!" # $ % & '" " ( )*+, ' - " . ) " , ,
.0 1... 13 4 20
)
21
(
)
!
MS
"# $ (
MR
&'
( ) * )
1 (
, * - .
) 2 (
) 1 9 : 9 (
–
;< = 1> " ' !) ?@< A 1> " B ;< 1')C A - @< A 1A
( 1( !DE 1A
) 2 @< A (
–
( 1( !DE 1( !DE @< A 1' !) ' ?@< A 1'?9* ' !) B
:# A G 4 / 9 / 1393 :K L G
23 / 11 / 1393
:1234
MN" ) O ? PQ ) & R "
MS
S A ( MR
T ! A + " % U N@ " V+ . ) " *W)
) ' B X ) & Y) % A .# B Z A ) GMM
&YZ " ( & R "
MS S A MR . ) " . 9 A D
!" GMM
@ " ) 'R )
EM ') A D ( ') +, ' [) !" Y\< V " ]^ +&) % ) A 9
A 9
T_ ; < % !" A D ) ` 9 a \< 1 Y\< % A (A B ' , " ')
A .# b< ) A 9
b [) ( * ) M, . ) " % c+, .# 9 A D % #ZA M * d . ) " ' D ) e \) % #ZA M * EM
1 Fast Trimmed-Likelihood
1 A ! @ K '; b #! . 9 A D
= < S 9 "
= < S " A T K " ') ( @< PE = f < .# 9 \) USN ) A A g 9 "
b " A ! @ K ,A
82 " @ R " % Dice
& R ' ^ " #ZA " A A % # " Z 1 MS
MN" A A U N@ . +< "
- * :
MN" " 1 & R MS
' j W) @ S 1
) MRI ( @ " 1 'R )
) EM X ) 1( ' B a N) )
GMM (
.
A New Method for Segmentation of Multiple Sclerosis (MS)
Lesions on Brain MR Images
Simin Jafari(1) – Ali Reza Karimian(2)
(1) MSc. - Department of Electrical Engineering, Najafabad Branch, Islamic Azad University, Najafabad, Esfahan, Iran
(2) Associate Professor - Department of Engineering, Esfahan University, Esfahan, Iran
Automatic segmentation of multiple sclerosis (MS) lesions in brain MRI has been widely investigated in recent years with the goal of helping MS diagnosis and patient follow-up. In this study we applied gaussian mixture model (GMM) to segment MS lesions in MR images. Usually, GMM is optimized using expectation-maximization (EM) algorithm. One of the drawbacks of this optimization method is that, it does not convergence to optimal maximum or minimum. Starting from different initial points and saving best result, is a strategy which is used to reach the near optimal. This approach is time consuming and we used another way to initiate the EM algorithm. Also, FAST- Trimmed Likelihood Estimator (FAST-TLE) algorithm was applied to determine which voxels should be rejected. The automatically segmentation outputs were scored by two specialists and the results show that our method has capability to segment the MS lesions with Dice similarity coefficient (DSC) score of 0.82.
Index Terms: Segmentation, multiple sclerosis, imaging (MRI), expectation-maximization (EM), gaussian
mixture model (GMM), magnetic resonance.
:5678 9 !8 6:
;< = 1')C A - @< A 1 D& % + 1A
)
22
(
1 ; 1 <d 9 '? + " %
k " R = ( Z , A '< < A ;
') ( < 0 ST " ( < O ? PQ ) + " 1A 9
1 )
MS
(
') ( @< \ \[ .# M " ,A
5 / 2 + " A ) ( )
MS
A
A A A ( ! " j P" Z '! A ( ! A X = K B # ] 1 [. o j # " " ( " ' + + " % A
') += (( -) 'bS , b #+ Z N "
' ,
') p S " & R q < " 'j \< A ; k " % ) A .A 9
+ " MS " " = < 'E T *W) -A 1# '+rC P bZ ( \ 1' " 1 ; B ' A & ) < A 1% 1' " A
? 1dY\ ) XC T 'bS s ') , $ b .A 9 A
P= ) A =
1 + " f 1 j " ) !< B @t s * < . < ') A .
+ " q , A U N@ MS
% ( ) A - A !) ') B
') p [) + " % +!) '? .A 9
, U N@
' j W) @ A " S K A D 1 + " %
2
)
MRI
(
A ;= P [ *; " < " " ) .# A A , MR MN" 1 ) ( SSN ) g & R % ' A "
( ) 9A &YZ % .# "
& R A T " MS
A
S MR X A *W) .# B Z A ) T ,
K + MN" ,
& R " MS
') K " ( Souplet
M< ?+, ]
2 [ " *W) O j u bY g " !<- .A 9 1
S T1-w T2-w ( * ) 1 P , u &
3
P ? )
# " " ( &" T A ) 1 D A ) P) 9 *W) N) ,
' N< *W) d ) ')
" A \) % OQ . <A - # A " A 9
@ " A (
'R )
4
) EM . ) " (
@ " # " % " #, b9 (A <- ' B X ) *W) ,
. 9 A D !
