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
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
5
9
1
~
3
6
0
2
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
5
9
1
-
3
6
0
2
3591
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
YO
L
O
-
ba
sed
dee
p learning
f
o
r
t
o
o
th
de
tect
io
n
,
seg
menta
tion,
a
nd numbering
in
pano
ra
mic rad
io
g
ra
phs
Sri O
k
t
a
m
uli
a
ni
1
,
L
uq
y
a
na
M
a
hd
iy
a
h
1
,
H
a
rit
s
ul H
a
q
1
,
Nesa
P
er
da
na
P
utr
i
2
,
Wula
n
da
ni L
iza
P
utr
i
3
1
D
e
p
a
r
t
me
n
t
o
f
P
h
y
si
c
s,
F
a
c
u
l
t
y
o
f
M
a
t
h
e
m
a
t
i
c
s
a
n
d
N
a
t
u
r
a
l
S
c
i
e
n
c
e
s
,
U
n
i
v
e
r
si
t
a
s
A
n
d
a
l
a
s
,
P
a
d
a
n
g
,
I
n
d
o
n
e
s
i
a
2
D
i
v
i
si
o
n
o
f
D
e
n
t
i
st
r
y
,
D
e
n
t
a
l
C
l
i
n
i
c
,
D
r
.
M
u
h
a
mm
a
d
Ze
i
n
R
e
g
i
o
n
a
l
G
e
n
e
r
a
l
H
o
sp
i
t
a
l
,
P
a
i
n
a
n
,
I
n
d
o
n
e
s
i
a
3
D
e
p
a
r
t
me
n
t
o
f
O
r
t
h
o
d
o
n
t
i
c
s
,
F
a
c
u
l
t
y
o
f
D
e
n
t
i
st
r
y
,
U
n
i
v
e
r
s
i
t
a
s A
n
d
a
l
a
s,
P
a
d
a
n
g
,
I
n
d
o
n
e
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
1
4
,
2
0
2
5
R
ev
is
ed
Ma
y
1
3
,
2
0
2
6
Acc
ep
ted
Ma
y
2
4
,
2
0
2
6
P
a
n
o
ra
m
ic
d
e
n
tal
ra
d
io
g
ra
p
h
s
a
re
e
ss
e
n
ti
a
l
fo
r
d
iag
n
o
sis
a
n
d
trea
tme
n
t
p
lan
n
in
g
,
b
u
t
a
re
d
iff
icu
lt
t
o
in
terp
re
t
d
u
e
t
o
o
v
e
rlap
p
i
n
g
a
n
a
to
m
ica
l
stru
c
tu
re
s
a
n
d
li
m
it
e
d
ima
g
e
q
u
a
li
ty
.
Au
t
o
m
a
ti
n
g
to
o
th
d
e
tec
ti
o
n
,
se
g
m
e
n
tatio
n
,
a
n
d
n
u
m
b
e
rin
g
c
a
n
e
n
h
a
n
c
e
c
li
n
ica
l
e
fficie
n
c
y
a
n
d
re
d
u
c
e
o
b
se
rv
e
r
v
a
riab
il
it
y
.
Th
is
stu
d
y
d
e
v
e
lo
p
e
d
a
y
o
u
o
n
l
y
l
o
o
k
o
n
c
e
v
e
rsio
n
8
(
YO
LOv
8
)
-
b
a
se
d
d
e
e
p
lea
rn
i
n
g
m
o
d
e
l
fo
r
a
u
to
m
a
ti
c
d
e
tec
ti
o
n
a
n
d
se
g
m
e
n
tatio
n
o
f
tee
th
in
p
a
n
o
ra
m
ic
ra
d
io
g
ra
p
h
s.
A
d
a
tas
e
t
o
f
3
0
2
ima
g
e
s
c
o
n
tain
i
n
g
9
,
0
0
9
a
n
n
o
tate
d
tee
t
h
wa
s
d
iv
i
d
e
d
in
to
7
0
%
train
i
n
g
,
2
0
%
v
a
li
d
a
ti
o
n
,
a
n
d
1
0
%
tes
ti
n
g
se
ts.
Tee
th
we
re
lab
e
led
u
si
n
g
t
h
e
F
e
d
e
ra
ti
o
n
De
n
taire
In
tern
a
ti
o
n
a
le
(
F
DI
)
two
-
d
i
g
it
n
u
m
b
e
rin
g
sy
ste
m
with
b
o
u
n
d
i
n
g
b
o
x
e
s
a
n
d
se
g
m
e
n
tati
o
n
m
a
sk
s.
M
o
d
e
l
p
e
rfo
rm
a
n
c
e
wa
s
a
ss
e
s
se
d
u
sin
g
p
re
c
isio
n
,
re
c
a
ll
,
F
1
-
sc
o
re
,
a
n
d
m
e
a
n
a
v
e
ra
g
e
p
re
c
isio
n
(m
AP)
a
t
in
ters
e
c
ti
o
n
o
v
e
r
u
n
io
n
(
Io
U
)
t
h
re
sh
o
l
d
s
o
f
0
.
5
a
n
d
0
.
5
–
0
.
9
5
.
T
h
e
m
o
d
e
l
a
c
h
iev
e
d
b
o
u
n
d
in
g
b
o
x
p
re
c
isio
n
,
re
c
a
ll
,
a
n
d
F
1
-
sc
o
re
o
f
0
.
9
0
7
5
,
0
.
9
3
5
2
,
a
n
d
0
.
9
2
1
2
,
re
sp
e
c
ti
v
e
ly
,
a
n
d
se
g
m
e
n
tati
o
n
sc
o
re
s
o
f
0
.
9
0
7
8
,
0
.
9
3
4
4
,
a
n
d
0
.
9
2
0
9
.
Bo
u
n
d
i
n
g
b
o
x
m
AP@
0
.
5
re
a
c
h
e
d
0
.
9
5
1
0
a
n
d
m
a
sk
m
AP@0
.
5
w
a
s
0
.
9
5
0
4
,
wh
il
e
stricte
r
th
re
sh
o
ld
s
l
o
we
re
d
p
e
rfo
rm
a
n
c
e
to
0
.
7
7
6
3
a
n
d
0
.
6
8
9
6
.
Clas
s
-
wise
a
n
a
ly
sis
sh
o
we
d
h
i
g
h
a
c
c
u
ra
c
y
o
v
e
ra
ll
b
u
t
re
d
u
c
e
d
p
e
rfo
r
m
a
n
c
e
o
n
p
o
ste
rio
r
tee
t
h
d
u
e
t
o
a
n
a
to
m
ic
a
l
o
v
e
rlap
a
n
d
p
e
rip
h
e
ra
l
ima
g
e
q
u
a
li
ty
.
YO
LOv
8
e
n
a
b
les
fa
st,
a
c
c
u
r
a
te,
a
n
d
ro
b
u
st
to
o
t
h
d
e
tec
ti
o
n
a
n
d
se
g
m
e
n
tatio
n
,
su
p
p
o
r
ti
n
g
re
a
l
-
ti
m
e
c
o
m
p
u
ter
-
a
id
e
d
d
iag
n
o
sis
i
n
o
rth
o
d
o
n
ti
c
s,
p
r
o
sth
o
d
o
n
ti
c
s,
a
n
d
fo
re
n
sic
d
e
n
ti
stry
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
Den
tal
im
ag
in
g
Pan
o
r
am
ic
r
ad
i
o
g
r
ap
h
Seg
m
en
tatio
n
T
o
o
th
d
etec
tio
n
T
o
o
th
n
u
m
b
e
r
in
g
Y
o
u
o
n
ly
lo
o
k
o
n
ce
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
:
Sri
Ok
tam
u
lian
i
Dep
ar
tm
en
t o
f
Ph
y
s
ics,
Facu
lty
o
f
Ma
th
e
m
atics a
n
d
Natu
r
al
Scien
ce
s
,
Un
iv
er
s
itas
An
d
alas
L
im
au
Ma
n
is
C
am
p
u
s
,
Pad
an
g
2
5
1
6
3
,
I
n
d
o
n
esia
E
m
ail: sr
io
k
tam
u
lian
i@
s
ci.
u
n
an
d
.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
Pan
o
r
am
ic
r
ad
io
g
r
ap
h
y
is
an
e
x
tr
ao
r
al
im
ag
in
g
tech
n
i
q
u
e
th
at
ca
p
tu
r
es
a
co
m
p
r
eh
en
s
iv
e
v
iew
o
f
th
e
m
ax
illo
f
ac
ial
r
eg
io
n
,
in
clu
d
in
g
th
e
m
an
d
ib
le,
m
a
x
illa,
an
d
th
eir
s
u
p
p
o
r
tin
g
s
tr
u
ct
u
r
es.
I
t
is
wid
ely
u
s
ed
in
d
en
tis
tr
y
b
ec
au
s
e
it
is
s
im
p
le
to
p
er
f
o
r
m
,
p
r
o
v
id
es
co
v
er
ag
e
o
f
all
teeth
an
d
jaws,
an
d
d
eliv
er
s
a
r
elativ
ely
lo
w
r
ad
iatio
n
d
o
s
e.
C
lin
icall
y
,
p
an
o
r
am
ic
r
ad
io
g
r
ap
h
s
ar
e
em
p
lo
y
ed
to
d
iag
n
o
s
e
jaw
f
r
ac
tu
r
es,
ev
alu
ate
tem
p
o
r
o
m
a
n
d
ib
u
lar
jo
in
t
(
T
MJ)
s
y
m
m
etr
y
,
ex
am
in
e
to
o
t
h
r
o
o
t
m
o
r
p
h
o
l
o
g
y
,
d
etec
t
cy
s
ts
an
d
lesi
o
n
s
,
an
d
ass
es
s
d
en
titi
o
n
d
u
r
in
g
g
r
o
wth
an
d
d
ev
elo
p
m
en
t
[
1
]
–
[
3
]
.
Alm
o
s
t
all
d
en
tal
tr
ea
tm
en
ts
r
eq
u
ir
e
co
m
p
lem
en
ta
r
y
r
ad
io
g
r
a
p
h
ic
ex
a
m
in
atio
n
s
to
s
u
p
p
o
r
t a
cc
u
r
ate
d
iag
n
o
s
is
an
d
o
p
tim
al
tr
ea
tm
e
n
t p
lan
n
i
n
g
[
4
]
,
[
5
]
.
Desp
ite
th
ese
ad
v
an
tag
es,
p
an
o
r
am
ic
r
ad
io
g
r
ap
h
s
s
u
f
f
er
f
r
o
m
in
h
er
e
n
t
lim
itatio
n
s
r
elate
d
to
th
e
p
h
y
s
ics
o
f
im
ag
in
g
.
C
o
m
p
ar
e
d
with
i
n
tr
ao
r
al
r
ad
io
g
r
ap
h
y
,
p
an
o
r
am
ic
im
ag
es
h
av
e
lo
wer
s
p
atial
r
eso
lu
tio
n
,
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
9
1
-
3
6
0
2
3592
ar
e
s
u
b
ject
to
g
e
o
m
etr
ic
d
is
to
r
tio
n
,
an
d
o
f
ten
d
is
p
lay
o
v
er
lap
p
in
g
an
at
o
m
ical
s
tr
u
ctu
r
es
[
6
]
–
[
8
]
.
T
h
ese
f
ac
to
r
s
m
ak
e
in
ter
p
r
etatio
n
co
m
p
lex
an
d
tim
e
-
c
o
n
s
u
m
in
g
,
a
n
d
th
ey
m
ay
co
n
t
r
ib
u
te
to
v
ar
iab
ilit
y
b
etwe
en
clin
ician
s
.
As
a
r
esu
lt,
th
er
e
h
as
b
ee
n
g
r
o
win
g
in
ter
est
in
d
ev
elo
p
in
g
co
m
p
u
ter
-
aid
ed
d
iag
n
o
s
ti
c
s
y
s
tem
s
th
at
ca
n
au
to
m
atica
lly
in
ter
p
r
et
p
an
o
r
a
m
ic
im
ag
es a
n
d
s
u
p
p
o
r
t
d
ec
is
io
n
-
m
ak
i
n
g
in
d
en
tal
p
r
ac
tice
[
9
]
–
[
1
1
]
.
Ar
tific
ial
in
tellig
en
ce
(
AI
)
te
ch
n
o
lo
g
y
h
as
b
r
o
u
g
h
t
s
ig
n
if
i
ca
n
t
ad
v
a
n
ce
s
in
t
h
e
f
ield
o
f
m
ed
ical
im
ag
e
an
aly
s
is
.
AI
ca
n
h
elp
ac
ce
ler
ate
an
d
en
h
an
ce
th
e
im
ag
e
id
en
tific
atio
n
an
d
an
al
y
s
is
p
r
o
ce
s
s
.
Ma
ch
in
e
lear
n
in
g
is
a
b
r
an
c
h
o
f
s
cien
c
e
th
at
s
tu
d
ies
alg
o
r
ith
m
s
an
d
s
tatis
t
ical
m
o
d
els,
en
ab
lin
g
c
o
m
p
u
ter
s
y
s
tem
s
to
p
er
f
o
r
m
s
p
ec
if
ic
task
s
with
o
u
t
r
e
q
u
ir
in
g
ex
p
licit
in
s
tr
u
ctio
n
s
.
T
h
is
tech
n
o
lo
g
y
r
elies
o
n
p
atter
n
s
an
d
co
n
clu
s
io
n
s
d
e
r
iv
ed
f
r
o
m
d
ata
.
T
o
o
b
tain
th
ese
p
atter
n
s
a
n
d
co
n
clu
s
io
n
s
,
m
ac
h
i
n
e
lear
n
in
g
alg
o
r
ith
m
s
b
u
ild
m
ath
em
atica
l
m
o
d
els
b
ased
o
n
s
am
p
le
d
ata
k
n
o
wn
as
tr
ain
in
g
d
ata
[
1
2
]
.
Ma
c
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
ar
e
d
iv
id
ed
in
to
th
r
ee
ty
p
es.
