fsl feat analysis
-
Upload
rukavicebela -
Category
Documents
-
view
251 -
download
1
Transcript of fsl feat analysis
-
8/11/2019 fsl feat analysis
1/94
FMRI Group Analysis
GLM
Design matrix
Effect size subject-series
Voxel-wise group analysis
Group
effect size
statistics
Subject
groupings
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
Standard-space
brain atlas
subjects
Single-subject
effect size
statistics
Single-subject
effect sizestatistics
Single-subject
effect size
statistics
Single-subjecteffect size
statistics
subjects
Register
subjects into
a standard
space
Effect size
statistics
Statistic ImageSignificant
voxels/clustersContrast
Thresholding
-
8/11/2019 fsl feat analysis
2/94
uses GLM at both lower and higher levels
typically need to infer across multiple subjects,sometimes multiple groups and/or multiple sessions
questions of interest involve comparisons at the
highest level
Multi-Level FMRI analysis
Group 2
HannaJosephine Anna Sebastian Lydia Elisabeth
Group 1
Mark Steve Karl Will Tom Andrew
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 4
Difference?
-
8/11/2019 fsl feat analysis
3/94
Does the group activate on average?
A simple example
Group
Mark Steve Karl Will Tom Andrew
-
8/11/2019 fsl feat analysis
4/94
Does the group activate on average?
0 effect size
A simple example
Group
Mark Steve Karl Will Tom Andrew
Yk = Xk k + k
First-level GLMon Marks 4D FMRI
data set
-
8/11/2019 fsl feat analysis
5/94
-
8/11/2019 fsl feat analysis
6/94
Does the group activate on average?
0 effect size
A simple example
Group
Mark Steve Karl Will Tom Andrew
Yk = Xk k + k
Marks
within-subject
variance
-
8/11/2019 fsl feat analysis
7/94
-
8/11/2019 fsl feat analysis
8/94
Does the group activate on average?
What group mean are we after? Is it:
1. The group mean for those exact 6 subjects?
Fixed-Effects (FE) Analysis
2. The group mean for the population from which
these 6 subjects were drawn?
Mixed-Effects (ME) analysis
A simple example
Group
Mark Steve Karl Will Tom Andrew
-
8/11/2019 fsl feat analysis
9/94
Do these exact 6 subjects activate on average?
Fixed-Effects Analysis
Group
Mark Steve Karl Will Tom Andrew
0 effect size
g =1
6
6
k=1
k
estimate group effect size asstraight-forward meanacross lower-level estimates
-
8/11/2019 fsl feat analysis
10/94
Do these exact 6 subjects activate on average?
Fixed-Effects Analysis
Group
Mark Steve Karl Will Tom Andrew
0 effect size
YK= XKK+ K
K = Xgg
g =1
6
6
k=1
kXg =
1
1
1
1
1
1
Group
mean
-
8/11/2019 fsl feat analysis
11/94
Do these exact 6 subjects activate on average?
Fixed-Effects Analysis
Group
Mark Steve Karl Will Tom Andrew
0 effect size
K = Xgg
Xg =
1
1
1
1
1
1
Group
mean
-
8/11/2019 fsl feat analysis
12/94
Do these exact 6 subjects activate on average?
Consider only these 6 subjects estimate the mean across these subject only variance is within-subject variance
Fixed-Effects Analysis
Group
Mark Steve Karl Will Tom Andrew
YK= XKK+ KK = Xgg
Fixed Effects Analysis:
-
8/11/2019 fsl feat analysis
13/94
Does the group activate on average?
What group mean are we after? Is it:
1. The group mean for those exact 6 subjects?
Fixed-Effects (FE) Analysis
2. The group mean for the population from which
these 6 subjects were drawn?
Mixed-Effects (ME) analysis
A simple example
Group
Mark Steve Karl Keith Tom Andrew
-
8/11/2019 fsl feat analysis
14/94
0 effect size
Does the population activate on average?
Mixed-Effects Analysis
Group
Mark Steve Karl Keith Tom Andrew
0 effect size
YK= XK K+ K
gk
Consider the distribution over thepopulation from which our 6subjects were sampled:
2
g is the between-subject variance
g
-
8/11/2019 fsl feat analysis
15/94
0 effect size
Does the population activate on average?
