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.
15
,
No
.
4
,
A
u
g
u
s
t
20
26
,
p
p
.
3
6
2
5
~
3
6
3
6
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
6
2
5
-
3
6
3
6
3625
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Sy
mm
etry inde
x
-
ba
sed g
a
it
impro
v
ement
predic
tion
using
a
CNN
-
LST
M
-
a
tt
e
ntion fram
e
wo
rk
P
us
hp
a
la
t
ha
O
ba
nn
a
1,
2
,
P
re
m
k
um
a
r
Ra
m
esh
1
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
JA
I
N
(
D
e
e
m
e
d
-
to
-
be
-
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
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
J
a
i
n
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
D
a
v
a
n
a
g
e
r
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
12
,
2
0
2
5
R
ev
is
ed
May
16
,
2
0
2
6
Acc
ep
ted
May
24
,
2
0
2
6
Th
e
sh
o
rtco
m
i
n
g
s
o
f
e
x
isti
n
g
c
li
n
ica
l
a
ss
e
ss
m
e
n
ts
a
re
a
d
d
re
ss
e
d
b
y
u
si
n
g
a
d
e
e
p
lea
rn
in
g
(DL)
fra
m
e
wo
rk
.
T
h
is
stu
d
y
will
a
ss
e
ss
th
e
g
a
it
imp
ro
v
e
m
e
n
t
o
f
lo
we
r
li
m
b
fra
c
tu
re
d
p
a
ti
e
n
ts u
sin
g
t
h
e
p
u
b
l
icly
a
v
a
il
a
b
le
G
a
it
Re
c
d
a
tas
e
t.
Th
e
d
a
tas
e
t
c
o
n
tai
n
s
th
e
v
e
rti
c
a
l
g
r
o
u
n
d
re
a
c
ti
o
n
f
o
rc
e
(G
RF
)
d
a
ta
u
se
d
to
c
o
m
p
u
te
b
io
m
e
c
h
a
n
ica
l
k
i
n
e
ti
c
fe
a
tu
re
s.
Th
is
fra
m
e
wo
rk
c
o
n
sid
e
rs
h
ip
,
k
n
e
e
,
a
n
k
le
,
a
n
d
c
a
lca
n
e
u
s
fra
c
tu
re
s
a
s
l
o
we
r
li
m
b
fra
c
tu
re
c
las
s
,
a
n
d
t
h
e
s
y
m
m
e
try
in
d
ice
s
c
o
m
p
u
tati
o
n
is
d
o
n
e
b
e
twe
e
n
a
ffe
c
ted
a
n
d
u
n
a
ffe
c
ted
li
m
b
s
,
a
n
d
t
h
e
ir
tem
p
o
ra
l
c
h
a
n
g
e
s
we
re
e
x
a
m
in
e
d
t
o
a
n
a
ly
z
e
a
n
d
trac
k
t
h
e
re
h
a
b
il
it
a
ti
o
n
p
ro
g
re
ss
o
f
in
d
iv
i
d
u
a
l
su
b
jec
t.
Th
e
fra
m
e
wo
rk
su
c
c
e
ss
fu
ll
y
id
e
n
ti
fies
t
h
e
re
d
u
c
ti
o
n
i
n
th
e
a
sy
m
m
e
try
b
e
twe
e
n
left
a
n
d
rig
h
t
le
g
fe
a
tu
re
s.
Th
e
o
v
e
ra
ll
g
a
it
im
p
ro
v
e
m
e
n
t
i
n
p
a
ti
e
n
ts
is
c
o
m
p
u
ted
u
sin
g
in
i
ti
a
l
a
n
d
fin
a
l
se
ss
io
n
c
o
m
p
o
site
a
sy
m
m
e
try
i
n
d
e
x
(
ASI
)
v
a
l
u
e
s
a
n
d
if
it
is
a
t
lea
st
3
0
%
th
e
n
o
v
e
ra
ll
g
a
it
is
im
p
ro
v
e
d
,
d
e
m
o
n
stra
ti
n
g
it
s
r
o
b
u
stn
e
ss
a
n
d
c
li
n
ica
l
re
lev
a
n
c
e
.
Th
e
su
g
g
e
ste
d
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
(
CNN
)
-
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
(
LS
TM
)
-
a
tt
e
n
ti
o
n
h
y
b
rid
m
o
d
e
l
o
u
tp
e
rfo
rm
e
d
with
re
g
re
ss
io
n
m
e
tri
c
s
o
f
ro
o
t
m
e
a
n
sq
u
a
re
e
rro
r
(
RM
S
E
)
o
f
1
.
0
9
,
m
e
a
n
a
b
s
o
lu
te
e
rro
r
(
M
AE
)
b
y
0
.
6
4
,
a
n
d
R
2
sc
o
re
o
f
0
.
9
8
a
n
d
c
las
sifica
ti
o
n
m
e
tri
c
s
o
f
a
c
c
u
ra
c
y
9
7
.
1
%
,
p
re
c
isio
n
o
f
9
7
.
6
%
,
re
c
a
ll
o
f
9
6
%
,
a
n
d
F
1
-
sc
o
re
o
f
9
6
.
8
%
b
y
c
a
p
tu
rin
g
b
o
t
h
t
h
e
lo
c
a
l
p
a
t
tern
s a
n
d
tem
p
o
ra
l
d
y
n
a
m
ics
o
f
th
e
g
a
it
d
a
ta.
K
ey
w
o
r
d
s
:
Atten
tio
n
B
io
m
ec
h
an
ical
f
ea
tu
r
es
L
o
wer
lim
b
f
r
ac
tu
r
es
R
eh
ab
ilit
atio
n
Sy
m
m
etr
y
in
d
e
x
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
:
Pu
s
h
p
alath
a
Ob
an
n
a
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
i
n
g
,
J
AI
N
(
Dee
m
ed
-
to
-
be
-
Un
iv
er
s
ity
)
B
en
g
alu
r
u
,
Kar
n
atak
a,
I
n
d
ia
E
m
ail:
p
u
s
h
p
alath
a.
s
o
n
u
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
R
eh
ab
ilit
atio
n
p
lay
s
a
cr
itical
r
o
le
in
r
esto
r
in
g
m
o
b
ilit
y
,
f
u
n
ctio
n
al
in
d
e
p
en
d
e
n
ce
,
an
d
o
v
e
r
all
q
u
ality
o
f
life
a
f
ter
a
lo
wer
lim
b
f
r
ac
t
u
r
e.
T
r
ad
itio
n
al
clin
ical
ass
ess
m
en
t
r
elies
o
n
o
b
s
er
v
atio
n
al
g
ait
an
aly
s
is
,
wh
ich
is
s
u
b
jectiv
e
an
d
o
f
ten
u
n
ab
le
to
d
etec
t
s
u
b
tle
b
io
m
ec
h
a
n
ical
im
p
r
o
v
em
en
ts
.
W
ith
th
e
ad
v
an
ce
m
en
ts
in
wea
r
ab
le
s
en
s
o
r
s
an
d
f
o
r
ce
p
l
atf
o
r
m
s
,
clin
ician
s
ca
n
n
o
w
o
b
tain
q
u
an
titativ
e
g
ait
m
ea
s
u
r
es su
ch
as p
ea
k
f
o
r
ce
(
PF
)
,
im
p
u
ls
e
,
an
d
s
y
m
m
etr
y
.
Ho
wev
er
,
a
s
ig
n
if
ica
n
t
ch
a
llen
g
e
r
em
ain
s
in
ef
f
ec
tiv
ely
in
ter
p
r
etin
g
t
h
ese
co
m
p
lex
,
lar
g
e
-
s
ca
le
d
atasets
to
ac
cu
r
ately
tr
ac
k
a
p
atien
t's
f
u
n
ctio
n
al
r
ec
o
v
e
r
y
.
T
h
is
s
tu
d
y
d
esig
n
s
an
d
v
alid
ates
a
d
ee
p
lear
n
in
g
(
D
L
)
r
eh
a
b
ilit
atio
n
f
r
am
ewo
r
k
th
at
co
m
b
in
es
a
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN
)
-
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
m
o
d
ellin
g
with
an
atten
tio
n
la
y
er
a
n
d
s
y
m
m
etr
y
in
d
ex
an
aly
s
is
t
o
o
b
jectiv
ely
m
o
n
ito
r
g
ait
r
ec
o
v
er
y
i
n
lo
wer
lim
b
f
r
ac
tu
r
e
p
atien
ts
.
T
h
e
f
r
am
ewo
r
k
aim
s
to
co
m
p
u
te
k
e
y
b
io
m
ec
h
an
ical
f
ea
tu
r
es
f
r
o
m
v
er
tical
g
r
o
u
n
d
r
ea
ctio
n
f
o
r
ce
(
GR
F)
s
ig
n
als,
ca
lcu
late
s
y
m
m
etr
y
in
d
ices
ac
r
o
s
s
m
u
ltip
le
r
eh
a
b
ilit
atio
n
s
ess
io
n
s
,
an
d
u
s
e
a
C
NN
-
L
STM
h
y
b
r
id
ar
ch
itectu
r
e
to
class
if
y
a
p
atien
t's
p
r
o
g
r
ess
as
eith
er
im
p
r
o
v
ed
o
r
n
o
t im
p
r
o
v
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.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
6
2
5
-
3
6
3
6
3626
T
h
e
ass
ess
m
en
t
o
f
g
ait
an
d
b
a
lan
ce
in
th
e
a
g
ed
p
eo
p
le
with
m
ild
im
p
air
m
e
n
t
is
d
o
n
e
u
s
in
g
a
g
ait
an
d
b
alan
ce
s
m
ar
tp
h
o
n
e
a
p
p
.
T
h
e
g
ait
p
ar
am
eter
s
wer
e
m
ea
s
u
r
ed
u
s
in
g
th
e
ap
p
.
T
h
is
m
eth
o
d
is
n
o
t
g
en
er
alize
d
f
o
r
d
if
f
er
en
t a
g
e
g
r
o
u
p
s
[
1
]
.
T
h
e
g
ait
an
aly
s
is
an
d
m
o
tio
n
s
c
o
r
e
to
o
l is u
s
ed
o
n
ly
f
o
r
1
0
o
b
s
er
v
atio
n
al
an
d
f
iv
e
o
th
er
g
ait
p
ar
am
ete
r
s
ar
e
u
s
ed
to
d
iv
id
e
th
e
p
at
h
o
lo
g
ical
an
d
p
h
y
s
io
lo
g
ical
g
ait
an
d
is
u
s
ed
f
o
r
clin
ical
p
r
ac
tice
[
2
]
.
T
ec
h
n
o
lo
g
ies
lik
e
wea
r
ab
le
s
en
s
o
r
s
an
d
m
ac
h
in
e
lear
n
in
g
(
ML
)
m
eth
o
d
s
ar
e
u
s
ed
f
o
r
m
o
d
er
n
g
ait
an
aly
s
is
.