MN" A D " OQ "
" < - V 1 B E ,
S FLAIR A " % " O <^ , ) E (- O . 9 X +
P 9
O A B b [) :C , % < ) A " ,
S & R 1 < - \) " \) FLAIR
' 9
. < 9 '< +, < * < A " & R ' 9 A K % t B & R A & '< + " A 1O j A D P A " ) A " u )
# R f < " A " 9 ' - tA *W) A . @< ; ) '@N"
B #Z ) A A = u bY - " K % % c+, M t * < .# Subbanna ( ?+, ] 3 [ " A T ]C) K V 1
' 9
& R MS < t S A MR
MN" S . <A r "
# " & R q B , , A +< N " ' A 9 ,
Z A D A ) *W)
. B
OQ , A +<
q B ,
X ) ,
\bj ) ' A S ( ) g < 9 A % +N ' B "
MN" K % A .# B E *W) ! " & R "
, b<A " ' " # " 9 " ' , YT P) 9 # %?+) ' A
;) " 1 A S b<A A D E A 0 T S ] A
( ?+, 'j T X . < 9 A A K )-]
4 [ " 1
\bj 1 ) ' A S ( )
K V % c+, % * " "
A ( )*+, j " & R *W) ' E # " < 1'\ S FLAIR \bj :C !t " *W) # N< !<- . "
) 1 T A ) 1 D A )
MN" < + Z " ' N< *W) d "
& R OQ <A MS
< + Z " :C a \< ( "
S b<A ! K % A . <A +< ' 9 FLAIR
"
vC ' E A 9 A D & R ' 9 ,
MS ') <
. < 9 , $ N) S A P\ ) j " Boudraa ( ?+, ] 5 [1 A " % < )
c 5 FCM) " ( ' 9 & R MS A D <A . FCM ' " # A " (A -A 9 T " B 9 T 9 X
' N< *W) d ) P) 9 ') & R
9 9 T * < q A D A )
A ) T ') " A
# B . OQ FCM q A " U N@ & R % " X + ' N< *W) d )
A . 9 % K w " % # A = ) X & R q + " , A V 9 T 0 T B " ') < 9 A = ) q A % 9 T " U N@ % " & R # " ] A ;)
MN" " ') 9 .A Sajja ( ?+, ] 6 [1 MN" " & R ' N< *W) d ) "
\bj A D MN" OQ ( "
" D A ) A )
T <A A ! @ V ) K V A D "
" .
% w
' 9 9
D A )
d V T A )
') X b<A ' B " & R ' N< *W) d ) " % " " 1
\bj A D g < + Z " OQ ( "
EM
\bj S ) ' A S ( ) " T2-w
PD-w "
D A ) T A )
MN" < 9 " .
K C?@) " b s " 1 & Y) % A V 9 _ % @ ,
& R U N@ A T K MS
MN" # " " A *W) ,
N) S b<A MR
') ' &) A D " < AA B
# " a \< E " & R A < *W) O j *W) ,
' 9
') T ) A A . < 9 K
A ! @ +, " *
(-y 9 ') A A .A 9 f < '" ; K r 9 ; < B " A MN" , 3 4 _ ') < 9 . 2 $ != > ? 4 ! MS 2 ; 1 ; "# 1 @:$ A :
MR
P?9 A A ! @ K q B A v " )
1 .# 9 A A ( @< (
S b<A p N < = ) % MR
A V , # " Z " "
# " 'T " U N@ ') 0 T & R ,
9 " % A .
S b<A FLAIR
1
T2-w
1
> " #& E A +9 , !9 X
@9 4 +9 V # " 4
!" 1394
)
23
(
) ,* ( + " A " !" ( , Q A " S *
. S
A g MR
( " 5
/
1 K " " 'R +< A C " ' ,
#) NR
mm
5 9 PE = . <
S %
7 A ) + " 18
+ "
% " ' A A ( 11
45 9 ! X . <
2 ; 1 ; ? $ > >
#! = ) % A A b!"
# D MN" #[E M * S 1 "
K A M s yCE S +;+ oL= P) 9 q ^ ,
S A ' 9 9 ' T ?