Mixed-Effects Analysis
Group
Mark Steve Karl Keith Tom Andrew
YK= XK K+ K
gk
Consider the distribution over thepopulation from which our 6subjects were sampled:
2
g is the between-subject variance
g
-
8/11/2019 fsl feat analysis
16/94
Does the population activate on average?
Mixed-Effects Analysis
Group
Mark Steve Karl Keith Tom Andrew
YK= XKK+ K
0 effect sizeg
k
gK= Xgg + g
Xg =
1
1
1
1
1
1
Population
mean
between-subject
variation
-
8/11/2019 fsl feat analysis
17/94
Does the population activate on average?
Mixed-Effects Analysis
Group
Mark Steve Karl Keith Tom Andrew
0 effect sizeg
k
gK= Xgg + g
Xg =
1
1
1
1
1
1
Population
mean
between-subject
variation
-
8/11/2019 fsl feat analysis
18/94
Does the population activate on average?
Consider the 6 subjects as samples from a wider population estimate the mean across the population between-subject variance accounts for random sampling
Mixed-Effects Analysis
Group
Mark Steve Karl Keith Tom Andrew
YK= XKK+ K
Mixed-Effects Analysis:
K= Xgg + g
-
8/11/2019 fsl feat analysis
19/94
All-in-One Approach
Could use one (huge) GLM to infer group difference
difficult to ask sub-questions in isolation
computationally demanding
need to process again when new data is acquired
Group 2
HannaJosephine Anna Sebastian Lydia Elisabeth
Group 1
Mark Steve Karl Will Tom Andrew
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 1
session 2
session 3
session 4
session 1
session 2
session 3
session 4
Difference?
-
8/11/2019 fsl feat analysis
20/94
Summary Statistics Approach
At each level:
Inputs are summary stats from levels
below (or FMRI data at the lowestlevel)
Outputs are summary stats or
statistic maps for inference
Need to ensure formal equivalence
between different approaches!
In FEAT estimate levels one stage at a time
Group
Subject
Session
Groupdifference
-
8/11/2019 fsl feat analysis
21/94
FLAME
Fully Bayesian framework
use non-central t-distributions:
Input COPES, VARCOPES & DOFsfrom lower-level
estimate COPES, VARCOPES &
DOFs at current level
pass these up
Infer at top level
Equivalent to All-in-One approach
FMRIBs Local Analysis of Mixed Effects
Group
Subject
Session
Groupdifference
COPES
VARCOPESDOFs
Z-Stats
COPES
VARCOPESDOFs
COPESVARCOPES
DOFs
-
8/11/2019 fsl feat analysis
22/94
FLAME Inference
Default is:
FLAME1: fast approximation for all voxels (using
marginal variance MAP estimates)
Optional slower, slightly more accurate approach:
FLAME1+2:
FLAME1 for all voxels, FLAME2 for voxels close to
threshold
FLAME2: MCMC sampling technique
-
8/11/2019 fsl feat analysis
23/94
Choosing Inference Approach
1. Fixed Effects
Use for intermediate/top levels
2. Mixed Effects - OLSUse at top level: quick and less accurate
3. Mixed Effects - FLAME 1
Use at top level: less quick but more accurate
4. Mixed Effects - FLAME 1+2
Use at top level: slow but even more accurate
-
8/11/2019 fsl feat analysis
24/94
FLAME vs. OLS
allow different within-levelvariances (e.g. patients vs.
controls)
allow non-balanced designs
(e.g. containing behaviouralscores)
allow un-equal group sizes
solve the negative varianceproblem
0 effect size
pat ctl
GroupSubjectSession
< 0
>0?
0
subject
effect size
-
8/11/2019 fsl feat analysis
30/94
Single Group Average
We have 8 subjects - all in one group - and want themean group average:
Does the group activate on average?
estimate mean
estimate std-error
(FE or ME)
test significance of
mean > 0
>0?
0
subject
effect size
-
8/11/2019 fsl feat analysis
31/94
Single Group Average
Does the group activate on average?
0
subject
effect size
-
8/11/2019 fsl feat analysis
32/94
Single Group Average
Does the group activate on average?
0
subject
effect size
-
8/11/2019 fsl feat analysis
33/94
Single Group Average
Does the group activate on average?
0
subject
effect size
-
8/11/2019 fsl feat analysis
34/94
Single Group Average
Does the group activate on average?