T
h
e
ML
m
eth
o
d
s
im
p
r
o
v
e
t
h
e
ac
cu
r
ac
y
o
f
th
e
s
y
s
tem
[
3
]
.
Ma
n
y
ML
class
if
ier
s
ar
e
u
s
ed
,
a
n
d
p
o
s
t
-
in
ju
r
y
r
ec
o
v
er
y
o
f
g
ait
v
a
r
iab
les
is
co
m
p
u
ted
f
o
r
d
if
f
e
r
e
n
t
p
atien
ts
f
o
r
t
h
e
g
ait
an
aly
s
i
s
[
4
]
.
T
h
e
k
n
ee
an
d
an
k
le
b
io
m
ec
h
an
ics
wer
e
in
f
l
u
en
ce
d
b
y
th
e
h
ee
l
-
f
ir
s
t
s
tr
ik
e
g
ait
in
th
e
h
ea
lth
y
p
er
s
o
n
at
h
lete.
T
h
e
g
ait
p
atter
n
is
m
o
d
if
ied
to
r
ed
u
ce
th
e
lo
wer
k
n
ee
s
tr
ess
with
th
e
p
o
ten
tial
ap
p
licatio
n
in
r
e
h
ab
ilit
atio
n
[
5
]
.
T
h
e
g
ai
t
im
p
r
o
v
em
e
n
t
is
m
o
n
ito
r
e
d
b
y
u
s
in
g
v
er
tical
GR
F
as
m
ar
k
er
af
ter
lo
wer
lim
b
in
ju
r
y
,
with
t
h
e
s
p
atio
-
tem
p
o
r
al,
jo
in
t
k
in
em
atics
,
a
n
d
k
in
etic
m
ea
s
u
r
es
im
p
r
o
v
ed
af
ter
s
o
m
e
m
o
n
t
h
s
.
T
h
e
walk
in
g
s
p
ee
d
is
ex
am
in
ed
d
u
r
in
g
r
eh
ab
ilit
atio
n
[
6
]
.
T
h
e
f
o
r
ce
p
late
is
u
s
ed
to
g
et
th
e
GR
F
d
a
ta
f
o
r
p
atien
ts
.
T
h
e
in
tr
a
-
in
ter
leg
weig
h
t
b
ea
r
in
g
an
d
s
tab
ilit
y
ar
e
esti
m
ated
u
s
i
n
g
th
e
g
ath
er
ed
d
ata
with
all
f
r
ac
tu
r
e
ty
p
es
[
7
]
.
T
h
e
p
atien
ts
ar
e
watc
h
ed
o
n
th
e
An
ter
io
r
cr
u
ciate
lig
am
en
t
f
o
r
6
to
1
2
m
o
n
t
h
s
af
ter
o
p
er
atio
n
an
d
ex
p
l
o
r
ed
t
h
at
b
y
ch
a
n
g
i
n
g
th
e
v
er
tical
GR
F
with
b
io
f
ee
d
b
ac
k
,
th
at
im
p
ac
t
th
e
k
n
ee
d
u
r
i
n
g
walk
i
n
g
[
8
]
.
T
h
e
g
ait
s
y
m
m
etr
y
ev
alu
at
io
n
o
f
p
o
s
t
-
s
tr
o
k
e
p
atien
ts
is
d
o
n
e
u
s
in
g
3
D
ac
ce
ler
o
m
eter
[
9
]
.
T
h
e
lo
wer
li
m
b
o
p
tim
izatio
n
is
d
o
n
e
u
s
in
g
ML
d
ata
-
d
r
iv
en
ap
p
r
o
ac
h
es
[
1
0
]
.
R
ea
l
-
tim
e
g
ait
an
d
p
o
s
itio
n
m
o
n
ito
r
in
g
in
en
ab
led
b
y
u
s
in
g
th
e
s
en
s
Flo
o
r
s
m
ar
t
ca
r
p
et
with
an
in
er
tial m
ea
s
u
r
em
en
t
u
n
it
(
I
MU
)
.
T
h
e
clo
u
d
-
b
ased
d
ash
b
o
ar
d
s
ar
e
p
r
o
v
id
e
d
to
th
e
t
h
er
a
p
is
t a
n
d
p
atien
ts
b
y
an
aly
zin
g
th
e
d
ata
u
s
in
g
ML
alg
o
r
ith
m
s
,
s
u
p
p
o
r
tin
g
tr
ain
i
n
g
id
ea
s
an
d
im
p
r
o
v
em
en
ts
in
th
e
r
eh
ab
ilit
atio
n
o
u
tco
m
es
[
1
1
]
.
T
h
e
we
ar
a
b
l
e
I
M
U
p
r
o
v
i
d
e
r
el
ia
b
le
g
ait
p
a
r
a
m
e
te
r
s
t
o
t
r
ac
k
t
h
e
l
o
n
g
-
te
r
m
r
eh
a
b
ili
tat
io
n
o
u
tc
o
m
es
o
f
ti
b
ia
l
f
r
a
ct
u
r
e
an
d
l
o
we
r
e
x
t
r
e
m
it
y
tr
a
u
m
a
p
a
tie
n
ts
[
1
2
]
–
[
1
4
]
.
T
h
e
r
esear
ch
g
ap
s
id
en
tifie
d
f
r
o
m
th
e
ex
is
tin
g
wo
r
k
ar
e
th
at
th
e
g
ait
an
aly
s
is
i
s
im
p
r
o
v
ed
b
y
th
e
DL
an
d
th
e
m
ajo
r
ity
o
f
DL
ap
p
li
ca
tio
n
s
h
av
e
b
ee
n
lim
ited
t
o
d
is
cr
ete
class
if
icatio
n
task
s
s
u
ch
as
id
en
tif
y
in
g
h
ea
lth
y
an
d
d
is
o
r
d
er
g
aits
.
W
h
ile
th
ese
d
iag
n
o
s
tic
to
o
ls
ar
e
v
alu
a
b
le,
th
ey
d
o
n
o
t
ad
d
r
ess
th
e
co
n
tin
u
o
u
s
,
f
in
e
-
g
r
ain
e
d
m
o
n
ito
r
in
g
r
eq
u
ir
ed
f
o
r
r
eh
a
b
ilit
atio
n
.
C
o
n
v
en
ti
o
n
al
m
o
d
els o
f
ten
co
n
s
id
er
ea
ch
g
ait
s
ess
io
n
as a
d
is
cr
ete
ev
en
t,
in
s
tead
as
p
ar
t
o
f
a
p
r
o
g
r
ess
iv
e
tim
e
s
er
ies.
F
o
r
lo
wer
lim
b
f
r
ac
tu
r
es
,
a
f
r
a
m
ewo
r
k
is
r
eq
u
ir
ed
to
q
u
an
tif
y
th
e
f
u
n
ctio
n
al
p
r
o
g
r
ess
o
v
er
tim
e.
T
h
e
n
o
v
el
co
n
t
r
ib
u
tio
n
o
f
th
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
d
i
r
ec
tly
ad
d
r
ess
es
th
is
g
ap
.
T
h
e
C
NN
-
L
STM
ar
ch
itectu
r
e
is
in
teg
r
ate
d
with
s
y
m
m
etr
y
in
d
ex
an
aly
s
is
,
p
r
o
v
id
es
a
q
u
an
titativ
e
ap
p
r
o
ac
h
to
tr
ac
k
t
h
e
r
eh
ab
ilit
atio
n
p
r
o
g
r
ess
.
Fro
m
th
e
v
er
tical
GR
F
d
ata
th
e
m
o
d
el
d
er
iv
es
t
h
e
k
i
n
etic
m
etr
ic
s
PF
,
tim
e
to
p
ea
k
(
T
T
P)
,
im
p
u
ls
e,
lo
ad
i
n
g
r
at
e
(
L
R
)
,
a
n
d
u
n
lo
a
d
in
g
r
ate
(
UR
)
f
o
r
b
o
th
th
e
le
g
s
t
o
ch
ar
a
cter
ize
g
ait
im
p
r
o
v
em
e
n
t.
T
h
e
co
n
v
o
lu
ti
o
n
1
D
lay
e
r
tr
an
s
f
o
r
m
s
ea
c
h
f
ea
tu
r
e
i
n
d
ep
e
n
d
en
tly
m
ea
n
s
th
e
p
o
in
t
wis
e
co
n
v
o
l
u
tio
n
is
p
e
r
f
o
r
m
ed
.
T
h
e
L
STM
lay
er
ca
p
tu
r
es
th
e
te
m
p
o
r
al
d
ep
en
d
en
cies
an
d
m
o
d
els
th
e
p
r
o
g
r
ess
io
n
o
f
g
ait
f
ea
tu
r
es
ac
r
o
s
s
tim
e.
T
h
e
atten
tio
n
lay
er
im
p
r
o
v
es
th
e
p
r
ed
ictiv
e
p
er
f
o
r
m
a
n
ce
an
d
also
in
ter
p
r
etab
ilit
y
an
d
allo
win
g
clin
ician
s
to
u
n
d
er
s
tan
d
wh
ich
p
ar
t
o
f
th
e
g
a
it
cy
cle
co
n
t
r
ib
u
tes
m
o
s
t
to
t
h
e
m
o
d
el
d
ec
is
io
n
.
T
h
e
lo
n
g
itu
d
in
al
tr
ac
k
i
n
g
o
f
g
ait
im
p
r
o
v
em
en
t
is
d
o
n
e
f
r
o
m
s
ess
io
n
to
s
ess
io
n
with
th
e
r
ed
u
ctio
n
o
f
3
%
i
n
asy
m
m
etr
y
an
d
f
in
ally
,
th
e
m
o
d
el
p
r
ed
icts
g
ait
im
p
r
o
v
e
m
en
t
if
th
er
e
is
a
r
ed
u
ctio
n
in
asy
m
m
etr
y
g
r
ea
ter
th
a
n
o
r
eq
u
al
to
3
0
%
with
th
e
ch
an
g
e
in
c
o
m
p
o
s
ite
asy
m
m
etr
y
in
d
ex
(
ASI
)
v
al
u
e
f
r
o
m
in
i
tial
to
f
in
al
s
ess
io
n
.
T
h
e
m
o
d
el
in
tr
o
d
u
ce
s
a
r
eg
r
ess
io
n
to
class
if
icatio
n
b
r
id
g
e,
allo
win
g
co
n
tin
u
o
u
s
p
r
e
d
ictio
n
an
d
b
in
a
r
y
d
ec
is
io
n
-
m
ak
in
g
with
in
t
h
e
s
a
m
e
f
r
am
ewo
r
k
.
T
h
e
p
u
b
licly
av
ailab
le
GaitR
ec
d
ataset
to
c
an
b
e
u
s
ed
t
o
tr
ac
k
th
e
lo
n
g
-
te
r
m
r
ec
o
v
er
y
o
f
lo
w
er
lim
b
f
r
ac
tu
r
e
p
atien
ts
.
2.
P
RO
P
O
SE
D
M
E
T
H
O
D
2
.
