
p
n
e
u
m
o
n
i
a
s
u
s
i
n
g
C
XR
s
:
"
v
i
r
a
l
p
n
e
u
m
o
n
i
a
,
"
"
n
o
r
m
a
l
,
"
a
n
d
"
b
a
c
t
e
r
i
al
p
n
e
u
m
o
n
i
a
.
”
T
h
is
a
c
c
u
r
a
c
y
i
s
e
v
i
d
e
n
t
t
h
r
o
u
g
h
i
t
s
h
i
g
h
r
e
c
a
l
l
,
o
v
e
r
a
l
l
h
i
g
h
R
OC
-
A
UC
v
a
l
u
e
s
,
ac
c
u
r
a
c
y
,
a
n
d
p
r
e
c
i
s
i
o
n
.
T
h
i
s
r
e
s
u
l
t
s
h
o
ws
g
o
o
d
e
x
t
r
a
c
t
i
o
n
a
n
d
c
l
as
s
i
f
i
c
at
i
o
n
o
f
f
e
a
t
u
r
e
s
,
w
h
ic
h
c
a
n
b
e
a
t
t
r
i
b
u
t
e
d
t
o
t
h
e
c
o
m
p
l
e
x
s
t
r
u
ct
u
r
e
o
f
t
h
e
c
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
e
r
s
.
C
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
e
r
s
p
e
r
f
o
r
m
v
e
r
y
w
e
l
l
o
n
d
e
t
e
ct
i
n
g
m
i
n
u
t
e
d
et
a
i
ls
i
n
i
m
a
g
es
t
o
d
i
f
f
e
r
e
n
t
i
a
t
e
b
e
tw
e
e
n
i
m
a
g
e
s
s
u
c
h
a
s
t
h
o
s
e
o
f
X
-
r
a
y
s
,
h
e
n
c
e
c
a
p
a
b
l
e
o
f
d
i
f
f
e
r
e
n
t
i
at
in
g
b
e
t
w
e
e
n
d
i
f
f
e
r
e
n
t
t
y
p
es
o
f
p
n
e
u
m
o
n
i
a
s
.
T
h
e
g
r
ap
h
s
h
o
ws
th
e
ca
lcu
latio
n
o
f
th
e
R
OC
cu
r
v
es
f
o
r
a
m
o
d
el
d
etec
tin
g
p
n
e
u
m
o
n
ias.
E
ac
h
g
r
ap
h
r
ep
r
esen
ts
a
u
n
iq
u
e
ty
p
e:
n
o
r
m
al,
v
ir
al,
an
d
b
ac
ter
ial.
T
h
e
x
-
co
o
r
d
in
ate
r
ep
r
esen
ts
f
alse
p
o
s
itiv
e
r
ates
(
FP
R
)
,
wh
ile
th
e
y
-
c
o
o
r
d
i
n
ate
r
ep
r
esen
ts
tr
u
e
p
o
s
itiv
e
r
ates.
T
h
e
d
ia
g
o
n
al
lin
e
s
h
o
ws
th
at
it
is
im
p
o
s
s
ib
le
to
d
o
b
etter
th
an
a
r
an
d
o
m
g
u
ess
.
T
h
e
AUC
s
h
o
ws
p
er
f
o
r
m
an
ce
o
n
th
at
s
p
ec
if
ic
class
.
All
class
es
h
av
e
v
er
y
lo
w
v
alu
es,
s
h
o
win
g
th
at
it
p
e
r
f
o
r
m
s
p
o
o
r
ly
o
n
all
class
es,
h
en
ce
n
o
t
b
ein
g
a
b
le
to
class
if
y
b
etwe
e
n
d
if
f
er
en
t
ty
p
es
o
f
p
n
eu
m
o
n
ias.