First,
s
u
p
er
v
is
ed
lear
n
in
g
is
a
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
e
in
wh
ich
a
m
o
d
el
is
tr
ain
ed
to
m
a
k
e
p
r
ed
ictio
n
s
b
ased
o
n
lab
eled
tr
ain
in
g
d
ata.
Seco
n
d
,
u
n
s
u
p
er
v
is
ed
lear
n
i
n
g
s
tu
d
ies
p
atter
n
s
an
d
r
elatio
n
s
h
ip
s
in
d
ata
wit
h
o
u
t
lab
els.
T
h
ir
d
,
r
ein
f
o
r
ce
m
en
t
lear
n
in
g
is
a
m
ac
h
in
e
lear
n
in
g
m
o
d
el
th
at
m
ak
es d
ec
is
io
n
s
th
r
o
u
g
h
tr
ial
an
d
er
r
o
r
.
R
ec
en
t
ad
v
an
ce
s
i
n
AI
h
av
e
l
ed
to
s
ig
n
if
ican
t
p
r
o
g
r
ess
in
m
ed
ical
im
ag
e
an
aly
s
is
.
Ma
ch
in
e
lear
n
in
g
en
ab
les
co
m
p
u
ter
s
to
lear
n
f
r
o
m
d
ata
an
d
m
ak
e
p
r
ed
ictio
n
s
with
o
u
t
b
ein
g
ex
p
licitly
p
r
o
g
r
am
m
ed
[
1
3
]
–
[
1
5
]
.
Ma
ch
in
e
lear
n
in
g
tec
h
n
iq
u
e
s
ar
e
ty
p
ically
ca
teg
o
r
ized
as
s
u
p
er
v
is
ed
,
u
n
s
u
p
e
r
v
is
ed
,
o
r
r
ein
f
o
r
ce
m
en
t
lear
n
in
g
,
d
e
p
en
d
in
g
o
n
th
e
ty
p
e
o
f
f
ee
d
b
ac
k
p
r
o
v
id
ed
d
u
r
i
n
g
tr
ain
in
g
[
1
6
]
–
[
1
8
]
.
Am
o
n
g
th
ese
ap
p
r
o
ac
h
es,
d
ee
p
lear
n
in
g
,
b
ased
o
n
ar
tifi
cial
n
eu
r
al
n
etwo
r
k
s
h
as
b
ec
o
m
e
a
d
o
m
in
a
n
t
p
ar
ad
i
g
m
b
ec
au
s
e
o
f
its
ab
ilit
y
to
m
o
d
el
co
m
p
lex
,
n
o
n
li
n
ea
r
p
atter
n
s
in
lar
g
e
d
atasets
[
1
6
]
,
[
1
9
]
.
Dee
p
lear
n
in
g
m
eth
o
d
s
h
av
e
ac
h
iev
ed
s
tate
-
of
-
th
e
-
ar
t
r
esu
lts
in
task
s
s
u
ch
as
im
ag
e
class
if
icatio
n
,
s
eg
m
en
tatio
n
,
an
d
o
b
ject
d
ete
ctio
n
,
m
ak
in
g
th
em
p
ar
ticu
lar
ly
s
u
itab
le
f
o
r
m
e
d
ical
im
ag
in
g
ap
p
licatio
n
s
.
I
m
ag
e
s
eg
m
en
tatio
n
,
th
e
p
r
o
c
ess
o
f
d
elin
ea
tin
g
r
eg
i
o
n
s
o
f
i
n
ter
est
with
in
an
im
ag
e
is
a
k
ey
s
tep
in
d
en
tal
im
ag
e
an
aly
s
is
.
Ma
n
u
al
s
eg
m
en
tatio
n
,
wh
ile
ac
cu
r
ate,
is
lab
o
r
-
in
ten
s
iv
e,
r
eq
u
ir
es
ex
p
er
t
k
n
o
wled
g
e,
an
d
is
p
r
o
n
e
to
in
te
r
-
an
d
i
n
tr
a
-
o
p
er
ato
r
v
ar
iab
ilit
y
.
Au
to
m
at
ic
s
eg
m
en
tatio
n
m
eth
o
d
s
h
a
v
e
b
ee
n
d
ev
elo
p
ed
to
o
v
er
co
m
e
th
ese
c
h
allen
g
es
a
n
d
im
p
r
o
v
e
r
e
p
r
o
d
u
cib
ilit
y
.
Fo
r
ex
am
p
le,
s
ev
er
al
s
tu
d
ies
h
av
e
s
u
cc
ess
f
u
lly
s
eg
m
en
ted
teeth
an
d
jaw
s
tr
u
ctu
r
es
in
p
an
o
r
am
ic
r
a
d
io
g
r
a
p
h
s
u
s
in
g
tr
a
d
itio
n
al
im
a
g
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
s
u
ch
as
th
r
esh
o
ld
in
g
an
d
m
o
r
p
h
o
lo
g
ical
o
p
er
atio
n
s
[
2
0
]
–
[
2
3
]
.
Ho
wev
e
r
,
th
ese
ap
p
r
o
ac
h
es
ar
e
o
f
te
n
tim
e
-
co
n
s
u
m
in
g
,
s
en
s
itiv
e
to
v
ar
iatio
n
s
in
im
ag
e
q
u
ality
,
a
n
d
s
till
r
eq
u
ir
e
ex
p
er
t
s
u
p
er
v
i
s
io
n
,
wh
ich
lim
its
th
eir
s
ca
lab
ilit
y
in
clin
ical
p
r
ac
tice.
T
h
is
h
as
m
o
tiv
ated
a
s
h
if
t
to
war
d
d
ee
p
lear
n
in
g
-
b
ased
s
eg
m
en
tatio
n
ap
p
r
o
ac
h
es,
wh
ich
ca
n
p
r
o
v
id
e
f
ast,
r
o
b
u
s
t,
an
d
f
u
lly
au
to
m
ated
r
esu
lts
.
W
ith
th
e
ad
v
e
n
t
o
f
d
ee
p
lear
n
i
n
g
,
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etw
o
r
k
s
(
C
NNs)
h
av
e
b
ec
o
m
e
th
e
p
r
ef
e
r
r
ed
ap
p
r
o
ac
h
f
o
r
m
e
d
ical
im
ag
e
a
n
aly
s
is
,
in
clu
d
in
g
s
eg
m
en
tatio
n
an
d
class
if
icatio
n
.
Ar
ch
itect
u
r
es
s
u
ch
as
U
-
Net
an
d
its
v
ar
ian
ts
h
a
v
e
ac
h
iev
ed
ex
ce
llen
t
p
er
f
o
r
m
a
n
ce
i
n
s
eg
m
en
tin
g
o
r
g
an
s
a
n
d
lesi
o
n
s
in
r
ad
io
l
o
g
ical
im
ag
es.
I
n
d
en
tis
tr
y
,
d
ee
p
lea
r
n
in
g
h
as
b
ee
n
ap
p
lie
d
to
task
s
s
u
ch
as
to
o
th
n
u
m
b
e
r
in
g
,
ca
r
ies
d
etec
tio
n
,
an
d
m
an
d
ib
u
lar
ca
n
al
s
eg
m
en
tatio
n
.
A
r
ec
en
t
s
tu
d
y
p
r
o
p
o
s
ed
a
d
ee
p
lear
n
in
g
m
o
d
el
f
o
r
au
to
m
atic
to
o
th
d
etec
tio
n
an
d
n
u
m
b
e
r
in
g
in
p
an
o
r
am
i
c
r
ad
io
g
r
ap
h
s
,
ac
h
iev
i
n
g
h
ig
h
ac
cu
r
ac
y
a
n
d
d
em
o
n
s
tr
atin
g
th
e
p
o
ten
tial
o
f
AI
-
b
ased
s
y
s
tem
s
to
ass
is
t
d
en
tal
p
r
ac
titi
o
n
er
s
[
2
4
]
.
Ho
wev
er
,
U
-
Net
-
b
ased
m
eth
o
d
s
o
f
ten
r
eq
u
ir
e
s
u
b
s
tan
tial
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
an
d
a
r
e
n
o
t o
p
tim
ized
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
.
Ob
ject
d
etec
tio
n
m
o
d
els,
p
a
r
ticu
lar
ly
th
o
s
e
b
ased
o
n
th
e
y
o
u
o
n
ly
l
o
o
k
o
n
ce
(
YOL
O
)
f
a
m
ily
,
h
av
e
g
ain
ed
atten
tio
n
f
o
r
th
ei
r
s
p
ee
d
an
d
ac
cu
r
ac
y
.
YOL
O,
f
ir
s
t
in
tr
o
d
u
ce
d
in
2
0
2
5
,
r
ed
ef
i
n
ed
o
b
ject
d
etec
tio
n
b
y
f
r
am
in
g
it
as
a
s
in
g
le
r
eg
r
ess
io
n
task
,
s
ig
n
if
ican
tly
r
ed
u
ci
n
g
in
f
er
en
ce
tim
e
[
2
5
]
.
R
ec
en
t
s
tu
d
ies
h
av
e
ap
p
lied
YOL
O
v
ar
ian
ts
to
d
e
n
tal
an
d
m
ed
ical
im
a
g
in
g
task
s
.
d
e
m
o
n
s
tr
ated
th
e
s
u
cc
ess
f
u
l
ap
p
licatio
n
o
f
YOL
O
v
ar
ian
ts
in
d
en
tal
an
d
m
ed
ic
al
im
ag
in
g
task
s
.
YOL
Ov
8
h
as
b
ee
n
u
tili
ze
d
f
o
r
d
etec
ti
n
g
an
d
s
eg
m
en
tin
g
r
ad
io
lu
ce
n
t
m
an
d
i
b
u
lar
lesi
o
n
s
o
n
p
an
o
r
am
ic
r
a
d
io
g
r
ap
h
s
,
ac
h
iev
in
g
m
ea
n
d
etec
tio
n
a
cc
u
r
ac
ies
o
f
0
.
9
1
8
with
o
u
t
au
g
m
en
tatio
n
a
n
d
0
.
9
5
2
with
au
g
m
en
tatio
n
,
o
u
tp
er
f
o
r
m
in
g
p
r
ev
i
o
u
s
m
o
d
els
tr
ain
ed
o
n
s
m
aller
d
atasets
[
2
6
]
.
T
h
ese
f
in
d
in
g
s
h
ig
h
lig
h
t
YOL
Ov
8
’
s
ca
p
a
b
ilit
y
to
m
an
ag
e
ch
allen
g
i
n
g
c
ases
in
v
o
lv
in
g
lo
w
co
n
tr
ast an
d
o
v
er
lap
p
in
g
a
n
ato
m
ical
s
tr
u
ctu
r
es.
Ou
ts
id
e
d
en
tis
tr
y
,
YOL
Ov
8
h
as
s
h
o
wn
s
u
p
er
io
r
p
e
r
f
o
r
m
an
ce
in
v
ar
io
u
s
s
eg
m
en
tatio
n
task
s
.
A
co
m
p
ar
ativ
e
s
tu
d
y
b
etwe
en
YOL
Ov
5
a
n
d
YOL
Ov
8
f
o
r
co
r
r
o
s
io
n
s
eg
m
e
n
tatio
n
o
n
m
etal
s
u
r
f
ac
es
an
d
f
o
u
n
d
th
at
YOL
Ov
8
p
r
o
d
u
ce
d
m
o
r
e
ac
cu
r
ate
r
esu
lts
,
p
ar
ticu
lar
ly
o
n
co
m
p
lex
an
d
ir
r
e
g
u
lar
p
atter
n
s
[
2
7
]
.
Similar
ly
,
YOL
Ov
8
ap
p
lied
t
o
p
lan
t
leaf
d
is
ea
s
e
d
etec
tio
n
u
s
in
g
2
,
8
5
0
a
n
n
o
tate
d
im
ag
e
s
ac
r
o
s
s
1
9
class
es
ac
h
iev
ed
9
8
%
p
r
ec
is
io
n
an
d
a
9
7
%
F1
-
s
co
r
e,
f
u
r
th
er
co
n
f
ir
m
in
g
its
v
er
s
atility
an
d
h
ig
h
p
er
f
o
r
m
an
ce
ac
r
o
s
s
d
if
f
er
en
t
d
o
m
ain
s
[
2
8
]
.
Desp
ite
th
ese
p
r
o
m
is
in
g
r
esu
lt
s
,
r
elativ
ely
f
ew
s
tu
d
ies
h
av
e
im
p
lem
en
ted
YOL
O
f
o
r
co
m
b
i
n
ed
to
o
th
d
etec
tio
n
,
s
eg
m
en
tatio
n
,
a
n
d
n
u
m
b
er
in
g
in
p
an
o
r
am
ic
r
a
d
io
g
r
ap
h
s
.
Mo
s
t
p
r
io
r
wo
r
k
h
as
f
o
cu
s
ed
o
n
d
etec
tio
n
o
r
lan
d
m
ar
k
id
en
tific
atio
n
al
o
n
e,
an
d
ch
allen
g
es
r
em
ain
in
s
eg
m
en
tin
g
p
o
s
ter
io
r
teeth
d
u
e
to
r
ed
u
ce
d
c
o
n
tr
ast
an
d
a
n
ato
m
ical
o
v
er
lap
.
Fu
r
th
er
m
o
r
e,
th
e
in
teg
r
atio
n
o
f
s
tan
d
ar
d
ized
n
u
m
b
e
r
in
g
s
y
s
tem
s
,
s
u
ch
as
th
e
Fed
er
atio
n
Den
tair
e
I
n
ter
n
atio
n
ale
(
FDI
)
two
-
d
ig
it
s
y
s
tem
,
in
to
d
ee
p
lear
n
in
g
p
ip
elin
es
r
e
m
ain
s
lim
ited
,
y
et
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
YOLO
-
b
a
s
ed
d
ee
p
lea
r
n
in
g
fo
r
to
o
th
d
etec
tio
n
,
s
eg
me
n
ta
tio
n
,
a
n
d
n
u
m
b
erin
g
in
…
(
S
r
i O
kta
mu
lia
n
i
)
3593
is
es
s
en
tial
f
o
r
clin
ical
ad
o
p
tio
n
.