1
.
B
io
m
e
cha
nica
l
f
ea
t
ure
e
x
t
ra
ct
i
on
Key
b
io
m
ec
h
an
ical
f
ea
tu
r
es
wer
e
co
m
p
u
ted
f
o
r
ea
ch
r
ec
o
r
d
ed
s
ess
io
n
an
d
tr
ial
f
o
r
b
o
t
h
t
h
e
af
f
ec
ted
an
d
u
n
af
f
ec
ted
lim
b
s
.
T
h
e
(
1
)
to
(
5
)
r
ep
r
esen
t
th
e
m
ath
em
ati
ca
l f
o
r
m
u
las u
s
ed
t
o
co
m
p
u
te
f
ea
tu
r
es.
−
PF
: th
e
m
ax
im
u
m
v
er
tical
f
o
r
c
e
v
er
tical
v
alu
e
r
ec
o
r
d
e
d
d
u
r
in
g
th
e
s
tan
ce
p
h
ase
o
f
a
g
ait
c
y
cle.
=
ma
x
(
(
)
)
(
1
)
−
T
T
P:
th
e
tim
e
elap
s
ed
f
r
o
m
in
itial f
o
o
t c
o
n
tact
to
t
h
e
o
cc
u
r
r
en
ce
o
f
th
e
PF
.
=
−
(
2
)
−
I
m
p
u
ls
e:
th
e
in
te
g
r
al
o
f
th
e
v
e
r
tical
f
o
r
ce
o
v
er
th
e
d
u
r
atio
n
o
f
th
e
s
tan
ce
p
h
ase
o
f
g
ait,
r
e
p
r
esen
tin
g
th
e
to
tal
lo
ad
ap
p
lied
o
v
e
r
tim
e.
=
∫
(
)
(
3
)
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
S
ymm
etry in
d
ex
-
b
a
s
ed
g
a
it imp
r
o
ve
men
t p
r
ed
ictio
n
u
s
in
g
a
C
N
N
-
LS
TM
…
(
P
u
s
h
p
a
la
th
a
Ob
a
n
n
a
)
3627
−
L
R
:
th
e
r
ate
o
f
f
o
r
ce
in
cr
ea
s
es d
u
r
in
g
th
e
weig
h
t a
cc
e
p
tan
ce
p
h
ase.
=
(
10%
)
−
(
0%
)
10%
−
0%
(
4)
−
UR
:
t
h
e
r
ate
o
f
f
o
r
ce
d
ec
r
ea
s
es
d
u
r
in
g
th
e
p
u
s
h
-
o
f
f
p
h
ase.
T
h
e
u
n
it
o
f
PF
is
%B
W
,
T
T
P
is
in
s
tan
ce
p
h
ase,
i
m
p
u
ls
e
in
(
%B
W
/
s
)
,
L
R
an
d
UR
in
(
%B
W
/
s
)
.
=
(
100%
)
−
(
90%
)
100%
−
90%
(
5
)
2
.
2
.
Sy
m
m
e
t
ry
ind
ex
mo
dellin
g
T
h
e
b
io
m
ec
h
a
n
ical
k
in
etic
f
ea
tu
r
es
ar
e
co
m
p
u
ted
f
o
r
b
o
th
th
e
lef
t
an
d
r
ig
h
t
leg
.
T
h
e
af
f
ec
ted
an
d
u
n
af
f
ec
ted
leg
s
ar
e
id
en
tifie
d
b
y
m
ap
p
in
g
v
alu
es
f
r
o
m
th
e
GR
F_
m
etad
ata.
csv
f
ile.
T
o
p
r
o
v
id
e
a
q
u
an
tifia
b
le
m
ea
s
u
r
e
o
f
g
ait
s
y
m
m
etr
y
,
t
h
e
ab
s
o
lu
te
s
y
m
m
etr
y
in
d
ex
was
co
m
p
u
ted
u
s
in
g
(
6
)
f
o
r
ea
ch
b
io
m
ec
h
an
ical
f
ea
tu
r
e
(
X)
.
T
h
e
g
ait
im
p
r
o
v
em
en
t
is
c
h
ec
k
ed
u
s
in
g
th
e
s
y
m
m
etr
y
in
d
e
x
a
f
ter
r
eh
a
b
ilit
atio
n
.
T
h
e
g
ait
s
y
m
m
etr
y
is
ca
p
tu
r
ed
o
n
ly
at
s
in
g
le
p
o
in
t
[
1
5
]
.
wh
er
e
Xaf
f
ec
ted
an
d
Xu
n
af
f
ec
ted
ar
e
th
e
f
ea
tu
r
e
v
alu
es
f
o
r
th
e
af
f
ec
ted
a
n
d
u
n
af
f
ec
ted
lim
b
s
,
r
esp
ec
tiv
ely
.
=
|
−
0
.
5
×
(
+
)
|
×
100
(
6
)
F
o
r
b
i
l
a
t
e
r
al
f
r
a
c
t
u
r
e
p
a
t
i
e
n
t
s
,
t
h
e
A
S
I
i
s
c
o
m
p
u
t
e
d
u
s
i
n
g
(
7
)
.
T
h
i
s
d
a
t
a
t
r
a
n
s
l
at
es
t
h
e
b
i
l
a
t
e
r
al
b
i
o
m
e
c
h
a
n
i
c
a
l
d
i
f
f
e
r
e
n
c
e
s
i
n
to
a
q
u
a
n
t
i
f
i
a
b
l
e
m
et
r
i
c
o
f
as
y
m
m
e
t
r
y
,
w
h
i
c
h
is
ess
e
n
t
i
al
f
o
r
r
e
h
a
b
i
l
it
a
t
i
o
n
a
s
s
e
s
s
m
e
n
t
.
T
h
e
d
a
ta
s
e
r
v
es
as
a
r
e
p
r
e
s
e
n
t
at
i
v
e
e
x
a
m
p
l
e
o
f
t
h
e
a
n
a
l
y
s
is
p
e
r
f
o
r
m
e
d
f
o
r
a
l
l
s
u
b
je
c
t
s
i
n
t
h
e
s
t
u
d
y
.
=
|
−
ℎ
0
.
5
×
(
+
ℎ
)
|
×
100
(
7
)
2
.
3
.
L
o
ng
it
ud
ina
l
g
a
it
pro
g
r
ess
a
na
ly
s
is
T
h
e
l
o
n
g
it
u
d
in
al
t
r
ac
k
i
n
g
o
v
e
r
tim
e
is
co
n
d
u
ct
ed
t
o
e
v
a
lu
ate
f
u
n
cti
o
n
al
r
ec
o
v
e
r
y
o
f
k
n
ee
r
e
p
la
ce
m
e
n
t
s
u
r
g
er
y
b
y
c
o
ll
ec
t
in
g
g
ait
d
at
a
a
n
d
p
ati
en
t
-
r
e
p
o
r
te
d
o
u
t
c
o
m
e
m
ea
s
u
r
es
(
PR
O
M)
o
f
2
0
p
at
ie
n
ts
[
1
6
]
.
I
n
th
e
p
r
o
p
o
s
ed
wo
r
k
to
m
o
n
ito
r
th
e
tem
p
o
r
al
p
r
o
g
r
ess
io
n
o
f
g
a
it
r
eh
ab
ilit
atio
n
,
a
weig
h
ted
co
m
p
o
s
ite
ASI
is
co
m
p
u
ted
b
y
co
m
b
in
in
g
d
i
f
f
e
r
en
t
k
in
etic
f
ea
tu
r
es.
Sin
ce
ea
ch
f
ea
tu
r
e
co
n
tr
ib
u
tes
d
if
f
er
e
n
tly
to
o
v
e
r
all
g
ait
s
y
m
m
etr
y
,
ev
id
en
ce
-
b
ased
we
ig
h
ts
wer
e
ass
ig
n
ed
to
r
ef
lect
th
eir
r
elativ
e
b
io
m
ec
h
an
ical
i
m
p
o
r
tan
ce
:
0
.
3
5
f
o
r
PF
,
0
.
2
5
f
o
r
im
p
u
ls
e,
0
.
1
5
f
o
r
TTP
,
0
.
1
5
f
o
r
LR
,
an
d
0
.
1
0
f
o
r
UR
.
T
h
e
weig
h
ted
co
m
p
o
s
it
e
s
co
r
e
is
co
m
p
u
ted
u
s
in
g
(
8
)
b
y
tak
i
n
g
th
e
s
u
m
o
f
p
r
o
d
u
ct
o
f
ASI
v
alu
e
o
f
f
ea
tu
r
e
b
y
its
co
r
r
esp
o
n
d
in
g
weig
h
t
.
T
h
e
im
p
r
o
v
e
m
en
t
in
th
e
s
y
m
m
etr
y
is
q
u
an
tifie
d
b
y
co
m
p
u
tin
g
s
ess
io
n
to
s
ess
io
n
ch
an
g
e
f
r
o
m
th
e
c
o
n
s
ec
u
ti
v
e
s
ess
io
n
s
u
s
in
g
a
s
lid
in
g
win
d
o
w
ap
p
r
o
ac
h
u
s
in
g
(
9
)
.
T
h
e
(
1
0
)
is
u
s
ed
to
co
m
p
u
te
th
e
o
v
er
all
im
p
r
o
v
em
en
t in
th
e
g
ait
f
r
o
m
th
e
co
m
p
o
s
ite
ASI
v
alu
es o
f
t
h
e
i
n
itial a
n
d
f
in
al
s
ess
io
n
.
=
∑
=
1
×
(
8
)
△
(
)
=
(
+
1
)
−
(
)
(
9
)
=
(
−
|
|
)
×
100
(
1
0
)
I
n
(
8
)
,
i
s
t
h
e
c
o
m
p
o
s
i
t
e
AS
I
f
o
r
t
h
e
i
th
s
a
m
p
l
e
,
r
e
p
r
e
s
e
n
ts
t
h
e
A
S
I
v
a
l
u
e
o
f
j
th
f
e
a
t
u
r
e
f
o
r
t
h
e
i
th
s
a
m
p
l
e
,
is
t
h
e
w
e
i
g
h
t
a
s
s
i
g
n
ed
t
o
t
h
e
j
th
s
a
m
p
l
e
,
a
n
d
t
h
e
s
u
m
m
a
t
i
o
n
r
u
n
s
o
v
e
r
al
l
f
ea
t
u
r
e
s
.
I
n
(
9
)
,
△
(
)
r
e
p
r
e
s
e
n
t
s
t
h
e
c
h
a
n
g
e
i
n
A
S
I
f
o
r
t
h
e
f
e
a
t
u
r
e
f
b
e
t
w
e
e
n
c
o
n
s
e
c
u
t
i
v
e
s
e
s
s
i
o
n
s
(
+
1
)
,
(
)
a
t
s
e
s
s
i
o
n
s
+
1
a
n
d
r
e
s
p
e
c
ti
v
e
l
y
.
