Transcript BOLD fMRI - Duke University

1. Preprocessing of
FMRI Data
fMRI Graduate Course
October 22, 2003
What is preprocessing?
• Correcting for non-task-related variability
in experimental data
– Usually done without consideration of
experimental design; thus, pre-analysis
– Occasionally called post-processing, in
reference to being after acquisition
• Attempts to remove, rather than model,
data variability
Signal, noise, and the General
Linear Model
Y  M  
Amplitude (solve for)
Measured Data
Noise
Design Model
Cf. Boynton et al., 1996
Signal-Noise-Ratio (SNR)
Task-Related
Variability
Non-task-related
Variability
Preprocessing Steps
•
•
•
•
•
•
•
Slice Timing Correction
Motion Correction
Coregistration
Normalization
Spatial Smoothing
Segmentation
Region of Interest Identification
• Bias field correction
Tools for Preprocessing
•
•
•
•
•
SPM
Brain Voyager
VoxBo
AFNI
Custom BIAC scripts
Slice Timing Correction
Why do we correct for slice timing?
• Corrects for differences in acquisition time within a TR
– Especially important for long TRs (where expected HDR
amplitude may vary significantly)
– Accuracy of interpolation also decreases with increasing TR
• When should it be done?
– Before motion correction: interpolates data from (potentially)
different voxels
• Better for interleaved acquisition
– After motion correction: changes in slice of voxels results in
changes in time within TR
• Better for sequential acquisition
Effects of uncorrected slice timing
•
•
•
•
Base Hemodynamic Response
Base HDR + Noise
Base HDR + Slice Timing Errors
Base HDR + Noise + Slice Timing Errors
Base HDR: 2s TR
1.2
1
Slice1
0.8
0.6
0.4
0.2
0
TR:-1
TR:0
TR:1
TR:2
TR:3
TR:4
TR:5
Base HDR + Noise
1.8
1.6
r = 0.77
Noise1
1.4
r = 0.81
Noise2
Noise3
1.2
1
r = 0.80
0.8
0.6
0.4
0.2
0
TR:-1
TR:0
TR:1
TR:2
TR:3
TR:4
TR:5
Base HDR + Slice Timing Errors
1.2
Slice1
Slice11
r = 0.92
r = 0.85
1
Slice12
0.8
r = 0.62
0.6
0.4
0.2
0
TR:-1
TR:0
TR:1
TR:2
TR:3
TR:4
TR:5
HDR + Noise + Slice Timing
1.8
Slice1
1.6
Slice11
r = 0.65
r = 0.67
Slice12
1.4
1.2
1
r = 0.19
0.8
0.6
0.4
0.2
0
TR:-1
TR:0
TR:1
TR:2
TR:3
TR:4
TR:5
Interpolation Strategies
• Linear interpolation
• Spline interpolation
• Sinc interpolation
Motion Correction
Head Motion: Good, Bad,…
… and catastrophically bad
Why does head motion introduce
problems?
A
B
C
507
89
154
663
507
89
119
171
83
520
119
171
179
117
53
137
179
117
Simulated Head Motion
Severe Head Motion: Simulation
Two 4s movements of
8mm in -Y direction
(during task epochs)
Motion
Severe Head Motion: Real Data
Two 4s movements of
8mm in –Y direction
(during task epochs)
Motion
Correcting Head Motion
• Rigid body transformation
– 6 parameters: 3 translation, 3 rotation
• Minimization of some cost function
– E.g., sum of squared differences
Effects of Head Motion Correction
Limitations of Motion Correction
• Artifact-related limitations
– Loss of data at edges of imaging volume
– Ghosts in image do not change in same manner as
real data
• Distortions in fMRI images
– Distortions may be dependent on position in field, not
position in head
• Intrinsic problems with correction of both slice
timing and head motion
Prevention is the best medicine
A
C
B
D
Coregistration
Should you Coregister?
• Advantages
–
–
–
–
Aids in normalization
Allows display of activation on anatomical images
Allows comparison across modalities
Necessary if no coplanar anatomical images
• Disadvantages
– May severely distort functional data
– May reduce correspondence between functional and
anatomical images
Normalization
Standardized Spaces
• Talairach space (proportional grid system)
–
–
–
–
From atlas of Talairach and Tournoux (1988)
Based on single subject (60y, Female, Cadaver)
Single hemisphere
Related to Brodmann coordinates
• Montreal Neurological Institute (MNI) space
– Combination of many MRI scans on normal controls
• All right-handed subjects
– Approximated to Talaraich space
• Slightly larger
• Taller from AC to top by 5mm; deeper from AC to bottom by 10mm
– Used by SPM, National fMRI Database, International Consortium
for Brain Mapping
Normalization to Template
Normalization Template
Normalized Data
Anterior and Posterior Commissures
Anterior
Commissure
Posterior
Commissure
Should you normalize?
