I
nte
rna
t
io
na
l J
o
urna
l o
f
I
nfo
rm
a
t
ics a
nd
Co
m
m
un
ica
t
io
n T
ec
hn
o
lo
g
y
(
I
J
-
I
CT
)
Vo
l.
1
4
,
No
.
1
,
A
p
r
il
20
2
5
,
p
p
.
9
1
~
10
1
I
SS
N:
2252
-
8
7
7
6
,
DOI
:
1
0
.
1
1
5
9
1
/iji
ct
.
v
1
4
i
1
.
pp
91
-
10
1
91
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ict.
ia
esco
r
e.
co
m
Explain
a
ble ze
ro
-
sho
t
lea
rning
and
t
ra
nsfer
learning
f
o
r re
a
l
time India
n
h
ea
lt
hca
re
Swa
t
i Sa
ig
a
o
nk
a
r
,
Va
ibh
a
v
Na
ra
wa
de
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
R
a
mr
a
o
A
d
i
k
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
D
Y
P
a
t
i
l
D
e
e
me
d
t
o
b
e
U
n
i
v
e
r
si
t
y
,
N
e
r
u
l
,
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
Au
g
2
8
,
2
0
2
4
R
ev
is
ed
Oct
1
5
,
2
0
2
4
Acc
ep
ted
No
v
1
9
,
2
0
2
4
Cli
n
ica
l
n
o
te
re
se
a
rc
h
is
g
lo
b
a
ll
y
re
c
o
g
n
ize
d
,
b
u
t
wo
r
k
o
n
re
a
l
-
ti
m
e
d
a
ta,
p
a
rti
c
u
larly
fro
m
I
n
d
ia,
is
st
il
l
lag
g
i
n
g
.
T
h
is
st
u
d
y
i
n
it
iate
d
b
y
trai
n
in
g
m
o
d
e
ls
o
n
m
e
d
ica
l
in
fo
rm
a
ti
o
n
m
a
rt
fo
r
in
ten
siv
e
c
a
re
(M
IM
IC)
c
li
n
ica
l
n
o
tes
,
fo
c
u
sin
g
o
n
c
o
n
d
it
i
o
n
s l
ik
e
c
h
ro
n
ic
k
id
n
e
y
d
ise
a
se
(CKD
),
m
y
o
c
a
rd
ial
in
fa
rc
ti
o
n
(M
I),
a
n
d
a
sth
m
a
u
sin
g
th
e
str
u
c
tu
re
d
m
e
d
ica
l
d
o
m
a
in
b
id
irec
ti
o
n
a
l
e
n
c
o
d
e
r
re
p
re
se
n
t
a
ti
o
n
s
fro
m
tran
sfo
rm
e
rs
(
S
M
DBERT
)
m
o
d
e
l.
S
u
b
se
q
u
e
n
tl
y
,
th
e
se
m
o
d
e
ls
we
re
a
p
p
li
e
d
to
a
n
In
d
ia
n
d
a
tas
e
t
o
b
tai
n
e
d
fr
o
m
tw
o
h
o
s
p
it
a
ls.
Th
e
k
e
y
d
iffere
n
c
e
b
e
twe
e
n
p
u
b
li
c
l
y
a
v
a
il
a
b
le
d
a
tas
e
ts
a
n
d
re
a
l
-
ti
m
e
d
a
ta
li
e
s
in
t
h
e
p
re
v
a
len
c
e
o
f
c
e
rtain
d
ise
a
se
s.
F
o
r
e
x
a
m
p
le,
in
a
re
a
l
-
ti
m
e
se
tt
in
g
,
tu
b
e
rc
u
l
o
sis
m
a
y
e
x
ist,
b
u
t
th
e
M
IM
IC
d
a
tas
e
t
lac
k
s
c
o
rre
sp
o
n
d
in
g
c
li
n
ica
l
n
o
tes
.
T
h
u
s,
a
n
in
n
o
v
a
ti
v
e
a
p
p
r
o
a
c
h
wa
s
d
e
v
e
lo
p
e
d
b
y
c
o
m
b
i
n
in
g
a
fin
e
-
t
u
n
e
d
S
M
DBERT
m
o
d
e
l
wit
h
a
c
u
sto
m
ize
d
z
e
ro
-
sh
o
t
lea
rn
i
n
g
m
e
th
o
d
to
e
ffe
c
ti
v
e
ly
a
n
a
l
y
z
e
tu
b
e
rc
u
l
o
sis
-
re
late
d
c
li
n
ica
l
n
o
tes
.
A
n
o
t
h
e
r
re
se
a
rc
h
g
a
p
is
th
e
lac
k
o
f
e
x
p
lain
a
b
il
it
y
b
e
c
a
u
se
d
e
e
p
lea
rn
in
g
(D
L)
m
o
d
e
ls
a
re
in
h
e
r
e
n
tl
y
b
lac
k
-
b
o
x
.
T
o
fu
rt
h
e
r
stre
n
g
t
h
e
n
th
e
re
li
a
b
il
it
y
o
f
t
h
e
m
o
d
e
ls,
lo
c
a
l
in
terp
re
tab
le
m
o
d
e
l
-
a
g
n
o
st
ic
e
x
p
lan
a
ti
o
n
s
(LI
M
E)
a
n
d
sh
a
p
ley
a
d
d
it
i
v
e
e
x
p
lan
a
ti
o
n
s
(S
HA
P
)
e
x
p
lan
a
ti
o
n
s
we
re
p
ro
jec
ted
a
lo
n
g
with
n
a
rra
ti
v
e
e
x
p
lan
a
ti
o
n
s
w
h
ich
g
e
n
e
ra
ted
e
x
p
lan
a
ti
o
n
s
i
n
a
n
a
tu
ra
l
lan
g
u
a
g
e
f
o
rm
a
t.
Th
u
s,
th
e
re
se
a
rc
h
p
r
o
v
i
d
e
s
a
sig
n
ifi
c
a
n
t
c
o
n
tri
b
u
ti
o
n
wi
th
e
n
se
m
b
le
tec
h
n
iq
u
e
o
f
z
e
ro
-
sh
o
t
lea
rn
in
g
a
n
d
S
M
DBERT
m
o
d
e
l
wit
h
a
n
a
c
c
u
ra
c
y
o
f
0
.
9
2
a
s
a
g
a
in
st
th
e
sp
e
c
ialize
d
m
o
d
e
ls
li
k
e
s
c
ien
ti
fic
BERT
(
S
CIBERT
)
,
b
i
o
m
e
d
ica
l
BERT
(
BIOBERT
)
a
n
d
c
li
n
ica
l
Bio
BERT
.
K
ey
w
o
r
d
s
:
Dis
til
B
E
R
T
E
x
p
lain
ab
ilit
y
I
n
d
ian
c
lin
ical
n
o
tes
T
r
an
s
f
er
lear
n
i
n
g
Z
er
o
-
s
h
o
t le
ar
n
in
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
:
Swati Saig
ao
n
k
ar
Dep
ar
tm
en
t
o
f
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
R
am
r
a
o
Ad
ik
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
D Y P
a
ti
l
De
e
m
e
d
to
b
e
Un
i
v
e
rsity
Ne
ru
l,
In
d
ia
E
m
ail:
sw
a
.
sa
i.
rt2
1
@d
y
p
a
ti
l
.
e
d
u
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
ad
o
p
tio
n
o
f
elec
tr
o
n
ic
h
ea
lth
r
ec
o
r
d
(
E
HR
)
s
y
s
tem
s
is
in
cr
ea
s
in
g
s
ig
n
if
ican
tly
[
1
]
.
Pu
b
licly
ac
ce
s
s
ib
le
r
eso
u
r
ce
s
s
u
ch
as
t
h
e
m
ed
ical
in
f
o
r
m
atio
n
m
ar
t
f
o
r
in
ten
s
iv
e
ca
r
e
(
MI
MI
C
)
,
an
d
en
h
an
ce
d
cr
itical
ca
r
e
u
n
it
(
eI
C
U
)
co
llab
o
r
ativ
e
r
esear
ch
d
atab
ases
p
r
o
v
id
e
v
alu
ab
le
in
f
o
r
m
atio
n
o
n
cr
itical
ly
ill
I
C
U
p
atien
ts
.
Sp
ec
if
ically
,
th
e
MI
MI
C
I
I
I
d
atab
ase
in
clu
d
es
d
ata
f
r
o
m
o
v
er
5
0
,
0
0
0
I
C
U
ad
m
is
s
io
n
s
[
2
]
,
a
n
d
t
h
e
eI
C
U
d
ataset
co
v
er
s
n
ea
r
ly
two
m
illi
o
n
ad
m
is
s
io
n
s
[
3
]
.
Acc
o
r
d
in
g
to
[
4
]
,
wh
ile
th
ese
d
atasets
ar
e
wid
ely
av
ailab
le,
th
er
e
is
a
n
o
tab
le
lack
o
f
d
atasets
f
r
o
m
I
n
d
ian
p
atien
t
p
o
p
u
la
tio
n
s
.
C
h
r
o
n
ic
d
is
ea
s
es
ar
e
d
ef
in
ed
a
s
co
n
d
itio
n
s
la
s
tin
g
m
o
r
e
th
an
th
r
ee
m
o
n
th
s
[
5
]
,
an
d
ea
r
ly
d
etec
tio
n
is
cr
u
cial
f
o
r
im
p
r
o
v
in
g
d
iag
n
o
s
is
an
d
m
an
ag
em
en
t.
Glo
b
all
y
,
h
ea
r
t
f
ailu
r
e
af
f
ec
ts
ap
p
r
o
x
im
ately
2
6
m
illi
o
n
p
eo
p
le
an
n
u
ally
,
p
o
s
in
g
s
ig
n
i
f
ican
t
ch
allen
g
es
f
o
r
s
u
r
g
eo
n
s
,
ca
r
d
io
lo
g
is
ts
,
an
d
clin
ician
s
in
p
r
ed
ictin
g
its
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
4
,
No
.
1
,
A
p
r
il
20
2
5
:
91
-
10
1
92
s
tar
t
[
6
]
.
Ast
h
m
a,
id
en
tifie
d
as
th
e
m
o
s
t
p
r
ev
alen
t
ch
r
o
n
ic
p
u
lm
o
n
ar
y
co
n
d
itio
n
wo
r
ld
wid
e,
f
u
r
th
e
r
u
n
d
er
s
co
r
es
th
e
n
ee
d
f
o
r
ef
f
e
ctiv
e
m
an
ag
em
e
n
t
[
7
]
.