I
n
(
1
0
)
,
,
d
e
n
o
t
e
t
h
e
f
i
r
s
t
a
n
d
l
a
s
t
s
e
s
s
i
o
n
c
o
m
p
o
s
i
t
e
A
S
I
v
a
l
u
es
o
f
e
a
c
h
p
a
t
ie
n
t
.
I
t
m
e
a
s
u
r
e
s
t
h
e
p
e
r
c
e
n
t
a
g
e
d
e
c
r
ea
s
e
i
n
t
h
e
a
s
y
m
m
e
t
r
y
o
v
e
r
ti
m
e.
2
.
4
.
Dee
p
lea
rning
cla
s
s
if
ica
t
io
n m
o
del
T
h
e
lo
ca
l
s
eq
u
e
n
tial
p
atter
n
s
ar
e
ca
p
tu
r
ed
u
s
in
g
DL
m
eth
o
d
s
lik
e
C
NN,
C
NN
-
L
STM
,
an
d
o
t
h
er
h
y
b
r
id
m
o
d
els
b
ased
o
n
GR
F
tim
e
s
er
ies
d
ata
f
o
r
g
ait
class
if
icatio
n
,
an
d
to
p
r
ed
ict
th
e
im
p
r
o
v
em
e
n
t
af
te
r
r
eh
ab
ilit
atio
n
.
T
h
e
r
esu
lts
ar
e
ex
p
lain
ed
u
s
in
g
th
e
la
y
er
-
wis
e
r
elev
an
ce
p
r
o
p
a
g
atio
n
to
o
l.
I
n
r
ec
en
t
s
tu
d
ies
,
th
e
v
er
tical
GR
F
i
s
esti
m
ated
f
r
o
m
I
MU
s
,
2
D
v
id
eo
u
s
in
g
ML
f
o
r
th
e
ass
ess
m
en
t
o
f
th
e
g
ait
o
u
ts
id
e
th
e
lab
an
d
to
tr
ac
k
th
e
lo
n
g
itu
d
in
al
r
eh
a
b
ilit
atio
n
[
1
7
]
–
[
1
9
]
.
T
h
is
s
tu
d
y
u
s
ed
th
e
C
NN
-
L
STM
-
atten
tio
n
h
y
b
r
id
class
if
ier
s
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.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
6
2
5
-
3
6
3
6
3628
an
d
ev
al
u
ated
to
au
to
m
ate
t
h
e
class
if
icatio
n
o
f
r
eh
a
b
ilit
atio
n
o
f
o
u
tco
m
es
b
ased
o
n
th
eir
ca
p
ac
ity
to
a
n
aly
ze
th
e
tem
p
o
r
al
ASI
s
eq
u
e
n
ce
s
.
T
h
e
m
ath
em
atica
l u
n
d
er
p
i
n
n
in
g
s
f
o
r
th
ese
m
o
d
els ar
e
as f
o
ll
o
ws.
2
.
4
.
1
.
T
he
CNN
-
L
S
T
M
-
a
t
t
ent
io
n hy
brid m
o
del
T
h
is
h
y
b
r
i
d
ap
p
r
o
ac
h
en
ab
les
th
e
m
o
d
el
to
r
ec
o
g
n
ize
co
m
p
lex
g
ait
p
atter
n
s
o
v
er
tim
e,
a
ca
p
ab
ilit
y
th
at
d
is
tin
g
u
is
h
es its
p
er
f
o
r
m
a
n
ce
.
Alg
o
r
ith
m
:
f
o
r
war
d
p
r
o
p
a
g
atio
n
f
o
r
th
e
C
NN
-
L
STM
h
y
b
r
i
d
DL
m
o
d
e
l
.
i)
Mo
d
el
in
p
u
t:
in
(
1
1
)
wh
e
r
e
n
is
th
e
b
atch
s
ize,
d
is
th
e
n
u
m
b
er
o
f
tim
e
s
tep
s
(
e.
g
.
,
r
eh
ab
ilit
atio
n
s
ess
io
n
s
)
,
an
d
1
r
ep
r
esen
ts
th
e
n
u
m
b
e
r
o
f
ch
an
n
els (
u
n
iv
ar
iat
e
ASI
tim
e
s
er
ies)
.
∈
×
×
1
(
1
1
)
ii)
T
h
e
p
r
o
ce
s
s
in
g
o
f
i
n
p
u
t
X
o
f
(
1
1
)
in
to
p
r
ed
ictio
n
ŷ
is
d
ef
in
ed
b
y
th
e
f
o
llo
win
g
s
eq
u
en
ce
o
f
m
ath
em
atica
l o
p
er
atio
n
s
.
−
Step
1
:
f
ea
tu
r
e
e
x
tr
ac
tio
n
v
ia
1
D
co
n
v
o
l
u
ti
on
.
I
n
(
1
2
)
an
d
(
1
3
)
(
)
∈
1
×
is
th
e
k
th
co
n
v
o
lu
tio
n
f
ilter
,
is
th
e
b
ias
f
o
r
f
ilter
k
,
k
=1
,
2
,
3
….
.
.
,
3
2
f
ilter
s
,
a
n
d
C
=1
m
ea
n
s
it
r
ed
u
ce
s
to
a
lin
ea
r
tr
an
s
f
o
r
m
with
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
.
T
h
is
ex
tr
ac
ts
th
e
lo
ca
l
f
ea
tu
r
e
r
ep
r
esen
tatio
n
f
o
r
ea
ch
tim
e
s
tep
.
,
,
=
∑
∑
,
(
)
,
+
,
+
−
1
=
0
=
1
(
1
2
)
∈
×
×
32
(
1
3
)
−
Step
2
:
d
im
en
s
io
n
ality
r
ed
u
ctio
n
v
ia
m
ax
p
o
o
lin
g
,
m
ax
p
o
o
l
in
g
1
D
with
p
o
o
lin
g
s
ize
=2
s
ca
n
s
th
e
tem
p
o
r
al
f
ea
tu
r
e
m
ap
two
-
ti
m
e
s
tep
s
at
a
tim
e
an
d
k
ee
p
s
o
n
ly
th
e
s
tr
o
n
g
est
ac
tiv
atio
n
f
o
r
ea
ch
ch
an
n
el
,
a
n
d
th
is
r
ed
u
ce
s
th
e
s
eq
u
en
ce
len
g
th
is
s
h
o
wn
in
(
1
4
)
.
W
h
er
e
T
=d
/
2
.
,
,
=
ma
x
(
,
2
,
,
,
2
+
1
,
)
∈
×
×
32
(
1
4
)
−
Step
3
:
L
STM
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
L
STM
is
u
s
ed
to
ex
tr
ac
t
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies
an
d
tem
p
o
r
al
ev
o
lu
tio
n
o
f
r
eh
ab
il
itatio
n
o
u
tco
m
es
u
n
d
er
m
u
lti
p
le
s
ess
io
n
s
u
s
in
g
(
1
5
)
to
(
2
1
)
.
W
h
er
e
ℎ
∈
32
is
th
e
h
id
d
en
s
tate,
∈
32
is
th
e
c
ell
s
tate,
⨀
r
ep
r
esen
ts
elem
en
t
wis
e
m
u
ltip
licatio
n
,
an
d
d
en
o
te
th
e
s
ig
m
o
id
ac
tiv
a
tio
n
.
=
(
∙
[
ℎ
−
1
;
]
+
)
(
1
5
)
=
(
∙
[
ℎ
−
1
;
]
+
)
(
1
6
)
̃
=
ta
n
h
(
[
ℎ
−
1
;
]
+
(
1
7
)
=
⨀
−
1
+
⨀
̃
(
1
8
)
=
(
[
ℎ
−
1
;
]
+
)
(
1
9
)
ℎ
=
⨀
ta
n
h
(
)
(
2
0
)
=
[
ℎ
1
,
ℎ
2
,
…
…
.
ℎ
]
∈
×
×
32
(
2
1
)
−
Step
4
:
atten
tio
n
,
lin
ea
r
p
r
o
jectio
n
s
ar
e
=
,
=
,
=
.
W
h
er
e
,
,
∈
32
×
32
.
T
h
e
s
im
ilar
ity
s
co
r
e
b
etwe
en
q
u
er
y
at
tim
estep
i
an
d
k
ey
at
s
tep
j
is
g
iv
en
b
y
=
√
32
.
Fo
r
ea
ch
tim
estep
th
e
s
co
r
es
1
,
2
,
…
…
.
in
to
p
r
o
b
ab
ilit
ies
th
at
s
u
m
to
1
.
I
n
(
2
2
)
,
r
ep
r
esen
ts
h
o
w
m
u
ch
tim
estep
co
n
tr
ib
u
tes to
th
e
tim
estep
.
,
=
ex
p
(
)
∑
ex
p
(
)
=
1
(
2
2
)
−
Fin
ally
,
th
e
tim
estep
I
b
ec
o
m
e
th
e
weig
h
ted
s
u
m
o
f
all
v
e
cto
r
s
u
s
in
g
atten
tio
n
weig
h
ts
an
d
is
s
h
o
wn
in
(
2
3
)
.
=
∑
∈
×
×
32
=
1
(
2
3
)
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
S
ymm
etry in
d
ex
-
b
a
s
ed
g
a
it imp
r
o
ve
men
t p
r
ed
ictio
n
u
s
in
g
a
C
N
N
-
LS
TM
…
(
P
u
s
h
p
a
la
th
a
Ob
a
n
n
a
)
3629
−
Step
5
:
tr
an
s
itio
n
to
f
u
lly
co
n
n
ec
ted
lay
er
s
(
Flatten
)
,
in
(
2
4
)
is
u
s
ed
to
r
esh
ap
e
th
e
m
u
ltid
im
en
s
io
n
al
f
ea
tu
r
e
m
ap
g
en
er
ated
b
y
p
r
ev
io
u
s
lay
er
in
to
a
o
n
e
l
o
n
g
v
ec
to
r
f
o
r
ea
c
h
s
am
p
le.
=
ℎ
(
,
(
,
×
32
)
)
∈
×
1600
(
2
4
)
−
Step
6
:
d
r
o
p
o
u
t,
in
th
is
lay
er
a
r
eg
u
lar
izatio
n
tech
n
iq
u
e
is
ap
p
lied
to
r
ed
u
ce
o
v
er
f
i
ttin
g
b
y
d
is
ab
lin
g
a
p
o
r
tio
n
o
f
n
e
u
r
o
n
s
d
u
r
in
g
tr
ain
in
g
u
s
in
g
(
2
5
)
,
d
r
o
p
o
u
t =
0
.
4
an
d
k
ee
p
p
r
o
b
ab
ilit
y
=0
.
6
.
=
⨀
0
.
6
,
∼
(
0
.