MR
') E
A B .
P?9 ) 1 MN" #! A ! @ K q B A v " :( & R "
MS
Fig. (1): Block diagram of the proposed method for segmentation of MS lesions
2 ; 2 ; 1 ;
! 4 6! B C32
' 9 9
S MRI A ' T ? s V ')
9 "
') k " A " : " ( ) q - W '9 <
P A 9
" ,
9 " 9 A ' D ) ' 9 9 1'< ?) # &Z ) V % c+, .
S O< % < ) 1' T ? s % b A A W
\bj A ; \ !9 A D q , " # A < " "
') ' 9 A 9 yCE " ; < A A 9 % % B .< A " .
O< = A W ') - " : " ( ) ?< % ,
(A " #
6
O< 1 S ^ " ,
') % " o , O 1A
O< % ( < A B " 1' T ? s { [S A ' !< " ^ " ,
') ' E S 9 "
.
K '? % A
% | ,
LB
7
) LPF A D (
A ] 7 [.
# 9 w K % A ') ( ) %
(A (
" ' E S A A D Y" A b " A - # A
A +< N ' !< S .
u exp log v lpf log v (1 ) (- A
1
u
X " $ ) T {Y
RF ' @B "
') # " 9 " # = A " \) ( +,) S A
(A 9 , @)
v
S A # ' 9 9 \) * < MR
') A @t " A 9
.
') ECT E " K % A P= ) A 9
:
') b [) A T j " < - {Y V ( . ) % " .A 9
9 q B , "
P ' 9
" 9 b [) 1 ,
; Hann ') +, #+ + * ) % OQ .A 9
< - {Y \) 9 A A U N@ q B , % % A \) " 15 / 0 ') A A #b < \) % 9
A.
p
P ' 9 9 % < ) \) b [) (
{Y ' ,
^ " ' 9 . < A < - {Y " #b <
' 9 9 " 9 b [) % < ) \) (A A 0 S T ( P . < A Z < - {Y ' ,
A ( P< V A D " A A ;= P ' 9 9 (A +,
< " A &" A &" ' B
8 3
') A A S A &" 9 "
.
A . ) " ') = ) % &
d P b (
8
)
FFT
') E " > g" .A +< A D ( :A ECT (
[
F(SI (x,y,z))F(G(x,y,z))]
F
) z , y , x ( G ) z , y , x ( SI SI
raw 1
raw smooth
− =
⊗ =
) 2 (
(- A
SI
&" ;= A q T S
G
P<
1 .< A ) ' B
P b
⊗
') %9 < + 9 "
.
( Y" .< A ) ' !< S b [) :
SI i SI "#$$%&!! (3)
(- A
I'())*+
9 +, S
I
S % < )
') 9 " . P?9 A < +< S V " (- f < P= ) )
2 (
# 9 A A ( @< " #b < A " } ) K % A D .
K !" K V ]^ +&) , : " ( ) % "
') A D # &
.
2 ; 2 ; 2 ; 1*D* EFG
' ;<- # " A +;+ P A " # %?+) *W) s ,
MN" A ; k " *W) , T " ' 9 " @ # A < "
' b [) ( ) M, . ) " 1 = ) % A # !" 1 < 9 < 9 oL= !" M * S 1*W) +;+ oL= " .
MR
` < T1-W A " !" A .
+;+ ' 9 9 1 S %
A *W) [) ' < " # D ) *W) a \< " ') . ) % " .A 9
.# 9 A D ~ ) ,
') 1 +;+ P[) % +N V " (
a " ) ,
D •) (A " b t , " ,
A A D S
% .
'9 q + 1 , D •) ,
1 S ) , " (- * ' ,
S b<A p N <
M K A
\bj # " " *W) ,
a \< N
" v ?@) a \<
MS
)
24
(
') D •) V ( " O 1A < A (- (
oL= " .
" S D •) % ( " +;+ P?9
3 4 ( ) ') .
P?9 Y< +, 3
4 ( ) D •) , 1# A ' R ,
% " " . < A ^ " ' 9 9 , *W) o j "
A D
~ ) +
erosion ') (- A = ( oL= ,
A +< P?9) 3 4 R a \< ( c+, 1 + % X + O ) .((A) '
') < A A *W) o j + % " . < 9 oL= # "
~ )
dilation OQ .# 9 A D
a \< (A "
A D (A " a " ) , + 1*W) PT A 9 oL= P?9 A 9 PE = *W) ' !< V ) # 9 A A M +< ( )
.