C
h
r
o
n
ic
k
id
n
ey
d
is
ea
s
e
(
C
KD)
,
a
p
r
o
g
r
ess
iv
e
ailm
en
t
af
f
ec
tin
g
o
v
er
1
0
%
o
f
th
e
wo
r
ld
'
s
p
o
p
u
latio
n
,
af
f
ec
tin
g
o
v
er
8
0
0
m
illi
o
n
i
n
d
iv
id
u
als
[
8
]
.
Giv
e
n
th
eir
h
i
g
h
p
r
ev
alen
ce
,
th
ese
th
r
ee
c
o
n
d
it
io
n
s
ar
e
th
e
p
r
im
ar
y
f
o
cu
s
o
f
o
u
r
r
esear
c
h
.
Du
r
in
g
o
u
r
an
al
y
s
is
o
f
I
n
d
ia
n
d
ata,
we
en
co
u
n
te
r
ed
ca
s
es
o
f
tu
b
e
r
cu
lo
s
is
.
Ho
wev
er
,
wh
e
n
ap
p
l
y
in
g
th
e
s
tr
u
ctu
r
ed
m
ed
ical
d
o
m
ain
b
i
d
ir
ec
tio
n
al
en
co
d
er
r
ep
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
e
r
s
(
SMDBER
T
)
m
o
d
el,
wh
ic
h
was
tr
ain
ed
o
n
th
e
MI
MI
C
d
ataset
lack
in
g
tu
b
er
c
u
lo
s
is
ca
s
es,
th
ese
in
s
tan
ce
s
wer
e
m
i
s
clas
s
if
i
ed
.
T
h
is
h
ig
h
lig
h
ts
th
e
ch
alle
n
g
e
o
f
tr
an
s
f
er
r
in
g
m
o
d
els
ac
r
o
s
s
d
atase
ts
wi
th
d
if
f
er
en
t
d
is
tr
ib
u
tio
n
s
.
As
d
em
o
n
s
tr
ated
in
[
9
]
,
wh
ile
m
o
d
els
tr
ain
ed
o
n
o
n
e
d
ataset
o
f
ten
p
er
f
o
r
m
well
o
n
s
im
ilar
d
atasets
,
ze
r
o
-
s
h
o
t
lear
n
in
g
tech
n
iq
u
es
ca
n
b
e
lev
er
ag
ed
to
ad
d
r
ess
th
e
lim
itatio
n
s
o
f
d
atasets
lack
in
g
s
u
f
f
icien
t e
x
am
p
les.
Han
d
lin
g
u
n
s
tr
u
ct
u
r
ed
d
ata,
s
u
ch
as
m
ed
ical
n
o
tes,
p
r
esen
t
s
s
ev
er
al
u
n
iq
u
e
d
if
f
icu
lties
.
On
e
o
f
th
e
m
ain
d
if
f
icu
lties
ar
is
es
f
r
o
m
t
h
e
d
iv
er
s
ity
an
d
lack
o
f
o
r
g
a
n
izatio
n
with
in
th
e
d
ata
[
1
0
]
,
[
1
1
]
.
C
lin
ical
n
o
tes
ar
e
o
f
ten
wr
itten
in
a
p
h
r
ase
-
lik
e
m
an
n
er
with
o
u
t
s
t
an
d
ar
d
ized
g
r
am
m
atica
l
s
tr
u
ct
u
r
es,
wh
ich
co
m
p
licates
th
eir
an
aly
s
is
.
I
n
th
e
ca
s
e
o
f
I
n
d
ian
d
ata,
t
h
er
e
is
an
ad
d
ed
c
o
m
p
lex
ity
d
u
e
to
th
e
p
r
ed
o
m
i
n
an
ce
o
f
n
o
n
-
d
ig
ital
f
o
r
m
ats,
wh
ich
m
ak
es
it
ch
a
llen
g
in
g
to
in
teg
r
ate
t
h
is
in
f
o
r
m
atio
n
s
ea
m
less
ly
in
to
d
ig
ital
s
y
s
tem
s
.
T
h
ese
clin
ical
n
o
tes
wer
e
n
o
t
o
r
ig
in
a
lly
d
ig
ital,
in
co
n
tr
ast
to
d
atasets
th
at
ar
e
ac
ce
s
s
ib
le
to
th
e
p
u
b
lic,
co
m
p
licatin
g
th
eir
ac
ce
s
s
ib
ilit
y
an
d
an
aly
s
is
.
T
h
e
p
r
ac
tice
o
f
tr
a
n
s
f
er
lear
n
in
g
,
wh
ich
in
v
o
lv
es
u
s
in
g
a
m
o
d
el
cr
ea
ted
f
o
r
o
n
e
task
as
th
e
f
o
u
n
d
atio
n
f
o
r
a
m
o
d
el
o
n
a
d
if
f
er
en
t
task
,
h
as
d
em
o
n
s
tr
ate
d
p
ar
ticu
lar
l
y
ef
f
ec
tiv
e
in
h
ea
l
th
ca
r
e
ap
p
licatio
n
s
.
T
r
an
s
f
o
r
m
e
r
-
b
ased
m
o
d
els,
s
u
ch
as
B
E
R
T
,
h
av
e
r
ev
o
lu
tio
n
ized
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
NL
P
)
by
en
ab
lin
g
m
o
d
els
to
u
n
d
er
s
tan
d
co
n
tex
t
m
o
r
e
d
ee
p
ly
th
r
o
u
g
h
atten
tio
n
m
ec
h
an
is
m
s
.
T
h
ese
m
o
d
els
ex
ce
l
in
task
s
in
v
o
lv
in
g
u
n
s
tr
u
ctu
r
e
d
clin
ical
d
ata,
s
u
ch
as
ex
tr
ac
ti
n
g
m
ea
n
i
n
g
f
u
l
i
n
s
ig
h
ts
f
r
o
m
E
HR
an
d
clin
ical
n
o
tes,
m
ak
in
g
th
em
in
v
alu
ab
le
to
o
ls
f
o
r
a
d
v
an
cin
g
p
atien
t c
a
r
e.
T
h
e
in
ter
p
r
etab
ilit
y
o
f
p
r
ed
i
ctiv
e
m
o
d
els
u
s
in
g
u
n
s
tr
u
ct
u
r
ed
clin
ical
d
ata
h
as
g
r
o
w
n
to
b
e
an
im
p
o
r
tan
t
f
ield
o
f
s
tu
d
y
in
r
ec
en
t
y
ea
r
s
.
As
m
ac
h
in
e
lea
r
n
in
g
(
ML
)
a
n
d
NL
P
tech
n
i
q
u
es
c
o
n
tin
u
e
to
ev
o
lv
e
,
th
eir
ap
p
licatio
n
to
clin
ical
n
o
tes,
a
r
ich
s
o
u
r
ce
o
f
u
n
s
tr
u
ctu
r
ed
d
ata,
p
o
s
s
ess
es
th
e
ab
ilit
y
to
tr
an
s
f
o
r
m
h
ea
lth
ca
r
e.
Ho
we
v
er
,
th
e
c
o
m
p
lex
ity
an
d
o
p
ac
ity
o
f
th
ese
m
o
d
els,
o
f
ten
r
e
f
er
r
e
d
to
as
"
b
lack
b
o
x
es,"
p
o
s
e
s
ig
n
if
ican
t c
h
allen
g
es in
clin
ic
al
s
ettin
g
s
wh
er
e
tr
an
s
p
ar
en
cy
an
d
u
n
d
er
s
tan
d
i
n
g
ar
e
p
ar
am
o
u
n
t.
T
r
an
s
f
o
r
m
e
r
-
b
ased
m
o
d
els
ar
e
in
cr
ea
s
in
g
ly
u
s
ed
i
n
h
ea
lth
ca
r
e
f
o
r
v
ar
io
u
s
p
r
ed
ictiv
e
task
s
,
in
clu
d
in
g
f
o
r
ec
asti
n
g
m
o
r
tality
r
ates
[
12
]
,
p
r
ed
ictin
g
p
atien
t
r
ea
d
m
is
s
io
n
s
[
13
]
,
a
n
d
esti
m
atin
g
h
o
s
p
ital
s
tay
d
u
r
atio
n
s
[
14
]
.
T
h
ese
m
o
d
els
h
av
e
als
o
p
r
o
v
en
ef
f
ec
tiv
e
i
n
task
s
s
u
ch
as
ex
tr
ac
tin
g
en
titi
es
[
1
5
]
,
[
1
6
]
,
i
d
en
tify
in
g
p
h
en
o
t
y
p
ic
ch
ar
ac
ter
is
tics
[1
7]
,
[
1
8
]
,
m
o
d
elin
g
p
atien
t
tr
ajec
to
r
ies,
an
d
elu
cid
atin
g
r
e
latio
n
s
h
ip
s
am
o
n
g
d
if
f
er
en
t
m
ed
ical
e
n
titi
es.
T
h
e
in
teg
r
atio
n
o
f
tr
an
s
f
o
r
m
er
m
o
d
els
in
to
h
ea
lth
ca
r
e
r
esear
ch
d
em
o
n
s
tr
ates
th
eir
ad
ap
tab
ilit
y
an
d
ef
f
icac
y
in
tack
lin
g
a
wid
e
r
an
g
e
o
f
is
s
u
es
,
ad
v
an
cin
g
clin
ical
d
ec
is
io
n
-
m
ak
in
g
s
y
s
tem
s
,
an
d
en
h
an
cin
g
o
u
r
u
n
d
er
s
tan
d
in
g
o
f
co
m
p
lex
m
ed
ical
c
o
n
d
itio
n
s
.
I
n
h
ea
lth
ca
r
e
d
ata
a
n
aly
s
is
,
a
r
an
g
e
o
f
m
eth
o
d
o
lo
g
ies
ar
e
em
p
lo
y
e
d
,
f
r
o
m
r
u
le
-
b
ased
s
y
s
tem
s
to
ad
v
an
ce
d
ML
an
d
d
ee
p
lear
n
in
g
(
DL
)
tech
n
i
q
u
es.
R
u
le
-
b
ased
m
eth
o
d
s
d
ep
e
n
d
o
n
es
tab
lis
h
ed
g
u
id
elin
es
d
er
iv
ed
f
r
o
m
s
p
ec
ialized
k
n
o
wled
g
e
,
b
u
t
th
ese
s
y
s
tem
s
ca
n
b
e
in
f
lex
ib
le
wh
en
f
ac
ed
with
n
o
v
el
d
ata,
o
f
te
n
r
eq
u
ir
in
g
u
p
d
ates
o
r
m
o
d
if
ica
tio
n
s
to
h
an
d
le
n
ew
co
n
ce
p
ts
.
ML
m
eth
o
d
s
,
o
n
th
e
o
th
er
h
a
n
d
,
lear
n
f
r
o
m
d
at
a
d
ir
ec
tly
.