6
)
(
2
5
)
−
Step
7
:
h
ig
h
-
le
v
el
r
ea
s
o
n
in
g
v
ia
f
u
lly
c
o
n
n
ec
te
d
(
Den
s
e)
l
ay
er
s
,
th
e
n
o
n
lin
ea
r
c
o
m
p
o
s
itio
n
o
f
ex
tr
ac
ted
f
ea
tu
r
es
is
lear
n
t
b
y
p
r
o
ce
s
s
in
g
th
e
f
latten
ed
v
ec
t
o
r
u
s
in
g
o
n
e
o
r
m
o
r
e
d
e
n
s
e
lay
er
s
.
T
h
e
(
2
6
)
r
ep
r
esen
ts
f
latten
e
d
in
p
u
t
f
r
o
m
th
e
p
r
ev
i
o
u
s
s
tep
.
W
h
er
e
(
1
)
∈
32
×
(
∙
32
)
is
th
e
weig
h
t
m
atr
ix
an
d
(
1
)
∈
×
32
is
th
e
b
ias v
ec
to
r
.
=
(
(
1
)
+
(
1
)
)
(
2
6
)
−
Step
8
:
o
u
tp
u
t r
eg
r
ess
io
n
lay
e
r
,
th
e
o
u
t
p
u
t la
y
er
p
r
ed
icts
th
e
ASI
u
s
in
g
(
2
7
)
.
̂
=
(
(
)
+
(
)
)
∈
×
1
(
2
7
)
3.
M
E
T
H
O
D
3
.
1
.
P
r
o
po
s
ed
m
o
del a
rc
hite
ct
ure
T
h
e
m
o
d
el
s
y
n
e
r
g
is
t
ica
ll
y
c
o
m
b
in
es
t
h
e
s
t
r
en
g
t
h
s
o
f
C
NNs
f
o
r
l
o
ca
l
te
m
p
o
r
al
p
att
er
n
e
x
t
r
a
cti
o
n
a
n
d
f
u
ll
y
co
n
n
e
ct
ed
n
et
wo
r
k
s
f
o
r
non
-
li
n
e
ar
c
lass
i
f
i
ca
t
io
n
,
en
a
b
li
n
g
r
o
b
u
s
t
an
al
y
s
is
o
f
te
m
p
o
r
al
g
ait
s
y
m
m
e
tr
y
p
at
te
r
n
s
f
o
r
p
r
e
d
i
cti
n
g
r
e
h
a
b
il
i
tati
o
n
o
u
t
c
o
m
es.
Fi
g
u
r
e
1
d
esc
r
i
b
es
t
h
e
s
u
g
g
est
ed
C
N
N
-
L
ST
M
-
a
tte
n
t
io
n
m
o
d
el
to
b
e
u
s
e
d
i
n
f
e
at
u
r
e
e
x
t
r
a
cti
o
n
a
n
d
s
e
q
u
e
n
ti
al
le
ar
n
i
n
g
.
T
h
e
wo
r
k
f
lo
w
s
t
ar
ts
wi
th
ASI
in
p
u
t
d
at
a
w
h
i
c
h
is
p
ass
ed
in
to
th
e
c
o
n
v
o
l
u
t
io
n
al
l
ay
er
s
w
h
er
e
l
o
c
al
s
e
q
u
e
n
ti
al
p
atte
r
n
s
a
r
e
ex
tr
ac
te
d
.
T
h
e
e
x
t
r
a
cte
d
c
h
a
r
ac
t
er
is
ti
cs
ar
e
c
o
n
v
e
y
e
d
t
o
L
ST
M
l
ay
e
r
s
t
h
a
t
a
ll
o
w
l
ea
r
n
i
n
g
d
e
p
e
n
d
e
n
c
ies
an
d
s
e
q
u
e
n
ce
m
o
d
el
s
ac
r
o
s
s
ti
m
e
.
T
h
e
atte
n
t
io
n
l
a
y
e
r
ass
i
g
n
s
we
ig
h
ts
t
o
t
h
e
f
e
at
u
r
es
f
o
r
b
et
te
r
m
o
d
el
p
r
ed
icti
o
n
.
Fo
r
c
lass
i
f
i
ca
t
io
n
o
r
p
r
e
d
ic
ti
o
n
,
t
h
e
o
u
t
p
u
t l
a
y
e
r
co
m
es
a
f
t
er
t
h
e
f
u
lly
c
o
n
n
ec
te
d
la
y
e
r
s
,
w
h
i
ch
a
g
g
r
e
g
at
e
th
e
p
r
e
v
i
o
u
s
l
y
o
b
ta
in
e
d
r
e
p
r
es
en
tat
io
n
s
.
Fig
u
r
e
1
.
C
NN
-
L
STM
m
o
d
el
ar
ch
itectu
r
e
3
.
2
.
Sy
s
t
e
m
wo
r
k
f
lo
w:
o
v
er
v
iew
a
nd
s
y
s
t
em
wo
rk
f
l
o
w
T
h
e
Fig
u
r
e
2
s
h
o
ws
th
e
g
ait
a
n
aly
s
is
d
u
r
in
g
r
eh
ab
ilit
atio
n
.
Her
e
th
e
p
r
o
ce
s
s
ed
GR
F
d
ata
is
co
llected
an
d
th
e
u
n
wan
te
d
n
o
is
e
was
f
ilter
ed
an
d
is
s
tan
d
ar
d
ized
th
e
m
ea
s
u
r
em
en
ts
in
th
e
GaitR
ec
d
ataset.
T
h
e
h
ea
lth
y
c
o
n
tr
o
l
r
elate
d
d
ata
a
n
d
th
e
p
atien
t
with
s
in
g
le
s
ess
io
n
ar
e
r
em
o
v
ed
f
r
o
m
th
e
d
a
taset.
T
h
e
af
f
ec
ted
an
d
u
n
af
f
ec
te
d
s
id
e
is
m
a
p
p
e
d
f
r
o
m
th
e
GR
F
m
etad
ata
d
at
aset.
T
h
en
th
e
b
io
m
ec
h
an
ical
k
in
etic
f
ea
tu
r
es
lik
e
PF
,
T
T
P,
i
m
p
u
ls
e,
L
R
,
UR
ar
e
co
m
p
u
ted
f
o
r
b
o
th
th
e
li
m
b
s
.
T
h
e
ASI
is
c
o
m
p
u
te
d
f
r
o
m
th
ese
f
ea
t
u
r
es
b
etwe
en
two
lim
b
s
.
T
h
e
co
m
p
o
s
ite
ASI
is
d
eter
m
in
ed
b
y
ass
ig
n
in
g
weig
h
ts
to
th
e
f
ea
tu
r
es.
T
h
e
m
o
d
el
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.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
6
2
5
-
3
6
3
6
3630
p
er
f
o
r
m
s
lo
n
g
itu
d
in
al
tr
ac
k
i
n
g
f
r
o
m
s
ess
io
n
to
s
ess
io
n
an
d
if
th
e
asy
m
m
etr
y
is
r
ed
u
ce
d
b
y
at
least
3
0
%
f
r
o
m
in
itial
to
f
in
al
s
ess
io
n
v
alu
es,
t
h
en
th
e
g
ait
is
r
ec
o
v
er
ed
an
d
t
h
at
is
th
e
en
d
o
f
r
e
h
ab
ilit
atio
n
.
I
f
n
o
t,
th
e
p
atien
t
will b
e
s
u
g
g
ested
f
o
r
n
ex
t ses
s
io
n
.
Fig
u
r
e
2
.
W
o
r
k
f
lo
w
o
f
g
ait
an
aly
s
is
d
u
r
in
g
r
eh
ab
ilit
atio
n
u
s
in
g
th
e
p
r
o
ce
s
s
ed
GaitR
ec
d
ataset
3
.
3
.
E
x
perim
ent
a
l
s
et
up
3
.
3
.
1
.
Da
t
a
s
et
des
cr
iptio
n
T
h
is
r
esear
ch
wo
r
k
lev
er
ag
es
th
e
GaitR
ec
d
ataset,
a
co
m
p
r
e
h
en
s
iv
e
an
d
o
p
en
s
o
u
r
ce
f
o
r
h
u
m
an
g
ait
an
aly
s
is
,
to
in
v
esti
g
ate
o
b
je
ctiv
e
r
eh
ab
ilit
atio
n
o
u
tco
m
es
in
p
atien
ts
with
lo
wer
lim
b
f
r
ac
tu
r
es.
As
a
lar
g
e
-
s
ca
le
d
ataset,
it
co
n
tain
s
7
5
,
7
3
2
b
ilater
al
walk
in
g
tr
ial
s
o
f
2
,
0
8
4
p
atien
ts
with
d
if
f
er
en
t
m
u
s
cu
lo
s
k
eleta
l
im
p
air
m
en
ts
.
T
h
e
af
f
ec
ted
lim
b
was
id
en
tifie
d
f
o
r
ea
ch
p
atien
t
u
s
in
g
th
e
GR
F_
m
etad
ata.
csv
f
ile,
wh
er
e
a
v
alu
e
o
f
'
0
'
in
d
icate
s
a
lef
t
af
f
ec
ted
leg
,
'
1
'
in
d
icate
s
a
r
ig
h
t
af
f
ec
ted
leg
,
an
d
'
2
'
s
ig
n
if
ies
a
b
ilater
al
f
r
ac
tu
r
e.
T
h
e
d
ataset
co
m
p
r
is
es
g
ait
r
ec
o
r
d
s
f
r
o
m
p
atien
ts
with
f
r
ac
t
u
r
es
class
if
ied
in
to
f
o
u
r
g
r
o
u
p
s
:
ca
lcan
eu
s
,
an
k
le,
h
ip
,
an
d
k
n
ee
.
T
h
is
r
ec
o
r
d
u
s
es
th
e
d
ata
co
llected
d
u
r
in
g
t
h
e
co
m
p
lete
r
eh
ab
ilit
atio
n
s
ta
y
o
f
p
atien
ts
v
ar
ies
f
r
o
m
wee
k
s
to
m
o
n
th
s
.
T
h
e
ef
f
ec
t
o
f
n
o
is
e
is
r
e
d
u
ce
d
b
y
u
s
in
g
2
n
d
o
r
d
er
B
u
tter
wo
r
t
h
L
PF
with
c
u
to
f
f
f
r
eq
u
e
n
cy
2
0
Hz
is
u
s
ed
with
am
p
litu
d
e
an
d
tim
e
n
o
r
m
aliza
tio
n
.
T
h
e
tim
e
is
n
o
r
m
alize
d
to
1
0
1
d
ata
p
o
in
ts
,
wh
ich
r
ed
u
ce
s
tem
p
o
r
al
d
if
f
e
r
en
ce
s
an
d
allo
ws
co
m
p
ar
ativ
e
an
aly
s
is
o
f
d
iv
er
s
e
s
tep
s
.
T
h
is
r
o
b
u
s
t
d
ataset
is
well
-
s
u
ited
f
o
r
ML
m
in
in
g
ap
p
licatio
n
s
aim
ed
at
ca
teg
o
r
izi
n
g
s
p
ec
if
ic
g
ait
p
atter
n
s
an
d
t
r
ac
k
in
g
f
u
n
ctio
n
al
r
ec
o
v
er
y
o
v
er
tim
e.