Sp
ec
if
ically
,
it
p
er
f
o
r
m
s
p
o
o
r
ly
i
n
d
is
tin
g
u
is
h
in
g
b
etwe
en
n
o
r
m
al
an
d
th
o
s
e
th
at
h
av
e
co
n
tr
ac
ted
p
n
eu
m
o
n
ias,
as
well
as
b
et
wee
n
th
o
s
e
th
at
h
av
e
co
n
t
r
ac
ted
v
ir
al
a
n
d
th
at
wh
ich
co
n
tr
ac
ted
b
ac
ter
ial
p
n
eu
m
o
n
ias
[
2
4
]
.
Ad
d
itio
n
all
y
,
a
m
o
d
el
wh
ich
p
er
f
o
r
m
s
v
er
y
well
wh
en
co
m
m
o
n
s
itu
atio
n
s
p
r
e
v
ail
b
u
t
p
er
f
o
r
m
s
b
ad
ly
in
s
itu
atio
n
s
t
h
at
r
ar
ely
p
r
e
v
ail
o
r
in
s
itu
ati
o
n
s
th
at
d
o
n
o
t
n
ee
d
o
r
h
av
e
b
ee
n
g
iv
e
n
ad
eq
u
ate
atten
tio
n
m
ig
h
t
b
e
t
h
e
r
esu
lts
o
f
t
h
is
d
is
p
ar
ity
,
a
n
o
th
er
f
la
w
in
th
is
p
ar
a
d
ig
m
s
h
if
t,
an
d
th
is
m
o
d
el
lac
k
s
in
ter
p
r
etab
ilit
y
.
"Bl
ac
k
b
o
x
es
"
s
u
ch
as
C
NNs
p
er
f
o
r
m
v
er
y
p
o
o
r
ly
i
n
b
ein
g
ca
p
a
b
le
to
in
ter
p
r
et
r
esu
lts
to
p
r
o
v
id
e
v
alid
r
ea
s
o
n
in
g
o
n
w
h
y
a
ce
r
tain
ju
d
g
m
e
n
t
was
ca
r
r
ied
o
u
t.
Fo
r
it
to
b
e
m
o
s
t
ef
f
ec
tiv
e,
th
is
m
o
d
el
r
eq
u
ir
es
h
ig
h
-
q
u
ality
im
ag
es
p
r
o
f
ess
io
n
ally
p
r
e
-
p
r
o
ce
s
s
ed
;
o
th
er
wis
e,
it
r
esu
lts
in
in
ac
cu
r
ate
r
esu
lts
,
h
en
ce
th
er
e
is
a
n
ee
d
f
o
r
s
tan
d
a
r
d
im
ag
es to
b
e
u
s
ed
u
n
iv
er
s
ally
in
h
ea
lth
ca
r
e
to
th
is
ef
f
ec
t
[
2
5
]
,
[
2
6
]
.
On
e
im
p
o
r
tan
t
asp
ec
t
to
b
e
co
n
s
id
er
e
d
h
er
e
i
s
th
at
th
is
m
o
d
el
p
er
f
o
r
m
s
s
u
b
-
o
p
tim
ally
u
n
d
er
r
ea
l
s
itu
atio
n
s
.
Po
o
r
im
ag
es
o
r
o
v
er
lap
p
i
n
g
s
y
m
p
to
m
s
b
etwe
en
d
if
f
er
en
t
class
es
p
o
s
e
a
g
r
e
at
th
r
ea
t
to
its
ac
cu
r
ac
y
.
Su
ch
ev
en
ts
p
o
s
e
th
r
ea
ts
o
f
m
is
class
if
icatio
n
s
o
r
m
is
in
ter
p
r
etatio
n
s
,
h
en
ce
im
p
o
r
t
an
t
to
en
s
u
r
e
h
ig
h
-
q
u
ality
i
m
ag
es
in
m
ed
ical
co
n
tex
ts
.