Su
p
er
v
is
ed
lear
n
in
g
,
f
o
r
ex
am
p
le,
u
s
es
lab
eled
d
a
tasets
to
p
er
f
o
r
m
task
s
lik
e
class
if
icatio
n
,
wh
ile
u
n
s
u
p
er
v
is
ed
lear
n
in
g
id
en
tifi
es
p
atter
n
s
in
u
n
lab
eled
d
ata.
ML
alg
o
r
ith
m
s
s
u
ch
as
lo
g
is
tic
r
eg
r
ess
io
n
(
L
R
)
,
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
i
n
es
(
SVM)
,
an
d
r
a
n
d
o
m
f
o
r
ests
ar
e
o
f
ten
ch
allen
g
ed
b
y
h
ig
h
-
d
im
e
n
s
io
n
al,
m
u
ltim
o
d
al
d
atasets
.
DL
tech
n
iq
u
es,
in
clu
d
in
g
co
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN
s
)
,
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs
)
,
an
d
tr
an
s
f
o
r
m
e
r
m
o
d
els
lik
e
B
E
R
T
an
d
d
is
till
ed
B
E
R
T
(
Dis
t
ilB
E
R
T
)
,
h
av
e
d
e
m
o
n
s
tr
ated
s
u
p
er
i
o
r
p
er
f
o
r
m
an
ce
in
e
x
tr
ac
tin
g
f
ea
tu
r
es
f
r
o
m
r
aw
d
ata
a
n
d
u
n
d
e
r
s
tan
d
in
g
co
n
tex
t.
B
E
R
T
is
p
ar
ticu
lar
ly
ef
f
ec
tiv
e
f
o
r
g
e
n
er
atin
g
p
o
wer
f
u
l
r
ep
r
e
s
en
tatio
n
s
f
r
o
m
u
n
lab
ele
d
d
ata
d
u
e
to
its
ab
ilit
y
to
u
n
d
e
r
s
tan
d
co
n
tex
t
in
b
o
th
d
ir
ec
tio
n
s
,
m
a
k
in
g
it
h
ig
h
ly
s
u
itab
le
f
o
r
v
ar
io
u
s
ML
task
s
,
i
n
clu
d
in
g
tex
t
class
if
icatio
n
.
Dis
tilB
E
R
T
,
a
lig
h
ter
an
d
m
o
r
e
ef
f
icie
n
t
v
er
s
io
n
o
f
B
E
R
T
,
ac
h
iev
es
s
im
ilar
p
er
f
o
r
m
an
ce
le
v
els
wh
ile
b
ein
g
c
o
m
p
u
tatio
n
ally
less
d
em
an
d
in
g
[
19
]
.
I
n
s
p
ec
if
ic
d
o
m
ain
s
s
u
ch
as
h
ea
lth
ca
r
e,
d
o
m
ain
-
a
d
ap
ted
m
o
d
els
lik
e
SMDBE
R
T
,
wh
ich
in
co
r
p
o
r
ate
s
p
ec
ialized
m
ed
ical
k
n
o
wled
g
e,
h
a
v
e
s
h
o
wn
to
o
u
tp
er
f
o
r
m
m
o
r
e
g
en
er
al
m
o
d
els
lik
e
Di
s
til
B
E
R
T
d
u
e
to
th
eir
en
h
a
n
ce
d
ca
p
ab
ilit
y
to
lev
er
a
g
e
d
o
m
ain
-
s
p
ec
if
ic
in
f
o
r
m
atio
n
[
20
]
.
R
esear
ch
o
n
u
s
in
g
ze
r
o
-
s
h
o
t
lear
n
in
g
m
o
d
els
s
p
ec
if
ically
f
o
r
p
r
e
d
ictin
g
tu
b
e
r
cu
lo
s
is
(
T
B
)
is
r
elativ
ely
lim
ited
.
Mo
s
t
s
tu
d
ies
f
o
cu
s
o
n
m
o
r
e
tr
ad
itio
n
a
l
ML
ap
p
r
o
ac
h
es
o
r
DL
tech
n
iq
u
es
ap
p
lied
t
o
m
ed
ical
im
ag
in
g
d
ata
lik
e
c
h
est
X
-
r
ay
s
.
W
h
ile
th
er
e
h
a
v
e
b
ee
n
s
o
m
e
p
r
o
g
r
ess
,
th
e
a
p
p
licatio
n
o
f
th
ese
m
o
d
els
f
o
r
T
B
p
r
ed
ictio
n
u
s
in
g
clin
ical
d
ata
o
r
n
o
tes
r
em
a
in
s
a
r
elativ
ely
u
n
ex
p
l
o
r
ed
ar
ea
.
C
ap
ellan
-
Ma
r
tin
et
a
l.
[
2
1
]
d
is
cu
s
s
es
th
e
u
s
e
o
f
v
is
io
n
tr
a
n
s
f
o
r
m
e
r
s
(
ViT
)
i
n
a
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
p
a
r
ad
ig
m
f
o
r
im
p
r
o
v
ed
TB
d
etec
tio
n
in
c
h
est
X
-
r
ay
s
.
Z
er
o
-
s
h
o
t
p
ed
iatr
ic
T
B
d
etec
tio
n
is
a
m
eth
o
d
,
wh
ich
h
as
s
h
o
wn
ap
p
r
ec
iab
le
g
ain
s
in
T
B
d
etec
tio
n
ef
f
icac
y
.
wh
en
co
m
p
a
r
ed
t
o
f
u
lly
-
s
u
p
er
v
is
ed
m
o
d
els.
T
h
e
wo
r
k
in
[
2
2
]
p
r
o
p
o
s
es
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
z
ero
-
s
h
o
t le
a
r
n
in
g
a
n
d
tr
a
n
s
fer lea
r
n
in
g
fo
r
r
ea
l time
…
(
S
w
a
ti S
a
ig
a
o
n
ka
r
)
93
g
en
er
alize
d
ze
r
o
-
s
h
o
t
lea
r
n
in
g
(
GZ
SL)
m
eth
o
d
u
s
in
g
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
(
SS
L
)
f
o
r
m
ed
ical
im
a
g
e
class
if
icatio
n
.
I
t
f
o
cu
s
es
o
n
tr
ain
in
g
a
f
ea
tu
r
e
g
en
er
at
o
r
an
d
ch
o
o
s
in
g
an
ch
o
r
v
ec
to
r
s
f
r
o
m
v
ar
io
u
s
d
is
ea
s
e
class
es.
T
h
e
wo
r
k
in
[
2
3
]
in
tr
o
d
u
ce
s
a
m
u
lti
-
lab
el
GZ
SL
n
etwo
r
k
th
at
u
s
es
ch
est
X
-
r
ay
im
ag
es
to
s
im
u
ltan
eo
u
s
ly
p
r
e
d
ict
s
ev
er
a
l
d
is
ea
s
es,
b
o
th
s
ee
n
an
d
u
n
s
ee
n
.
I
t
u
s
es
a
v
is
u
al
r
ep
r
esen
tatio
n
g
u
i
d
ed
b
y
s
em
an
tics
tak
en
f
r
o
m
a
s
u
b
s
tan
tial c
o
r
p
u
s
o
f
m
ed
ical
liter
atu
r
e.
T
h
er
e
ar
e
s
o
m
e
wo
r
k
s
u
s
in
g
p
r
o
m
p
t
-
b
ased
lar
g
e
la
n
g
u
a
g
e
m
o
d
els
(
LLMs
)
.
Z
h
u
et
a
l.
[
2
4
]
lo
o
k
s
in
to
th
e
f
lex
ib
ilit
y
o
f
L
L
Ms
lik
e
GPT
-
4
to
E
HR
d
ata
f
o
r
ze
r
o
-
s
h
o
t
clin
ical
p
r
ed
ictio
n
.
I
n
cr
u
cial
task
s
lik
e
m
o
r
tality
,
len
g
th
-
of
-
s
tay
,
an
d
3
0
-
d
ay
r
ea
d
m
is
s
io
n
,
it
e
x
h
ib
its
en
h
an
ce
d
p
r
ed
ictio
n
p
er
f
o
r
m
an
ce
.
Similar
ly
,
Hea
lth
Pro
m
p
t
[
2
5
]
,
u
tili
ze
s
p
r
o
m
p
t
-
b
ased
lear
n
in
g
,
allo
win
g
p
r
e
-
tr
ain
e
d
lan
g
u
ag
e
m
o
d
el
s
th
at
d
o
n
'
t
r
eq
u
ir
e
m
o
r
e
tr
ain
in
g
d
ata
to
ad
ju
s
t
to
n
ew
task
s
.
T
h
e
s
tu
d
y
d
em
o
n
s
tr
ates
th
at
Hea
lth
Pro
m
p
t
ca
n
f
u
n
ctio
n
ef
f
ec
tiv
ely
an
d
ef
f
icien
tly
in
ca
p
tu
r
in
g
th
e
co
n
tex
t
o
f
clin
ical
tex
ts
in
v
ar
io
u
s
clin
ical
NL
P
task
s
,
s
h
o
wca
s
in
g
th
e
Z
SL'
s
p
o
ten
tial to
im
p
r
o
v
e
clin
ical
j
u
d
g
m
e
n
t
an
d
r
ed
u
ce
d
ep
en
d
en
cy
o
n
lar
g
e
an
n
o
tated
d
atasets
.
T
h
is
o
v
er
v
iew
h
ig
h
lig
h
ts
th
e
u
n
tap
p
ed
p
o
te
n
tial
o
f
ze
r
o
-
s
h
o
t
lear
n
in
g
f
o
r
T
B
p
r
e
d
ictio
n
u
s
in
g
clin
ical
d
ata,
s
u
g
g
esti
n
g
a
p
r
o
m
is
in
g
d
ir
ec
tio
n
f
o
r
f
u
tu
r
e
r
es
ea
r
ch
b
e
y
o
n
d
th
e
m
o
r
e
c
o
m
m
o
n
ly
ex
p
lo
r
e
d
ar
ea
s
o
f
m
ed
ical
im
ag
in
g
a
n
d
E
HR
-
b
ased
p
r
ed
ictio
n
s
.
Pre
d
ictiv
e
m
o
d
els
u
s
in
g
clin
ical
n
o
tes
f
ac
e
u
n
iq
u
e
ch
allen
g
es
d
u
e
to
th
e
n
atu
r
e
o
f
th
e
d
ata.
Sev
e
r
al
s
tu
d
i
es
h
av
e
h
ig
h
lig
h
ted
t
h
e
i
m
p
o
r
tan
ce
o
f
m
o
d
el
in
ter
p
r
etab
ilit
y
in
h
ea
lth
ca
r
e.