• Advantages
–
–
–
–
Allows generalization of results to larger population
Improves comparison with other studies
Provides coordinate space for reporting results
Enables averaging across subjects
• Disadvantages
– Reduces spatial resolution
– May reduce activation strength by subject averaging
– Time consuming, potentially problematic
• Doing bad normalization is much worse than not normalizing
Slice-Based Normalization
Before Adjustment (15 Subjects)
After Adjustment to Reference Image
Registration courtesy Dr. Martin McKeown (BIAC)
Spatial Smoothing
Techniques for Smoothing
• Application of
Gaussian kernel
– Usually expressed in
#mm FWHM
– “Full Width – Half
Maximum”
– Typically ~2 times
voxel size
Effects of Smoothing on Activity
Unsmoothed Data
Smoothed Data (kernel width 5 voxels)
Should you spatially smooth?
• Advantages
– Increases Signal to Noise Ratio (SNR)
• Matched Filter Theorem: Maximum increase in SNR by filter with
same shape/size as signal
– Reduces number of comparisons
• Allows application of Gaussian Field Theory
– May improve comparisons across subjects
• Signal may be spread widely across cortex, due to intersubject
variability
• Disadvantages
– Reduces spatial resolution
– Challenging to smooth accurately if size/shape of signal is not
known
Segmentation
• Classifies voxels within an image into different
anatomical divisions
– Gray Matter
– White Matter
– Cerebro-spinal Fluid (CSF)
Image courtesy J. Bizzell & A. Belger
Histogram of Voxel Intensities
0.04
0.035
Anatomical
0.03
0.025
0.02
0.015
0.01
0.005
0
Functional
Region of Interest Drawing
Why use an ROI-based approach?
• Allows direct, unbiased measurement of activity
in an anatomical region
– Assumes functional divisions tend to follow
anatomical divisions
• Improves ability to identify topographic changes
– Motor mapping (central sulcus)
– Social perception mapping (superior temporal sulcus)
• Complements voxel-based analyses
Drawing ROIs
• Drawing Tools
– BIAC software (e.g., Overlay2)
– Analyze
– IRIS/SNAP (G. Gerig)
• Reference Works
– Print atlases
– Online atlases
• Analysis Tools
– roi_analysis_script.m
ROI Examples
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
-3
-1.5
0
1.5
3
4.5
6
7.5
9
10.5
12
13.5
15
16.5
Left Hemisphere - Gaze Shifts
Right Hemisphere - Gaze Shifts
4
3
2
1
0
-1
-2
60
55
50
45
40
35
30
25
20
15
10
5
0
80
75
70
65
60
55
50
45
40
35
30
25
20
Distance Posterior from the Anterior Commissure (in mm)
BIAC is studying biological motion and social perception – here by determining how context modulates brain activity
in elicited when a subject watches a character shift gaze toward or away from a target.
Additional Resources
• SPM website
– http://www.fil.ion.ucl.ac.uk/spm/course/notes01.html
– SPM Manual
• Brain viewers
– http://www.bic.mni.mcgill.ca/cgi/icbm_view/
2. Issues in
Experimental Design
fMRI Graduate Course
October 23, 2003
What is Experimental Design?
• Controlling the timing and quality of presented
stimuli to influence resulting brain processes
• What can we control?
–
–
–
–
Experimental comparisons (what is to be measured?)
Stimulus properties (what is presented?)
Stimulus timing (when is it presented?)
Subject instructions (what do subjects do with it?)
Goals of Experimental Design
• To maximize the ability to test hypotheses
• To facilitate generation of new hypotheses
What are hypotheses?
• Statements about the relations between
independent and dependent variables.
A
B
C
D
Hemodynamic Hypotheses
Neuronal Hypotheses
Psychological Hypotheses
Independent Variables
• Aspects of the experimental design that we want to
manipulate
– Often have multiple levels (e.g., experimental and control
conditions)
– Critical design choice lies in determining how to choose stimuli to
match independent variable
A
B
C
Dependent Variable: BOLD signal
5
4
3
2
1
0
-1
-2
-3
-4
-5
0
50
100
150
200
250
300
Causal and non-causal relations
between variables
A
B
Is the BOLD response epiphenomenal?