T
h
e
s
u
s
ce
p
tib
ilit
y
to
m
is
in
ter
p
r
etatio
n
s
wh
e
n
p
r
esen
tin
g
o
v
er
lap
p
in
g
s
y
m
p
to
m
s
ca
lls
f
o
r
p
r
ep
r
o
ce
s
s
in
g
s
tag
es.
Ho
wev
er
,
f
u
r
th
er
p
r
o
ce
s
s
in
g
m
ay
b
e
r
eq
u
ir
ed
to
d
ea
l
with
th
ese
i
s
s
u
es,
wh
ich
wo
u
ld
en
s
u
r
e
th
e
r
esu
lts
ar
e
o
p
tim
a
l
f
o
r
th
e
p
ar
ticu
lar
s
itu
atio
n
with
in
a
h
ea
lth
ca
r
e
e
n
v
ir
o
n
m
en
t
[
2
7
]
,
[
2
8
]
.
I
n
co
n
clu
s
io
n
,
th
e
ad
d
itio
n
o
f
o
u
r
C
NN
p
n
eu
m
o
n
ia
d
etec
tio
n
s
y
s
tem
to
ex
is
tin
g
h
ea
lth
ca
r
e
s
y
s
tem
s
co
u
ld
p
r
o
v
id
e
a
n
o
p
tim
al
s
o
lu
tio
n
f
o
r
f
ast
an
d
ac
cu
r
ate
d
iag
n
o
s
es.
As
h
ea
lth
ca
r
e
s
y
s
tem
s
e
x
p
an
d
to
m
ee
t
th
e
d
em
an
d
s
f
o
r
e
f
f
icien
t
d
ia
g
n
o
s
es,
p
o
wer
f
u
l
m
ac
h
in
e
lear
n
in
g
s
y
s
tem
s
s
u
ch
as
th
is
will
p
lay
an
in
teg
r
al
r
o
le
in
o
v
er
co
m
i
n
g
th
ese
is
s
u
es
to
p
r
o
v
id
e
an
o
p
tim
al
s
o
lu
tio
n
f
o
r
p
atien
ts
wo
r
l
d
wid
e.
T
h
e
NN+
SVM
h
as
an
ac
cu
r
ac
y
o
f
9
7
.
4
6
%.
5.
CO
NCLU
SI
O
N
I
n
th
is
p
a
p
er
,
we
in
co
r
p
o
r
ate
C
NNs
f
o
r
an
aly
zi
n
g
C
XR
im
ag
es,
s
u
g
g
esti
n
g
th
at
it
h
el
p
ed
an
al
y
ze
p
n
eu
m
o
n
ia
m
o
r
e
ac
c
u
r
ately
co
m
p
a
r
ed
to
th
e
ex
is
tin
g
m
eth
o
d
s
.
Ou
r
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
,
n
am
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
3
4
9
-
1
3
6
1
1358
NN
-
SVM,
with
a
h
ig
h
ac
c
u
r
a
cy
o
f
9
7
.
5
%,
class
if
ies
th
e
X
-
r
ay
co
r
r
ec
tl
y
am
o
n
g
n
o
r
m
al,
b
ac
ter
ial
p
n
eu
m
o
n
ia,
v
ir
al
p
n
eu
m
o
n
ia
,
b
ased
o
n
h
i
g
h
p
er
f
o
r
m
a
n
ce
m
etr
ics
s
u
ch
as
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
R
OC
-
AUC
v
alu
e.
T
h
is
will
h
ig
h
ly
aid
th
e
h
ea
lth
ca
r
e
in
d
u
s
tr
y
.
I
t
will
h
elp
r
ad
i
o
g
r
ap
h
er
s
to
an
aly
ze
X
-
r
a
y
s
im
m
ed
iately
,
th
u
s
r
esu
ltin
g
in
a
tim
e
-
s
av
in
g
ad
v
an
tag
e.