T
o
ad
d
r
ess
th
ese
g
ap
s
,
th
is
s
tu
d
y
p
r
o
p
o
s
es
a
YOL
Ov
8
m
-
s
eg
m
o
d
el
f
o
r
th
e
au
to
m
ated
d
etec
tio
n
,
s
eg
m
en
t
atio
n
,
an
d
n
u
m
b
e
r
in
g
o
f
teet
h
in
p
an
o
r
am
ic
d
e
n
tal
r
ad
io
g
r
ap
h
s
.
A
d
ataset
o
f
3
0
2
r
a
d
io
g
r
ap
h
s
co
m
p
r
is
in
g
9
,
0
0
9
an
n
o
tated
teeth
was u
s
ed
,
with
an
n
o
tatio
n
s
b
ased
o
n
FDI
two
-
d
ig
it sy
s
tem
.
2.
M
E
T
H
O
D
2
.
1
.
Da
t
a
s
et
p
re
pa
ra
t
io
n
T
h
is
s
tu
d
y
u
tili
ze
d
p
a
n
o
r
a
m
ic
d
en
tal
r
ad
io
g
r
ap
h
s
co
llecte
d
f
r
o
m
th
e
Den
tal
an
d
Or
al
Ho
s
p
ital
Un
iv
er
s
itas
An
d
alas
(
R
S
GM
Un
an
d
)
.
A
to
tal
o
f
3
0
2
im
a
g
es
wer
e
in
clu
d
ed
af
ter
s
cr
ee
n
in
g
f
o
r
q
u
ality
an
d
co
m
p
leten
ess
.
I
m
ag
in
g
p
ar
am
eter
s
wer
e
s
tan
d
ar
d
ized
at
8
0
k
V
tu
b
e
v
o
lta
g
e,
9
.
0
m
A
tu
b
e
cu
r
r
en
t,
a
n
d
1
4
.
0
s
ex
p
o
s
u
r
e
tim
e,
with
a
m
e
asu
r
ed
d
o
s
e
o
f
1
5
7
.
2
m
Gy
/cm
2
.
I
m
ag
es
with
s
ev
er
e
m
o
tio
n
ar
tifa
cts,
m
is
s
in
g
an
ato
m
ical
s
tr
u
ctu
r
es,
o
r
in
ad
eq
u
ate
ex
p
o
s
u
r
e
wer
e
ex
cl
u
d
e
d
to
en
s
u
r
e
a
co
n
s
is
ten
t
d
ataset.
All
im
ag
es
wer
e
in
itially
s
to
r
ed
in
DI
C
OM
f
o
r
m
at,
wh
ich
co
n
tain
s
b
o
th
r
a
w
p
ix
el
d
ata
an
d
ac
q
u
is
itio
n
m
etad
ata
s
u
ch
as
ex
p
o
s
u
r
e
p
ar
am
eter
s
.
Fo
r
m
o
d
el
tr
ain
in
g
,
th
e
DI
C
OM
f
iles
wer
e
ex
p
o
r
te
d
in
to
b
itm
ap
im
ag
e
f
ile
(
B
MP)
f
o
r
m
at
a
n
d
r
esized
to
6
4
0
×6
4
0
p
ix
els
to
m
ee
t
YOL
Ov
8
in
p
u
t
r
e
q
u
ir
em
e
n
ts
.
His
to
g
r
a
m
eq
u
aliza
tio
n
an
d
in
ten
s
ity
n
o
r
m
aliza
tio
n
wer
e
ap
p
lied
to
im
p
r
o
v
e
im
ag
e
c
o
n
tr
ast
an
d
en
h
an
ce
t
h
e
v
is
ib
ilit
y
o
f
an
ato
m
ical
s
tr
u
ctu
r
es.
Patien
t in
f
o
r
m
atio
n
was a
n
o
n
y
m
ized
b
ef
o
r
e
p
r
o
ce
s
s
in
g
.
2
.
2
.
Anno
t
a
t
io
n a
nd
l
a
belin
g
Prio
r
to
m
o
d
el
tr
ain
in
g
,
all
p
a
n
o
r
am
ic
r
ad
i
o
g
r
a
p
h
s
u
n
d
e
r
wen
t
a
s
tr
u
ctu
r
ed
r
eg
io
n
o
f
in
te
r
est
(
R
OI
)
an
n
o
tatio
n
p
r
o
ce
s
s
u
s
in
g
th
e
R
o
b
o
f
lo
w
p
latf
o
r
m
.
E
ac
h
to
o
th
was
m
an
u
ally
lab
eled
ac
co
r
d
in
g
t
o
th
e
FDI
two
-
d
ig
it
d
en
tal
n
o
m
en
clatu
r
e
s
y
s
tem
,
wh
er
e
th
e
f
ir
s
t
d
i
g
it
in
d
icate
s
th
e
q
u
ad
r
an
t
an
d
th
e
s
ec
o
n
d
d
ig
it
in
d
icate
s
th
e
to
o
th
n
u
m
b
er
wi
th
in
th
at
q
u
ad
r
an
t
as
s
h
o
wn
i
n
Fig
u
r
e
1
.
Fo
r
ea
ch
to
o
t
h
,
b
o
th
b
o
u
n
d
in
g
b
o
x
es
an
d
p
o
ly
g
o
n
al
s
eg
m
en
tatio
n
m
ask
s
wer
e
g
e
n
er
ated
,
en
ab
li
n
g
th
e
m
o
d
el
to
p
e
r
f
o
r
m
in
s
t
an
ce
s
eg
m
en
tatio
n
r
ath
er
th
an
s
im
p
le
d
etec
tio
n
.
Fig
u
r
e
1
.
FDI
two
-
d
ig
it to
o
th
n
u
m
b
er
i
n
g
s
y
s
tem
All
an
n
o
tatio
n
s
wer
e
in
d
ep
en
d
en
tly
r
e
v
iewe
d
b
y
o
n
e
ex
p
e
r
ie
n
ce
d
d
en
tis
t
to
en
s
u
r
e
lab
elin
g
ac
cu
r
ac
y
an
d
co
n
s
is
ten
cy
.
An
y
d
is
cr
e
p
an
cies
b
etwe
en
an
n
o
tat
o
r
s
wer
e
r
eso
lv
ed
th
r
o
u
g
h
co
n
s
e
n
s
u
s
d
is
cu
s
s
io
n
s
to
m
in
im
ize
in
ter
-
o
p
er
ato
r
v
a
r
iab
ilit
y
.
T
h
e
f
in
al
d
ataset
co
m
p
r
is
ed
9
,
0
0
9
a
n
n
o
tated
teeth
ac
r
o
s
s
3
0
2
p
an
o
r
am
i
c
r
ad
io
g
r
a
p
h
s
,
with
a
b
alan
ce
d
r
ep
r
esen
tatio
n
o
f
m
a
x
illar
y
an
d
m
an
d
ib
u
lar
teeth
.
T
h
is
en
s
u
r
ed
th
at
th
e
m
o
d
el
was
ex
p
o
s
ed
to
s
u
f
f
icie
n
t
an
a
to
m
ical
v
ar
iab
ilit
y
ac
r
o
s
s
q
u
a
d
r
an
ts
.
An
ex
a
m
p
le
o
f
th
e
a
n
n
o
tatio
n
wo
r
k
f
lo
w
an
d
R
OI
lab
elin
g
is
illu
s
tr
a
ted
in
Fig
u
r
e
2
.
E
ac
h
to
o
t
h
was
m
an
u
ally
d
elin
e
ated
with
a
p
o
ly
g
o
n
al
s
eg
m
en
tatio
n
m
ask
an
d
lab
ele
d
ac
co
r
d
i
n
g
to
t
h
e
FDI
two
-
d
i
g
it sy
s
tem
.
2
.
3
.
YO
L
O
v
8
wo
r
k
a
rc
hite
c
t
ure
YOL
Ov
8
,
th
e
latest iter
atio
n
o
f
th
e
YOL
O
o
b
ject
d
etec
tio
n
f
am
ily
,
was selec
ted
f
o
r
th
is
s
t
u
d
y
d
u
e
to
its
h
ig
h
s
p
ee
d
,
ac
c
u
r
ac
y
,
a
n
d
r
o
b
u
s
tn
ess
in
h
a
n
d
lin
g
s
m
all
o
b
jects
co
m
m
o
n
ly
p
r
esen
t
i
n
d
en
tal
p
a
n
o
r
a
m
ic
r
ad
io
g
r
a
p
h
s
.
Un
lik
e
ea
r
lier
YOL
O
v
er
s
io
n
s
,
YOL
Ov
8
em
p
lo
y
s
an
ch
o
r
-
f
r
ee
d
etec
tio
n
,
a
d
ec
o
u
p
led
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
9
1
-
3
6
0
2
3594
class
if
icatio
n
/r
eg
r
ess
io
n
h
ea
d
.
I
t
also
u
s
es
a
lig
h
tweig
h
t
C
SP
Dar
k
n
et
-
in
s
p
ir
ed
b
ac
k
b
o
n
e,
r
esu
ltin
g
in
im
p
r
o
v
e
d
g
en
e
r
aliza
tio
n
an
d
r
ed
u
ce
d
c
o
m
p
u
tatio
n
al
o
v
e
r
h
ea
d
.
T
h
e
YOL
Ov
8
ar
c
h
itectu
r
e
is
d
iv
id
ed
in
to
th
r
ee
m
ain
c
o
m
p
o
n
e
n
ts
:
b
ac
k
b
o
n
e,
n
ec
k
,
a
n
d
h
ea
d
as
s
h
o
wn
in
Fig
u
r
e
3
.
T
h
e
m
o
d
e
l
co
n
s
is
ts
o
f
a
b
ac
k
b
o
n
e
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
a
n
ec
k
f
o
r
m
u
lti
-
s
ca
le
f
ea
tu
r
e
f
u
s
io
n
,
an
d
a
h
ea
d
f
o
r
an
c
h
o
r
-
f
r
ee
d
etec
tio
n
a
n
d
in
s
tan
ce
s
eg
m
en
tatio
n
.
i)
B
ac
k
b
o
n
e:
r
esp
o
n
s
ib
le
f
o
r
ex
t
r
ac
tin
g
im
a
g
e
f
ea
t
u
r
es
s
u
ch
as
ed
g
es,
tex
tu
r
es,
an
d
s
h
ap
es
f
r
o
m
th
e
in
p
u
t
im
ag
e.
I
t
c
o
n
s
is
ts
o
f
co
n
v
o
l
u
tio
n
al
lay
er
s
(
C
o
n
v
)
,
s
p
atial
p
y
r
am
i
d
d
ilated
c
o
n
v
o
lu
tio
n
(
SP
D
-
C
o
n
v
)
,
cr
o
s
s
s
tag
e
p
ar
tial
co
n
n
ec
tio
n
s
with
f
u
s
io
n
(
C
2
f
)
,
an
d
s
p
atial
p
y
r
am
id
p
o
o
lin
g
-
f
ast
(
SP
PF
)
.
T
h
ese
m
o
d
u
les
en
h
a
n
ce
th
e
m
o
d
el'
s
ab
ilit
y
to
ca
p
tu
r
e
m
u
lti
-
s
ca
le
f
ea
tu
r
es,
wh
ich
is
cr
itical
f
o
r
d
etec
tin
g
s
m
all
an
d
o
v
er
lap
p
in
g
teeth
.
ii)
Nec
k
:
ag
g
r
eg
ates
an
d
r
ef
in
e
s
f
ea
tu
r
es
f
r
o
m
m
u
ltip
le
s
ca
les
o
f
th
e
b
ac
k
b
o
n
e
u
s
in
g
u
p
s
am
p
lin
g
,
co
n
ca
ten
atio
n
,
c
o
n
v
o
lu
tio
n
al
lay
er
s
,
an
d
th
e
g
l
o
b
al
atten
ti
o
n
m
ec
h
a
n
is
m
(
GAM
)
to
p
r
eser
v
e
s
p
atial
in
f
o
r
m
atio
n
an
d
im
p
r
o
v
e
d
ete
ctio
n
r
o
b
u
s
tn
ess
.
iii)
Hea
d
:
p
r
o
d
u
ce
s
th
e
f
in
al
o
u
t
p
u
ts
,
in
clu
d
i
n
g
b
o
u
n
d
in
g
b
o
x
es,
o
b
ject
class
if
icatio
n
,
a
n
d
s
eg
m
en
tatio
n
m
ask
s
.
YOL
Ov
8
u
s
es a
n
ch
o
r
-
f
r
ee
h
ea
d
,
wh
ich
im
p
r
o
v
es f
le
x
ib
ilit
y
an
d
lo
ca
lizatio
n
ac
cu
r
ac
y
[
2
6
]
,
[
2
9
]
.
Fig
u
r
e
2
.
E
x
am
p
le
o
f
R
OI
an
n
o
tatio
n
p
r
o
ce
s
s
p
er
f
o
r
m
ed
in
R
o
b
o
f
lo
w
Fig
u
r
e
3
.
YOL
Ov
8
wo
r
k
a
r
ch
i
tectu
r
e
T
h
is
s
tu
d
y
em
p
lo
y
ed
th
e
YOL
Ov
8
-
s
eg
v
ar
ian
t,
wh
ic
h
p
er
f
o
r
m
s
in
s
tan
ce
s
eg
m
en
tatio
n
b
y
p
r
ed
ictin
g
b
o
th
b
o
u
n
d
i
n
g
b
o
x
es
an
d
s
eg
m
en
tatio
n
m
ask
s
f
o
r
ea
ch
to
o
th
.