T
h
e
in
ter
p
r
etab
ilit
y
o
f
ML
m
o
d
els
is
a
cr
itical
asp
ec
t
o
f
h
ea
lth
ca
r
e,
as
u
n
d
er
s
tan
d
i
n
g
t
h
e
r
atio
n
ale
b
eh
in
d
m
o
d
el
p
r
e
d
ictio
n
s
is
ess
en
tial
f
o
r
e
n
s
u
r
in
g
tr
u
s
t
a
n
d
im
p
r
o
v
in
g
p
atien
t
ca
r
e.
T
wo
p
r
o
m
in
e
n
t
tech
n
iq
u
es
f
o
r
in
ter
p
r
etin
g
c
o
m
p
lex
m
o
d
els
ar
e
lo
ca
l
in
ter
p
r
eta
b
le
m
o
d
el
-
a
g
n
o
s
tic
ex
p
lan
atio
n
s
(
L
I
ME
)
an
d
s
h
ap
ley
ad
d
itiv
e
ex
p
lan
atio
n
s
(
SH
AP)
.
T
h
e
wo
r
k
in
[
2
6
]
h
ig
h
lig
h
ts
th
e
ap
p
licatio
n
o
f
L
I
ME
an
d
SHAP
in
d
etec
t
in
g
a
lzh
eim
er
's
d
is
ea
s
e,
d
em
o
n
s
tr
atin
g
h
o
w
th
es
e
m
eth
o
d
s
ca
n
in
cr
ea
s
e
th
e
tr
an
s
p
ar
en
cy
an
d
r
eliab
ilit
y
o
f
ar
tific
i
al
in
tellig
en
ce
(
AI
)
-
b
ased
p
r
ed
ictio
n
s
.
T
h
e
r
e
v
ie
w
em
p
h
asizes
th
at
alth
o
u
g
h
L
I
ME
an
d
SHAP
p
r
o
v
id
e
s
ig
n
if
ican
t
in
s
ig
h
ts
in
to
m
o
d
el
d
ec
is
io
n
s
,
th
ey
also
h
av
e
lim
itatio
n
s
,
o
n
e
b
ein
g
th
e
n
ee
d
f
o
r
tailo
r
in
g
to
s
p
ec
if
i
c
m
ed
ical
co
n
tex
ts
.
A
n
o
th
er
s
tu
d
y
[
2
7
]
in
v
esti
g
ates
th
e
u
s
e
o
f
L
I
ME
a
n
d
SHA
P
f
o
r
a
u
to
n
o
m
o
u
s
d
is
ea
s
e
p
r
e
d
ictio
n
,
n
o
tin
g
t
h
at
wh
ile
th
ese
tech
n
iq
u
es
e
x
ce
l
in
g
en
er
atin
g
lo
ca
l
ex
p
lan
ati
o
n
s
,
th
ey
f
ac
e
c
h
allen
g
es
wh
en
ap
p
lied
to
m
o
r
e
co
m
p
lex
m
o
d
els
o
r
lar
g
e
r
d
ata
s
ets.
T
h
e
r
esear
ch
u
n
d
er
s
co
r
es
th
e
i
m
p
o
r
tan
ce
o
f
th
ese
in
ter
p
r
etab
ilit
y
m
eth
o
d
s
in
m
ak
in
g
AI
-
d
r
iv
en
d
ec
is
io
n
s
m
o
r
e
u
n
d
er
s
tan
d
ab
le,
b
u
t
also
p
o
in
ts
o
u
t
th
e
d
if
f
icu
lties
in
s
ca
lin
g
th
ese
ex
p
lan
atio
n
s
ef
f
ec
tiv
ely
.
As
a
r
esu
lt,
o
u
r
r
esear
ch
is
m
o
tiv
ated
b
y
t
h
e
n
ee
d
t
o
s
im
p
lify
ex
p
lai
n
ab
ilit
y
b
y
u
s
in
g
clea
r
an
d
ac
ce
s
s
ib
le
lan
g
u
ag
e.
Ad
d
itio
n
ally
,
we
aim
to
ad
d
r
ess
th
e
g
ap
s
ca
u
s
ed
b
y
v
ar
y
in
g
d
is
ea
s
e
ch
ar
ac
ter
is
tics
ac
r
o
s
s
d
if
f
er
en
t
c
o
u
n
tr
ies,
en
s
u
r
in
g
th
at
th
e
m
o
d
els
co
u
ld
b
e
ef
f
ec
tiv
ely
ap
p
lied
u
s
in
g
tr
an
s
f
er
lear
n
in
g
.
Ob
jectiv
es o
f
th
e
p
a
p
er
:
1)
Dev
elo
p
m
en
t
o
f
en
s
em
b
le
ap
p
r
o
ac
h
u
s
in
g
SMDBER
T
an
d
z
er
o
s
h
o
t
m
o
d
el
f
o
r
tu
b
er
cu
lo
s
is
class
if
icatio
n
.
2)
Dev
elo
p
m
en
t
o
f
s
im
p
le
y
et
ef
f
ec
tiv
e
n
ar
r
ativ
e
b
ased
in
ter
p
r
etab
ilit
y
m
o
d
e
l
f
o
r
tr
an
s
f
o
r
m
er
b
ased
m
o
d
els.
2.
RE
S
E
ARCH
M
E
T
H
O
D
2
.
1
.
Da
t
a
p
re
-
pro
ce
s
s
ing
T
h
e
d
is
ea
s
es
tar
g
eted
wer
e
Ast
h
m
a,
m
y
o
ca
r
d
ial
in
f
ar
ctio
n
(
MI
)
a
n
d
C
KD
with
I
C
D
-
9
co
d
es
‘
4
9
3
2
0
’
,
‘
5
8
4
9
’
,
‘
4
1
0
0
1
’
,
‘
4
1
0
1
1
’
,
a
n
d
‘
4
1
0
2
1
’
,
r
esp
ec
tiv
e
ly
,
wh
er
e
th
e
co
d
es
'
4
1
0
0
1
'
,
'4
1
0
1
1
'
,
an
d
'
4
1
0
2
1
'
wer
e
g
r
o
u
p
ed
to
g
eth
e
r
u
n
d
er
MI
.
T
h
e
d
ata
u
s
ed
f
o
r
g
en
er
atin
g
th
e
b
ase
m
o
d
el
was
MI
MI
C
[
2
]
.
I
n
itially
,
1
7
0
,
4
4
6
s
am
p
les
wer
e
co
llect
ed
an
d
th
en
r
ef
in
e
d
,
n
a
r
r
o
win
g
th
e
d
ataset
d
o
wn
to
5
,
2
3
4
s
am
p
les
s
p
ec
if
ically
f
o
cu
s
ed
o
n
d
is
ch
ar
g
e
s
u
m
m
ar
ies
f
o
r
an
al
y
s
is
.
SMDBE
R
T
m
o
d
el
was
f
in
e
-
tu
n
ed
u
s
in
g
th
ese
clin
ical
n
o
tes,
ex
tr
ac
ted
f
r
o
m
th
e
NOT
E
VE
NT
S
tab
le
o
f
th
e
MI
MI
C
d
ataset,
alo
n
g
with
o
th
er
s
tr
u
ctu
r
ed
d
ata.
B
asic
p
r
ep
r
o
ce
s
s
in
g
was
ap
p
lied
b
e
f
o
r
e
m
o
d
el
tr
ain
in
g
,
in
cl
u
d
in
g
co
n
v
er
tin
g
te
x
t
to
lo
wer
ca
s
e,
r
em
o
v
i
n
g
s
p
ec
ial
ch
ar
ac
ter
s
,
UR
L
s
,
an
d
n
o
n
-
alp
h
an
u
m
er
ic
elem
en
ts
.
Af
ter
th
e
in
itial
p
h
ase,
th
e
m
o
d
els
wer
e
u
s
ed
o
n
clin
ical
n
o
t
es
f
r
o
m
two
n
u
r
s
in
g
h
o
m
es
o
f
Mu
m
b
ai,
I
n
d
ia.
T
h
e
n
o
tes
wer
e
g
ath
er
e
d
d
u
r
i
n
g
two
p
er
i
o
d
s
:
J
an
u
ar
y
to
Ap
r
il
2
0
2
3
a
n
d
Octo
b
er
t
o
No
v
em
b
er
2
0
2
2
.
A
to
tal
o
f
1
4
5
clin
ical
n
o
tes,
s
p
ec
if
ically
r
elate
d
to
th
e
d
is
ea
s
es
u
s
ed
f
o
r
tr
ain
in
g
an
d
TB
,
wer
e
s
elec
ted
f
o
r
an
aly
s
is
.
Fig
u
r
es 1
an
d
2
s
h
o
w
s
am
p
les o
f
clin
ical
n
o
tes
b
ef
o
r
e
an
d
af
ter
b
asic p
r
e
-
pr
o
ce
s
s
in
g
.
An
ex
am
p
le
o
f
an
an
o
n
y
m
o
u
s
I
n
d
ian
clin
ical
n
o
te
is
s
h
o
wn
in
F
ig
u
r
e
3
.
2
.
2
.
Arc
hite
ct
ure
As
th
e
SMD
B
E
R
T
m
o
d
el
g
av
e
b
etter
p
er
f
o
r
m
an
ce
t
h
an
Dis
tilB
E
R
T
,
it
wa
s
ch
o
s
en
as
a
b
ase
m
o
d
el.
Af
ter
tr
ain
in
g
th
e
m
o
d
el,
th
e
m
o
d
el
was
th
en
ap
p
lied
o
n
r
e
al
tim
e
I
n
d
ian
d
ata
co
llected
f
r
o
m
2
h
o
s
p
itals
o
f
Mu
m
b
ai,
I
n
d
ia
u
s
in
g
tr
a
n
s
f
er
lear
n
in
g
.
Ov
er
all,
m
o
r
e
th
an
1
,
5
0
0
clin
ical
n
o
tes,
esp
ec
ially
d
is
ch
ar
g
e
s
u
m
m
ar
ies
wer
e
co
llected
,
o
u
t
o
f
w
h
ich
1
4
5
n
o
tes
wer
e
u
s
ed
.
T
h
e
d
is
ea
s
es
tar
g
eted
wer
e
th
e
s
am
e
as
th
o
s
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
4
,
No
.
1
,
A
p
r
il
20
2
5
:
91
-
10
1
94
u
s
ed
f
o
r
t
r
ain
in
g
,
with
th
e
ad
d
itio
n
o
f
TB
,
wh
ich
is
p
r
ev
al
en
t
in
I
n
d
ia
b
u
t
n
o
tab
ly
ab
s
en
t
f
r
o
m
th
e
p
u
b
licly
av
ailab
le
d
ataset,
wh
ich
p
r
ed
o
m
in
an
tly
r
ep
r
esen
ts
th
e
Un
ited
States
.
