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 Single-subject Single-subject Single-subject effect size statistics subjects Register subjects into a standard space Effect size statistics Statistic Image Significant voxels/clusters Contrast Thresholding
63
Embed
FMRI Group Analysis - FSLfsl.fmrib.ox.ac.uk/fslcourse/lectures/feat2_part1.pdfFMRI Group Analysis GLM Design matrix Effect size subject-series Voxel-wise group analysis Group effect
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
FMRI Group Analysis
GLM
Design matrix
Effect size subject-series
Voxel-wise group analysis
Groupeffect sizestatistics
Subjectgroupings
111111000000
000000111111
Standard-spacebrain atlas
subjects
Single-subject effect sizestatistics
Single-subject effect sizestatistics
Single-subject effect sizestatistics
Single-subject effect sizestatistics
subjects
Registersubjects intoa standardspace
Effect sizestatistics
Statistic ImageSignificant
voxels/clusters
Contrast
Thresholding
• 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
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?
Does the group activate on average?
A simple example
Group
Mark Steve Karl Will Tom Andrew
Does the group activate on average?
0 effect size
A simple example
Group
Mark Steve Karl Will Tom Andrew
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 Mark’s 4D FMRIdata set
Does the group activate on average?
0 effect size
A simple example
Group
Mark Steve Karl Will Tom Andrew
Yk = Xk�k + ⇥k
Mark’s effect size
Does the group activate on average?
0 effect size
A simple example
Group
Mark Steve Karl Will Tom Andrew
Yk = Xk�k + ⇥k
Mark’s within-subject
variance
Does the group activate on average?
0 effect size
A simple example
Group
Mark Steve Karl Will Tom Andrew
All first-level GLMson 6 FMRI data set
YK = XK�K + ⇥K
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
Do these exact 6 subjects activate on average?
Fixed-Effects Analysis
Group
Mark Steve Karl Will Tom Andrew
0 effect size
�g =16
6�
k=1
�k
estimate group effect size as straight-forward mean
across lower-level estimates
Do these exact 6 subjects activate on average?
Fixed-Effects Analysis
Group
Mark Steve Karl Will Tom Andrew
0 effect size
YK = XK�K + ⇥K
�K = Xg�g
�g =16
6�
k=1
�kXg =
�
⇧⇧⇧⇧⇧⇧⇤
111111
⇥
⌃⌃⌃⌃⌃⌃⌅Group mean
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 = XK�K + ⇥K
�K = Xg�g
Fixed Effects Analysis:
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
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
�g�k
Consider the distribution over the population from which our 6 subjects were sampled:
�2g• is the between-subject variance
�g
Does the population activate on average?
Mixed-Effects Analysis
Group
Mark Steve Karl Keith Tom Andrew
YK = XK�K + ⇥K
0 effect size�g�k
�g�K = Xg�g + ⇥g
Xg =
�
⇧⇧⇧⇧⇧⇧⇤
111111
⇥
⌃⌃⌃⌃⌃⌃⌅Population
mean
between-subject
variation
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 = XK�K + ⇥K
Mixed-Effects Analysis:
�K = Xg�g + ⇥g
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?
Summary Statistics Approach
• At each level:
• Inputs are summary stats from levels below (or FMRI data at the lowest level)
• 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
FLAME
• Fully Bayesian framework
• use non-central t-distributions: Input COPES, VARCOPES & DOFs from lower-level
• estimate COPES, VARCOPES & DOFs at current level
• pass these up
• Infer at top level
• Equivalent to All-in-One approach
FMRIB’s Local Analysis of Mixed Effects
Group
Subject
Session
Groupdifference
COPESVARCOPES
DOFs
Z-Stats
COPESVARCOPES
DOFs
COPESVARCOPES
DOFs
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
Choosing Inference Approach
1.Fixed Effects
Use for intermediate/top levels
2.Mixed Effects - OLS
Use 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
FLAME vs. OLS
• allow different within-level variances (e.g. patients vs. controls)