Am
o
n
g
th
e
av
ailab
le
m
o
d
el
s
izes
(
n
an
o
,
s
m
all,
m
ed
iu
m
,
lar
g
e,
an
d
x
-
lar
g
e)
,
YOL
Ov
8
m
-
s
e
g
was
ch
o
s
en
to
b
alan
ce
in
f
er
en
ce
s
p
ee
d
with
ac
cu
r
ac
y
,
m
a
k
in
g
it
s
u
itab
le
f
o
r
p
o
ten
tial
r
ea
l
-
tim
e
ap
p
licatio
n
s
in
clin
ical
s
ettin
g
s
.
Mo
d
el
tr
ain
in
g
was
co
n
d
u
cte
d
o
n
Go
o
g
le
C
o
lab
,
lev
er
ag
in
g
g
r
ap
h
ics
p
r
o
c
ess
in
g
u
n
it
(
GPU)
ac
ce
ler
a
tio
n
f
o
r
ef
f
icien
t
co
m
p
u
tatio
n
.
T
h
e
tr
ain
in
g
d
u
r
atio
n
v
ar
ied
d
ep
e
n
d
in
g
o
n
th
e
d
ataset
s
ize,
th
e
n
u
m
b
er
o
f
o
u
tp
u
t
class
es,
an
d
th
e
b
atch
co
n
f
ig
u
r
atio
n
.
Hy
p
er
p
ar
am
eter
s
,
s
u
ch
as
th
e
lear
n
in
g
r
ate,
o
p
tim
izer
s
ettin
g
s
,
an
d
m
ax
im
u
m
n
u
m
b
er
o
f
e
p
o
ch
s
,
wer
e
t
u
n
ed
iter
ativ
ely
to
ac
h
iev
e
o
p
tim
al
m
o
d
el
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
YOLO
-
b
a
s
ed
d
ee
p
lea
r
n
in
g
fo
r
to
o
th
d
etec
tio
n
,
s
eg
me
n
ta
tio
n
,
a
n
d
n
u
m
b
erin
g
in
…
(
S
r
i O
kta
mu
lia
n
i
)
3595
2
.
4
.
T
ra
ini
ng
s
et
up
Mo
d
el
tr
ain
in
g
was
p
e
r
f
o
r
m
e
d
o
n
a
s
y
s
tem
eq
u
ip
p
e
d
with
an
NVI
DI
A
SMI
GPU
(
d
r
iv
er
v
er
s
io
n
:
5
5
0
.
5
4
.
1
5
,
C
UDA
v
er
s
io
n
:
1
2
.
4
)
,
6
4
GB
o
f
R
AM
,
r
u
n
n
in
g
m
ac
OS
Mo
n
ter
ey
,
an
d
ad
d
itio
n
ally
s
u
p
p
o
r
ted
b
y
Go
o
g
le
C
o
lab
with
GPU
ac
ce
ler
atio
n
f
o
r
en
h
a
n
ce
d
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
T
h
e
d
at
aset
was
r
an
d
o
m
ly
d
iv
id
ed
in
to
7
0
% tr
ain
in
g
(
2
1
1
im
ag
es),
2
0
% v
alid
atio
n
(
6
1
im
ag
es),
an
d
1
0
% testi
n
g
(
3
0
im
ag
es),
en
s
u
r
in
g
a
b
alan
ce
d
d
is
tr
ib
u
tio
n
o
f
to
o
th
class
es
ac
r
o
s
s
all
s
u
b
s
et
s
.
T
h
is
s
p
lit
m
in
im
ized
s
am
p
lin
g
b
ias
an
d
m
ain
tain
ed
class
-
lev
el
r
ep
r
esen
tatio
n
in
ea
ch
p
ar
titi
o
n
.
No
tab
ly
,
d
ata
a
u
g
m
en
tatio
n
tech
n
iq
u
es
wer
e
n
o
t
ap
p
lied
in
th
is
s
tu
d
y
to
p
r
eser
v
e
th
e
n
atu
r
al
d
is
tr
ib
u
tio
n
o
f
p
ix
el
in
ten
s
ities
an
d
an
ato
m
ical
v
ar
iatio
n
s
in
h
er
en
t
to
p
an
o
r
am
ic
r
ad
io
g
r
a
p
h
s
.
T
h
is
d
ec
is
io
n
was
m
ad
e
to
av
o
id
i
n
tr
o
d
u
cin
g
ar
tific
ial
ar
tifa
cts
th
at
co
u
ld
d
is
to
r
t
to
o
th
b
o
u
n
d
ar
ies,
wh
ich
ar
e
alr
ea
d
y
af
f
ec
ted
b
y
im
ag
in
g
p
h
y
s
ics (
e.
g
.
,
s
ca
tter
an
d
g
eo
m
etr
ic
d
is
to
r
tio
n
)
.
Fu
tu
r
e
wo
r
k
will
e
v
alu
ate
au
g
m
en
tatio
n
s
tr
ateg
ies
s
u
c
h
as r
o
tatio
n
,
s
ca
lin
g
,
an
d
c
o
n
tr
ast
a
d
ju
s
tm
en
ts
to
f
u
r
th
e
r
en
h
a
n
ce
m
o
d
el
g
en
e
r
aliza
tio
n
.
T
h
e
m
o
d
el
was
tr
ain
ed
f
o
r
3
0
0
ep
o
c
h
s
with
a
b
at
ch
s
ize
o
f
1
6
,
u
s
in
g
th
e
d
ef
au
lt
YOL
Ov
8
co
m
p
o
s
i
te
lo
s
s
f
u
n
ctio
n
,
wh
ich
co
m
b
i
n
es
class
if
icatio
n
,
b
o
u
n
d
in
g
b
o
x
r
eg
r
ess
io
n
,
an
d
m
ask
lo
s
s
es
to
jo
in
tly
o
p
tim
i
ze
d
etec
tio
n
an
d
s
eg
m
e
n
tatio
n
p
er
f
o
r
m
a
n
ce
.
E
ar
l
y
s
to
p
p
in
g
was
im
p
lem
en
te
d
b
ased
o
n
v
alid
atio
n
m
AP to
p
r
ev
en
t o
v
e
r
f
itti
n
g
an
d
r
ed
u
ce
u
n
n
ec
ess
ar
y
co
m
p
u
tatio
n
.
2.
5
.
E
v
a
lua
t
i
o
n
m
et
rics
Mo
d
el
p
er
f
o
r
m
an
ce
was
ass
e
s
s
ed
o
n
th
e
in
d
ep
en
d
e
n
t
test
s
et
u
s
in
g
s
tan
d
ar
d
o
b
ject
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
m
etr
ics
[
3
0
]
,
[
3
1
]
.
T
h
e
p
r
im
ar
y
m
etr
ic
was
m
ea
n
av
er
ag
e
p
r
ec
is
io
n
(
m
A
P)
,
ca
lcu
lated
at
an
in
ter
s
ec
tio
n
o
v
e
r
u
n
io
n
(
I
o
U
)
t
h
r
esh
o
ld
o
f
0
.
5
(
m
AP@
0
.
5
)
a
n
d
av
e
r
ag
ed
ac
r
o
s
s
th
r
esh
o
ld
s
f
r
o
m
0
.
5
to
0
.
9
5
in
in
cr
em
en
ts
o
f
0
.
0
5
(
m
AP@
0
.
5
:0
.
9
5
)
,
p
r
o
v
id
i
n
g
b
o
th
len
ie
n
t
an
d
s
tr
ict
m
ea
s
u
r
es
o
f
d
et
ec
tio
n
ac
cu
r
ac
y
.
I
n
ad
d
itio
n
,
p
r
ec
is
io
n
,
d
ef
in
e
d
as
th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
p
r
ed
icted
teeth
a
m
o
n
g
all
p
r
ed
ic
tio
n
s
,
was
u
s
ed
to
ev
alu
ate
th
e
m
o
d
el
’
s
ab
ilit
y
to
av
o
i
d
f
alse
p
o
s
itiv
es,
wh
ile
r
ec
all
,
d
ef
in
ed
as
th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
d
etec
ted
teeth
am
o
n
g
all
g
r
o
u
n
d
-
tr
u
t
h
an
n
o
tatio
n
s
,
r
ef
lecte
d
its
ab
ilit
y
to
m
in
im
ize
f
alse n
eg
ativ
es.
T
o
ca
p
tu
r
e
th
e
tr
ad
e
-
o
f
f
b
etwe
en
p
r
ec
is
io
n
an
d
r
ec
all,
th
e
F1
-
s
co
r
e
was
ca
lcu
lated
as
th
e
h
ar
m
o
n
ic
m
ea
n
o
f
th
e
two
,
r
ep
r
esen
tin
g
a
b
alan
ce
d
m
ea
s
u
r
e
o
f
o
v
er
all
m
o
d
el
p
er
f
o
r
m
an
ce
.
Per
f
o
r
m
an
ce
was
r
e
p
o
r
ted
b
o
th
at
th
e
p
er
-
class
lev
el
(
f
o
r
i
n
d
iv
id
u
al
teeth
)
an
d
as
a
n
av
e
r
ag
e
ac
r
o
s
s
all
class
es
,
en
ab
lin
g
a
co
m
p
r
eh
e
n
s
iv
e
ass
es
s
m
en
t o
f
d
etec
tio
n
a
n
d
s
e
g
m
en
tatio
n
q
u
ality
at
b
o
th
g
r
a
n
u
lar
an
d
h
o
lis
tic
lev
els.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
All
ex
p
er
im
en
tal
s
ettin
g
s
an
d
im
p
lem
en
tatio
n
d
etails
ar
e
s
u
m
m
ar
ized
in
T
ab
le
1
.
T
h
ese
in
clu
d
e
m
o
d
el
co
n
f
ig
u
r
atio
n
,
tr
ain
in
g
h
y
p
er
p
ar
am
eter
s
,
d
ataset
s
p
lits
,
s
o
f
twar
e
v
er
s
io
n
s
,
an
d
h
ar
d
war
e
s
p
ec
if
icatio
n
s
.
T
ab
le
1
is
in
ten
d
e
d
to
s
u
p
p
o
r
t
r
ep
r
o
d
u
cib
ilit
y
with
o
u
t c
h
an
g
i
n
g
th
e
m
ea
n
in
g
o
f
th
e
o
r
ig
in
al
s
tatem
en
t.
T
ab
le
1
.
E
x
p
er
im
en
t
co
n
f
i
g
u
r
a
tio
n
an
d
tr
ai
n
in
g
d
etails
C
a
t
e
g
o
r
y
C
o
n
f
i
g
u
r
a
t
i
o
n
M
o
d
e
l
a
r
c
h
i
t
e
c
t
u
r
e
Y
O
LO
v
8
(
u
l
t
r
a
l
y
t
i
c
s
)
–
i
n
st
a
n
c
e
se
g
m
e
n
t
a
t
i
o
n
Y
O
LO
v
8
v
a
r
i
a
n
t
Y
O
LO
v
8
m
-
s
e
g
Y
O
LO
v
8
v
e
r
si
o
n
/
p
a
c
k
a
g
e
U
l
t
r
a
l
y
t
i
c
s YO
LO
v
8
(
v
8
.
x
,
p
i
p
p
a
c
k
a
g
e
u
l
t
r
a
l
y
t
i
c
s)
F
r
a
mew
o
r
k
P
y
T
o
r
c
h
P
y
t
h
o
n
v
e
r
s
i
o
n
P
y
t
h
o
n
3
.
1
2
O
p
t
i
mi
z
e
r
A
d
a
mW
(
l
r
=
0
.
0
0
0
2
7
,
m
o
m
e
n
t
u
m
=
0
.
9
)
I
n
i
t
i
a
l
l
e
a
r
n
i
n
g
r
a
t
e
0
.
0
1
W
e
i
g
h
t
d
e
c
a
y
0
.
0
0
0
5
W
a
r
m
-
u
p
st
r
a
t
e
g
y
Li
n
e
a
r
w
a
r
m
-
u
p
(
Y
O
LO
v
8
d
e
f
a
u
l
t
)
B
a
t
c
h
si
z
e
16
I
n
p
u
t
i
m
a
g
e
s
i
z
e
6
4
0
×
6
4
0
p
i
x
e
l
s
N
u
mb
e
r
o
f
e
p
o
c
h
s
3
0
0
Ea
r
l
y
st
o
p
p
i
n
g
En
a
b
l
e
d
Ea
r
l
y
st
o
p
p
i
n
g
p
a
t
i
e
n
c
e
5
0
e
p
o
c
h
s (b
a
se
d
o
n
v
a
l
i
d
a
t
i
o
n
mA
P
)
Lo
ss f
u
n
c
t
i
o
n
Y
O
LO
v
8
c
o
mp
o
s
i
t
e
l
o
ss (
c
l
a
ssi
f
i
c
a
t
i
o
n
+
b
o
u
n
d
i
n
g
b
o
x
r
e
g
r
e
ssi
o
n
+
mas
k
l
o
ss)
C
h
e
c
k
p
o
i
n
t
i
n
g
B
e
st
m
o
d
e
l
sa
v
e
d
b
a
s
e
d
o
n
h
i
g
h
e
s
t
v
a
l
i
d
a
t
i
o
n
mA
P
D
a
t
a
s
e
t
s
p
l
i
t
7
0
%
t
r
a
i
n
i
n
g
(
2
1
1
i
ma
g
e
s)
,
2
0
%
v
a
l
i
d
a
t
i
o
n
(
6
1
i
m
a
g
e
s)
,
1
0
%
t
e
s
t
i
n
g
(
3
0
i
m
a
g
e
s)
C
l
a
s
s
b
a
l
a
n
c
i
n
g
B
a
l
a
n
c
e
d
d
i
s
t
r
i
b
u
t
i
o
n
o
f
t
o
o
t
h
c
l
a
sses
a
c
r
o
ss
a
l
l
s
u
b
se
t
s
D
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
N
o
t
a
p
p
l
i
e
d
R
a
n
d
o
m s
e
e
d
F
i
x
e
d
see
d
u
se
d
f
o
r
d
a
t
a
se
t
s
p
l
i
t
t
i
n
g
a
n
d
t
r
a
i
n
i
n
g
t
o
e
n
s
u
r
e
r
e
p
r
o
d
u
c
i
b
i
l
i
t
y
A
n
n
o
t
a
t
i
o
n
f
o
r
ma
t
Y
O
LO
seg
m
e
n
t
a
t
i
o
n
f
o
r
m
a
t
G
P
U
N
V
I
D
I
A
G
P
U
(
d
r
i
v
e
r
5
5
0
.