-
8/11/2019 fsl feat analysis
35/94
Unpaired Two-Group Difference
We have two groups (e.g. 9 patients, 7 controls)
with different between-subject variance
Is there a significant group difference?
estimate means
estimate std-errors(FE or ME)
test significance of
difference in means
>0?
0
subject
effect size
-
8/11/2019 fsl feat analysis
36/94
Unpaired Two-Group Difference
We have two groups (e.g. 9 patients, 7 controls)
with different between-subject variance
Is there a significant group difference?
estimate means
estimate std-errors(FE or ME)
test significance of
difference in means
>0?
0
subject
effect size
-
8/11/2019 fsl feat analysis
37/94
Unpaired Two-Group Difference
Is there a significant group difference?
0
subject
effect size
-
8/11/2019 fsl feat analysis
38/94
Unpaired Two-Group Difference
Is there a significant group difference?
0
subject
effect size
-
8/11/2019 fsl feat analysis
39/94
Unpaired Two-Group Difference
Is there a significant group difference?
0
subject
effect size
-
8/11/2019 fsl feat analysis
40/94
Unpaired Two-Group Difference
Is there a significant group difference?
0
subject
effect size
-
8/11/2019 fsl feat analysis
41/94
Unpaired Two-Group Difference
Is there a significant group difference?
-
8/11/2019 fsl feat analysis
42/94
Unpaired Two-Group Difference
Is there a significant group difference?
-
8/11/2019 fsl feat analysis
43/94
Paired T-Test
8 subjects scanned under 2 conditions (A,B)
Is there a significant difference between conditions?
0
subject
effect size
-
8/11/2019 fsl feat analysis
44/94
Paired T-Test
8 subjects scanned under 2 conditions (A,B)
Is there a significant difference between conditions?
0
subject
effect size
-
8/11/2019 fsl feat analysis
45/94
>0?
Paired T-Test
8 subjects scanned under 2 conditions (A,B)
Is there a significant difference between conditions?
0
subject
effect size
try non-paired t-test
-
8/11/2019 fsl feat analysis
46/94
8 subjects scanned under 2 conditions (A,B)
Is there a significant difference between conditions?
data
0
subject
effect size
Paired T-Test
-
8/11/2019 fsl feat analysis
47/94
8 subjects scanned under 2 conditions (A,B)
Is there a significant difference between conditions?
data
0
subject
effect size
Paired T-Test
subject meanaccounts for large prop.
of the overall variance
-
8/11/2019 fsl feat analysis
48/94
8 subjects scanned under 2 conditions (A,B)
Is there a significant difference between conditions?
de-meaned data
0
subject
effect size
data
0
subject
effect size
Paired T-Test
subject meanaccounts for large prop.
of the overall variance
-
8/11/2019 fsl feat analysis
49/94
8 subjects scanned under 2 conditions (A,B)
Is there a significant difference between conditions?
de-meaned data
0
subject
effect size
data
0
subject
effect size
Paired T-Test
>0?subject mean
accounts for large prop.
of the overall variance
-
8/11/2019 fsl feat analysis
50/94
subject
effect size00
Paired T-Test
Is there a significant difference between conditions?
-
8/11/2019 fsl feat analysis
51/94
subject
effect size00
Paired T-Test
Is there a significant difference between conditions?
-
8/11/2019 fsl feat analysis
52/94
subject
effect size00
Paired T-Test
Is there a significant difference between conditions?
-
8/11/2019 fsl feat analysis
53/94
subject
effect size00
Paired T-Test
Is there a significant difference between conditions?
-
8/11/2019 fsl feat analysis
54/94
subject
effect size00
Paired T-Test
Is there a significant difference between conditions?
-
8/11/2019 fsl feat analysis
55/94
subject
effect size00
Paired T-Test
Is there a significant difference between conditions?
-
8/11/2019 fsl feat analysis
56/94
Paired T-Test
Is there a significant difference between conditions?
EV1models the A-B
paired difference; EVs2-9 are confounds
which model out eachsubjects mean
-
8/11/2019 fsl feat analysis
57/94
Paired T-Test
Is there a significant difference between conditions?
-
8/11/2019 fsl feat analysis
58/94
Multi-Session & Multi-Subject
5 subjects each have three sessions.
Does the group activate on average?