PE = # !< A S 1 A S A 9 PE = V ) p R
') # A " +;+ ( " P?9)
-3 4 .(( )
2 ; 3 ; 1<@I J ! &' K
*W) d ) T A ) 1 D A ) ' E # " ( < *W) # 9 P ?@ ' N< # " q , ? w " .
A *W) ,
' 9 9 d
' B ') , P\ ) % " " 1 9 "
')
' 9 9 (
# " V , d X ) V g ,
A \ ' B
' B a N) X ) . ) % " .
9
)
(GMM A
# 9 A D \) % ]
8 [. E % " K % A ? +
A & " 1 # n
" # Z A A k
D ) :C
\bj 1 < 9 "
k
' B W ) % < ) ")
UN@) O< (
') & \bj + A 9 ') q ;< j "
X + = A L
d A .< A ) A A A T X + = ) :C V " A A V u & A o , O .A 9 @ " ('SN@) GMM
A \) !" b [) 1
% < ) O<
K
A " , T ' B W ) \bj " .
S "
MR D ) S )
T1-w,T2-w,FLAIR A D (
A , j " ?
10
, S A " @) ? " ] \ ZA
S A P , " 9 ubY )
i
V ) '? * <A A
&" A " V " .#9 A , T ( &" Y\< V )
Y\< &"
x!
A :
x! ,x-. !, x-0 !, x-012 !3 (4)
) P?9 2 1' E S ( .' 9 9 ' T ? s { [S f < :(
P (A ( 1 < - {Y X + (p \) " < - {Y ,
+, (A 1 S % < ) ' !< S ( 1 S ' 9 (A
Fig. (2): Result of intensity inhomogeneity correction, (a) Original image, (b)Automatically determined segmentation of the input image in
to data (white) and background (black), (c) Filling the noise locations with the average image intensity of the non-noise locations, (d)
Smoothing image intensity, (e) Final image
A & B O K
S ? * " M*N 9 A S , A
A & " 1 9 " M*N*K
n= " #9 A , T &" Y\<
A D " GMM
' " :C " \bj
< 9 " % A .
A " a \< # % " w \)
x!
E "
i.i.d
MN
9 , < O< ( * ) " P\ ) ' B d A q
# UN@) % < ) , ) .
GMM UN@) 9 g
θ
"
') ( " Z A X + = ' t d" V E A 9
:
p x!|θ ∑ α=8>. 8. N,x!; μ8, ∑83 (5)
N x!; μ=, ∑= .
0?@A|∑|BAexp C
. 0 x!
μ= -∑=1. x! μ= D (6)
^ " Y" A
k
( @< :C A & ,A 1 ,
j
α
, A T X + =
1:C N d" ' t ' B 1:C , (X ) < d ) j
A "
% < )
∑8
# :C , O< O ) * < O ) .
O< * <
A :C V ' 9 9 d XC\ w ")
') & E " :C V " ( D ) S :A 9
( )
p) ( )
(
) A ( )
> " #& E A +9 , !9 X
@9 4 +9 V # " 4
!" 1394
)
25
(
) P?9 3 ( . +;+ oL= f < :( ) % (p 1' E S
• 1 S D
~ ) + X + O S (A 1 S D •) % oL= (
erosion
' !< S ( 1*W) V ) ( 1
Fig. (3): Result of skull removing, (a) Input image, (b) The first component of the image, (c) Removing the first component of
the image,(d) The processed image after applying erosion operator, (e) Brain mask, (f) Output image
∑ E
σ-.,*0 0 0
0 σ-0,*0 0
0 0 σ-012,*0
H (7)
% < ) T 9 < A \) O< 1 ,
, ) A % @ ^ + =
θ
q sA ') < 9
Iμ8, ∑8, α8J
=
θ
.
K )-GMM b # % +N , ) +;)
θ
"
< +< , '9
)-x!
A ? dZ A . ' E
A K )-X )
% +N 1' B a N) , )
" &) % @ " X + =
') . 9 "
" % . ) % +N
#, b9 @ " B
11
)
MLE
( A D
') ') # A " Y" :
-θK argmax
O L θ argmaxO∏ p x!|θ
R
!>. (8)
(- A
L
d" ') X + = 9 " . b [) "
MLE 1 EM
A D ') ] 9 [ .
EM A " , ) b [) #! ? " ' b) V 1
q B P) 9 ') , 9 "
:
X q B A \) " : A \) " ' A S
θ
') A A 0 S T A 9
.
'R ) q B )
E
A :( P ') + X + = \) q B % " ,
A \) : " ' B d , Y" u j 1 , ) A )
') b [) A 9
:
p!=>∑TSTU VSY.U,VW;XTW;X,∑YT,∑Y3
YZB
(9)
@ " q B M)
)- # A " ^ + = A D " :( " A
% < ) A \) ,
O< ') " % @ ^ + = , < 9
.