I
t
will
also
wo
r
k
as
a
s
ec
o
n
d
o
p
i
n
io
n
,
s
p
ec
if
ically
h
elp
f
u
l
in
a
b
u
s
y
h
ea
lth
ca
r
e
ce
n
ter
o
r
a
d
is
tan
t
r
eg
io
n
.
I
t
will
h
elp
d
o
cto
r
s
to
d
etec
t
d
is
ea
s
e
s
m
o
r
e
ac
cu
r
ately
,
th
u
s
lead
in
g
to
q
u
ick
er
d
iag
n
o
s
is
with
h
ig
h
er
co
n
f
id
en
ce
,
f
in
ally
h
av
in
g
a
p
o
s
itiv
e
ef
f
ec
t
o
n
th
e
p
atien
t.
H
o
wev
er
,
th
er
e
ar
e
a
f
ew
d
if
f
icu
lties
in
th
e
ap
p
r
o
ac
h
.
W
e
p
r
o
p
o
s
e
a
n
ew
ap
p
r
o
ac
h
ca
lled
NN
-
SVM,
a
s
m
aller
,
co
m
b
in
ed
ap
p
r
o
ac
h
to
3
-
class
p
n
e
u
m
o
n
ia
d
etec
tio
n
in
C
XR
im
ag
es
th
at
co
m
b
in
es
a
C
NN
with
an
R
B
F
-
SVM
class
if
ier
.
I
t
attain
s
a
n
ac
cu
r
ac
y
o
f
9
7
.
4
6
%
with
g
r
ea
t
p
e
r
f
o
r
m
an
ce
m
etr
ics.
T
h
e
tech
n
i
q
u
e
is
ea
s
ily
r
ep
r
o
d
u
cib
le,
s
ca
lab
le,
th
u
s
m
ak
in
g
it a
f
ea
s
ib
le
s
o
lu
tio
n
f
o
r
r
ad
i
o
lo
g
y
ass
is
tan
ce
.
6.
L
I
M
I
T
AT
I
O
NS
AN
D
F
UT
U
RE
WO
RK
S
Ho
wev
er
,
d
esp
ite
its
ef
f
icien
c
y
,
th
er
e
ex
is
t
s
ev
er
al
m
ajo
r
c
o
n
s
in
th
e
m
o
d
el.
T
h
e
p
r
esen
ce
o
f
d
ata
b
iases
is
a
s
er
io
u
s
co
n
b
ec
a
u
s
e
it
m
ig
h
t n
o
t
b
e
ab
le
t
o
r
e
p
r
esen
t
ap
p
r
o
p
r
iately
well
th
e
d
i
v
e
r
s
ity
o
f
p
n
eu
m
o
n
ia
in
f
ec
tio
n
s
in
h
ea
lth
ca
r
e
p
r
ac
tical
ap
p
licatio
n
s
o
r
s
ettin
g
s
.
T
h
e
ef
f
icie
n
cy
o
f
its
o
p
er
ati
o
n
s
m
ig
h
t
g
et
d
is
tu
r
b
ed
an
d
its
p
er
f
o
r
m
an
ce
s
co
u
l
d
b
e
r
ed
u
ce
d
b
ec
a
u
s
e
o
f
im
b
alan
c
es
in
d
ata
r
elate
d
to
d
if
f
er
en
t
k
in
d
s
o
f
p
n
e
u
m
o
n
ia
in
f
ec
tio
n
s
,
f
o
r
ex
am
p
le,
t
h
er
e
co
u
ld
b
e
a
n
im
b
alan
ce
i
n
im
a
g
es
o
f
p
atien
ts
with
eith
er
b
ac
ter
ial
p
n
eu
m
o
n
ia
o
r
v
ir
al
in
f
ec
tio
n
s
,
m
ea
n
in
g
m
o
r
e
im
ag
es o
f
th
o
s
e
with
b
ac
ter
ial
in
f
ec
tio
n
s
th
an
t
h
o
s
e
with
v
ir
al
in
f
ec
tio
n
s
.