T
h
e
tr
ai
n
-
test
s
p
lit
f
o
llo
ws
th
e
p
r
ed
e
f
i
n
ed
s
p
lit
p
r
o
v
id
e
d
in
t
h
e
GR
F_
m
etad
ata.
csv
f
ile.
I
t is in
th
e
r
atio
7
0
:3
0
f
o
r
wo
r
k
.
T
h
e
s
u
b
ject
-
wis
e
s
p
litt
in
g
is
u
s
ed
in
th
e
p
r
o
p
o
s
ed
wo
r
k
.
3
.
3
.
2
.
Da
t
a
prepro
ce
s
s
ing
T
h
e
p
r
o
p
o
s
ed
wo
r
k
u
tili
ze
d
th
e
p
r
o
ce
s
s
ed
v
er
tical
GR
F
d
ata
f
iles
f
r
o
m
th
e
GaitR
ec
d
ata
s
et,
wh
ich
h
ad
alr
ea
d
y
u
n
d
er
g
o
n
e
d
e
n
o
is
in
g
,
s
eg
m
en
tatio
n
,
an
d
n
o
r
m
a
lizatio
n
as
p
ar
t
o
f
th
e
d
ataset's
o
f
f
icial
p
r
o
to
c
o
l.
T
h
e
d
ataset
c
o
n
tain
s
1
0
1
s
am
p
les
o
f
v
er
tical
GR
F
v
alu
es
f
o
r
ea
ch
p
atien
t,
ex
clu
s
iv
ely
r
ep
r
esen
tin
g
th
e
s
tan
ce
p
h
ase
o
f
th
e
g
ait
c
y
cle
[
2
0
]
.
I
t
was
o
b
s
er
v
ed
th
at
t
h
e
cl
ass
d
is
tr
ib
u
tio
n
o
f
th
e
d
atase
t
was
u
n
b
ala
n
ce
d
.
T
o
ad
d
r
ess
th
is
,
a
d
ata
b
ala
n
cin
g
p
r
o
ce
d
u
r
e
was
p
er
f
o
r
m
e
d
,
wh
ich
in
v
o
lv
ed
r
em
o
v
in
g
h
ea
lth
y
c
o
n
tr
o
l
d
ata
an
d
s
u
b
jects with
o
n
ly
a
s
in
g
l
e
r
ec
o
r
d
e
d
s
ess
io
n
.
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
S
ymm
etry in
d
ex
-
b
a
s
ed
g
a
it imp
r
o
ve
men
t p
r
ed
ictio
n
u
s
in
g
a
C
N
N
-
LS
TM
…
(
P
u
s
h
p
a
la
th
a
Ob
a
n
n
a
)
3631
T
h
e
B
o
r
d
er
li
n
e
s
y
n
th
etic
m
in
o
r
ity
o
v
er
s
am
p
lin
g
tech
n
i
q
u
e
(
SMOT
E
)
is
an
o
v
er
s
am
p
lin
g
tech
n
iq
u
e
f
o
cu
s
s
es
th
e
in
s
tan
ce
s
n
ea
r
th
e
d
ec
is
io
n
b
o
u
n
d
ar
ies
an
d
g
e
n
er
ates
th
e
s
y
n
th
etic
s
am
p
les
f
o
r
m
in
o
r
ity
class
es
to
b
alan
ce
th
e
class
d
is
tr
ib
u
tio
n
o
f
th
e
d
ataset.
T
h
e
m
in
o
r
ity
class
s
am
p
les th
at
ar
e
s
u
r
r
o
u
n
d
ed
b
y
th
e
m
ajo
r
ity
class
n
eig
h
b
o
r
s
th
at
ar
e
lik
e
ly
to
b
e
m
is
class
if
ied
ar
e
id
en
tifie
d
b
y
t
h
e
B
o
r
d
er
lin
e
S
MO
T
E
alg
o
r
ith
m
.
I
t
s
y
n
th
esizes
th
e
n
ew
s
am
p
l
es
b
y
in
co
r
p
o
r
atin
g
b
etwe
en
t
h
e
m
in
o
r
ity
s
am
p
les
an
d
n
eig
h
b
o
r
s
an
d
e
n
r
ich
es
th
e
d
ec
is
io
n
b
o
u
n
d
ar
y
.
T
h
e
s
y
n
th
etic
s
am
p
les
ar
e
g
e
n
er
ated
u
s
in
g
th
e
m
in
o
r
ity
s
am
p
le
an
d
o
n
e
o
f
t
h
e
k
-
n
ea
r
est
n
eig
h
b
o
r
s
is
s
elec
t
ed
f
r
o
m
t
h
e
m
in
o
r
ity
class
an
d
is
g
iv
en
b
y
(
2
8
)
.
W
h
er
e
λ
∈
[
0
,
1
]
r
ep
r
esen
ts
r
an
d
o
m
s
ca
le.
So
,
th
e
lo
ca
l
s
tr
u
ctu
r
e
o
f
t
h
e
m
in
o
r
ity
class
is
s
av
ed
an
d
th
e
n
ew
s
am
p
les
r
esid
es
alo
n
g
th
e
lin
e
s
eg
m
en
t b
etwe
en
an
d
.
=
+
.
(
−
)
(
2
8
)
3
.
3
.
3
.
I
m
plem
ent
a
t
i
o
n det
a
ils
All
co
m
p
u
tatio
n
al
ex
p
e
r
im
en
t
s
co
n
d
u
cte
d
u
s
in
g
Py
th
o
n
3
.
1
0
,
C
PU
o
f
2
.
4
GHz
,
T
e
n
s
o
r
Flo
w
o
f
2
.
1
5
with
Nu
m
Py
,
Pan
d
as.
T
h
e
ar
ch
itectu
r
e
was
im
p
lem
en
te
d
u
s
in
g
Ker
as
h
ig
h
lev
el
A
PIs.
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
u
s
ed
1
D
c
o
n
v
o
lu
t
io
n
al
lay
er
(
3
2
f
ilter
s
,
k
er
n
el
s
ize
=1
,
R
eL
U
ac
tiv
atio
n
)
an
d
L
STM
lay
er
with
3
2
u
n
it.
T
h
e
atten
tio
n
m
ec
h
a
n
is
m
is
u
s
ed
to
ass
ig
n
ad
ap
ti
v
e
weig
h
ts
to
v
a
r
io
u
s
tim
e
s
tep
s
an
d
allo
ws
th
e
m
o
d
el
to
co
n
ce
n
tr
ate
m
o
r
e
th
e
s
ess
io
n
s
th
at
co
n
tr
ib
u
tes
to
war
d
s
im
p
r
o
v
em
e
n
t.
T
h
e
o
u
tp
u
t
o
f
atten
tio
n
lay
e
r
is
f
latten
ed
a
n
d
p
ass
ed
th
r
o
u
g
h
d
e
n
s
e
lay
er
o
f
3
2
u
n
its
with
R
eL
U
ac
tiv
atio
n
.
A
d
r
o
p
o
u
t
lay
er
o
f
r
ate
0
.
4
is
u
s
ed
to
e
n
h
an
ce
th
e
g
en
er
aliza
tio
n
.
T
h
e
o
u
t
p
u
t
lay
e
r
s
ig
m
o
id
is
u
s
ed
f
o
r
b
i
n
ar
y
p
r
ed
ictio
n
wh
eth
er
th
e
p
atien
t
g
ait
is
im
p
r
o
v
e
d
o
r
n
o
t.
An
A
d
am
o
p
tim
izer
with
th
e
lear
n
i
n
g
r
ate
o
f
0
.
0
0
0
1
an
d
b
in
ar
y
c
r
o
s
s
en
tr
o
p
y
lo
s
s
is
u
s
ed
to
tr
ain
th
e
m
o
d
el
en
s
u
r
e
s
s
tab
le
co
n
v
er
g
en
ce
.
T
h
e
tr
ai
n
in
g
was p
er
f
o
r
m
ed
with
a
b
at
ch
s
ize
o
f
3
2
f
o
r
2
5
ep
o
ch
s
u
s
in
g
E
ar
l
y
s
to
p
p
in
g
with
p
atien
ce
o
f
1
0
t
o
av
o
i
d
o
v
er
f
itti
n
g
.
T
h
e
z
-
s
co
r
e
n
o
r
m
aliza
tio
n
is
u
s
ed
s
tan
d
ar
d
ize
all
th
e
f
ea
tu
r
es a
n
d
th
e
d
ata
s
p
litt
in
g
is
d
o
n
e
s
u
b
ject
wis
e
to
av
o
id
d
ata
leak
a
g
e
ac
r
o
s
s
s
ess
io
n
s
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
e
p
r
o
p
o
s
ed
s
y
s
tem
,
th
e
k
in
etic
f
ea
tu
r
es
ar
e
ex
tr
ac
te
d
f
r
o
m
th
e
GR
F
d
ata
f
o
r
b
o
th
le
f
t
an
d
r
ig
h
t
lim
b
f
o
r
m
u
ltip
le
s
ess
io
n
s
an
d
th
e
co
m
p
o
s
ite
ASI
is
co
m
p
u
t
ed
f
o
r
ea
ch
s
ess
io
n
.
T
h
e
p
atie
n
ts
with
at
least
two
s
ess
io
n
s
ar
e
co
n
s
id
er
ed
f
o
r
an
aly
s
is
.
T
h
e
ASI
d
ata
o
f
th
e
k
i
n
etic
f
ea
tu
r
es
ar
e
f
ed
t
o
th
e
p
r
o
p
o
s
ed
s
y
s
tem
.
T
h
e
m
o
d
el
p
er
f
o
r
m
s
lo
n
g
itu
d
in
al
tr
ac
k
in
g
o
f
3
%
r
ed
u
ctio
n
in
asy
m
m
etr
y
f
r
o
m
s
ess
io
n
to
s
ess
io
n
.
I
f
th
e
p
er
ce
n
tag
e
o
f
r
ed
u
ctio
n
in
th
e
asy
m
m
etr
y
is
g
r
ea
ter
th
an
o
r
e
q
u
al
to
3
0
% f
r
o
m
in
itial to
f
in
al
s
es
s
io
n
,
th
en
th
e
o
v
er
all
s
tatu
s
o
f
g
ait
is
co
n
s
id
er
ed
as
im
p
r
o
v
ed
o
r
n
o
t
im
p
r
o
v
ed
.
T
ab
le
1
s
h
o
ws
th
e
in
itial
an
d
f
in
al
s
ess
io
n
co
m
p
o
s
ite
ASI
v
alu
es,
th
e
p
er
ce
n
tag
e
o
f
r
ed
u
ctio
n
in
th
e
asy
m
m
etr
y
,
an
d
th
e
o
v
er
all
s
tatu
s
o
f
th
e
g
ait
wh
eth
er
it is
im
p
r
o
v
e
d
o
r
n
o
t i
m
p
r
o
v
e
d
.