A \)
θ
') " g" A D " < 9
:
μ* ∑\WZBVW[WT
∑\WZB[WT
(10)
∑*
= ∑\WZB.[WT] P!=,x! μ=
*3,x
! μ=*3
-R
!>. (11 )
α=* ∑ [WT
\ WZB
R
(12)
^ " g" A
t
( @< ') ? A & ,A 9 "
.
&" q B !) B : E M +, ( 9 " A \) ' !< ? , )
') < 9 . # E % " % A A ? + [< ECT j "
\) V
θ_
\)
θ_`.
< B " ') b [)
A 9
a bc`. d a bc
') ? ' A 9
A \) A 9
θ_ θ_`.
A b " < 9 V A*< , " "
|θ_`. θ_| e ϵ
A 9 .
') +, ' [) !" Y\< V " ]^ +&) 1 % ) t .A 9
D ) A \) p N < "
θ
, T ' D ) f < " ; )
9 " % t ]^ +&) 1' [) !" Y\< A (A B ' , " .
D ) ` 9 a \< < ; < % !" A
') # b< ) < A " ' D ) e 1 \) % A .
#ZA M * d
A D %
# 9 ] 10 [.
2 ; 3 ; 1 ; @# 1 8 1 K$ <
A D 'b ) ',A \) K V . ) % "
9
# . " j " EM
" 100 ' A S ` 9 Y\<
') . < " OQ .A 9
O 1' +, " (
50
') Z ) ?
% !" ' , ) &" = ) A .A 9
') p N < <A " A - # A " ^ + = ( +, # !< A < 9
" ` 9 a \< ( " , ) EM
9 A A
') ' +, " ( ] A ;) .A 9
a \< +;) K % A
P= '& j 9 " A " ` 9 M9 %?+) ,
. ,A M * b [) ;= M, . ) " " \) % A
',A \) K 1# b<A '? " ! 9 _
) ( )
p (
) ( )
A (
) ( )
)
26
(
\bj " # # " "
') A ; *W) '& bj , 1
" % " " GMM
S b<A " ! T1-W
A A q ;< A & " S V > + .
, T M, ',A \) b [) ( ) ) 1A [) ?
A A :C 9 A % +N , ) A D " OQ 1 ,
S A ' " T1-w
\bj S V 1 "
# " , 9
') b [) A 9
&" S A , ) % +N " .
\bj Cj * < "
T1-w A D ') A
S T2-w FLAIR
S < +, T1
'b \ E "
:C " ') < 9 . A , " a " ) :C , " OQ
A & " ') B , 1 S 256
% "
12
% 9 b [)
A < o [< " ' B P< V g q B , 5
+, % "
') P[) " " 1 S A % " :C , % < ) .A 9 % A V
q B ,
A " , T , Y" * < # " , O< \) .
') b [) A 9
:
σs,t2=g1.4918 med,hx
! μ*,'h3i
2
(13) (- A
t
1:C +<
s
S +<
med
% < ) +
# . S A , ) A \) % " T2-w
FLAIR
') A % +N ]CbZ S " a " ) O< % < )) < 9
V+ 1O< % < ) A \) % X = .(# 9 b [) A , T EM
A 9 +, ' !" Y\< " ) .
' ;<-MLE
a \< A E A # : = a \< "
L 1A < '" T A ? + % +N
{ [S X + = B
9
13
(TLE)
') A D ]
11 [.
2 ; 3 ; 2 ; 9 L M"# 5 * G N ) * # (TLE)
K A TLE A A A & # % " w 1 A " 1 ,
.A 9 9 LB ' B W ) , ) b [) q , " %
#" } \) V ]^ +&) A & )
h
A ( ') B .< % ubj O .A 9
!" V * < A
A A q ? ,
') q ;< < 9 9 LB A L
X + = d" X + = d" " .
') A D 9 { [S ') ( " E " (- Y" A 9
:A 9
TL θ ∏R1+!>.p,x !hθ3 (14)
n
P P A & ,
h
P A & ') < 9 oL= ,
9 " .
d"
v i
P * < (- X + = : " , E " ,
A ') " :
p,x .hθ3 d p,x 0hθ3 d ⋯ d p,x Rhθ3 (15)
b [) " TLE
FAST-TLE ') A D
] 11 [ .