T
h
e
ef
f
icien
cy
o
f
its
o
p
er
atio
n
s
m
ig
h
t
g
et
d
is
tu
r
b
ed
an
d
its
p
er
f
o
r
m
a
n
ce
s
co
u
ld
b
e
r
ed
u
c
ed
b
ec
au
s
e
it
co
u
ld
b
e
a
b
le
to
p
er
f
o
r
m
w
ell
o
n
c
o
m
m
o
n
k
i
n
d
s
o
r
cir
cu
m
s
tan
ce
s
b
u
t
m
ig
h
t
f
ail
to
p
e
r
f
o
r
m
well
wh
en
it
in
v
o
lv
es
u
n
co
m
m
o
n
a
n
d
n
o
t
as
well
-
r
ep
r
esen
ted
k
in
d
s
o
r
cir
cu
m
s
tan
ce
s
.
A
n
o
th
er
co
n
is
r
elate
d
to
its
in
ter
p
r
etab
ilit
y
b
ec
a
u
s
e
an
o
t
h
er
s
er
io
u
s
co
n
is
r
elate
d
to
its
in
ter
p
r
etab
ilit
y
b
ec
au
s
e
C
NNs,
wh
ich
b
elo
n
g
to
d
ee
p
lear
n
in
g
an
d
p
er
f
o
r
m
ex
ce
llen
tly
well
in
h
ea
lth
ca
r
e
ap
p
licatio
n
s
,
wo
r
k
in
an
i
n
ef
f
ec
t
iv
e
way
b
ec
au
s
e
o
f
its
d
eg
r
ee
o
f
o
b
s
cu
r
ity
b
ec
au
s
e
it c
o
u
ld
ac
t a
s
a
'
b
lack
b
o
x
'
b
ec
au
s
e
its
o
p
er
atio
n
s
an
d
p
er
f
o
r
m
an
ce
s
m
ig
h
t n
o
t
b
e
in
ter
p
r
etab
le
i
n
a
n
ef
f
ec
tiv
e
way
b
ec
au
s
e
it
co
u
l
d
lack
clea
r
tr
a
n
s
p
ar
en
c
y
r
e
g
ar
d
in
g
its
o
p
er
atio
n
s
an
d
p
er
f
o
r
m
an
ce
s
.
T
o
b
e
a
b
le
to
p
er
f
o
r
m
its
o
p
er
atio
n
s
in
an
ef
f
ec
tiv
e
way
b
ec
au
s
e
it
r
eq
u
ir
es
q
u
ality
an
d
well
-
p
r
o
ce
s
s
ed
im
ag
es
o
f
p
atien
ts
'
lu
n
g
s
b
ec
au
s
e
it
m
ig
h
t
g
et
d
is
tu
r
b
ed
b
ec
a
u
s
e
o
f
q
u
ality
-
r
elate
d
p
r
o
b
lem
s
.