5
4
.
1
5
,
C
U
D
A
1
2
.
4
)
S
y
st
e
m
R
A
M
6
4
G
B
O
p
e
r
a
t
i
n
g
sy
st
e
m
mac
O
S
M
o
n
t
e
r
e
y
C
l
o
u
d
e
n
v
i
r
o
n
me
n
t
G
o
o
g
l
e
C
o
l
a
b
w
i
t
h
G
P
U
a
c
c
e
l
e
r
a
t
i
o
n
C
o
d
e
a
n
d
d
a
t
a
a
v
a
i
l
a
b
i
l
i
t
y
Tr
a
i
n
i
n
g
/
v
a
l
i
d
a
t
i
o
n
/
t
e
st
sp
l
i
t
s,
a
n
n
o
t
a
t
i
o
n
e
x
p
o
r
t
s
c
r
i
p
t
s,
a
n
d
c
o
n
f
i
g
u
r
a
t
i
o
n
f
i
l
e
s
a
r
e
a
v
a
i
l
a
b
l
e
u
p
o
n
r
e
a
so
n
a
b
l
e
r
e
q
u
e
s
t
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
9
1
-
3
6
0
2
3596
3
.
1
.
T
ra
ini
ng
a
nd
v
a
lid
a
t
io
n
perf
o
rm
a
nce
T
h
e
YOL
Ov
8
m
-
s
eg
m
o
d
el
w
as
tr
ain
ed
f
o
r
3
0
0
ep
o
ch
s
u
s
i
n
g
Go
o
g
le
C
o
lab
with
ea
r
ly
s
to
p
p
in
g
to
p
r
ev
en
t
o
v
er
f
itti
n
g
.
Fig
u
r
e
4
s
h
o
ws
th
e
tr
ain
in
g
an
d
v
alid
ati
o
n
cu
r
v
es
f
o
r
b
o
x
lo
s
s
,
class
if
icatio
n
lo
s
s
,
m
ask
lo
s
s
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
m
AP
m
etr
ics.
B
o
th
tr
ain
in
g
an
d
v
alid
atio
n
l
o
s
s
es
d
ec
r
ea
s
ed
co
n
s
is
ten
tly
,
co
n
v
er
g
in
g
to
war
d
th
e
f
i
n
al
ep
o
ch
s
,
in
d
icatin
g
th
at
th
e
m
o
d
el
s
u
cc
ess
f
u
lly
m
in
im
ized
er
r
o
r
s
.
Pre
cisi
o
n
,
r
ec
all,
an
d
m
AP
s
tead
ily
i
n
cr
ea
s
ed
d
u
r
in
g
tr
ain
in
g
,
r
e
ac
h
in
g
a
p
latea
u
n
ea
r
th
e
f
in
al
ep
o
ch
s
.
T
h
is
d
em
o
n
s
tr
ates th
at
th
e
m
o
d
el
a
ch
iev
ed
s
tab
le
g
en
e
r
aliza
tio
n
with
o
u
t o
v
er
f
itti
n
g
.
Fig
u
r
e
4
.
T
r
ain
in
g
a
n
d
v
alid
atio
n
lo
s
s
f
u
n
ctio
n
s
an
d
p
e
r
f
o
r
m
an
ce
m
etr
ics o
f
th
e
YOL
O
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
ac
r
o
s
s
3
0
0
ep
o
ch
s
Fig
u
r
e
4
illu
s
tr
ates
th
e
tr
ain
i
n
g
an
d
v
alid
atio
n
b
eh
av
io
r
o
f
th
e
YOL
O
-
b
ased
s
eg
m
en
tatio
n
m
o
d
el
o
v
er
3
0
0
ep
o
ch
s
.
T
h
e
lo
s
s
f
u
n
ctio
n
s
(
b
o
x
,
s
eg
m
en
tatio
n
,
class
if
icatio
n
,
an
d
d
is
tr
ib
u
tio
n
f
o
ca
l
lo
s
s
)
co
n
s
is
ten
tly
d
ec
r
ea
s
ed
th
r
o
u
g
h
o
u
t
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
with
a
r
a
p
id
r
e
d
u
ctio
n
d
u
r
in
g
t
h
e
f
ir
s
t
5
0
–
8
0
e
p
o
ch
s
f
o
llo
wed
b
y
g
r
ad
u
al
co
n
v
er
g
en
ce
.
I
m
p
o
r
tan
tly
,
th
e
tr
ain
in
g
an
d
v
alid
atio
n
cu
r
v
es
r
em
a
in
clo
s
ely
alig
n
ed
,
in
d
icatin
g
g
o
o
d
g
en
er
aliza
ti
o
n
with
o
u
t
o
v
er
f
itti
n
g
.
T
h
e
p
er
f
o
r
m
an
ce
m
et
r
ics
s
h
o
w
th
at
b
o
u
n
d
in
g
b
o
x
p
r
ec
is
io
n
an
d
r
ec
all
s
tab
ilized
at
ap
p
r
o
x
im
ately
0
.
9
3
–
0
.
9
5
,
wh
ile
s
eg
m
en
tatio
n
p
r
ec
is
io
n
an
d
r
ec
all
r
ea
ch
ed
s
im
ilar
lev
els
with
o
n
ly
m
in
o
r
f
lu
ctu
atio
n
s
.
T
h
e
m
AP
cu
r
v
e
s
f
u
r
th
er
d
em
o
n
s
tr
ate
s
tr
o
n
g
m
o
d
el
p
er
f
o
r
m
an
ce
,
with
b
o
u
n
d
in
g
b
o
x
m
AP@
0
.
5
ap
p
r
o
ac
h
in
g
0
.
9
6
a
n
d
m
AP@
0
.
5
–
0
.
9
5
s
tab
ilizin
g
n
ea
r
0
.
7
9
.
I
n
co
n
tr
ast,
m
ask
m
AP@
0
.
5
r
ea
ch
ed
~0
.
9
6
b
u
t
d
r
o
p
p
ed
to
~0
.
7
0
at
s
tr
icter
I
o
U
th
r
esh
o
ld
s
(
m
AP@
0
.
5
–
0
.
9
5
)
.
T
h
is
g
ap
h
ig
h
lig
h
ts
th
at
wh
ile
o
b
ject
lo
ca
lizatio
n
is
h
ig
h
ly
ac
cu
r
ate,
c
ap
tu
r
in
g
p
r
ec
is
e
s
eg
m
en
tatio
n
b
o
u
n
d
ar
ies is
m
o
r
e
ch
allen
g
in
g
.
Ov
er
all,
th
e
cu
r
v
es
co
n
f
ir
m
s
tab
le
c
o
n
v
er
g
en
c
e,
ef
f
icien
t
o
p
tim
izatio
n
,
a
n
d
ex
ce
llen
t
d
etec
tio
n
p
er
f
o
r
m
an
ce
,
with
s
eg
m
en
tati
o
n
ac
cu
r
ac
y
lim
ited
p
r
im
ar
ily
b
y
t
h
e
p
r
ec
is
io
n
o
f
f
in
e
b
o
u
n
d
ar
ies.
T
h
e
tr
ai
n
in
g
an
d
v
alid
atio
n
lo
s
s
cu
r
v
es
d
e
m
o
n
s
tr
ate
a
co
n
s
is
ten
t
an
d
g
r
ad
u
al
d
ec
lin
e
ac
r
o
s
s
all
co
m
p
o
n
en
ts
,
in
clu
d
in
g
b
o
u
n
d
in
g
b
o
x
r
e
g
r
ess
io
n
,
s
eg
m
en
tatio
n
,
class
if
icatio
n
,
a
n
d
d
is
tr
ib
u
tio
n
f
o
ca
l
lo
s
s
.
A
r
ap
i
d
r
ed
u
ctio
n
in
lo
s
s
was
o
b
s
er
v
ed
d
u
r
in
g
th
e
in
itial
5
0
–
8
0
e
p
o
ch
s
,
f
o
llo
wed
b
y
a
s
tab
le
co
n
v
er
g
e
n
ce
p
h
ase
ar
o
u
n
d
ep
o
ch
1
5
0
.
I
m
p
o
r
ta
n
tly
,
t
h
e
tr
ain
i
n
g
an
d
v
alid
atio
n
c
u
r
v
es
r
em
ain
ed
cl
o
s
ely
alig
n
ed
th
r
o
u
g
h
o
u
t
th
e
p
r
o
ce
s
s
,
in
d
icatin
g
th
at
th
e
m
o
d
el
ac
h
iev
e
d
s
tab
le
lear
n
in
g
with
o
u
t
s
ig
n
s
o
f
o
v
e
r
f
itti
n
g
.
T
h
is
co
n
v
er
g
e
n
ce
b
eh
av
io
r
s
u
g
g
ests
th
at
th
e
tr
ain
in
g
p
r
o
ce
d
u
r
e
was
e
f
f
icien
t
an
d
t
h
at
th
e
m
o
d
el
r
ea
ch
ed
a
n
o
p
tim
al
b
alan
ce
b
etwe
en
f
itti
n
g
th
e
tr
ain
in
g
d
ata
a
n
d
g
e
n
er
alizin
g
to
u
n
s
ee
n
ca
s
es.
T
h
e
tr
ain
in
g
b
eh
av
io
r
o
b
s
er
v
ed
in
th
is
s
tu
d
y
is
co
n
s
is
ten
t
with
r
ep
o
r
ts
f
r
o
m
p
r
io
r
wo
r
k
s
in
d
en
tal
im
ag
in
g
u
s
in
g
d
ee
p
lear
n
in
g
m
o
d
els.
Kar
ao
g
lu
et
a
l.
[
2
4
]
ap
p
lied
a
C
NN
-
b
ased
m
o
d
el
f
o
r
a
u
to
m
atic
to
o
th
d
etec
tio
n
an
d
n
u
m
b
er
i
n
g
o
n
p
an
o
r
am
ic
r
a
d
io
g
r
a
p
h
s
,
r
ep
o
r
tin
g
s
tab
le
co
n
v
e
r
g
en
ce
af
ter
ap
p
r
o
x
im
ately
2
0
0
ep
o
c
h
s
,
with
p
r
ec
is
io
n
an
d
r
ec
all
v
alu
es
o
f
0
.
9
0
–
0
.
9
2
.
Similar
ly
,
R
ašić
et
a
l.
[
2
6
]
r
ep
o
r
ted
th
at
YOL
Ov
8
ac
h
iev
ed
a
r
a
p
id
r
e
d
u
ctio
n
in
tr
ain
in
g
lo
s
s
d
u
r
in
g
th
e
ea
r
ly
ep
o
ch
s
,
with
f
in
al
m
AP@
0
.
5
v
alu
es
ar
o
u
n
d
0
.
9
5
f
o
r
m
an
d
i
b
u
lar
lesi
o
n
d
etec
tio
n
,
alig
n
in
g
clo
s
ely
with
th
e
p
r
esen
t
b
o
u
n
d
i
n
g
b
o
x
r
esu
lts
(
m
AP@
0
.
5
≈
0
.
9
6
)
.
Ho
wev
er
,
as
in
o
u
r
s
tu
d
y
,
t
h
ey
o
b
s
er
v
e
d
a
p
er
f
o
r
m
a
n
ce
d
r
o
p
wh
e
n
ev
alu
ated
at
s
tr
i
cter
I
o
U
th
r
esh
o
ld
s
(
m
AP@
0
.
5
–
0
.
9
5
,
≈
0
.
7
5
–
0
.
8
0
)
,
h
ig
h
lig
h
tin
g
th
e
ch
allen
g
e
o
f
f
in
e
b
o
u
n
d
ar
y
d
elin
ea
tio
n
in
co
m
p
le
x
r
ad
io
g
r
a
p
h
ic
im
a
g
es.
C
o
m
p
ar
ed
to
ea
r
lier
YOL
Ov
8
s
tu
d
ie
s
o
n
p
a
n
o
r
a
m
ic
to
o
t
h
lan
d
m
ar
k
d
etec
tio
n
[
3
2
]
,
wh
ich
ty
p
ically
r
ep
o
r
ted
m
AP@
0
.
5
v
alu
es
b
etwe
en
0
.
9
0
an
d
0
.
9
3
,
th
e
YOL
Ov
8
m
-
s
eg
m
o
d
el
u
s
ed
in
t
h
is
s
tu
d
y
d
em
o
n
s
tr
ates
im
p
r
o
v
ed
d
etec
tio
n
ac
cu
r
ac
y
an
d
m
o
r
e
s
tab
le
co
n
v
er
g
en
ce
.
Nev
e
r
th
eless
,
th
e
r
ed
u
ce
d
s
eg
m
en
tatio
n
p
e
r
f
o
r
m
an
ce
at
h
ig
h
I
o
U
th
r
esh
o
ld
s
(
~
0
.
7
0
f
o
r
m
AP@
0
.
5
–
0
.
9
5
)
r
e
m
ain
s
a
lim
itatio
n
,
as
also
n
o
ted
b
y
L
ee
et
a
l.
[
2
2
]
in
U
-
N
et
-
b
ased
s
eg
m
en
tatio
n
o
f
p
a
n
o
r
am
ic
r
ad
io
g
r
ap
h
s
,
wh
er
e
b
o
u
n
d
a
r
y
a
m
b
ig
u
ity
in
p
o
s
ter
io
r
teeth
lim
ited
f
in
e
-
g
r
ain
ed
ac
c
u
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
YOLO
-
b
a
s
ed
d
ee
p
lea
r
n
in
g
fo
r
to
o
th
d
etec
tio
n
,
s
eg
me
n
ta
tio
n
,
a
n
d
n
u
m
b
erin
g
in
…
(
S
r
i O
kta
mu
lia
n
i
)
3597
3.
