repetitions or more items? The effect of repeating

641 MT
More repetitions or more items?
The effect of repeating stimuli on MVPA for fMRI and E/MEG
Alexandra Woolgar Mirjana Bozic Caroline Whiting
Li Su Cai Wingfield William Marslen-Wilson
Elisabeth Fonteneau
ARC Centre of Excellence in Cognition and its Disorders, Department of Cognitive Science,
Macquarie University, Sydney, Australia; University of Cambridge, Cambridge, UK; MRC Cognition and Brain
Sciences Unit, Cambridge, UK
Introduction
Representational Similarity Analysis
How much data is needed for Multivariate Pattern Analysis (MVPA)?
MR: I walk (acoustic baseline)
“I walk”
Example ROI: left Middle
Temporal Gyrus (MTG)
To reduce noise, psychological paradigms commonly average responses
across multiple trials. Two common schemes:
Ö
fMRI: BOLD response at each voxel in a ROI:
...
...
1) Average the responses to many items belonging to the same experimental condition
1- correlation of patterns
I walk
he walks
2) Average the responses to many repetitions of the same items
Representational
Dissimilarity Matrix
(RDM)
he walked
I fall
he falls
We compare the effects of including more items per condition to including more repetitions of each item in multivariate analyses of empirical
fMRI and E/MEG data.
Correlation
Inference from
condition-label randomisation: how well
does the model
describe the data?
he fell
MR I walk
MR he walks
1.0
more
dissimilar
MR he walked
1- correlation:
MR I fall
MR he falls
MR he fell
Design
fMRI: model - data correlation for each ROI
E/MEG: model - data correlation for each ROI, at each time point4
12 experimental conditions (6 verbs in short phrases, 6 “Musical Rain” (MR) acousticallymatched baselines which cannot be interpretted as speech1)
10 items per condition
8 repetitions of each item (fMRI), 12 repetitions of each item (E/MEG)
20 participants, acoustic presentation: infrequent semantic completion task
Repetitions:
- Estimate response to each condition for each repetitionp (collapse items)
- Ramdomly sample from 1 to n repetitions
Items:
- Estimate response to each item separately (collapse blocks)
- Ramdomly sample from 1 to 10 items
Compare to simulated data (random noise)
fMRI: 3T, continuous “quiet” sequence, TR = 2, event-related GLM using HRF in SPM5
E/MEG: 306-channel vector view MEG, simultaneous 70 channel EEG, eyeblink artefacts removed
with ICA, source level estimation from minimum norm estimation (MNE) using individual structural MRIs
fMRI
Repetitions
Number of repetitions
1
2
Items
Number of items
3
4
5
6
7
:
;
:
;
:
;
8

2
3
;
<
;
:
;
:
;
4
5
6
7
8
9

=
>
=
G
>
=
F
>
=
E
>
<;
:
;
:
;

£
:

:
10

?
:
:
0
Model
Data

"
"
¤
¬
<
Example ROI:
L MTG

:
;
:
;
:
;
:
;
:
;
=
D
>
=
C
>
=
B
>
=
A
>
=
@
>

ª
©
L
K
¨
J

J
§
$
$
#
#
I

M
L
K

«
Ö
M
Example ROI:
L MTG
Ö
¦
I
"
"
!
!
¥
!
!
:
;
=
>
?

H
:
=
£
Average of 12
fronto-temporal
anatomical ROIs
:
;
:
<
;
:
;
:
;
:
;
`
V
q
W
\
]
[
\
W
^
W
r
r
]
_

p
¤

:
<
:
Y
p
`
V
q
W
\
]
[
\ W
^
W
r
r
]
_
Y
p
Average of 12
fronto-temporal
anatomical ROIs
:
<
`
V
q
W
\
]
[
U
a
W
V
p
Y
`
V
q
W
\
]
[
U
a
W
V
<
:
Y
:;
:
$
"
:
<;
:
;
:
;
:
;
<
M
L
K
:
;
:
;
:
;
:
;
:
;
:
;
:
;
:
;
:
;
J
$
$
#
#
I
"
"
!
!
!
!
:
;
:
:

:
H
:
Simulated Data
(noise - correlation
with grand mean)
`
V
q
W
\
]
[
\ W
^
W
r
r
]
_
Y
p
`
V
q
W
\
]
[
\ W
^
W
r
r
]
_
Simulated Data
(noise - correlation
with grand mean)
|
:;
w
<;
:
;
:
;
:
;
<
:
{
x
$
{
"
w
v
z
L
V
q
W
\
]
[
U
a
W
V
Y
p
V
q
W
\
q
W
\
]
[
U
a
]
[
U
a
W
V
<
:
W
V
Y
}
~
}