Use three levels: in the second levelwe
model the within-subject repeated measure
-
8/11/2019 fsl feat analysis
59/94
Multi-Session & Multi-Subject
5 subjects each have three sessions.
Does the group activate on average?
Use three levels: in the third levelwe model
the between-subjects variance
-
8/11/2019 fsl feat analysis
60/94
Multi-Session & Multi-Subject
5 subjects each have three sessions.
Does the group activate on average?
Use three levels:
in the second level we model the within subject
repeated measure typically using fixed effects(!)
as #sessions are small
in the third level we model the between subjects
variance using fixed or mixed effects
-
8/11/2019 fsl feat analysis
61/94
Reducing variance
Does the group activate on average?
mean effect size small
relative to std errormean effect size large
relative to std error
0
s
ubject
effect size
>0?
0
s
ubject
effect size
>0?
-
8/11/2019 fsl feat analysis
62/94
mean effect size large
relative to std error
Reducing variance
Does the group activate on average?
mean effect size large
relative to std error
0
s
ubject
effect size
>0?
0
s
ubject
effect size
>0?
-
8/11/2019 fsl feat analysis
63/94
We have 7 subjects - all in one group. We also have
additional measurements (e.g. age; disability score;
behavioural measures like reaction times):
Does the group activate on average?
use covariates to
explain variation
estimate mean
estimate std-error(FE or ME)
0
subject
effect size
Single Group Average & Covariates
-
8/11/2019 fsl feat analysis
64/94
We have 7 subjects - all in one group. We also have
additional measurements (e.g. age; disability score;
behavioural measures like reaction times):
0
subject
effect size
Single Group Average & Covariates
fastRTslow
-
8/11/2019 fsl feat analysis
65/94
We have 7 subjects - all in one group. We also have
additional measurements (e.g. age; disability score;
behavioural measures like reaction times):
Does the group activate on average?
0
subject
effect size
Single Group Average & Covariates
fastRTslow
-
8/11/2019 fsl feat analysis
66/94
We have 7 subjects - all in one group. We also have
additional measurements (e.g. age; disability score;
behavioural measures like reaction times):
Does the group activate on average?
use covariates to
explain variation
estimate mean
estimate std-error(FE or ME)
0
subject
effect size
Single Group Average & Covariates
fastRTslow
S l G A C
-
8/11/2019 fsl feat analysis
67/94
We have 7 subjects - all in one group. We also have
additional measurements (e.g. age; disability score;
behavioural measures like reaction times):
Does the group activate on average?
use covariates to
explain variation
estimate mean
estimate std-error(FE or ME)
0
subject
effect size
Single Group Average & Covariates
fastRTslow
S l G A & C
-
8/11/2019 fsl feat analysis
68/94
We have 7 subjects - all in one group. We also have
additional measurements (e.g. age; disability score;
behavioural measures like reaction times):
Does the group activate on average?
use covariates to
explain variation
estimate mean
estimate std-error(FE or ME)
0
subject
effect size
Single Group Average & Covariates
fastRTslow
S l G A & C
-
8/11/2019 fsl feat analysis
69/94
We have 7 subjects - all in one group. We also have
additional measurements (e.g. age; disability score;behavioural measures like reaction times):
Does the group activate on average?
use covariates to
explain variation
estimate mean
estimate std-error(FE or ME)
0
subject
effect size
Single Group Average & Covariates
fastRTslow
Si l G A & C i
-
8/11/2019 fsl feat analysis
70/94
We have 7 subjects - all in one group. We also have
additional measurements (e.g. age; disability score;behavioural measures like reaction times):
Does the group activate on average?
use covariates toexplain variation
estimate mean
estimate std-error(FE or ME)
0
subject
effect size
Single Group Average & Covariates
fastRTslow
Si l G A & C i
-
8/11/2019 fsl feat analysis
71/94
Does the group activate on average?
use covariates to explain variation
need to de-mean additional covariates!