2
.
Q
ua
ntit
a
t
iv
e
perf
o
rma
nce
m
et
rics
T
h
e
o
v
er
all
ev
al
u
atio
n
m
etr
ics
co
n
f
i
r
m
ed
th
e
r
o
b
u
s
tn
ess
o
f
t
h
e
YOL
O
-
b
ased
s
eg
m
e
n
tatio
n
m
o
d
el
as
s
h
o
wn
in
T
ab
le
2
.
B
o
u
n
d
in
g
b
o
x
p
r
ec
is
io
n
an
d
r
ec
all
ac
h
iev
ed
v
alu
es
o
f
0
.
9
0
7
5
a
n
d
0
.
9
3
5
2
,
r
esp
ec
tiv
ely
,
wh
ile
s
eg
m
en
tatio
n
p
r
ec
is
io
n
an
d
r
ec
all
r
ea
ch
ed
0
.
9
0
7
8
an
d
0
.
9
3
4
4
.
T
h
e
co
r
r
esp
o
n
d
i
n
g
F1
-
s
co
r
es
wer
e
0
.
9
2
1
2
(
b
o
u
n
d
in
g
b
o
x
)
an
d
0
.
9
2
0
9
(
m
ask
)
,
r
e
f
lectin
g
a
s
tr
o
n
g
b
ala
n
ce
b
etwe
en
p
r
ec
is
io
n
an
d
r
ec
all
f
o
r
b
o
th
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
ta
s
k
s
.
T
h
e
m
AP
p
r
o
v
i
d
ed
f
u
r
th
er
i
n
s
ig
h
ts
in
to
m
o
d
el
p
e
r
f
o
r
m
a
n
ce
.
At
an
I
o
U
th
r
esh
o
ld
o
f
0
.
5
,
th
e
b
o
u
n
d
in
g
b
o
x
m
AP
was
0
.
9
5
1
0
,
wh
ile
th
e
m
ask
m
AP
r
ea
c
h
ed
0
.
9
5
0
4
,
co
n
f
ir
m
in
g
e
x
ce
lle
n
t
o
v
er
all
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
ca
p
a
b
ilit
y
.
W
h
en
ev
alu
ated
ac
r
o
s
s
s
tr
icter
th
r
esh
o
ld
s
(
m
AP@
0
.
5
:0
.
9
5
)
,
p
e
r
f
o
r
m
an
ce
d
ec
lin
ed
to
0
.
7
7
6
3
f
o
r
b
o
u
n
d
in
g
b
o
x
es
an
d
0
.
6
8
9
6
f
o
r
m
a
s
k
s
,
in
d
icatin
g
th
at
p
r
ec
is
e
b
o
u
n
d
a
r
y
alig
n
m
en
t
r
em
ain
s
m
o
r
e
c
h
allen
g
in
g
f
o
r
s
eg
m
en
tatio
n
co
m
p
ar
ed
t
o
o
b
j
ec
t d
etec
tio
n
.
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
o
f
d
ee
p
l
ea
r
n
in
g
a
p
p
r
o
ac
h
es f
o
r
d
en
tal
im
ag
e
an
aly
s
is
S
t
u
d
y
M
e
t
h
o
d
D
a
t
a
s
e
t
(
i
mag
e
s)
Ta
sk
R
e
p
o
r
t
e
d
P
e
r
f
o
r
ma
n
c
e
R
e
mar
k
s
Le
e
e
t
a
l
.
[
2
2
]
C
N
N
(
F
C
N
-
b
a
se
d
)
1
,
0
0
0
p
a
n
o
r
a
mi
c
s
To
o
t
h
seg
m
e
n
t
a
t
i
o
n
mI
o
U
≈
0
.
8
2
Ef
f
e
c
t
i
v
e
b
u
t
c
o
m
p
u
t
a
t
i
o
n
a
l
l
y
h
e
a
v
y
,
l
i
m
i
t
e
d
r
e
a
l
-
t
i
m
e
u
se
K
i
m
e
t
a
l
.
[
1
1
]
C
N
N
+
h
e
u
r
i
s
t
i
c
5
0
0
p
a
n
o
r
a
mi
c
s
To
o
t
h
d
e
t
e
c
t
i
o
n
a
n
d
n
u
m
b
e
r
i
n
g
A
c
c
u
r
a
c
y
=
0
.
9
0
R
e
l
i
e
d
p
a
r
t
l
y
o
n
h
e
u
r
i
st
i
c
r
u
l
e
s
,
l
e
ss r
o
b
u
s
t
t
o
i
m
a
g
e
v
a
r
i
a
t
i
o
n
K
a
r
a
o
g
l
u
e
t
a
l
.
[
2
4
]
U
-
N
e
t
+
h
e
u
r
i
st
i
c
6
5
0
p
a
n
o
r
a
mi
c
s
To
o
t
h
d
e
t
e
c
t
i
o
n
a
n
d
n
u
m
b
e
r
i
n
g
A
c
c
u
r
a
c
y
≈
0
.
9
4
H
i
g
h
p
e
r
f
o
r
m
a
n
c
e
b
u
t
r
e
q
u
i
r
e
s
c
o
mp
l
e
x
p
r
e
p
r
o
c
e
ssi
n
g
R
a
ši
ć
e
t
a
l
.
[
2
6
]
Y
O
LO
v
8
2
,
0
0
0
p
a
n
o
r
a
mi
c
s
Le
si
o
n
d
e
t
e
c
t
i
o
n
P
r
e
c
i
s
i
o
n
=
0
.
9
1
8
(
n
o
a
u
g
)
,
0
.
9
5
2
(
a
u
g
)
D
e
mo
n
st
r
a
t
e
d
Y
O
LO
v
8
st
r
e
n
g
t
h
i
n
d
e
n
t
a
l
t
a
s
k
s
Ta
f
a
l
a
e
t
a
l
.
[
3
2
]
Y
O
LO
v
8
4
0
0
c
e
p
h
a
l
o
g
r
a
ms
La
n
d
m
a
r
k
d
e
t
e
c
t
i
o
n
mA
P
@
0
.
5
=
0
.
9
5
C
o
n
f
i
r
me
d
h
i
g
h
a
c
c
u
r
a
c
y
i
n
d
e
n
t
a
l
l
a
n
d
m
a
r
k
t
a
s
k
s
Th
i
s
st
u
d
y
Y
O
LO
v
8
m
-
s
e
g
3
0
2
p
a
n
o
r
a
mi
c
s
(
9
,
0
0
9
a
n
n
o
t
a
t
e
d
t
e
e
t
h
)
To
o
t
h
d
e
t
e
c
t
i
o
n
,
seg
m
e
n
t
a
t
i
o
n
,
n
u
m
b
e
r
i
n
g
B
o
u
n
d
i
n
g
b
o
x
:
p
r
e
c
i
si
o
n
0
.
9
0
7
5
,
r
e
c
a
l
l
0
.
9
3
5
2
,
F1
-
sc
o
r
e
0
.
9
2
1
2
,
mA
P
@
0
.
5
=
0
.
9
5
1
0
,
mA
P
@
0
.
5
:
0
.
9
5
=
0
.
7
7
6
3
;
mask
:
p
r
e
c
i
si
o
n
0
.
9
0
7
8
,
R
e
c
a
l
l
0
.
9
3
4
4
,
F
1
0
.
9
2
0
9
,
mA
P
@
0
.
5
=
0
.
9
5
0
4
,
mA
P
@
0
.
5
:
0
.
9
5
=
0
.
6
8
9
6
O
u
t
p
e
r
f
o
r
ms
mo
st
p
r
i
o
r
w
o
r
k
s
i
n
c
o
m
b
i
n
e
d
d
e
t
e
c
t
i
o
n
+
se
g
me
n
t
a
t
i
o
n
;
seg
m
e
n
t
a
t
i
o
n
mA
P
@
0
.
5
:
0
.
9
5
l
o
w
e
r
(
0
.
6
8
9
6
)
,
e
s
p
e
c
i
a
l
l
y
i
n
p
o
s
t
e
r
i
o
r
t
e
e
t
h
C
las
s
-
wis
e
an
aly
s
is
f
u
r
th
er
h
ig
h
lig
h
ted
t
h
at
ce
r
tain
to
o
th
,
s
u
ch
as
4
5
,
4
8
,
an
d
3
8
,
co
n
s
is
ten
tly
ac
h
iev
ed
h
ig
h
er
co
n
f
id
en
ce
a
n
d
ac
cu
r
ac
y
,
wh
er
ea
s
p
o
s
ter
io
r
teeth
,
in
clu
d
in
g
1
8
,
2
4
,
2
6
,
an
d
2
8
,
e
x
h
ib
ited
s
lig
h
tly
lo
wer
p
er
f
o
r
m
a
n
ce
.
T
h
is
v
ar
iatio
n
m
ay
b
e
attr
ib
u
ted
to
r
ed
u
ce
d
v
is
ib
ilit
y
at
th
e
p
er
ip
h
er
y
o
f
p
an
o
r
am
ic
im
a
g
es,
an
ato
m
ical
o
v
er
lap
,
o
r
lim
ited
s
am
p
le
r
e
p
r
esen
tatio
n
o
f
th
ese
teeth
in
t
h
e
tr
ain
in
g
d
ataset.
T
h
e
b
o
u
n
d
in
g
b
o
x
m
AP@
0
.
5
(
0
.
9
5
1
0
)
is
c
o
m
p
a
r
ab
le
t
o
t
h
e
p
e
r
f
o
r
m
an
ce
r
ep
o
r
ted
[
3
2
]
,
wh
o
ac
h
iev
e
d
a
n
m
AP
o
f
0
.
9
5
u
s
in
g
YOL
Ov
8
f
o
r
ce
p
h
alo
m
etr
ic
lan
d
m
ar
k
d
etec
tio
n
.
Ho
wev
e
r
,
th
e
s
eg
m
en
tatio
n
m
AP@
0
.
5
:0
.
9
5
(
0
.
6
8
9
6
)
is
s
o
m
ewh
at
lo
wer
t
h
an
th
e
r
esu
lts
in
g
en
er
al
o
b
ject
s
eg
m
e
n
tatio
n
task
s
.
T
h
is
d
is
cr
ep
an
cy
h
ig
h
lig
h
ts
th
e
in
cr
ea
s
ed
co
m
p
le
x
ity
o
f
d
e
n
tal
s
eg
m
en
tatio
n
,
wh
e
r
e
f
in
e
an
a
to
m
ical
b
o
u
n
d
ar
ies
an
d
o
v
er
lap
p
in
g
s
tr
u
ctu
r
es in
tr
o
d
u
ce
a
d
d
itio
n
al
ch
allen
g
es c
o
m
p
ar
ed
to
g
e
n
er
al
o
b
ject
d
et
ec
tio
n
.
3.
3
.
Co
nfusi
o
n
m
a
t
rix
a
na
l
y
s
is
T
h
e
co
n
f
u
s
io
n
m
atr
ix
in
Fig
u
r
e
5
p
r
o
v
id
es
f
u
r
th
er
i
n
s
ig
h
ts
in
to
th
e
d
is
tr
ib
u
tio
n
o
f
cla
s
s
if
icatio
n
er
r
o
r
s
ac
r
o
s
s
all
to
o
th
class
es.
T
h
e
s
tr
o
n
g
d
ia
g
o
n
al
d
o
m
in
a
n
ce
in
d
icate
s
th
at
th
e
m
ajo
r
ity
o
f
p
r
ed
ictio
n
s
wer
e
co
r
r
ec
tly
class
if
ied
.
Ho
wev
er
,
a
s
m
all
n
u
m
b
er
o
f
o
f
f
-
d
iag
o
n
al
elem
en
ts
r
ef
lect
m
is
cla
s
s
if
i
ca
tio
n
s
,
p
ar
ticu
lar
ly
am
o
n
g
teeth
in
th
e
p
o
s
ter
io
r
r
eg
io
n
(
class
es
2
4
–
2
8
)
.
T
h
ese
f
in
d
in
g
s
ar
e
co
n
s
is
ten
t
with
th
e
lo
wer
p
r
ec
is
io
n
an
d
r
ec
all
o
b
s
er
v
e
d
f
o
r
th
ese
ca
teg
o
r
ies
in
t
h
e
class
-
wis
e
an
aly
s
is
.
Ad
d
itio
n
ally
,
a
m
in
o
r
d
eg
r
ee
o
f
c
o
n
f
u
s
io
n
was
o
b
s
er
v
ed
with
th
e
b
ac
k
g
r
o
u
n
d
class
,
p
ar
ticu
lar
l
y
f
o
r
tee
th
with
less
d
is
tin
ct
b
o
u
n
d
ar
ie
s
.
No
n
eth
eless
,
th
e
o
v
er
all
s
tr
u
ctu
r
e
o
f
th
e
m
atr
ix
co
n
f
ir
m
s
h
ig
h
class
if
icatio
n
ac
cu
r
ac
y
an
d
m
in
im
al
c
r
o
s
s
-
class
er
r
o
r
s
.
T
h
ese
f
in
d
in
g
s
ar
e
c
o
n
s
is
ten
t
with
p
r
e
v
io
u
s
s
tu
d
ies
th
at
ap
p
lied
d
ee
p
lear
n
in
g
to
p
an
o
r
am
ic
r
ad
io
g
r
a
p
h
s
an
d
r
e
p
o
r
ted
h
ig
h
er
m
is
class
if
icatio
n
r
ates
in
p
o
s
ter
io
r
teeth
d
u
e
to
a
n
ato
m
ical
co
m
p
lex
ity
an
d
o
v
er
lap
p
i
n
g
s
h
ad
o
ws
[
2
2
]
.