T
ab
le
1
.
T
h
e
in
itial ses
s
io
n
an
d
f
in
al
s
ess
io
n
ASI
v
alu
es with
o
v
er
all
s
tatu
s
S
u
b
j
e
c
t
_
I
D
I
n
i
t
i
a
l
s
e
ssi
o
n
c
o
m
p
o
s
i
t
e
A
S
I
F
i
n
a
l
s
e
ssi
o
n
c
o
m
p
o
si
t
e
A
S
I
A
sy
mm
e
t
r
y
r
e
d
u
c
t
i
o
n
p
e
r
c
e
n
t
a
g
e
O
v
e
r
a
l
l
_
s
t
a
t
u
s
S
0
1
1
9
.
8
2
8
.
9
8
5
4
.
6
9
%
I
mp
r
o
v
e
d
S
0
2
3
8
.
4
3
2
2
.
3
9
4
1
.
7
2
%
I
mp
r
o
v
e
d
S
0
3
3
5
.
2
3
2
8
.
0
9
2
0
.
2
8
N
o
t
i
m
p
r
o
v
e
d
S
0
4
2
0
.
7
8
1
3
.
5
6
3
4
.
7
4
I
mp
r
o
v
e
d
S
0
5
3
3
.
0
2
2
2
.
5
7
3
1
.
6
3
I
mp
r
o
v
e
d
Fig
u
r
e
3
illu
s
tr
ates
th
e
r
elatio
n
s
h
ip
b
etwe
en
th
e
ac
tu
al
ASI
im
p
r
o
v
em
e
n
t
co
m
p
u
ted
f
r
o
m
p
atien
t
g
ait
d
ata
an
d
th
e
p
r
o
p
o
s
ed
m
o
d
el
-
p
r
ed
icted
ASI
im
p
r
o
v
e
m
en
t
b
etwe
en
th
e
in
itial
a
n
d
f
in
al
s
ess
io
n
o
f
r
eh
ab
ilit
atio
n
.
E
ac
h
p
o
in
t in
th
e
p
lo
t r
ep
r
esen
ts
o
n
e
r
eh
ab
ilit
atio
n
s
u
b
ject.
A
lin
ea
r
tr
en
d
is
o
b
s
er
v
ed
alo
n
g
th
e
d
iag
o
n
al
d
i
r
ec
tio
n
,
in
d
icatin
g
th
e
m
o
d
el
p
r
ed
ictio
n
s
clo
s
ely
m
atch
th
e
ac
tu
al
im
p
r
o
v
em
e
n
t
v
alu
es
,
d
em
o
n
s
tr
atin
g
s
tr
o
n
g
m
o
d
el
a
cc
u
r
ac
y
an
d
g
e
n
er
aliza
b
ilit
y
.
Fig
u
r
e
4
illu
s
tr
ates
th
e
m
o
d
el’
s
p
r
ed
ictiv
e
im
p
r
o
v
em
en
t
ac
r
o
s
s
th
e
f
ir
s
t
2
0
p
atien
ts
with
th
e
ac
tu
al
im
p
r
o
v
em
e
n
t
p
e
r
ce
n
tag
es
with
th
e
p
r
ed
icted
v
alu
es
d
er
i
v
ed
f
r
o
m
GR
F
-
b
ased
f
ea
tu
r
es.
T
h
e
g
r
o
u
n
d
tr
u
th
im
p
r
o
v
em
e
n
t
is
s
h
o
wn
b
y
th
e
b
lu
e
lin
e,
an
d
th
e
m
o
d
el’
s
o
u
tp
u
t
is
s
h
o
wn
b
y
th
e
o
r
a
n
g
e
lin
e.
T
h
e
clo
s
e
alig
n
m
en
t
o
f
two
c
u
r
v
es
s
h
o
ws
s
tr
o
n
g
p
r
ed
ictiv
e
f
id
elity
,
with
m
in
im
al
d
ev
iatio
n
in
t
h
e
ca
s
e
o
f
m
o
d
e
r
ate
im
p
r
o
v
em
e
n
t.
T
h
is
v
is
u
al
co
m
p
ar
is
o
n
s
u
p
p
o
r
ts
th
e
m
o
d
e
l’
s
clin
ical
ap
p
licab
ilit
y
f
o
r
in
d
iv
id
u
alize
d
g
ait
r
ec
o
v
er
y
ass
ess
m
en
t.
Fig
u
r
e
5
s
h
o
ws
th
e
co
n
f
u
s
i
o
n
m
atr
ix
th
at
illu
s
tr
ates
th
e
m
o
d
el’
s
class
if
icatio
n
p
er
f
o
r
m
an
ce
in
p
r
ed
ictin
g
th
e
s
tatu
s
o
f
th
e
r
eh
ab
ilit
atio
n
im
p
r
o
v
e
m
en
t
a
m
o
n
g
lo
wer
lim
b
f
r
ac
tu
r
e
p
atien
ts
.
T
h
e
m
o
d
el
co
r
r
ec
tly
id
en
tifie
s
2
4
2
p
atien
ts
as
im
p
r
o
v
ed
an
d
2
9
4
p
atie
n
ts
as
n
o
t
im
p
r
o
v
ed
.
T
h
e
m
is
class
if
icatio
n
r
ate
i
s
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.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
6
2
5
-
3
6
3
6
3632
v
er
y
lo
w;
1
0
p
atien
ts
ar
e
id
en
tifie
d
as
f
alse
n
eg
ativ
es
an
d
6
p
atien
ts
as
f
alse
p
o
s
itiv
es.
T
h
e
m
o
d
el
ca
p
tu
r
e
s
g
ait
r
ec
o
v
er
y
p
atter
n
s
with
h
ig
h
ac
cu
r
ac
y
an
d
m
ain
tain
s
b
ala
n
ce
d
s
en
s
itiv
ity
an
d
s
p
ec
if
icit
y
.
Fig
u
r
e
3
.
Pre
d
icte
d
ASI
im
p
r
o
v
em
en
t v
er
s
u
s
ac
tu
al
ASI
im
p
r
o
v
em
en
t
Fig
u
r
e
4
.
Mo
d
el
p
r
e
d
ictiv
e
im
p
r
o
v
em
e
n
t a
cr
o
s
s
th
e
f
ir
s
t 2
0
p
atien
ts
Fig
u
r
e
5
.
C
o
n
f
u
s
io
n
m
atr
i
x
s
h
o
win
g
m
o
d
el
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
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
S
ymm
etry in
d
ex
-
b
a
s
ed
g
a
it imp
r
o
ve
men
t p
r
ed
ictio
n
u
s
in
g
a
C
N
N
-
LS
TM
…
(
P
u
s
h
p
a
la
th
a
Ob
a
n
n
a
)
3633
T
h
e
m
o
d
el
’
s
class
if
icatio
n
m
etr
ics o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
ar
e
co
m
p
u
te
d
u
s
in
g
(
2
9
)
to
(
3
2
)
.
=
+
+
+
+
(
2
9
)
=
+
(
3
0
)
=
+
(
3
1
)
1
−
=
2
×
+
(
3
2
)
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
r
eg
r
ess
io
n
m
etr
ics ar
e
co
m
p
u
ted
u
s
in
g
th
e
(
3
3
)
to
(
3
5
)
.
=
√
1
∑
(
−
̂
)
2
=
1
(
3
3
)
=
1
∑
|
−
̂
|
=
1
(
3
4
)
W
h
er
e
is
th
e
ac
tu
al
tar
g
et
v
al
u
e,
̂
is
th
e
m
o
d
el
p
r
e
d
icted
v
al
u
e
,
an
d
N
is
th
e
n
u
m
b
er
o
f
test
s
am
p
les.
2
=
1
−
∑
(
−
̂
)
2
=
1
∑
(
−
̅
̅
̅
)
2
=
1
(
3
5
)
W
h
er
e
∑
(
−
̂
)
2
=
1
is
th
e
r
esid
u
al
s
u
m
o
f
s
q
u
ar
es,
∑
(
−
̅
)
2
=
1
is
th
e
to
tal
v
ar
ian
ce
an
d
̅
is
th
e
m
ea
n
o
f
th
e
to
tal
ac
t
u
al
v
al
u
es.
T
h
e
2
=
1
r
ep
r
esen
ts
p
e
r
f
ec
t
p
r
ed
ictio
n
,
2
=
0
in
d
icate
s
th
e
m
o
d
el
is
n
o
b
etter
th
an
p
r
e
d
ictin
g
th
e
m
ea
n
an
d
2
<
0
r
ep
r
esen
ts
is
wo
r
s
e
th
an
p
r
e
d
ictin
g
th
e
m
ea
n
.
Fig
u
r
e
6
s
h
o
ws
th
e
r
e
g
r
ess
io
n
m
o
d
el
lear
n
in
g
b
e
h
av
io
r
s
o
f
tr
ain
in
g
a
n
d
v
alid
atio
n
lo
s
s
in
F
ig
u
r
e
6
(
a)
an
d
m
ea
n
ab
s
o
lu
te
er
r
o
r
(
MA
E
)
cu
r
v
es
in
F
ig
u
r
e
6
(
b
)
ac
r
o
s
s
2
5
ep
o
ch
s
.
Her
e
,
b
o
th
th
e
l
o
s
s
an
d
MA
E
p
lo
ts
ar
e
d
ec
r
ea
s
in
g
m
o
n
o
to
n
ically
with
th
e
in
cr
ea
s
in
g
e
p
o
ch
s
,
in
d
icatin
g
s
tab
le
an
d
ef
f
e
ctiv
e
lear
n
in
g
.
T
h
e
v
alid
atio
n
cu
r
v
es
clo
s
ely
f
o
ll
o
w
an
d
r
em
ain
b
elo
w
th
e
tr
a
in
in
g
cu
r
v
es,
d
em
o
n
s
tr
atin
g
g
o
o
d
g
e
n
er
aliza
tio
n
an
d
ab
s
en
ce
o
f
o
v
er
f
itti
n
g
.
T
h
e
s
m
o
o
th
co
n
v
er
g
e
n
ce
o
f
b
o
th
m
etr
ics
co
n
f
ir
m
s
th
at
th
e
m
o
d
el
lear
n
s
d
is
cr
im
in
ativ
e
p
atter
n
s
f
r
o
m
t
h
e
d
ata
an
d
ac
h
iev
es
r
eliab
le
p
r
ed
ictiv
e
p
er
f
o
r
m
an
ce
.
T
h
e
r
eg
r
ess
io
n
m
etr
ics
lik
e
r
o
o
t
m
ea
n
s
q
u
ar
e
er
r
o
r
(
R
MSE
)
o
f
1
.
0
9
,
MA
E
o
f
0
.
6
4
,
an
d
R
2
s
co
r
e
o
f
0
.
9
8
ar
e
o
b
tain
ed
f
r
o
m
th
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
class
if
icatio
n
m
etr
ics
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
,
lik
e
ac
cu
r
ac
y
o
f
9
7
.