H
en
ce
,
it
is
a
s
er
io
u
s
co
n
b
e
ca
u
s
e
its
ap
p
licatio
n
s
wo
u
ld
b
e
lim
ited
b
ec
au
s
e
it
m
ig
h
t
f
ail
to
p
er
f
o
r
m
well
b
ec
au
s
e
h
ea
lth
ca
r
e
ap
p
licatio
n
s
r
eg
ar
d
i
n
g
h
ea
lth
ca
r
e
a
n
d
r
elate
d
s
ec
to
r
s
wo
u
ld
b
e
lim
it
ed
b
ec
au
s
e
it
m
ig
h
t
n
o
t
b
e
ab
le
to
p
er
f
o
r
m
in
a
n
e
f
f
ec
tiv
e
way
b
ec
au
s
e
its
ap
p
licatio
n
s
m
ig
h
t
b
e
lim
ited
b
ec
a
u
s
e
it
m
ig
h
t
f
ail
to
p
er
f
o
r
m
in
an
e
f
f
ec
tiv
e
wa
y
b
ec
au
s
e
its
o
p
er
atio
n
s
m
i
g
h
t
g
et
d
is
tu
r
b
e
d
b
ec
a
u
s
e
it
m
ig
h
t
lack
clea
r
tr
an
s
p
ar
en
cy
r
eg
ar
d
in
g
its
o
p
er
atio
n
s
b
ec
au
s
e
its
o
p
er
atio
n
s
an
d
p
er
f
o
r
m
an
ce
s
m
i
g
h
t
n
o
t
b
e
in
ter
p
r
etab
le
i
n
an
ef
f
ec
tiv
e
way
b
ec
au
s
e
it
c
o
u
ld
ac
t
as
a
'
b
lack
b
o
x
'
b
ec
au
s
e
its
d
eg
r
ee
o
f
o
b
s
cu
r
ity
m
ig
h
t
b
e
in
cr
ea
s
ed
b
ec
au
s
e
its
o
p
er
atio
n
s
an
d
p
er
f
o
r
m
an
ce
s
co
u
ld
b
e
in
e
f
f
ec
tiv
e
b
ec
au
s
e
it c
o
u
ld
lack
clea
r
tr
an
s
p
ar
en
cy
b
ec
au
s
e
it
co
u
ld
b
e
s
er
io
u
s
b
ec
au
s
e
it
m
ig
h
t
f
ail
to
p
er
f
o
r
m
in
an
ef
f
ec
tiv
e
way
b
ec
au
s
e
its
ap
p
licatio
n
s
wo
u
ld
b
e
lim
ited
b
ec
au
s
e
it
m
ig
h
t
f
ail
to
b
e
ab
le
to
p
e
r
f
o
r
m
its
o
p
e
r
atio
n
s
b
ec
au
s
e
it
m
ig
h
t
lack
q
u
ality
an
d
well
-
p
r
o
ce
s
s
in
g
b
ec
au
s
e
it m
ig
h
t la
ck
clea
r
tr
an
s
p
ar
e
n
cy
.
Mo
r
eo
v
er
,
an
o
th
e
r
co
n
is
r
ela
ted
to
its
ef
f
ec
ts
o
n
h
ea
lth
ca
r
e
ap
p
licatio
n
s
.
H
en
ce
,
its
q
u
a
lity
m
ig
h
t
g
et
d
is
tu
r
b
e
d
b
ec
a
u
s
e
it
m
ig
h
t
f
ail
to
p
er
f
o
r
m
in
a
n
e
f
f
ec
tiv
e
way
b
ec
au
s
e
d
is
ea
s
es
o
r
i
n
f
ec
tio
n
s
r
elate
d
to
lu
n
g
s
m
ig
h
t
b
e
s
er
io
u
s
b
ec
a
u
s
e
it
co
u
ld
b
e
ab
le
to
d
eter
m
in
e
b
ec
au
s
e
its
ap
p
licatio
n
s
wo
u
ld
b
e
lim
ited
b
ec
au
s
e
it
m
i
g
h
t
f
ail
to
p
er
f
o
r
m
in
an
ef
f
ec
tiv
e
way
b
ec
au
s
e
its
o
p
er
atio
n
s
m
ig
h
t
g
et
d
i
s
tu
r
b
ed
b
ec
au
s
e
it
m
ig
h
t
lack
clea
r
tr
an
s
p
ar
en
c
y
b
ec
au
s
e
it
co
u
ld
b
e
s
er
io
u
s
b
e
ca
u
s
e
it
m
ig
h
t
ac
t
as
a
b
ar
r
ier
b
ec
au
s
e
it
co
u
ld
b
e
f
u
n
ctio
n
in
g
b
e
ca
u
s
e
it
m
ig
h
t
f
ail
to
p
er
f
o
r
m
in
an
ef
f
ec
ti
v
e
way
b
ec
au
s
e
its
q
u
ality
m
ig
h
t
g
et
d
is
tu
r
b
ed
.