') P+ E % " %
% +N A \) " "
(O< % < )) 9 A A +, A T X + =
A b [) , ')
< 9 A \) &" = ) A . ) O<
A A " % < ' ,
b [) < A @ " X + = A A " A ( ")
,
') 9 LB ? = ) A % .A 9
') A A ' < 9 ,
% +N b [) A W € , 9 #" } +;) V . , < ‚ , )
\) b
h
') + 9 %ƒ+Y) A 9 p N < „ *" Z " "
& R MS #?D - :C , 9 oL= ' " ,
< % A .
\) \)
h
A A Z j " 1
10 % P ? P .< A *W) ,
# 9 B ]
12 [ '& bj :C = ) % ' ' T .
( ") MS ') A A A & ( " v ?@) 9 "
MS , .
2 ; 4 ; 9$ $ P > K
9 oL= a \< ' ;<-
h
dZ A # 'j \< P) 9
:C " ) ') ' " ,
9 " a \< { [S " % " " .
') A D O <^ , ) E 1 9 ' 9 " E % .
" A A , , " < B j
') b [) :C A 9
.
„ *" < - \) V E % B ( " A A % 1 9 "
') ' 9 A A
') & E " E % .A 9 :A 9
d! min∀ 8op,x! μ83-∑ ,x1.8 ! μ83q (16)
2 ; 5 ; 1 R6 = S <: MS
> S <: )
A A & R P) 9 1O <^ , ) E g 9 ' 9 ,
MS
#?D - 1> 1
') s * < 1( ,
9 " #! .
P U N@
& R ,
MS
P A
g ' < Z 1 ,
P X S ' 9 9 : " p S *W) ( SSN )
,
A D " # 9 & a \< N " A Z % < Z %
MS
A " , T ' !< :# E " K % N) P= ) .
( ' 9 9 " ' b) ( < Z S A :
T2 FLAIR 9
! ' 9
MS # D A ) ! ^ " = .
P %
P " , ,
Hyper , o &) 9 B % " " .
< - {Y V + > S A A .< A ) a \< ' 9 " v ?@) A A A * < A A % 1 9 " MS
A " , N< .
(p = < < " ' b) ( < Z # % " w 1( < Z % A :
" v ?@) '= < < MS
9 " + = b< .
+ '= <
mm3
9 ) y R " "
14
A = S %9
5 ' 10
A A # (P ? ,
MS ') T < 9 ] 13 [.
( " .< A ) A A - ? " ' b) ( < Z PS ) D A )
T # , " .< A ) A A B :
9 b< PS ) D A )
& R * 1 9 " PS ) *W) " MS
% A @< p [)
') oL= = ) A 9
> " #& E A +9 , !9 X
@9 4 +9 V # " 4
!" 1394
)
27
(
(A ? Cj " ' b) ( < Z +, ,
% % B .< A " :
& R ?< MS
] b \ ') ;= A A " P?9
1 9 "
? Cj
A A D +, , A A B j " .
? V A " v ?@) S
MS ? q € , A A " ,
" v ?@) +,
MS
A A # A A % L 1 9 b< ,
MS
') T A 9 .
( ?9 " ' b) ( < Z P
& R : MS A ]^ +&) P?9 … "
') 9 " " v ?@) '= < % " " . MS
* 9 < ') % t
& R MS '+< p [) < 9
V (A " 'YT ( * ) b [) " .
#! A A .< A ) = < S N) O< = <
x,y
A D
A . A A # !< A ) 'Z " ,
& R ( " 1 < MS
' 9
9 , T .
P?9 A ) 4 (
S A A ! @ P= )
FLAIR M +<
# 9 A A .
3 ; 1 J T : K
# #ZA M * . ) " 1 9 ( " PbZ MN" A j ( +, EM K 1 TLE 9 A D Y" ) ) 14 ( Y" % A .( \) % &
h
# " % +N A }•) M\< % 9 LB *W) ,
A A a \< a \< A & + \) % B t .
1 9 "
') k " :C A @< ' 9 '" T " a \< A 9 ,
\bj ' " \) B < 9 "
h
a \< A & @ " ' T * <
P 9 b 1A 9 p N < :C " u & ) ' ,
,
') ' " < 9 ' 9 a \< ( " b 9 " 1 9 " .
MN" S 1 ) % " !" \) % & . ) " 9 "
MN" S " A ! @ K ) A g 9 "
\) USN
" @ R ( * ) < 9 Dice
15
) DSC \) W " S " (
h
% " 0 5 / 0 )
' +, =
A " % < " f < . 9 b [) (
\) DSC ') M * A & R " " S " "
% A) .