Similar
ly
,
YOL
O
-
b
ased
ap
p
r
o
a
ch
es
h
av
e
also
d
em
o
n
s
tr
ated
a
p
er
f
o
r
m
a
n
ce
g
ap
b
etwe
en
an
ter
io
r
an
d
p
o
s
ter
io
r
teeth
,
with
f
r
e
q
u
en
t
c
o
n
f
u
s
io
n
o
cc
u
r
r
in
g
in
m
a
x
illar
y
p
r
e
m
o
lar
s
an
d
m
o
lar
s
wh
er
e
r
o
o
t
o
v
er
la
p
is
c
o
m
m
o
n
[
2
6
]
,
[
3
2
]
.
T
h
e
r
esu
lts
co
r
r
o
b
o
r
ate
t
h
ese
ch
allen
g
es,
r
ein
f
o
r
cin
g
th
at
p
o
s
ter
io
r
teeth
r
em
ain
th
e
m
o
s
t d
if
f
icu
lt
to
class
if
y
ac
cu
r
ately
.
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
9
1
-
3
6
0
2
3598
Fig
u
r
e
5
.
C
o
n
f
u
s
io
n
m
atr
i
x
o
f
th
e
YOL
O
s
eg
m
en
tatio
n
m
o
d
el
3.
4
.
Ana
ly
s
is
o
f
det
ec
t
i
o
n a
nd
s
eg
m
ent
a
t
io
n r
esu
lt
s
Fig
u
r
e
6
p
r
esen
ts
th
e
o
u
tp
u
t
o
f
th
e
YOL
O
s
eg
m
en
tatio
n
m
o
d
el
ap
p
lied
to
a
s
tan
d
ar
d
p
an
o
r
am
i
c
d
en
tal
r
a
d
io
g
r
a
p
h
.
T
h
is
f
i
g
u
r
e
p
r
esen
t
class
lab
els,
co
n
f
i
d
en
ce
s
co
r
es,
an
d
co
l
o
r
ed
s
eg
m
en
tatio
n
m
ask
s
.
Hig
h
p
er
f
o
r
m
a
n
ce
was
ac
h
ie
v
ed
ac
r
o
s
s
m
o
s
t
teeth
,
with
s
lig
h
tly
lo
wer
co
n
f
id
en
ce
in
p
o
s
ter
io
r
r
eg
io
n
s
.
E
ac
h
to
o
th
is
co
r
r
ec
tly
id
en
tifie
d
an
d
lab
eled
with
its
co
r
r
esp
o
n
d
i
n
g
class
n
u
m
b
er
,
ac
co
m
p
an
ied
b
y
co
n
f
id
en
ce
s
co
r
es
r
an
g
in
g
f
r
o
m
0
.
8
7
–
0
.
9
7
.
T
h
e
b
o
u
n
d
in
g
b
o
x
es
(
r
ec
tan
g
u
lar
o
u
tlin
es)
an
d
s
eg
m
en
tatio
n
m
ask
s
(
co
lo
r
ed
o
v
e
r
lay
s
)
d
e
m
o
n
s
tr
ate
p
r
ec
is
e
lo
ca
lizatio
n
o
f
in
d
iv
id
u
al
teeth
ac
r
o
s
s
b
o
th
th
e
m
ax
illar
y
an
d
m
an
d
ib
u
lar
ar
ch
es.
Fig
u
r
e
6
.
E
x
am
p
le
o
f
YOL
O
-
b
ased
to
o
th
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
o
n
a
p
a
n
o
r
am
ic
d
en
tal
r
ad
io
g
r
a
p
h
Mo
s
t
d
etec
tio
n
s
ex
h
ib
it
h
ig
h
co
n
f
id
en
ce
(
>
0
.
9
0
)
,
r
ef
lec
tin
g
s
tr
o
n
g
m
o
d
el
r
eliab
ilit
y
in
to
o
th
r
ec
o
g
n
itio
n
.
Fo
r
ex
am
p
le,
te
eth
4
2
,
4
5
,
4
8
,
an
d
3
7
ac
h
i
ev
ed
co
n
f
id
en
ce
s
co
r
es
b
etw
ee
n
0
.
9
4
an
d
0
.
9
7
,
in
d
icatin
g
ex
ce
llen
t
a
g
r
ee
m
e
n
t
b
etwe
en
p
r
e
d
icted
an
d
tr
u
e
b
o
u
n
d
ar
ies.
L
o
wer
co
n
f
i
d
e
n
ce
v
alu
es,
s
u
ch
as
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
YOLO
-
b
a
s
ed
d
ee
p
lea
r
n
in
g
fo
r
to
o
th
d
etec
tio
n
,
s
eg
me
n
ta
tio
n
,
a
n
d
n
u
m
b
erin
g
in
…
(
S
r
i O
kta
mu
lia
n
i
)
3599
to
o
th
1
8
(
0
.
8
7
)
a
n
d
to
o
th
2
8
(
0
.
8
8
)
,
co
r
r
esp
o
n
d
to
r
e
g
io
n
s
wh
er
e
v
is
ib
ilit
y
is
r
ed
u
ce
d
d
u
e
to
a
n
ato
m
ical
o
v
er
lap
,
p
o
s
itio
n
in
g
,
o
r
im
ag
e
co
n
tr
ast,
h
ig
h
lig
h
tin
g
ar
ea
s
wh
er
e
th
e
m
o
d
el
f
ac
es
g
r
ea
ter
c
h
allen
g
es.
Desp
ite
th
e
d
en
s
e
ar
r
a
n
g
em
en
t
o
f
teet
h
an
d
o
v
er
lap
p
in
g
b
o
u
n
d
in
g
b
o
x
es
in
th
e
ce
n
tr
al
r
eg
io
n
,
t
h
e
m
o
d
el
s
u
cc
ess
f
u
lly
d
is
tin
g
u
is
h
es
in
d
iv
id
u
al
s
tr
u
c
tu
r
es
with
o
u
t
s
ig
n
i
f
ican
t
m
is
class
if
icatio
n
.
T
h
is
s
u
g
g
ests
r
o
b
u
s
t
ca
p
a
b
ilit
y
in
h
an
d
lin
g
co
m
p
lex
a
n
ato
m
ical
p
r
esen
tatio
n
s
ty
p
ical
o
f
p
an
o
r
am
ic
im
ag
i
n
g
.
H
o
wev
er
,
o
cc
asio
n
al
o
v
er
lap
s
b
etwe
en
n
eig
h
b
o
r
in
g
teeth
(
e
.
g
.
,
teeth
2
1
–
2
6
an
d
3
3
–
3
6
)
m
ay
in
tr
o
d
u
ce
am
b
ig
u
ity
an
d
s
h
o
u
ld
b
e
f
u
r
th
e
r
r
ef
in
ed
th
r
o
u
g
h
im
p
r
o
v
e
d
b
o
u
n
d
ar
y
-
awa
r
e
s
eg
m
en
tatio
n
m
e
th
o
d
s
.
Fig
u
r
e
6
is
co
n
s
is
ten
t
with
p
r
ev
io
u
s
s
tu
d
ies
ap
p
l
y
in
g
d
ee
p
lear
n
in
g
m
o
d
els
to
p
a
n
o
r
a
m
ic
d
en
tal
im
ag
in
g
.
E
ar
lier
r
esear
ch
u
s
in
g
a
U
-
Net
-
b
ased
s
eg
m
en
tatio
n
ap
p
r
o
ac
h
d
e
m
o
n
s
tr
a
ted
th
at
p
o
s
ter
io
r
r
eg
io
n
s
o
f
ten
ex
h
ib
ited
b
o
u
n
d
ar
y
am
b
ig
u
ities
,
lead
in
g
t
o
u
n
d
er
-
o
r
o
v
er
-
s
eg
m
e
n
tatio
n
[
2
2
]
.
L
ik
ewise,
YOL
Ov
8
-
b
ased
s
tu
d
ies
r
ep
o
r
t
ed
h
ig
h
-
co
n
f
id
e
n
ce
d
etec
tio
n
s
ac
r
o
s
s
m
o
s
t
tee
th
,
alth
o
u
g
h
p
o
s
ter
io
r
lan
d
m
ar
k
s
g
en
er
ally
s
h
o
we
d
s
lig
h
tly
lo
w
er
p
r
ec
is
io
n
d
u
e
t
o
an
ato
m
ical
co
m
p
lex
ity
[
3
2
]
.
An
o
th
er
s
tu
d
y
o
n
YOL
Ov
8
f
o
r
m
an
d
ib
u
lar
lesi
o
n
d
etec
tio
n
h
ig
h
lig
h
ted
th
at
ir
r
eg
u
lar
o
r
lo
w
-
co
n
tr
ast
r
eg
io
n
s
r
em
ai
n
ed
s
u
s
ce
p
tib
le
to
r
ed
u
ce
d
co
n
f
id
en
ce
s
co
r
es
d
e
s
p
ite
th
e
m
o
d
el’
s
o
v
er
all
s
tr
o
n
g
p
er
f
o
r
m
an
ce
[
2
6
]
.
Ov
er
all,
Fig
u
r
e
6
illu
s
tr
ates
th
e
m
o
d
el’
s
ca
p
ac
ity
f
o
r
h
i
g
h
-
ac
cu
r
ac
y
d
e
n
tal
lan
d
m
ar
k
id
en
tific
atio
n
an
d
s
eg
m
en
tati
o
n
,
co
n
f
ir
m
in
g
its
p
o
ten
tial f
o
r
au
to
m
ated
to
o
th
n
u
m
b
er
i
n
g
,
clin
ical
ass
ess
m
en
t,
an
d
d
ec
is
io
n
s
u
p
p
o
r
t in
d
e
n
tal
im
ag
in
g
.
3.
5
.
Clini
ca
l a
nd
pra
ct
ica
l i
m
pli
ca
t
io
ns
T
h
e
r
esu
lts
s
h
o
w
th
e
p
o
ten
ti
al
o
f
YOL
O
-
b
ased
s
eg
m
e
n
tatio
n
f
o
r
au
to
m
ate
d
to
o
th
d
et
ec
tio
n
an
d
n
u
m
b
er
i
n
g
in
p
an
o
r
am
ic
d
en
t
al
r
ad
io
g
r
ap
h
s
.
Hig
h
ac
cu
r
ac
y
ac
r
o
s
s
m
o
s
t
class
es
s
u
g
g
ests
th
at
th
e
m
o
d
el
ca
n
s
er
v
e
as
a
r
eliab
le
to
o
l
f
o
r
cl
in
ical
ap
p
licatio
n
s
,
i
n
clu
d
in
g
o
r
th
o
d
o
n
tic
ass
ess
m
en
t,
tr
ea
t
m
en
t
p
lan
n
in
g
,
an
d
f
o
r
en
s
ic
id
en
tific
atio
n
.
I
m
p
o
r
t
an
tly
,
th
e
m
o
d
el
also
d
em
o
n
s
tr
ates h
ig
h
co
m
p
u
tatio
n
al
ef
f
ici
en
cy
.
T
r
ain
in
g
was
co
m
p
leted
in
0
.
9
4
1
h
o
u
r
s
f
o
r
3
0
0
e
p
o
ch
s
,
r
e
f
lectin
g
r
a
p
id
co
n
v
e
r
g
en
ce
c
o
m
p
ar
e
d
to
m
an
y
C
NN
-
b
ased
f
r
am
ewo
r
k
s
.
Du
r
in
g
i
n
f
er
en
ce
,
th
e
m
o
d
el
ac
h
iev
ed
an
av
e
r
a
g
e
p
r
o
ce
s
s
in
g
tim
e
o
f
0
.
2
m
s
f
o
r
p
r
ep
r
o
ce
s
s
in
g
,
3
.
0
m
s
f
o
r
in
f
er
e
n
ce
,
a
n
d
4
.
9
m
s
f
o
r
p
o
s
t
-
p
r
o
ce
s
s
in
g
p
er
im
ag
e,
co
n
f
ir
m
in
g
its
ca
p
a
b
ilit
y
f
o
r
n
ea
r
r
ea
l
-
tim
e
p
er
f
o
r
m
an
ce
in
clin
ical
s
ettin
g
s
.
Su
ch
ef
f
icien
cy
is
p
ar
ticu
lar
ly
v
alu
ab
le
in
b
u
s
y
d
e
n
tal
p
r
ac
tices,
wh
er
e
f
ast
tu
r
n
ar
o
u
n
d
o
f
r
a
d
io
g
r
ap
h
ic
a
n
aly
s
is
ca
n
im
p
r
o
v
e
wo
r
k
f
lo
w
an
d
d
ec
is
io
n
-
m
ak
in
g
.
Pre
v
io
u
s
clin
ical
ev
alu
atio
n
s
o
f
AI
in
d
e
n
tal
im
ag
in
g
h
av
e
alr
ea
d
y
s
h
o
w
n
th
e
im
p
o
r
tan
ce
o
f
r
o
b
u
s
t
au
to
m
ated
s
y
s
tem
s
f
o
r
d
ia
g
n
o
s
is
an
d
tr
ea
tm
en
t
p
lan
n
in
g
[
1
1
]
,
[
2
2
]
,
[
2
4
]
,
[
2
6
]
,
[
3
2
]
.
T
h
e
r
esu
lts
ex
ten
d
th
ese
f
in
d
in
g
s
b
y
d
em
o
n
s
tr
atin
g
th
at
YOL
O
-
b
ased
s
eg
m
en
tatio
n
ac
h
iev
es
co
m
p
ar
a
b
le
ac
cu
r
ac
y
to
ea
r
lie
r
C
NN
-
b
ased
m
eth
o
d
s
,
wh
ile
o
f
f
er
in
g
f
aster
in
f
e
r
en
ce
a
n
d
p
o
ten
tial
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
.
Ho
wev
er
,
lik
e
p
r
io
r
s
tu
d
ies,
we
co
n
f
i
r
m
th
at
im
ag
e
q
u
ality
an
d
an
ato
m
ica
l
co
m
p
lex
ity
at
th
e
p
er
ip
h
e
r
y
r
em
ain
s
ig
n
if
ican
t
lim
itatio
n
s
th
at
n
ee
d
to
b
e
ad
d
r
ess
ed
in
f
u
tu
r
e
r
esear
ch
.