T
h
is
h
ig
h
lig
h
ts
th
e
r
eg
io
n
al
v
ar
iatio
n
in
d
is
ea
s
e
p
r
ev
alen
ce
ac
r
o
s
s
d
if
f
er
en
t
co
u
n
tr
ies.
Fig
u
r
e
4
d
ep
i
cts
th
e
ar
ch
itectu
r
e
o
f
ap
p
lica
tio
n
o
f
SMDBER
T
m
o
d
el
o
n
th
e
I
n
d
ian
d
ataset.
Fig
u
r
e
1.
C
lin
ical
n
o
te
f
r
o
m
MI
MI
C
2
.
3
.
M
et
ho
d
T
h
e
SMDBER
T
m
o
d
el
was
s
elec
ted
f
o
r
its
s
u
p
er
io
r
p
e
r
f
o
r
m
an
ce
co
m
p
ar
ed
t
o
o
th
e
r
tr
a
n
s
f
o
r
m
er
-
b
ased
m
o
d
els.
SMDBER
T
in
co
r
p
o
r
ates
ad
d
itio
n
al
k
n
o
wled
g
e
b
y
in
teg
r
atin
g
s
y
m
p
to
m
an
d
d
is
ea
s
e
in
f
o
r
m
atio
n
.
T
h
e
m
o
d
el
was
f
in
e
-
tu
n
e
d
o
n
d
ata
f
r
o
m
th
e
MI
MI
C
d
atab
ase
an
d
th
en
a
p
p
lied
to
r
ea
l
-
tim
e
I
n
d
ian
d
ata
th
r
o
u
g
h
tr
an
s
f
er
le
ar
n
in
g
.
Sin
ce
TB
was
n
o
t
in
clu
d
ed
in
th
e
tr
ain
i
n
g
s
et
d
u
e
to
a
lack
o
f
av
ailab
le
d
ata,
a
ze
r
o
-
s
h
o
t
m
o
d
el
wa
s
em
p
lo
y
ed
f
o
r
its
class
if
ic
atio
n
.
T
h
e
ar
ch
itectu
r
e
o
f
e
n
s
em
b
le
m
o
d
el
o
f
SMDBE
R
T
an
d
cu
s
to
m
is
ed
Z
er
o
s
h
o
t m
o
d
el
was c
o
n
s
tr
u
cte
d
as g
iv
en
in
F
ig
u
r
e
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
z
ero
-
s
h
o
t le
a
r
n
in
g
a
n
d
tr
a
n
s
fer lea
r
n
in
g
fo
r
r
ea
l time
…
(
S
w
a
ti S
a
ig
a
o
n
ka
r
)
95
Fig
u
r
e
2.
Pre
-
p
r
o
cc
ess
ed
clin
i
ca
l n
o
te
f
r
o
m
MI
MI
C
Fig
u
r
e
3.
Sy
m
p
to
m
e
x
tr
ac
ted
clin
ical
n
o
te
o
f
I
n
d
ian
d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
4
,
No
.
1
,
A
p
r
il
20
2
5
:
91
-
10
1
96
Fig
u
r
e
4
.
Ar
c
h
itectu
r
e
Fig
u
r
e
5
.
E
n
s
em
b
le
a
r
ch
itectu
r
e
T
h
e
m
o
d
el
ca
n
b
e
e
x
p
r
ess
ed
m
ath
em
atica
lly
as
:
L
et
S
:
s
et
o
f
in
p
u
t sen
ten
ce
s
(
c
lin
ical
n
o
tes).
D
:
s
et
o
f
d
is
ea
s
e
lab
els,
wh
er
e
D
=
{0
,
1
,
2
,
3
}
SMD
pred
(
s
)
:
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
o
v
er
th
e
d
is
ea
s
e
lab
els
p
r
ed
icted
b
y
th
e
SM
-
DB
E
R
T
m
o
d
el
f
o
r
in
p
u
t
s
en
ten
ce
s
.
E
H
R
D
a
t
a
s
e
t
(
M
I
M
I
C
)
C
oh
or
t
S
e
l
e
c
t
i
o
n
D
a
t
a
P
r
e
p
r
oc
e
s
s
i
n
g
S
M
D
B
E
R
T
m
od
e
l
Cli
n
ical
No
t
e
+
S
tructur
ed
C
li
n
ical
D
ata
CL
S
D
ata
M
ask
D
ata
SEP
D
IS
TIL
BE
RT
D
ense,
Fu
ll
y
c
o
n
n
ec
te
d
L
ay
er
D
ro
p
o
u
t
L
a
y
er
Outp
u
t
L
a
y
er
f
o
r
Classi
ficat
io
n
In
p
u
t
Id
s +
Att
entio
n
M
as
k
+
Ext
r
a E
m
b
edd
in
g
Ext
erna
l
Kno
wledg
e
D
i
s
e
a
s
e
S
pe
c
i
f
i
c
d
o
m
a
i
n
k
no
w
l
e
d
g
e
F
i
ne
t
un
i
ng
o
f
S
M
D
B
E
R
T
f
o
r
s
pe
c
i
f
i
c
d
i
s
e
a
s
e
s
I
n
d
i
a
n
E
H
R
D
a
t
a
A
s
t
h
m
a
MI
C
K
D
T
ub
e
r
c
ul
os
i
s
F
i
ne
-
t
u
ne
d
m
od
e
l
P
r
e
d
i
c
t
i
on
E
H
R
D
a
t
a
s
e
t
(
I
nd
i
a
n
R
e
a
l
t
i
m
e
D
a
t
a
)
S
M
D
B
E
R
T
m
od
e
l
Cli
n
ical
No
t
e
+
S
tructur
ed
C
li
n
ical
D
ata
CL
S
D
ata
M
ask
D
ata
SEP
D
IS
TIL
BE
RT
D
ense,
Fu
ll
y
c
o
n
n
ec
te
d
L
ay
er
D
ro
p
o
u
t
L
a
y
er
Outp
u
t
L
a
y
er
f
o
r
Classi
ficat
io
n
In
p
u
t
Id
s +
Att
entio
n
M
as
k
+
Ext
r
a E
m
b
edd
in
g
Ext
erna
l
Kno
wledg
e
T
ub
e
r
c
ul
os
i
s
C
l
i
ni
c
a
l
N
ot
e
Z
e
r
o
S
h
ot
C
l
a
s
s
i
f
i
e
r
C
on
d
i
t
i
on
D
e
s
c
r
i
p
t
i
on
S
M
D
B
E
R
T
D
i
s
e
a
s
e
Sc
ore
s
Z
e
r
o
S
h
ot
D
i
s
e
a
s
e
Sc
ore
s
W
e
i
g
ht
e
d
Av
e
r
a
g
e
D
i
s
e
a
s
e
Sc
ore
s
D
i
s
e
a
s
e
P
r
e
d
i
c
t
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
z
ero
-
s
h
o
t le
a
r
n
in
g
a
n
d
tr
a
n
s
fer lea
r
n
in
g
fo
r
r
ea
l time
…
(
S
w
a
ti S
a
ig
a
o
n
ka
r
)
97
ZS
p
r
ed
(
s
)
:
p
r
o
b
a
b
i
li
t
y
d
is
t
r
i
b
u
t
io
n
o
v
e
r
t
h
e
d
i
s
e
a
s
e
l
a
b
e
ls
p
r
e
d
ic
t
e
d
b
y
t
h
e
z
e
r
o
-
s
h
o
t
m
o
d
e
l
f
o
r
i
n
p
u
t
s
e
n
t
e
n
c
e
s
.
α,
β:
w
eig
h
ts
ass
ig
n
ed
to
SM
-
DB
E
R
T
an
d
ze
r
o
-
s
h
o
t p
r
e
d
ictio
n
s
,
r
esp
ec
tiv
ely
,
wh
e
r
e
α
+β=
1
.
C
o
m
b
in
ed
pred
(
s
)
:
f
in
al
co
m
b
in
ed
p
r
ed
ictio
n
s
co
r
e
f
o
r
ea
ch
d
i
s
ea
s
e
lab
el
d
∈
D
f
o
r
s
en
ten
ce
s
Fin
alPr
ed
(
s
)
:
t
h
e
f
in
al
p
r
e
d
icted
lab
el
f
o
r
s
en
ten
ce
s
.
Step
s
:
SMDBE
R
T
p
r
ed
ictio
n:
(
)
=
{
(
)
∣
∣
∈
}
w
h
er
e
(
)
is
th
e
p
r
o
b
ab
ilit
y
ass
ig
n
ed
b
y
th
e
SMDBER
T
m
o
d
el
to
d
is
ea
s
e
f
o
r
s
en
ten
ce
s
.
Z
er
o
-
s
h
o
t p
r
ed
ictio
n
:
(
)
=
{
(
)
∣
∣
∈
}
w
h
er
e
(
)
is
th
e
p
r
o
b
ab
ilit
y
ass
ig
n
ed
b
y
th
e
ze
r
o
s
h
o
t
m
o
d
el
to
d
is
ea
s
e
f
o
r
s
en
ten
ce
s
.
C
o
m
b
in
ed
p
r
ed
ictio
n
:
(
)
=
{
⋅
(
)
+
⋅
(
)
∣
∣
∈
}
Her
e,
α
an
d
β
r
ep
r
esen
t
th
e
r
elativ
e
im
p
o
r
tan
ce
o
r
weig
h
t
o
f
th
e
SMDBE
R
T
an
d
ze
r
o
-
s
h
o
t
p
r
ed
ictio
n
s
,
r
esp
ec
tiv
ely
.
Fin
al
p
r
ed
ictio
n
:
(
)
=
a
r
g
(
)
T
h
e
f
in
al
p
r
ed
icted
lab
el
is
th
e
o
n
e
with
th
e
h
ig
h
est co
m
b
in
e
d
p
r
ed
ictio
n
s
co
r
e.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
ab
le
1
s
h
o
ws
th
e
p
er
f
o
r
m
a
n
ce
o
f
d
if
f
er
e
n
t
m
o
d
els
o
n
th
e
I
n
d
ian
d
ata.
Mo
d
els,
n
am
ely
B
E
R
T
,
Dis
til
B
E
R
T
an
d
SMDBE
R
T
wer
e
in
itially
f
in
e
tu
n
ed
o
n
MI
MI
C
n
o
tes.
Fo
llo
win
g
th
e
tr
ain
in
g
p
h
ase,
th
e
m
o
d
els
u
n
d
er
wen
t
p
r
ac
tical
ap
p
licatio
n
u
s
in
g
I
n
d
ian
clin
ical
r
ec
o
r
d
s
o
b
tain
ed
f
r
o
m
h
o
s
p
itals
s
i
tu
ated
in
Mu
m
b
ai,
I
n
d
ia.