Single Group Average & Covariates
FEAT G A l i
-
8/11/2019 fsl feat analysis
72/94
Run FEAT on raw FMRI data to get first-level .featdirectories, each one with several (consistent) COPEs
low-res copeN/varcopeN .feat/stats
when higher-level FEAT is run, highres copeN/varcopeN .feat/reg_standard
FEAT Group Analysis
FEAT G A l i
-
8/11/2019 fsl feat analysis
73/94
FEAT Group Analysis
Run second-level FEAT to get one .gfeat directory
Inputs can be lower-
level .feat dirs or
lower-level COPEs
the second-level GLM analysis is run separately
for each first-level COPE
each lower-level COPE generates its own .featdirectory inside the .gfeat dir
Th t ll f lk
-
8/11/2019 fsl feat analysis
74/94
Thats all folks
A di
-
8/11/2019 fsl feat analysis
75/94
Appendix:
G F t t
-
8/11/2019 fsl feat analysis
76/94
Group F-tests
Gro F tests
-
8/11/2019 fsl feat analysis
77/94
Group F-tests
3 groups of subjects
Group F tests
-
8/11/2019 fsl feat analysis
78/94
Group F-tests
3 groups of subjects
Is any of the groups activating on average?
ANOVA: 1 factor 4 levels
-
8/11/2019 fsl feat analysis
79/94
ANOVA: 1-factor 4-levels
ANOVA: 1 factor 4 levels
-
8/11/2019 fsl feat analysis
80/94
ANOVA: 1-factor 4-levels
8 subjects, 1 factor at 4 levels
ANOVA: 1 factor 4 levels
-
8/11/2019 fsl feat analysis
81/94
ANOVA: 1-factor 4-levels
8 subjects, 1 factor at 4 levels
Is there any effect?
ANOVA: 1 factor 4 levels
-
8/11/2019 fsl feat analysis
82/94
ANOVA: 1-factor 4-levels
8 subjects, 1 factor at 4 levels
Is there any effect?
EV1 fits cond. D, EV2 fits cond A relative to D etc.
ANOVA: 1 factor 4 levels
-
8/11/2019 fsl feat analysis
83/94
ANOVA: 1-factor 4-levels
8 subjects, 1 factor at 4 levels
Is there any effect?
EV1 fits cond. D, EV2 fits cond A relative to D etc.
F-test shows any difference between levels
ANOVA: 2 factor 2 levels
-
8/11/2019 fsl feat analysis
84/94
ANOVA: 2-factor 2-levels
ANOVA: 2 factor 2 levels
-
8/11/2019 fsl feat analysis
85/94
ANOVA: 2-factor 2-levels
8 subjects, 2 factor at 2 levels. FE Anova: 3 F-tests give
standard results for factor A, B and interaction
ANOVA: 2 factor 2 levels
-
8/11/2019 fsl feat analysis
86/94
ANOVA: 2-factor 2-levels
8 subjects, 2 factor at 2 levels. FE Anova: 3 F-tests give
standard results for factor A, B and interaction
If both factors are random effects then Fa=fstat1/fstat3,
Fb=fstat2/fstat3ME
ANOVA: 2 factor 2 levels
-
8/11/2019 fsl feat analysis
87/94
ANOVA: 2-factor 2-levels
8 subjects, 2 factor at 2 levels. FE Anova: 3 F-tests give
standard results for factor A, B and interaction
If both factors are random effects then Fa=fstat1/fstat3,
Fb=fstat2/fstat3ME
ME: if fixed fact. is A, Fa=fstat1/fstat3
ANOVA: 3 factor 2 levels
-
8/11/2019 fsl feat analysis
88/94
ANOVA: 3-factor 2-levels
ANOVA: 3-factor 2-levels
-
8/11/2019 fsl feat analysis
89/94
ANOVA: 3-factor 2-levels
16 subjects, 3 factor at 2 levels.
ANOVA: 3-factor 2-levels
-
8/11/2019 fsl feat analysis
90/94
ANOVA: 3-factor 2-levels
16 subjects, 3 factor at 2 levels.
Fixed-Effects ANOVA:
ANOVA: 3-factor 2-levels
-
8/11/2019 fsl feat analysis
91/94
ANOVA: 3-factor 2-levels
16 subjects, 3 factor at 2 levels.
Fixed-Effects ANOVA:
For random/mixed effects need different Fs.
Understanding FEAT dirs
-
8/11/2019 fsl feat analysis
92/94
Understanding FEAT dirs
First-level analysis:
Understanding FEAT dirs
-
8/11/2019 fsl feat analysis
93/94
Understanding FEAT dirs
Second-level analysis:
Thats all folks
-
8/11/2019 fsl feat analysis
94/94
That s all folks