~
}

~
Ý
×
Ü
×
Û
×
"
v
L
u
K
K
t
t
J
J
:
;
:
;
Ú
z
}

~
}

~
w
$
y
x
#
I
:
rand()
×
rand()
`
$
x
u
`
Y
:
p
p
;
$
#
Ù
×
I
M
M
"
"
y
!
;
}

~
!
!
w
;
x
}
~

:
H
s
s
}
×
w
p
`
V
q
W
\
]
[
\ W
^
W
r
r
]
_
:
;
:
;
Average of
fronto-temporal
ROIs
9
1
7
-
`
V
q
W
\
]
[
U
a
W
V
<
:
;
:
;
:
;
Average of
fronto-temporal
ROIs
9
1
:
Y
p
E/MEG
7
-
p
Y
Repetitions
;
;
:
;
:
;
Average of
fronto-temporal
ROIs
9
1
7
-
*
/
&
;
:
;
:
;
&
S
Ì
Í
Î
Ï
Ð
Ñ
:
;
:
;
:
;
Ì
Í
Î
Ï
Ð
Ñ
O
Ó
Ô
Î
Õ
Ñ
:
Å
¾
Ä
½
¾
Ã
½
¾
Â
½
¾
¿
½
¾
À
½
¾
Average of
fronto-temporal
ROIs
Ò
Ì
Ó
Ô
Î
Õ
Ñ
Ò
R
(
P
:
<
)
&
O
.
.
;
<
Q
O
:
O
.
-
:
;
-
*
*
*
1
1
1
&
Y
)
&
*
1
1
S
Ì
P
.
;
;
:
<
)
&
:
;
(
P
-
;
:
R
Q
(
)
&
&
:
Ò
R
Q
(
P
:
S
Ò
R
¾
½
-
*
/
Q
½
9
1
*
/
&
V
7
*
S
`
Items
:
/
:
×
:
Ø
H
:

!
:
Á
1
1
&
1
&
&
:
:
N
:
N
N
N
½
!
¹!
º!
»!
!
!
!
!
µ
¶µ
·µ
¸µ
µ
µ
µ
µ
!
¹
!
º
!
»
!
!
!
!
!
Ç
Ç
:
;
:
:
;
;
ÉÇ
ÊÇ
ËÇ
Ç
Ç
Ç
È
Ç
Æ
³
´
¼
½
¾
Á
¼
½
¾
À
¼
½
¾
¿
³
³
:
;
:
;
³
³
T
U
V
W
X
V
Y
Z
[
\ ]
V
^
\ ]
_
]
`
_
]
_
Y
W
a
T
T
Average of
fronto-temporal
ROIs, collapsed
over entire 480ms
epoch

²®

±®
¯
U
;
:
;
;
:
;
V
W
X
V
Y
Z
[
\ ]
V
^
\ ]
_
]
`
_
]
_
Y
W

*
1
2
7
)

¯
.
*
(
1
.
&
0
S

±
®

;
:
;
;
:
;
/
R
*
Q
)
V
W
X
V
Y
Z
[
\ ]

Average of
fronto-temporal
ROIs, collapsed
over entire 480ms
epoch
®
°
¯
°

^
\
]
_
]
`
_
]
_
Y
W
a
b
c
d
e
f
d
g
h
i
j k
d
l
j k
m
k
n
m
k
m
g
e
o
²
®
¯
²®

®
¯
;
:;
;
:;

9
*
1
2
7

)
-
*

±
.
/
(
1
&
.
S
0

±®

;
/
R
*
Q
)
(

®
9
/
U
a

®
°
:;
¯
P
(
P
(
(
.
.
-
)
-
)
,
;
:;
,
9
9
&
&

8
+
®
°

8
+
O
*
*
O
7
7
.
.
6
)
6
)
-
-
6

®
*
(
¯

®
6
¯
(
*
'
5
:
:
1
;
'
;
5
1
1
'
'
1
&
&
&
&
3
4
3
4
N
%
N
:
1
:
%
:

<
`
V
q
W
\
]
[
\
W
^
W
r
r
]
_
<
:
Y
p
`
V
q
W
\
]
[
\ W
^
W
r
r
]
_
<
<

:
:
Y
p
Summary
`
V
q
W
\
]
[
U
a
W
V
Y
p
`
V
q
W
\
]
[
U
a
W
V
Y
Conclusion
fMRI
- Repetitions: correlation with model reached a plateau after ~4 repetitions, or 40
stimulus presentations.
- Similar effect for including more items, though less pronounced. Plateau at ~6
items, or 48 presentations.
E/MEG
- More data needed to reach plateau: 6 repetitions or 60 presentations.
- Items: did not reach a plateau within the number of items included.
References:

:
p

1
1 Uppenkamp et al 2006, NI
2 Kreigeskorte et al 2006, PNAS
3 Bozic et al, 2011, ICON
4 Su et al, 2012, PRNI
The optimal number of presentations was around 40 for fMRI, and
60 for MEG in this design. The effects of including more repetitions
and more items was similar, although responses tended to plateau
earlier after additional repetitions.
See also posters Bozic et al 327 MT and
Su et al 637 MT
Contact: [email protected]

Download Report

repetitions or more items? The effect of repeating

Paperzz.com

Your Paperzz