1
%,
r
ec
all
o
f
0
.
9
6
0
,
p
r
ec
is
io
n
o
f
0
.
9
7
6
,
a
n
d
F1
-
s
co
r
e
o
f
0
.
9
6
8
,
ar
e
o
b
tain
ed
.
(
a)
(
b
)
Fig
u
r
e
6
.
L
ea
r
n
in
g
cu
r
v
es o
f
t
h
e
r
eg
r
ess
io
n
m
o
d
el
of
(
a)
l
o
s
s
an
d
(
b
)
MA
E
Fig
u
r
e
7
illu
s
tr
ates th
e
lo
n
g
itu
d
in
al
p
r
o
g
r
ess
io
n
o
f
t
h
e
co
m
p
o
s
ite
ASI
f
o
r
s
u
b
ject
S0
4
ac
r
o
s
s
m
u
ltip
le
g
ait
s
ess
io
n
s
.
T
h
e
b
lu
e
m
ar
k
e
r
s
s
h
o
w
s
ess
io
n
-
wis
e
ASI
ch
an
g
es,
in
d
icatin
g
th
e
p
atien
t’
s
r
ec
o
v
er
y
tr
ajec
to
r
y
o
v
er
tim
e.
T
h
e
g
r
ee
n
p
o
in
ts
h
ig
h
lig
h
t
s
ess
io
n
s
wh
er
e
a
≥
3
%
r
ed
u
cti
o
n
i
n
ASI
f
r
o
m
t
h
e
p
r
e
v
io
u
s
s
ess
io
n
r
ef
lects
a
m
ea
s
u
r
ab
le
im
p
r
o
v
e
m
en
t
in
g
ait
s
y
m
m
etr
y
,
wh
er
e
as
th
e
r
ed
d
ash
ed
lin
e
r
ep
r
esen
ts
th
e
3
%
clin
ical
th
r
esh
o
ld
u
s
ed
to
id
en
tif
y
m
e
an
in
g
f
u
l
p
r
o
g
r
ess
.
Ov
er
all,
th
e
d
o
wn
war
d
tr
en
d
in
ASI
v
alu
es
d
em
o
n
s
tr
ates
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.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
6
2
5
-
3
6
3
6
3634
co
n
s
is
ten
t
r
ed
u
ctio
n
in
asy
m
m
etr
y
,
s
ig
n
if
y
in
g
f
u
n
ctio
n
al
r
ec
o
v
er
y
an
d
en
h
a
n
ce
d
g
ait
s
y
m
m
etr
y
th
r
o
u
g
h
o
u
t
th
e
r
eh
ab
ilit
atio
n
p
er
io
d
.
T
h
e
wea
r
ab
le
s
en
s
o
r
s
ar
e
in
teg
r
ated
with
th
e
ML
to
p
r
ed
ict
f
r
ac
tu
r
e
h
ea
lin
g
.
T
h
e
co
n
tin
u
o
u
s
u
n
d
er
f
o
o
t
d
ata
was
p
r
o
ce
s
s
ed
,
an
d
th
e
f
ea
tu
r
es
ar
e
e
x
tr
ac
ted
u
s
in
g
l
o
g
ic
r
e
g
r
ess
io
n
class
if
ier
.
T
h
e
class
if
icatio
n
m
etr
ics
ar
e
o
b
tai
n
ed
f
r
o
m
th
e
m
o
d
el
[
2
1
]
.
T
h
e
d
ep
th
v
i
d
eo
s
an
d
p
r
ess
u
r
e
d
at
a
ar
e
u
s
ed
f
r
o
m
th
e
o
n
lin
e
d
ataset
th
at
co
m
p
r
is
es
3
0
h
ea
lth
y
s
u
b
jects
p
er
f
o
r
m
in
g
s
ev
en
r
eh
ab
ilit
atio
n
ex
er
cis
es
u
s
in
g
a
3
D
C
N
N
class
if
ier
an
d
ac
h
iev
e
d
a
class
if
icatio
n
ac
cu
r
ac
y
o
f
9
5
.
7
1
%.
On
ly
th
e
class
if
icatio
n
m
etr
ics
ar
e
ex
t
r
ac
ted
f
r
o
m
th
e
m
o
d
el
[
2
2
]
.
T
h
e
r
e
h
ab
ilit
atio
n
o
u
tco
m
es
th
at
ar
e
ess
en
tial
f
o
r
th
e
clin
ical
d
ec
is
io
n
a
r
e
p
r
e
d
icted
u
s
in
g
C
NN
an
d
an
R
NN
cla
s
s
if
ier
u
s
in
g
clin
ician
-
r
ep
o
r
ted
o
u
tco
m
e
m
ea
s
u
r
es
(
C
R
OM
)
an
d
P
R
OM
.
T
h
e
r
eg
r
ess
io
n
m
etr
ics
an
d
class
if
icat
io
n
m
etr
ics
ar
e
b
o
th
ex
tr
ac
ted
f
r
o
m
t
h
e
m
o
d
el.
T
h
e
F1
-
s
co
r
e
is
im
p
r
o
v
e
d
b
y
9
%
an
d
1
2
%
r
ed
u
ctio
n
is
MA
E
co
m
p
ar
ed
to
tr
ad
itio
n
al
ML
m
eth
o
d
s
[
2
3
]
.
I
n
th
e
lo
n
g
itu
d
in
al
s
t
u
d
y
,
th
e
d
ata
wer
e
co
llected
f
r
o
m
t
h
e
GR
F
cu
r
v
e
o
f
1
3
tib
ial
f
r
ac
tu
r
e
p
atien
t
s
.
T
h
e
c
o
m
p
o
s
ite
ASI
is
u
s
ed
as
a
ta
r
g
et
f
o
r
th
e
m
o
d
el.
T
h
e
m
o
tio
n
ca
p
tu
r
e
an
d
f
o
r
ce
p
late
ar
e
u
s
ed
to
r
ec
o
r
d
th
e
d
ata
an
d
th
e
r
eg
r
ess
io
n
m
etr
ics
s
u
ch
as
R
M
SE,
MA
E
an
d
R
2
ar
e
ex
tr
ac
ted
[
2
4
]
.
A
x
L
STM
an
d
L
ag
-
L
lam
a
m
o
d
el
ar
ch
itectu
r
e
is
u
s
ed
to
tem
p
o
r
al
s
er
ies
an
aly
s
is
o
f
lo
wer
lim
b
r
eh
ab
ilit
atio
n
f
o
r
th
e
p
o
s
t
s
tr
o
k
e
p
atien
ts
in
th
e
lim
b
r
ec
o
v
er
y
,
an
d
ex
tr
ac
ted
th
e
R
MSE
an
d
MA
E
f
o
r
d
y
n
am
ic
n
eu
r
o
m
u
s
cu
lar
s
tab
ilizatio
n
(
DNS)
g
ait
an
d
u
p
p
er
-
lim
b
s
u
p
p
o
r
ted
(
UPS)
g
ait.
T
h
e
R
2
o
f
0
.
7
2
is
ac
h
iev
ed
f
r
o
m
th
is
m
o
d
el
[
2
5
]
.
T
ab
le
2
co
m
p
ar
es
th
e
p
er
f
o
r
m
a
n
ce
m
etr
ics
o
f
C
NN
-
L
STM
-
atten
tio
n
h
y
b
r
id
m
o
d
el
with
t
h
e
ex
i
s
tin
g
tr
ad
itio
n
al
a
r
ch
itectu
r
es.
T
ab
le
3
s
h
o
ws
th
e
5
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
r
esu
lt
o
f
o
u
r
m
o
d
el.
T
h
e
m
ea
n
s
o
f
R
MSE
,
MA
E
,
an
d
R
2
o
b
tai
n
ed
f
r
o
m
th
is
5
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
ar
e
1
.
4
3
9
3
,
0
.
8
8
1
9
,
an
d
0
.
9
7
7
3
a
n
d
a
r
e
v
er
y
clo
s
e
to
th
e
s
in
g
le
-
tim
e
ex
ec
u
tio
n
r
esu
lt.
Fig
u
r
e
7
.
L
o
n
g
itu
d
in
al
tr
en
d
s
o
f
co
m
p
o
s
ite
ASI
f
o
r
s
u
b
ject
S0
4
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
o
f
p
r
o
p
o
s
ed
wo
r
k
with
ex
is
tin
g
m
o
d
els
R
e
f
e
r
e
n
c
e
M
o
d
e
l
t
y
p
e
Ta
r
g
e
t
R
M
S
E
M
A
E
R
2
A
c
c
u
r
a
c
y
(
%)
F1
-
s
c
o
r
e
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
P
r
o
p
o
se
d
w
o
r
k
C
N
N
+
LST
M
+
a
t
t
e
n
t
i
o
n
v
G
R
F
1
.
0
9
0
.
6
4
0
.
9
8
9
7
.
1
9
6
.
8
9
7
.
6
96
[
2
1
]
L
o
g
i
s
t
i
c
r
e
g
r
e
ssi
o
n
G
R
F
f
e
a
t
u
r
e
s
-
-
-
7
5
.
6
7
5
.
4
7
6
.
1
7
5
.
6
[
2
2
]
3
D
C
N
N
P
r
e
ssu
r
e
d
a
t
a
-
-
-
9
5
.
7
1
9
5
.
7
4
9
5
.
8
3
9
5
.
7
1
[
2
3
]
H
y
b
r
i
d
C
N
N
+
R
N
N
P
R
O
M
a
n
d
C
R
O
M
-
4
.
2
5
0
.
8
7
8
4
.
2
86
88
85
[
2
4
]
M
L
(
f
o
r
c
e
P
l
a
t
e
)
C
o
m
p
o
si
t
e
A
S
I
4
.
7
8
3
.
5
6
0
.
7
2
-
-
-
-
[
2
5
]
x
LST
M
a
n
d
La
g
-
La
mm
a
D
N
S
g
a
i
t
1
0
.
5
4
6
.
2
8
0
.
7
2
-
-
-
-
U
P
S
g
a
i
t
1
2
.
2
8
7
.
5
4
-
-
-
-
T
ab
le
3
.
Fiv
e
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
r
esu
lt
F
o
l
d
R
M
S
E
M
A
E
R
2
1
1
.
3
6
0
3
5
5
0
.
8
3
1
0
9
7
0
.
9
7
6
2
2
9
2
1
.
4
4
7
6
7
9
0
.
8
6
4
3
1
3
0
.
9
8
0
1
4
8
3
1
.
4
5
8
9
4
3
0
.
9
2
7
9
0
4
0
.
9
7
8
9
0
6
4
1
.
3
5
9
3
3
5
0
.
8
4
9
5
5
3
0
.
9
7
8
0
5
5
1
.
5
7
0
6
6
5
0
.
9
3
7
0
4
0
.
9
7
3
5
9
9
Evaluation Warning : The document was created with Spire.PDF for Python.