A
to
tal
o
f
1
45
clin
ical
n
o
tes
wer
e
m
eticu
l
o
u
s
ly
s
elec
ted
,
co
n
ce
n
tr
atin
g
ex
clu
s
iv
ely
o
n
th
e
tar
g
eted
d
is
ea
s
es
.
T
h
e
tr
an
s
cr
i
p
tio
n
o
f
clin
ical
n
o
tes
was
c
o
n
d
u
cted
m
an
u
ally
d
u
e
t
o
th
e
p
r
o
v
is
io
n
o
f
s
ca
n
n
ed
co
p
ies
b
y
th
e
h
o
s
p
itals
.
I
t
is
wo
r
th
n
o
tin
g
th
at
tr
ea
tm
en
t
s
p
ec
if
ics
wer
e
in
ten
tio
n
ally
ex
clu
d
ed
f
r
o
m
t
h
e
d
ataset
to
alig
n
with
th
e
r
esea
r
ch
'
s
f
o
cu
s
o
n
ca
p
tu
r
in
g
s
y
m
p
to
m
s
.
Sp
ec
ialized
m
o
d
els
wer
e
also
test
ed
o
n
th
e
I
n
d
ian
d
ata
co
llected
.
SMDBER
T
g
av
e
b
etter
r
esu
lts
as
co
m
p
ar
ed
to
o
th
e
r
m
o
d
els
as
it
tak
es,
s
y
m
p
to
m
d
is
ea
s
e
in
f
o
r
m
atio
n
as
an
a
d
d
itio
n
al
em
b
e
d
d
in
g
.
As
it
ca
n
b
e
s
ee
n
o
u
r
m
eth
o
d
wi
th
an
en
s
em
b
le
o
f
SMDBE
R
T
an
d
ze
r
o
s
h
o
t le
ar
n
in
g
g
a
v
e
b
etter
r
esu
lts
with
an
ac
cu
r
ac
y
o
f
0
.
9
2
4
.
T
ab
le
1
.
Per
f
o
r
m
an
ce
m
etr
ics o
f
f
in
e
-
t
u
n
ed
m
o
d
els o
n
I
n
d
ian
d
ata
T
y
p
e
o
f
M
o
d
e
l
M
o
d
e
l
s
A
c
c
u
r
a
c
y
P
r
e
c
i
si
o
n
R
e
c
a
l
l
F1
-
s
c
o
r
e
F
i
n
e
-
t
u
n
e
d
B
ER
T
0
.
5
5
0
.
5
0
.
5
5
0
.
5
2
D
i
st
i
l
B
ER
T
0
.
6
4
0
.
6
5
0
.
6
4
0
.
6
2
S
M
D
B
E
R
T
0
.
7
5
0
.
6
9
0
.
7
5
0
.
7
2
S
p
e
c
i
a
l
i
s
e
d
S
C
I
B
ER
T
0
.
7
6
0
.
8
3
0
.
7
6
0
.
7
2
B
I
O
B
ER
T
0
.
6
2
0
.
5
2
0
.
6
2
0
.
5
4
C
l
i
n
i
c
a
l
B
i
o
B
ER
T
0
.
6
1
0
.
5
1
0
.
6
1
0
.
5
3
P
r
o
p
o
se
d
E
n
s
e
mb
l
e
a
p
p
r
o
a
c
h
0
.
9
2
4
0
.
9
2
7
0
.
9
2
4
0
.
9
2
5
In
(
1
)
,
(
2
)
,
a
n
d
(
3
)
d
escr
ib
e
s
th
e
co
m
p
u
tatio
n
al
f
o
r
m
u
latio
n
s
f
o
r
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
th
e
F1
-
s
co
r
e,
r
esp
ec
tiv
ely
.
=
+
(
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
4
,
No
.
1
,
A
p
r
il
20
2
5
:
91
-
10
1
98
=
+
(
2
)
1
−
=
2
∗
∗
+
(
3
)
Fig
u
r
e
6
d
is
p
lay
s
th
e
m
o
d
els'
p
er
f
o
r
m
an
ce
g
r
ap
h
ically
with
F
ig
u
r
e
6
(
a)
s
h
o
win
g
m
o
d
el
p
e
r
f
o
r
m
a
n
ce
with
r
esp
ec
t
to
th
e
s
p
ec
ialized
m
o
d
els
an
d
Fi
g
u
r
e
6
(
b
)
s
h
o
wi
n
g
th
e
p
r
o
p
o
s
ed
m
o
d
el
p
er
f
o
r
m
an
ce
with
r
esp
ec
t
to
f
in
e
-
tu
n
e
d
m
o
d
els
.
As
s
h
o
wn
in
th
e
d
iag
r
am
s
th
e
e
n
s
em
b
le
ap
p
r
o
ac
h
o
f
SMDBER
T
an
d
z
er
o
s
h
o
t
h
as
g
iv
en
b
etter
r
esu
lts
th
an
o
th
er
f
in
e
-
tu
n
ed
m
o
d
els
o
r
s
p
ec
ialized
m
o
d
els.
T
h
e
s
p
ec
ialized
m
o
d
els
co
n
s
id
er
ed
ar
e
s
cien
tific
B
E
R
T
(
S
C
I
B
E
R
T
)
,
b
io
m
ed
ical
B
E
R
T
(
B
I
O
B
E
R
T
)
an
d
B
io
_
C
lin
ical
B
E
R
T
,
wh
ile
f
in
e
-
tu
n
e
d
m
o
d
els ar
e
B
E
R
T
an
d
Dis
tilB
E
R
T
.
T
h
e
co
n
f
u
s
io
n
m
atr
i
x
d
em
o
n
s
tr
ate
d
in
Fig
u
r
e
7
,
s
h
o
ws
th
at
th
e
m
o
d
el
ef
f
ec
tiv
ely
class
if
ies
a
m
ajo
r
ity
o
f
in
s
tan
ce
s
co
r
r
ec
tly
,
in
d
icatin
g
s
tr
o
n
g
p
er
f
o
r
m
an
ce
an
d
r
eliab
le
p
r
ed
icti
v
e
ac
cu
r
ac
y
ac
r
o
s
s
m
u
ltip
le
class
es
.
Fig
u
r
e
8
s
h
o
w
s
th
e
o
u
tp
u
t
o
f
L
I
ME
,
w
h
ich
ex
p
lain
s
p
r
e
d
ictio
n
s
m
ad
e
b
y
th
e
en
s
em
b
le
ap
p
r
o
ac
h
b
y
h
ig
h
lig
h
tin
g
th
e
co
n
tr
ib
u
tio
n
s
o
f
d
if
f
er
e
n
t
in
p
u
t
f
ea
tu
r
es
.
Pre
d
ictio
n
p
r
o
b
ab
il
ities
s
ec
tio
n
s
h
o
ws
th
e
p
r
ed
icted
p
r
o
b
ab
ilit
ies
f
o
r
ea
ch
class
.
T
h
e
m
o
d
el
p
r
e
d
ict
s
th
is
class
with
a
p
r
o
b
ab
ilit
y
o
f
0
.
8
0
.
T
h
e
m
i
d
d
le
s
ec
tio
n
s
h
o
ws
th
e
f
ea
tu
r
es
th
at
co
n
tr
ib
u
te
to
th
e
p
r
e
d
ictio
n
o
f
th
e
class
.
T
h
e
r
ig
h
t
s
id
e
o
f
th
e
d
iag
r
am
s
h
o
w
th
e
s
p
ec
if
ic
wo
r
d
s
o
r
p
h
r
ases
,
wh
ich
a
r
e
h
ig
h
li
g
h
ted
,
r
e
f
le
ctin
g
th
eir
im
p
o
r
ta
n
ce
in
th
e
m
o
d
el'
s
d
ec
is
io
n
-
m
ak
in
g
p
r
o
ce
s
s
.
Fig
u
r
e
9
s
h
o
ws
th
e
clin
ical
n
o
te
g
iv
en
as
in
p
u
t
,
it
in
clu
d
es
v
ar
io
u
s
attr
ib
u
tes
lik
e
ag
e,
g
en
d
er
,
f
i
n
al
d
iag
n
o
s
is
,
co
m
p
lain
ts
o
n
ad
m
is
s
io
n
p
ast
m
ed
ical
h
i
s
to
r
y
,
v
ital
s
ig
n
s
,
an
d
o
th
er
r
elev
an
t
clin
ical
in
f
o
r
m
atio
n
.
T
h
e
h
o
r
izo
n
tal
b
ar
at
th
e
to
p
r
ep
r
esen
t
s
th
e
SHAP
v
alu
e
f
o
r
th
e
h
ig
h
lig
h
ted
o
u
tp
u
t
wh
ic
h
is
Ou
tp
u
t
0
.
T
h
e
v
alu
es
o
n
th
e
b
ar
r
an
g
e
f
r
o
m
0
.
0
to
1
.
0
,
in
d
icatin
g
th
e
p
r
o
b
ab
ilit
y
o
r
co
n
f
id
e
n
ce
o
f
th
e
m
o
d
el'
s
p
r
ed
ictio
n
f
o
r
ea
ch
o
u
tp
u
t.
(
a)
(
b
)
Fig
u
r
e
6
.
Gr
a
p
h
d
is
p
lay
i
n
g
p
e
r
f
o
r
m
a
n
ce
co
m
p
ar
is
o
n
o
f
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
(
a
)
with
s
p
ec
ialiaze
d
m
o
d
els
(
b
)
with
f
in
e
-
tu
n
e
d
m
o
d
els
Fig
u
r
e
7
.
C
o
n
f
u
s
io
n
m
atr
i
x
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
z
ero
-
s
h
o
t le
a
r
n
in
g
a
n
d
tr
a
n
s
fer lea
r
n
in
g
fo
r
r
ea
l time
…
(
S
w
a
ti S
a
ig
a
o
n
ka
r
)
99
Fig
u
r
e
8
.
L
I
ME
e
x
p
la
n
atio
n
Fig
u
r
e
1
0
s
h
o
ws
th
e
co
m
b
i
n
ed
in
ter
p
r
eta
b
ilit
y
tech
n
iq
u
e
wh
er
ein
Fig
u
r
e
1
0
(
a
)
p
r
esen
t
s
n
ar
r
ativ
e
ex
p
lan
atio
n
s
,
w
h
ile
F
ig
u
r
e
1
0
(
b
)
p
r
o
v
i
d
es
co
r
r
esp
o
n
d
i
n
g
L
I
ME
ex
p
lan
atio
n
s
.
T
h
ese
L
I
ME
ju
s
tific
atio
n
s
ar
e
p
r
o
d
u
ce
d
alo
n
g
s
id
e
th
e
n
ar
r
ativ
es
to
f
ac
ilit
ate
a
clea
r
er
u
n
d
e
r
s
tan
d
in
g
o
f
th
e
m
o
d
el
’
s
p
r
ed
ictio
n
s
.