& R P ;= # % A & R " . ) \) B *"
" "
cm3
10 & R P ;= 1 & R " % c+, 9 "
?t
st3
10 .( 9 " A \) % !" DSC
( + " S "
A \) " & R " " 15
/ 0
e h e
1 / 0 " & R " " ( + "
\) A 25 / 0
h=
# A "
.
) P?9 A 5 ( ) }†
h
\) "
DSC ( + " A MS
# 9 A A M +< A & R " " .
) P?9 4 MN" A ! @ K P= ) :( & R "
MS
\bj 1' E S :‡t " # 1^ " A E " a \< ' 9 1*W) # " "
1' 9 LB < - O a \< :‡t " # 1% A .O <^ , ) ' !< % < Z X + O a \<
& R)
MS
(
Fig. (4): Steps of proposed algorithm on a typical image, Top, from right to left: Original image, Brain tissue classification, Detection
of candidate lesions with the Mahalanobis distance,Bottom, from right to left: Outliers after applying intensity rules, Outliers after
applying other heuristic rules (MS lesions).
)
28
(
) P?9 5 ) }† :(
h
\) "
DSC
( + " A
MS
& R " "
A
Fig. (5): Effect of h on DSC factor for MS patients with low and high lesion load
( +, B " A ( +j #! 1 9 ( " M #+ Z A < B
) \) ' " a \< ') + ( 9
h
a \< A & B *"
A A ( ) % A ) 1A 9 B .< A < A A * < ' ,
') (- ( A , :C '? A ' " ,
" A "
< 9 oL= a \< % % " L= .
% & " " a \< % { [E o 1 = % " !" \) 1 B
4 )- # A " .
" v ?@) a \< " MS
{Y V 1 a \< % "
9 A D ' 9 < -'+!) M\< 1 < - ) \) % & .
') D & R { [E ' 9 A " M ) - O
N) S
T2-W FLAIR \) % 130 . )- # A "
MN" f < " (- f < 1 A ! @ K '" . ) " "
,
9 \) USN ) V9* A g ' A
. ) % " .
, &) DSC 1
# =
16
#[E 1
17
'Be
18
g" . 9 A D
) X A 9 b [) A \) 9 _ , &) 1
A A M +< (
.# 9 g" % A
TP P A & A , g # ' ,
1 A T ' A K MS
9 A A U N@ <
. TN A &
P 1K A , g # ' , MS
@< T 9 <
. FP
P A & ! # ' ,
A T K g
FN A &
P
1' A K g ! # ' ,
MS
9 ' 9
< .
Table (1): Statistics of similarity between proposed approach and manual segmentation derived from 25 MS patients
) X 1 MN" " A ! @ #, b9 ( * ) :( "
" USN ) V9* A g ' A , 25
" C b) + "
MS
& R "
Sensitivity TP/ TP v FN Specificity
TN/ TN v FP Accuracy
TN v TP / TN v TP v FN v FP DSC
2TP/ 2TP v FP v FN
0.761 0.998
0.942 0.77
A
0.852 0.996
0.97 0.875
% < ) \)
0.8065 0.997
0.956 0.8225
0 0.2 0.4 0.6 0.8
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
D
S
C
h
MN" #! '9 ) O ? PQ ) & R "
MS
S A (
MR
*W) .0 1 21 4 30
)
29
(
( +, & R " " " 9 PE = f < # UN@) j
') ') # % " % 9 " & R " " ( + " S A 9 "
P '+ A & c< t & R * ,
MS ') 1 9 "
\) < " # A TP
.=C) P" Z j " ') #
' E A
# A & R " '< ) TP
#9 A , T '+ M, .
4 ; UM N 1D :
',A \) K V 1 & Y) % A O j A D ( "
# " X ) A - " . ) " 9 B " *W) ,
',A \) .
A ,
.< ' EM 1 9 T " FCM K-means M\<
') D !<- ' +, A A ',A \) % " @) j " .
K TLE ' [) + * ) " ( p #! * <
# # +, *r = a \< % 9 LB " A ! @ K .
A D
9 A ) A Cj
# " ' 9 ,
% c+, b<A , A
M, ' [) + * ) " ' +, YT 1' A S ',A \) ') ,A . S K % A FLAIR
#! A < P? V ( "
MN" & R " MS
# 9 A D .
" S %
%Y" o j A A ) & R o j ) A " : = " *W) ,
" #b < A S
PD T2 # = *W) ' @ D= & R "
') ( @< + M * " & ) (- ,A
YT FP ') .
&YZ A b!" #! % ` < S "
T2-w " * < (
S P+?) Cj
FLAIR 9 A D
.
A < S " O j u bY " < A ! @ K .