Desp
ite
th
ese
p
r
o
m
is
in
g
r
esu
lts
,
th
is
s
tu
d
y
h
as sev
er
al
m
eth
o
d
o
l
o
g
ical
lim
itatio
n
s
th
at
s
h
o
u
ld
b
e
co
n
s
id
er
e
d
wh
e
n
in
ter
p
r
etin
g
th
e
f
in
d
in
g
s
.
I
n
p
ar
ticu
lar
,
m
o
d
el
p
er
f
o
r
m
an
ce
was
ev
alu
ate
d
u
s
in
g
a
s
in
g
le
f
ix
e
d
tr
ain
–
v
ali
d
atio
n
–
test
s
p
lit
r
ath
er
th
a
n
i
m
ag
e
-
lev
el
k
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
o
r
r
ep
ea
ted
m
u
lti
-
s
ee
d
e
x
p
er
im
e
n
ts
.
W
h
ile
th
is
s
tr
ateg
y
was
a
d
o
p
t
ed
to
m
ain
tain
a
n
in
d
ep
en
d
en
t
test
s
et
an
d
alig
n
with
co
m
m
o
n
p
r
ac
tice
i
n
YO
L
O
-
b
ased
s
eg
m
en
tatio
n
f
r
am
e
wo
r
k
s
,
it
m
ay
lim
it
th
e
r
o
b
u
s
tn
ess
o
f
p
er
f
o
r
m
an
c
e
esti
m
atio
n
f
o
r
r
elativ
ely
s
m
all
d
atasets
.
Fu
tu
r
e
s
tu
d
ies
s
h
o
u
ld
in
co
r
p
o
r
ate
k
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
o
r
r
ep
ea
ted
r
an
d
o
m
s
p
lits
to
q
u
an
ti
f
y
p
e
r
f
o
r
m
an
ce
v
ar
iab
ilit
y
ac
r
o
s
s
d
if
f
er
en
t
d
ata
p
ar
titi
o
n
s
an
d
p
r
o
v
i
d
e
m
o
r
e
s
tatis
tically
r
o
b
u
s
t e
s
tim
ates o
f
clin
ical
r
eliab
ilit
y
.
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
d
em
o
n
s
tr
ated
th
e
ef
f
ec
tiv
en
ess
o
f
a
YOL
Ov
8
-
b
ased
d
ee
p
lear
n
in
g
m
o
d
el
f
o
r
au
to
m
ated
to
o
th
d
etec
tio
n
,
s
eg
m
en
tatio
n
,
an
d
n
u
m
b
er
i
n
g
i
n
p
an
o
r
a
m
ic
d
en
tal
r
ad
io
g
r
ap
h
s
.
T
h
e
m
o
d
el
ac
h
ie
v
ed
h
ig
h
ac
cu
r
ac
y
,
with
b
o
u
n
d
in
g
b
o
x
m
etr
ics
o
f
p
r
ec
is
io
n
0
.
9
0
7
5
,
r
ec
all
0
.
9
3
5
2
,
F1
-
s
co
r
e
0
.
9
2
1
2
,
an
d
m
AP@
0
.
5
o
f
0
.
9
5
1
0
,
wh
ile
s
eg
m
e
n
tatio
n
p
er
f
o
r
m
a
n
ce
was
co
m
p
a
r
ab
le
(
p
r
ec
is
io
n
0
.
9
0
7
8
,
r
ec
all
0
.
9
3
4
4
,
F1
-
s
co
r
e
0
.
9
2
0
9
,
an
d
m
AP@
0
.
5
o
f
0
.
9
5
0
4
)
.
Al
th
o
u
g
h
p
er
f
o
r
m
an
ce
d
ec
lin
e
d
at
s
tr
icter
th
r
esh
o
ld
s
(
m
AP
@
0
.
5
:0
.
9
5
=0
.
7
7
6
3
f
o
r
b
o
u
n
d
in
g
b
o
x
es
an
d
0
.
6
8
9
6
f
o
r
m
ask
s
)
,
th
e
o
v
er
all
r
esu
lts
co
n
f
ir
m
r
o
b
u
s
t
l
o
ca
lizatio
n
an
d
s
eg
m
en
tatio
n
ca
p
ab
ilit
ies.
C
la
s
s
-
wi
s
e
an
d
q
u
alitativ
e
an
aly
s
es
f
u
r
th
er
c
o
n
f
ir
m
e
d
th
e
m
o
d
el’
s
r
eliab
i
lity
in
ce
n
tr
al
an
d
an
ter
io
r
teeth
,
w
h
ile
h
ig
h
lig
h
tin
g
ch
allen
g
es
in
p
o
s
ter
io
r
teeth
(
1
8
,
2
4
,
2
6
,
an
d
2
8
)
d
u
e
to
o
v
er
lap
p
in
g
an
ato
m
y
a
n
d
r
ed
u
ce
d
im
ag
e
q
u
ality
at
th
e
p
er
ip
h
er
y
.
D
esp
ite
th
ese
lim
itatio
n
s
,
th
e
m
o
d
el
co
n
s
is
ten
tly
ac
h
iev
ed
h
ig
h
co
n
f
i
d
en
ce
p
r
ed
ictio
n
s
(
>0
.
9
0
)
ac
r
o
s
s
m
o
s
t
class
es
an
d
m
ain
tain
ed
c
o
n
s
is
ten
t
n
u
m
b
er
in
g
alig
n
ed
with
th
e
g
r
o
u
n
d
tr
u
t
h
.
T
h
e
f
in
d
in
g
s
d
em
o
n
s
tr
ate
th
e
p
o
ten
tial
o
f
YOL
Ov
8
as
a
f
ast,
ac
cu
r
ate,
an
d
s
ca
lab
le
to
o
l
f
o
r
clin
ical
d
en
tis
tr
y
,
s
u
p
p
o
r
tin
g
ap
p
licatio
n
s
s
u
ch
as
o
r
th
o
d
o
n
tic
ass
ess
m
en
t,
p
r
o
s
th
o
d
o
n
tic
p
lan
n
in
g
,
an
d
f
o
r
en
s
ic
id
e
n
tific
atio
n
wh
ile
im
p
r
o
v
i
n
g
d
iag
n
o
s
tic
ef
f
icien
cy
.
Ho
wev
er
,
th
i
s
s
tu
d
y
was
lim
ited
b
y
th
e
u
s
e
o
f
a
s
in
g
le
tr
ain
–
v
a
lid
atio
n
–
test
s
p
lit
wi
th
o
u
t
cr
o
s
s
-
v
alid
atio
n
,
wh
ich
m
ay
af
f
ec
t
th
e
r
o
b
u
s
tn
ess
o
f
p
er
f
o
r
m
an
ce
ev
al
u
atio
n
.
Fu
tu
r
e
s
tu
d
ies
s
h
o
u
ld
in
co
r
p
o
r
ate
im
ag
e
-
lev
el
cr
o
s
s
-
v
alid
atio
n
,
ex
p
an
d
d
atasets
,
ap
p
ly
tar
g
ete
d
au
g
m
en
tatio
n
an
d
b
o
u
n
d
ar
y
r
ef
in
em
e
n
t
t
ec
h
n
iq
u
es,
an
d
v
alid
ate
t
h
e
m
o
d
el
o
n
lar
g
e
r
m
u
lti
-
in
s
titu
tio
n
al
d
atasets
to
im
p
r
o
v
e
s
eg
m
en
tatio
n
ac
c
u
r
ac
y
an
d
e
n
s
u
r
e
clin
ical
g
e
n
er
aliz
ab
ilit
y
.
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
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
5
9
1
-
3
6
0
2
3600
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
th
o
r
s
in
ce
r
ely
th
a
n
k
s
d
r
g
.
Har
f
in
d
o
Nis
m
al,
Sp
.
B
M,
Dir
ec
to
r
o
f
R
SGM
Un
an
d
,
a
lo
n
g
with
Mr
.
Ad
n
an
,
d
r
g
.
G
u
n
awa
n
,
S
p
.
R
KG,
Mr
.
I
d
h
il
I
h
s
an
,
Am
d
.
Kes.Rad
,
an
d
th
e
R
ad
io
lo
g
y
Dep
ar
tm
en
t
s
taf
f
o
f
R
SGM
Un
an
d
f
o
r
th
eir
v
alu
ab
le
s
u
p
p
o
r
t
an
d
ass
is
tan
ce
t
h
r
o
u
g
h
o
u
t
th
is
s
tu
d
y
.
T
h
e
au
th
o
r
also
g
r
atef
u
ll
y
ac
k
n
o
wled
g
es
Mr
.
R
ico
Ad
r
i
al,
M.
Si,
Dr
.
Af
d
h
al
Mu
ttaq
i
n
,
M.
Si
,
a
n
d
Ms.
Mu
ty
a
Vo
n
n
is
a,
M.
Sc.
f
o
r
t
h
eir
co
n
s
tr
u
ctiv
e
f
ee
d
b
ac
k
a
n
d
in
s
i
g
h
tf
u
l su
g
g
esti
o
n
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
was
f
u
n
d
e
d
b
y
t
h
e
Facu
lty
o
f
Ma
th
e
m
atics
an
d
Natu
r
al
Scien
ce
s
,
Un
iv
er
s
itas
An
d
alas,
u
n
d
er
t
h
e
R
KAT
s
ch
em
e
th
r
o
u
g
h
r
esear
c
h
co
n
t
r
ac
t n
o
.
1
1
/
UN.
1
6
.
0
3
.
D/PP
/FMI
PA/2
0
2
3
,
f
is
ca
l y
ea
r
2
0
2
3
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
i
d
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Sri
Ok
tam
u
lian
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
L
u
q
y
an
a
Ma
h
d
iy
a
h
✓
✓
✓
✓
✓
✓
✓
Har
its
u
l H
aq
✓
✓
✓
✓
✓
Nesa
Per
d
an
a
Pu
tr
i
✓
✓
✓
✓
✓
W
u
lan
d
an
i L
iza
Pu
tr
i
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
I
NF
O
RM
E
D
CO
NS
E
N
T
W
e
h
av
e
o
b
tain
ed
in
f
o
r
m
ed
c
o
n
s
en
t f
r
o
m
all
in
d
iv
id
u
als in
c
lu
d
ed
in
t
h
is
s
tu
d
y
.
E
T
H
I
CAL AP
P
RO
V
AL
E
th
ical
ap
p
r
o
v
al
was
o
b
tain
ed
f
r
o
m
th
e
Facu
lty
o
f
Me
d
icin
e,
Un
iv
e
r
s
itas
An
d
alas
(
No
:
5
1
3
/UN.
1
6
.
2
/KE
P
-
FK/2
0
2
4
)
.
T
h
e
s
tu
d
y
was
co
n
d
u
cte
d
in
ac
co
r
d
an
ce
with
th
e
Dec
l
ar
atio
n
o
f
Helsin
k
i,
g
u
ar
an
teein
g
eth
ical
u
s
e
o
f
h
u
m
an
im
ag
in
g
d
ata
f
o
r
r
esear
ch
p
u
r
p
o
s
es.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
s
u
p
p
o
r
tin
g
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
av
ailab
le
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
a
u
th
o
r
,
[
SO
]
.
Du
e
to
p
atien
t
p
r
iv
ac
y
co
n
s
id
er
atio
n
s
an
d
in
s
titu
tio
n
al
d
ata
g
o
v
er
n
an
ce
p
o
licies,
th
e
f
u
ll
-
r
eso
l
u
tio
n
p
an
o
r
am
ic
r
ad
io
g
r
ap
h
s
ca
n
n
o
t
b
e
m
ad
e
p
u
b
licly
av
ailab
le.
A
s
u
b
s
et
o
f
th
e
d
ataset
s
tr
u
ctu
r
e,
an
n
o
tatio
n
f
o
r
m
at,
an
d
p
r
o
ject
co
n
f
ig
u
r
atio
n
u
s
e
d
in
th
is
s
tu
d
y
is
ac
ce
s
s
ib
le
v
ia
th
e
R
o
b
o
f
lo
w
p
latf
o
r
m
(
h
ttp
s
://u
n
iv
er
s
e.
r
o
b
o
f
lo
w.
co
m
/tu
g
as
-
ak
h
ir
-
r
ly
x
r
/s
eg
m
en
tasi
-
g
ig
i)
to
f
ac
ilit
ate
tr
an
s
p
ar
en
cy
an
d
r
ep
r
o
d
u
cib
ilit
y
.
Acc
ess
to
th
e
co
m
p
lete
d
ataset,
in
clu
d
in
g
t
r
ain
in
g
,
v
alid
atio
n
,
a
n
d
test
s
p
l
its
,
is
r
estricte
d
an
d
m
ay
b
e
g
r
an
ted
u
p
o
n
r
e
q
u
est f
o
r
r
esear
ch
p
u
r
p
o
s
es,
s
u
b
ject
to
ap
p
r
o
v
al
o
f
eth
ics an
d
d
ata
u
s
e
ag
r
ee
m
en
ts
.
RE
F
E
R
E
NC
E
S
[
1
]
H
.
C
.
C
r
o
w
,
E.
P
a
r
k
s,
J.
H
.
C
a
mp
b
e
l
l
,
D
.
S
.
S
t
u
c
k
i
,
a
n
d
J.
D
a
g
g
y
,
“
Th
e
u
t
i
l
i
t
y
o
f
p
a
n
o
r
a
m
i
c
r
a
d
i
o
g
r
a
p
h
y
i
n
t
e
mp
o
r
o
ma
n
d
i
b
u
l
a
r
j
o
i
n
t
a
ssessm
e
n
t
,
”
D
e
n
t
o
m
a
x
i
l
l
o
f
a
c
i
a
l
Ra
d
i
o
l
o
g
y
,
v
o
l
.
3
4
,
n
o
.
2
,
p
p
.
9
1
–
9
5
,
2
0
0
5
,
d
o
i
:
1
0
.
1
2
5
9
/
d
mf
r
/
2
4
8
6
3
5
5
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.