T
h
e
n
ar
r
ativ
es
o
f
f
er
a
wr
itten
s
u
m
m
ar
y
,
d
etailin
g
h
o
w
s
p
ec
if
ic
wo
r
d
s
o
r
f
ea
tu
r
es
h
av
e
a
m
ajo
r
im
p
a
ct
o
n
th
e
m
o
d
el
’s
d
ec
is
io
n
s
.
B
y
p
r
esen
tin
g
b
o
th
n
ar
r
ativ
es
an
d
v
is
u
al
ex
p
lan
atio
n
s
to
g
eth
e
r
,
u
s
er
s
ca
n
m
o
r
e
ea
s
ily
in
ter
p
r
et
th
e
m
ai
n
elem
en
ts
th
at
th
e
m
o
d
el
c
o
n
s
id
er
s
wh
en
m
ak
in
g
its
p
r
e
d
ictio
n
s
.
Fig
u
r
e
9
.
SHAP
e
x
p
lan
atio
n
(
a)
(
b
)
Fig
u
r
e
1
0
.
C
o
m
b
in
ed
i
n
ter
p
r
et
ab
ilit
y
tech
n
iq
u
e
(
a)
n
a
r
r
ativ
es
f
o
r
im
p
o
r
tan
t w
o
r
d
s
(
b
)
L
I
M
E
ex
p
lan
atio
n
g
en
er
ated
alo
n
g
with
n
ar
r
ativ
e
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
4
,
No
.
1
,
A
p
r
il
20
2
5
:
91
-
10
1
100
4.
CO
NCLU
SI
O
N
I
n
th
is
s
tu
d
y
,
we
ad
d
r
ess
ed
th
e
c
h
allen
g
es
o
f
a
p
p
ly
in
g
ML
m
o
d
els
to
r
ea
l
-
tim
e
clin
ical
d
ata,
p
ar
ticu
lar
ly
in
th
e
I
n
d
ia
n
co
n
tex
t.
B
y
u
tili
zin
g
th
e
SMD
B
E
R
T
m
o
d
el
f
in
e
-
tu
n
ed
o
n
p
u
b
licly
a
v
ailab
le
d
atasets
,
MI
MI
C
,
th
e
p
o
ten
tial
o
f
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
els
f
o
r
th
e
class
if
icatio
n
o
f
ch
r
o
n
ic
d
is
ea
s
es
s
u
ch
as
Ast
h
m
a,
MI
,
a
n
d
C
KD
was
d
em
o
n
s
tr
ated
.
H
o
wev
er
,
th
e
lim
itatio
n
s
o
f
th
ese
m
o
d
els
wer
e
ev
id
en
t
wh
en
th
ey
wer
e
ap
p
lied
to
I
n
d
ia
n
clin
ica
l
d
ata,
wh
er
e
d
is
ea
s
es
lik
e
tu
b
er
cu
lo
s
is
ar
e
m
o
r
e
p
r
ev
alen
t
b
u
t
n
o
t
r
ep
r
esen
ted
in
th
e
tr
ain
in
g
d
atasets
.
T
o
o
v
er
co
m
e
th
is
,
an
in
n
o
v
ativ
e
e
n
s
em
b
le
ap
p
r
o
ac
h
th
at
co
m
b
i
n
es
th
e
s
tr
en
g
th
s
o
f
th
e
SMDBE
R
T
m
o
d
el
an
d
a
cu
s
to
m
ized
ze
r
o
-
s
h
o
t
lear
n
in
g
m
eth
o
d
was
d
ev
el
o
p
ed
,
en
ab
lin
g
ef
f
ec
ti
v
e
class
if
icatio
n
o
f
tu
b
e
r
cu
lo
s
is
c
ases
d
esp
ite
th
e
lack
o
f
co
r
r
esp
o
n
d
in
g
ex
a
m
p
les in
th
e
tr
ain
i
n
g
d
ata.
Ad
d
itio
n
ally
,
th
e
is
s
u
e
o
f
m
o
d
el
in
ter
p
r
etab
ilit
y
was
tack
led
,
wh
ich
is
cr
itical
f
o
r
clin
ica
l
ad
o
p
tio
n
.
T
h
e
r
esear
ch
in
t
r
o
d
u
ce
d
n
a
r
r
ativ
e
-
b
ased
in
ter
p
r
etab
ilit
y
t
ec
h
n
iq
u
e,
alo
n
g
with
L
I
ME
in
ter
p
r
ea
tb
ilit
y
to
p
r
o
v
id
e
n
atu
r
al
lan
g
u
ag
e
ex
p
lan
atio
n
s
o
f
th
e
m
o
d
el'
s
d
ec
is
io
n
-
m
ak
in
g
p
r
o
ce
s
s
.
T
h
is
ap
p
r
o
ac
h
n
o
t
o
n
ly
en
h
an
ce
s
th
e
tr
an
s
p
ar
e
n
cy
o
f
p
r
ed
ictio
n
s
b
u
t
also
h
elp
s
h
ea
lth
ca
r
e
p
r
o
f
ess
io
n
als
in
u
n
d
er
s
tan
d
i
n
g
th
e
r
atio
n
ale
b
eh
in
d
th
e
m
o
d
el'
s
o
u
tp
u
ts
,
m
ak
in
g
th
ese
to
o
ls
m
o
r
e
r
eliab
le
an
d
tr
u
s
two
r
th
y
i
n
clin
ical
p
r
ac
tice.
Ou
r
r
esear
ch
'
s
f
in
d
in
g
s
h
ig
h
lig
h
t
th
e
s
ig
n
if
ican
ce
o
f
cu
s
to
m
izin
g
m
ac
h
in
e
lear
n
in
g
m
o
d
els
to
th
e
u
n
iq
u
e
r
e
q
u
ir
em
en
ts
o
f
v
a
r
io
u
s
d
em
o
g
r
ap
h
ics
an
d
th
e
n
ee
d
o
f
cr
ea
tin
g
s
t
r
o
n
g
in
ter
p
r
etab
il
ity
f
r
am
ewo
r
k
s
f
o
r
in
tr
icate
m
o
d
els.
B
y
in
te
g
r
ati
n
g
an
en
s
em
b
le
o
f
ze
r
o
-
s
h
o
t
lear
n
in
g
with
SMD
-
B
E
R
T
,
o
u
r
ap
p
r
o
ac
h
ac
h
iev
ed
an
ac
cu
r
ac
y
o
f
0
.
9
2
,
o
u
t
p
er
f
o
r
m
in
g
s
p
ec
ialized
m
o
d
els
lik
e
SC
I
B
E
R
T
,
B
I
O
B
E
R
T
,
an
d
c
lin
ical
B
io
B
E
R
T
.
T
h
is
r
esear
ch
co
n
tr
i
b
u
tes
s
ig
n
if
ican
tly
to
th
e
f
ield
b
y
d
em
o
n
s
tr
atin
g
h
o
w
a
d
v
an
ce
d
NL
P
tech
n
iq
u
es
ca
n
b
e
ad
ap
ted
f
o
r
d
iv
er
s
e
h
ea
lth
ca
r
e
en
v
ir
o
n
m
e
n
ts
an
d
b
y
p
r
o
p
o
s
i
n
g
a
n
ew
d
ir
ec
tio
n
f
o
r
th
e
in
ter
p
r
etab
ilit
y
o
f
AI
m
o
d
els
in
clin
ical
s
u
r
r
o
u
n
d
in
g
s
.
Mo
v
in
g
f
o
r
war
d
,
ex
p
a
n
d
in
g
t
h
e
s
co
p
e
o
f
r
ea
l
-
tim
e
d
atasets
an
d
f
u
r
th
er
r
ef
in
in
g
i
n
ter
p
r
etab
ilit
y
m
et
h
o
d
s
will b
e
cr
u
cial
f
o
r
ad
v
a
n
cin
g
AI
'
s
r
o
le
in
h
ea
lth
ca
r
e
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
au
th
o
r
s
ar
e
th
an
k
f
u
l
t
o
th
e
Dep
ar
tm
en
t
o
f
C
o
m
p
u
ter
E
n
g
in
ee
r
in
g
,
R
am
r
ao
Ad
ik
I
n
s
t
itu
te,
D
Y
Patil De
em
ed
to
b
e
Un
iv
er
s
ity
,
to
p
r
o
v
id
e
f
ac
ilit
ies an
d
s
u
p
p
o
r
t to
ca
r
r
y
o
u
t th
e
r
esear
ch
wo
r
k
.
RE
F
E
R
E
NC
E
S
[
1
]
J.
H
e
n
r
y
,
Y
.
P
y
l
y
p
c
h
u
k
,
T.
S
e
a
r
c
y
,
a
n
d
V
.
P
a
t
e
l
,
“
A
d
o
p
t
i
o
n
o
f
e
l
e
c
t
r
o
n
i
c
h
e
a
l
t
h
r
e
c
o
r
d
s
y
st
e
ms a
m
o
n
g
U
.
S
.
n
o
n
-
f
e
d
e
r
a
l
a
c
u
t
e
c
a
r
e
h
o
s
p
i
t
a
l
s:
2
0
0
8
-
2
0
1
5
,
”
O
N
C
D
a
t
a
Br
i
e
f
,
v
o
l
.
5
,
n
o
.
3
5
,
2
0
1
6
.
[
2
]
A
.
E
.
W
.
J
o
h
n
s
o
n
e
t
a
l
.
,
“
M
I
M
I
C
-
I
I
I
,
a
f
r
e
e
l
y
a
c
c
e
ss
i
b
l
e
c
r
i
t
i
c
a
l
c
a
r
e
d
a
t
a
b
a
s
e
,
”
S
c
i
e
n
t
i
f
i
c
D
a
t
a
,
v
o
l
.
3
,
n
o
.
1
,
p
.
1
6
0
0
3
5
,
M
a
y
2
0
1
6
,
d
o
i
:
1
0
.
1
0
3
8
/
sd
a
t
a
.
2
0
1
6
.
3
5
.
[
3
]
T.
J.
P
o
l
l
a
r
d
,
A
.
E.
W
.
J
o
h
n
s
o
n
,
J.
D
.
R
a
f
f
a
,
L.
A
.
C
e
l
i
,
R
.
G
.
M
a
r
k
,
a
n
d
O
.