ˆ [
') 9 A D } ) ! Be Cj
" #DB (
') A D ' 9 9 Cj ! 9LB ,
#?D - * < A P A " # < ' ,
] 1 .[ Cj ) L
A A D ' 9 Cj " +, '< ?) MN" A b!" A
" " }•)
') 9 "
.
'Be " % " % c+, ,
" ' b)
? Cj
, +, P?9 & R A ) # B Z A D .
A \) (A - # A " " " A ! @ 1 , ) !"
A \) " 9 D ) & R " " ( + " S DSC
# A " @) & R " " ( + " " ' ^ " .
< D † ) " A P A
1 N)
\) u ZA A ! @ K "
A
!9 9A # . X ? ( \\[) N) ,
A A
D ) A D
A < K % c+, MN" ( B < B ,
"
& R MS P? " ]^ +&) ,
MR !" 'E T
9 < . " % X = A ] 4 [ ] 13 [ " DSC A A = 75 / 0
63 / 0 K *B 9 A K A ! @ ) % % < ) j " )
82 / 0 # M, e " \) A .# M * ' bZ ,
6: >
1. Multiple Sclerosis
2. Magnetic Resonance Imaging 3. Voxel
4. Expectation Maximization 5. Fuzzy C-Means
6. Blurring 7. Low Pass Filter 8. Fast Fourier Transform 9. Gaussian Mixture Model 10. Slice
11. Maximum Likelihood Estimator 12. Bin
13. Trimmed Likelihood Estimator 14. Resolution
15. Dice Similarity Coefficient 16. Sensitivity
17. Accuracy 18. Specificity
References
[1] X. Lladó, A. Oliver, M. Cabezas, J. Freixenet, J.C. Vilanova, A. Quiles, L. Valls, L. Ramió-Torrentà, À. Rovira, "Segmentation of multiple sclerosis lesions in brain MRI: A review of automated approaches", Information Sciences, Vol. 186, pp. 185-164, 2012.
[2] J.C. Souplet, C. Lebrun, N. Ayache, G. Malandain, "An automatic segmentation of T2-FLAIR multiple sclerosis lesions", In The MIDAS Journal-MS Lesion Segmentation (MICCAI 2008 Workshop), 2008.
[3] N. Subbanna, M. Shah, S. Francis, S. Narayanan, D. Collins, D. Arnold, T. Arbel, "MS lesion segmentation using Markov Random Fields", In Proceedings of International Conference on Medical Image Computing and Computer Assisted Intervention, London, UK, 2009.
[4] R. Khayati, M. Vafadust, F .Towhidkhah, M. Nabavi, "Fully automatic segmentation of multiple sclerosis lesions in brain MR FLAIR images using adaptive mixtures method and Markov random field model", Computers in biology and medicine, Vol. 38, pp. 379-390, 2008.
[5] A.O. Boudraa ,S.M.R. Dehak, Y.M. Zhu, C. Pachai, Y.G. Bao, J. Grimaud, "Automated segmentation of multiple sclerosis lesions in multispectral MR imaging using fuzzy clustering", Computers in biology and medicine, Vol. 30, pp. 23-40, 2000.
)
30
(
magnetic resonance imaging", Human Brain Mapping, Vol. 10, pp. 204-211, 2000.
[8] W.M. Wells III, W.E.L. Grimson, R. Kikinis, F.A. Jolesz, "Adaptive segmentation of MRI data", IEEE Trans. on Medical Imaging, Vol. 15, pp. 429-442, 1996.
[9] A.P. Dempster, N.M. Laird, D.B. Rubin, "Maximum likelihood from incomplete data via the EM algorithm", Journal of the Royal Statistical Society. Series B (Methodological), pp .1-38, 1977.
[10] D. García-Lorenzo, S. Prima, D.L. Arnold, D.L. Collins, C. Barillot, "Trimmed-likelihood estimation for focal lesions and tissue segmentation in multisequence MRI for multiple sclerosis", IEEE Trans. on Medical Imaging, pp. 1455-1467, 2011.
[11] N. Neykov, P. Filzmoser, R. Dimova, P. Neytchev, "Robust fitting of mixtures using the trimmed likelihood estimator", Computational Statistics & Data Analysis, Vol. 52, pp. 299-308, 2007.
[12] D. García-Lorenzo, S. Prima ,S.P. Morrissey, C. Barillot, "A robust Expectation-Maximization algorithm for Multiple Sclerosis lesion segmentation", In MICCAI Workshop: 3D Segmentation in the Clinic: A Grand Challenge II, MS lesion segmentation, 2008.