H
en
ce
,
an
o
th
er
c
o
n
is
th
at
its
q
u
ality
m
ig
h
t
g
et
d
is
tu
r
b
ed
b
e
ca
u
s
e
it
m
ig
h
t
f
ail
to
p
er
f
o
r
m
in
an
ef
f
ec
tiv
e
wa
y
b
ec
au
s
e
it
co
u
ld
b
e
a
b
le
to
d
eter
m
in
e
b
ec
a
u
s
e
it
co
u
ld
b
e
f
u
n
ctio
n
in
g
b
ec
a
u
s
e
its
o
p
er
atio
n
s
m
ig
h
t
g
et
d
is
tu
r
b
ed
b
ec
au
s
e
it
m
ig
h
t
lack
clea
r
tr
a
n
s
p
ar
en
cy
b
ec
au
s
e
i
t
co
u
ld
b
e
s
er
io
u
s
b
ec
au
s
e
it
c
o
u
ld
ac
t
as
a
b
ar
r
ie
r
b
ec
au
s
e
it
m
ig
h
t
f
ail
to
p
er
f
o
r
m
in
an
ef
f
ec
tiv
e
way
b
ec
au
s
e
it
m
ig
h
t
lack
q
u
ality
an
d
well
-
p
r
o
ce
s
s
in
g
.
Ho
wev
er
,
d
esp
ite
its
s
ev
er
al
c
o
n
s
,
it h
as sev
er
al
p
r
o
s
,
an
d
th
o
s
e
p
r
o
s
m
ig
h
t.
W
ith
th
e
ad
v
e
n
t
o
f
tr
an
s
f
e
r
l
ea
r
n
in
g
s
tr
ateg
ies,
f
u
r
t
h
er
p
r
o
g
r
ess
m
ay
b
e
ac
h
iev
e
d
.
Me
d
ical
im
ag
e
ca
teg
o
r
izatio
n
h
as
m
ad
e
u
s
e
o
f
m
o
d
els
lik
e
Den
s
eNe
t
an
d
R
esNet,
wh
ich
wer
e
p
r
e
-
tr
ain
ed
o
n
m
ass
iv
e
d
atasets
lik
e
I
m
ag
eNe
t.
R
esea
r
ch
h
as
d
em
o
n
s
tr
ated
th
at
th
e
s
e
p
r
e
-
tr
ain
ed
m
o
d
els
ca
n
b
e
r
ef
in
ed
o
n
m
ed
ical
p
ictu
r
es
to
g
et
g
r
ea
t
p
er
f
o
r
m
an
ce
with
co
m
p
a
r
ativ
ely
s
m
a
ller
d
atasets
:
i)
p
atien
t
-
wis
e
t
em
p
o
r
al
m
o
d
elin
g
ac
r
o
s
s
s
er
ial
C
X
R
s
,
ii)
m
u
lti
-
m
o
d
al
in
teg
r
atio
n
with
clin
ic
al
m
etad
ata,
iii)
d
o
m
ain
ad
ap
tatio
n
f
o
r
cr
o
s
s
-
s
ite
r
o
b
u
s
tn
ess
,
an
d
iv
)
ca
lib
r
ated
r
is
k
s
tr
atif
icatio
n
f
o
r
tr
iag
e
.
ACK
NO
WL
E
DG
M
E
N
T
S
We
w
o
u
l
d
l
ik
e
to
t
h
a
n
k
Da
y
a
n
an
d
a
Sa
g
ar
U
n
i
v
e
r
s
it
y
f
o
r
s
u
p
p
o
r
t
i
n
c
o
m
p
l
eti
n
g
th
is
r
es
ea
r
c
h
wo
r
k
.
Evaluation Warning : The document was created with Spire.PDF for Python.