B
a
d
a
w
i
,
“
Th
e
e
I
C
U
c
o
l
l
a
b
o
r
a
t
i
v
e
r
e
s
e
a
r
c
h
d
a
t
a
b
a
s
e
,
a
f
r
e
e
l
y
a
v
a
i
l
a
b
l
e
mu
l
t
i
-
c
e
n
t
e
r
d
a
t
a
b
a
s
e
f
o
r
c
r
i
t
i
c
a
l
c
a
r
e
r
e
se
a
r
c
h
,
”
S
c
i
e
n
t
i
f
i
c
D
a
t
a
,
v
o
l
.
5
,
n
o
.
1
,
p
.
1
8
0
1
7
8
,
S
e
p
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
3
8
/
s
d
a
t
a
.
2
0
1
8
.
1
7
8
.
[
4
]
J.
Y
a
n
g
e
t
a
l
.
,
“
B
r
i
e
f
i
n
t
r
o
d
u
c
t
i
o
n
o
f
med
i
c
a
l
d
a
t
a
b
a
se
a
n
d
d
a
t
a
m
i
n
i
n
g
t
e
c
h
n
o
l
o
g
y
i
n
b
i
g
d
a
t
a
e
r
a
,
”
J
o
u
rn
a
l
o
f
Ev
i
d
e
n
c
e
-
Ba
s
e
d
Me
d
i
c
i
n
e
,
v
o
l
.
1
3
,
n
o
.
1
,
p
p
.
5
7
–
6
9
,
F
e
b
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
1
1
/
j
e
b
m.
1
2
3
7
3
.
[
5
]
R
.
A
l
a
n
a
z
i
,
“
I
d
e
n
t
i
f
i
c
a
t
i
o
n
a
n
d
p
r
e
d
i
c
t
i
o
n
o
f
c
h
r
o
n
i
c
d
i
sea
ses
u
si
n
g
mac
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
J
o
u
r
n
a
l
o
f
H
e
a
l
t
h
c
a
re
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
0
2
2
,
p
p
.
1
–
9
,
F
e
b
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
2
/
2
8
2
6
1
2
7
.
[
6
]
S
.
A
h
me
d
e
t
a
l
.
,
“
P
r
e
d
i
c
t
i
o
n
o
f
c
a
r
d
i
o
v
a
sc
u
l
a
r
d
i
sea
se
o
n
se
l
f
-
a
u
g
m
e
n
t
e
d
d
a
t
a
se
t
s
o
f
h
e
a
r
t
p
a
t
i
e
n
t
s
u
si
n
g
mu
l
t
i
p
l
e
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
m
o
d
e
l
s
,
”
J
o
u
r
n
a
l
o
f
S
e
n
so
rs
,
v
o
l
.
2
0
2
2
,
p
p
.
1
–
2
1
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
2
/
3
7
3
0
3
0
3
.
[
7
]
L.
J.
S
p
e
n
c
e
r
,
A
.
D
e
d
u
,
H
.
A
.
K
a
l
k
i
d
a
n
,
M
.
A
.
S
o
l
o
m
o
n
,
A
.
C
r
i
st
i
a
n
a
,
a
n
d
N
o
o
s
h
i
n
A
b
b
a
s
i
,
“
G
l
o
b
a
l
,
r
e
g
i
o
n
a
l
,
a
n
d
n
a
t
i
o
n
a
l
i
n
c
i
d
e
n
c
e
,
p
r
e
v
a
l
e
n
c
e
,
a
n
d
y
e
a
r
s
l
i
v
e
d
w
i
t
h
d
i
s
a
b
i
l
i
t
y
f
o
r
3
5
4
d
i
s
e
a
ses
a
n
d
i
n
j
u
r
i
e
s
f
o
r
1
9
5
c
o
u
n
t
r
i
e
s
a
n
d
t
e
r
r
i
t
o
r
i
e
s,
1
9
9
0
–
2
0
1
7
:
a
sy
st
e
mat
i
c
a
n
a
l
y
s
i
s f
o
r
t
h
e
G
l
o
b
a
l
B
u
r
d
e
n
o
f
D
i
se
a
se
S
t
u
d
y
2
0
1
7
.
,
”
T
h
e
L
a
n
c
e
t
,
v
o
l
.
3
9
2
,
p
p
.
1
7
8
9
–
1
8
5
8
,
2
0
1
8
.
[
8
]
C
.
P
.
K
o
v
e
s
d
y
,
“
Ep
i
d
e
m
i
o
l
o
g
y
o
f
c
h
r
o
n
i
c
k
i
d
n
e
y
d
i
s
e
a
s
e
:
a
n
u
p
d
a
t
e
2
0
2
2
,
”
K
i
d
n
e
y
I
n
t
e
rn
a
t
i
o
n
a
l
S
u
p
p
l
e
m
e
n
t
s
,
v
o
l
.
1
2
,
n
o
.
1
,
p
p
.
7
–
1
1
,
A
p
r
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
k
i
su
.
2
0
2
1
.
1
1
.
0
0
3
.
[
9
]
S
.
S
a
i
g
a
o
n
k
a
r
a
n
d
V
.
N
a
r
a
w
a
d
e
,
“
D
o
mai
n
a
d
a
p
t
a
t
i
o
n
o
f
t
r
a
n
sf
o
r
mer
-
b
a
se
d
n
e
u
r
a
l
n
e
t
w
o
r
k
mo
d
e
l
f
o
r
c
l
i
n
i
c
a
l
n
o
t
e
c
l
a
s
si
f
i
c
a
t
i
o
n
i
n
I
n
d
i
a
n
h
e
a
l
t
h
c
a
r
e
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
(
S
i
n
g
a
p
o
r
e
)
,
A
u
g
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
1
8
7
0
-
0
2
4
-
0
2
0
5
3
-
z.
[
1
0
]
B
.
S
h
i
c
k
e
l
,
P
.
J.
T
i
g
h
e
,
A
.
B
i
h
o
r
a
c
,
a
n
d
P
.
R
a
s
h
i
d
i
,
“
D
e
e
p
EH
R
:
a
s
u
r
v
e
y
o
f
r
e
c
e
n
t
a
d
v
a
n
c
e
s
i
n
d
e
e
p
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
f
o
r
e
l
e
c
t
r
o
n
i
c
h
e
a
l
t
h
r
e
c
o
r
d
(
EH
R
)
a
n
a
l
y
si
s,”
I
EEE
J
o
u
rn
a
l
o
f
Bi
o
m
e
d
i
c
a
l
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
2
,
n
o
.
5
,
p
p
.
1
5
8
9
–
1
6
0
4
,
2
0
1
8
,
d
o
i
:
1
0
.
1
1
0
9
/
JB
H
I
.
2
0
1
7
.
2
7
6
7
0
6
3
.
[
1
1
]
Y
.
M
e
n
g
,
W
.
S
p
e
i
e
r
,
M
.
O
n
g
,
a
n
d
C
.
W
.
A
r
n
o
l
d
,
“
H
C
ET:
h
i
e
r
a
r
c
h
i
c
a
l
c
l
i
n
i
c
a
l
e
m
b
e
d
d
i
n
g
w
i
t
h
t
o
p
i
c
mo
d
e
l
i
n
g
o
n
e
l
e
c
t
r
o
n
i
c
h
e
a
l
t
h
r
e
c
o
r
d
s
f
o
r
p
r
e
d
i
c
t
i
n
g
f
u
t
u
r
e
d
e
p
r
e
ss
i
o
n
,
”
I
EE
E
J
o
u
rn
a
l
o
f
B
i
o
m
e
d
i
c
a
l
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
5
,
n
o
.
4
,
p
p
.
1
2
6
5
–
1
2
7
2
,
A
p
r
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
JB
H
I
.
2
0
2
0
.
3
0
0
4
0
7
2
.
[
1
2
]
J.
Y
e
,
L.
Y
a
o
,
J
.
S
h
e
n
,
R
.
J
a
n
a
r
t
h
a
n
a
m,
a
n
d
Y
.
L
u
o
,
“
P
r
e
d
i
c
t
i
n
g
m
o
r
t
a
l
i
t
y
i
n
c
r
i
t
i
c
a
l
l
y
i
l
l
p
a
t
i
e
n
t
s
w
i
t
h
d
i
a
b
e
t
e
s
u
si
n
g
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
c
l
i
n
i
c
a
l
n
o
t
e
s
,
”
BM
C
Me
d
i
c
a
l
I
n
f
o
rm
a
t
i
c
s
a
n
d
D
e
c
i
s
i
o
n
Ma
k
i
n
g
,
v
o
l
.
2
0
,
n
o
.
S
1
1
,
p
.
2
9
5
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
8
6
/
s
1
2
9
1
1
-
0
2
0
-
0
1
3
1
8
-
4.
[
1
3
]
K
.
H
u
a
n
g
,
J.
A
l
t
o
sa
a
r
,
a
n
d
R
.
R
a
n
g
a
n
a
t
h
,
“
C
l
i
n
i
c
a
l
B
E
R
T:
m
o
d
e
l
i
n
g
c
l
i
n
i
c
a
l
n
o
t
e
s
a
n
d
p
r
e
d
i
c
t
i
n
g
h
o
s
p
i
t
a
l
r
e
a
d
mi
ss
i
o
n
,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
1
9
0
4
.
0
5
3
4
2
,
2
0
1
9
,
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
:
/
/
a
r
x
i
v
.
o
r
g
/
a
b
s/
1
9
0
4
.
0
5
3
4
2
.
[
1
4
]
J.
W
u
,
Y
.
Li
n
,
P
.
Li
,
Y
.
H
u
,
L.
Z
h
a
n
g
,
a
n
d
G
.
K
o
n
g
,
“
P
r
e
d
i
c
t
i
n
g
p
r
o
l
o
n
g
e
d
l
e
n
g
t
h
o
f
I
C
U
st
a
y
t
h
r
o
u
g
h
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
”
D
i
a
g
n
o
s
t
i
c
s
,
v
o
l
.
1
1
,
n
o
.
1
2
,
p
.
2
2
4
2
,
N
o
v
.
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
s
t
i
c
s1
1
1
2
2
2
4
2
.
[
1
5
]
S
.
A
.
M
o
q
u
r
r
a
b
,
U
.
A
y
u
b
,
A
.
A
n
j
u
m,
S
.
A
sg
h
a
r
,
a
n
d
G
.
S
r
i
v
a
s
t
a
v
a
,
“
A
n
a
c
c
u
r
a
t
e
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
f
o
r
c
l
i
n
i
c
a
l
e
n
t
i
t
y
r
e
c
o
g
n
i
t
i
o
n
f
r
o
m
c
l
i
n
i
c
a
l
n
o
t
e
s,”
I
EE
E
J
o
u
rn
a
l
o
f
B
i
o
m
e
d
i
c
a
l
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
5
,
n
o
.
1
0
,
p
p
.
3
8
0
4
–
3
8
1
1
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
JB
H
I
.
2
0
2
1
.
3
0
9
9
7
5
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.