Detection vs. Estimation
• Detection: What is active?
• Estimation: How does its activity change
over time?
-8
-6
-4
-2
0
2
4
6
8
10
12
14
16
Detection
• Detection power defined by SNR
SNR = aM/
M = hemodynamic changes (unit)
a = measured amplitude
 = noise standard deviation
• Depends greatly on hemodynamic
response shape
Estimation
• Ability to determine the shape of fMRI response
• Accurate estimation relies on minimization of
variance in estimate of HDR at each time point
• Efficiency of estimation is generally independent
of HDR form
Optimal Experimental Design
• Maximizing both Detection and Estimation
– Maximal variance in stimulus timing
(increases estimation)
– Maximal variance in measured signal
(increases detection)
• Limitations
– Refractory effects
– Signal saturation
fMRI Design Types
1) Blocked Designs
2) Event-Related Designs
a) Periodic Single Trial
b) Jittered Single Trial
c) Staggered Single Trial
3) Mixed Designs
a) Combination blocked/event-related
b) Variable stimulus probability
1. Blocked Designs
What are Blocked Designs?
• Blocked designs segregate different
cognitive processes into distinct time
periods
Task A
Task B
Task A
Task B
Task A
Task B
Task A
Task B
Task A
REST
Task B
REST
Task A
REST
Task B
REST
PET Designs
• Measurements done
following injection of
radioactive bolus
• Uses total activity
throughout task
interval (~30s)
• Blocked designs
necessary
– Task 1 = Injection 1
– Task 2 = Injection 2
Choosing Length of Blocks
• Longer block lengths allow for stability of extended responses
– Hemodynamic response saturates following extended stimulation
• After about 10s, activation reaches max
– Many tasks require extended intervals
• Processing may differ throughout the task period
• Shorter block lengths allow for more transitions
– Task-related variability increases (relative to non-task) with increasing
numbers of transitions
• Periodic blocks may result in aliasing of other variance in the data
– Example: if the person breathes at a regular rate of 1 breath/5sec, and
the blocks occur every 10s
Effects of Block Interval upon HDR
14
14
40s
12
14
20s
12
14
15s
12
10
10
10
8
8
8
8
6
6
6
6
4
4
4
4
2
2
2
2
0
0
0
20
40
60
80
12
100
120
8s
10
0
0
20
40
60
80
10
100
120
6s
9
8
8
6
4
5
4
4
3
40
60
80
100
120
80
100
120
0
20
40
60
80
7
4s
100
120
2s
6
5
3
2
2
1
1
0
20
60
4
1
0
40
7
5
2
0
20
8
6
3
2
0
0
6
7
10s
12
10
0
0
20
40
60
80
100
120
0
0
20
40
60
80
100
120
0
20
40
60
80
100
120
What baseline should you choose?
• Task A vs. Task B
– Example: Squeezing Right Hand vs. Left Hand
– Allows you to distinguish differential activation
between conditions
– Does not allow identification of activity common to
both tasks
• Can control for uninteresting activity
• Task A vs. No-task
– Example: Squeezing Right Hand vs. Rest
– Shows you activity associated with task
– May introduce unwanted results
Interpreting Baseline Activity
From Gusnard & Raichle, 2001
Non-Task Processing
• In many experiments, activation is greater in
baseline conditions than in task conditions!
– Requires interpretations of significant activation
• Suggests the idea of baseline/resting mental
processes
–
–
–
–
–
Emotional processes
Gathering/evaluation about the world around you
Awareness (of self)
Online monitoring of sensory information
Daydreaming
From Shulman et al., 1997 (PET data)
From Binder et al., 1999
From Huettel et al., 2002
(Baseline > Target Detection)
From Huettel et al., 2001 (Change Detection)
Power in Blocked Designs
1. Summation of responses results in large
variance
Single, unit amplitude HDR,
convolved by 1, 2, 4 ,8, 12,
or 16 events (1s apart).
HDR Estimation: Blocked Designs
Power in Blocked Designs
-0.5
-0.5
-1
-1
-1.5
-1.5
Simulation of single run with either 2 or 10 blocks.
97
10
0
94
91
88
85
82
79
76
73
70
67
64
61
58
55
52
49
46
43
40
37
34
31
28
25
22
19
16
13
7
10
1
10
0
94
97
88
91
82
85
76
79
70
73
64
67
0
58
61
0
52
55
0.5
46
49
0.5
40
43
1
34
37
1
28
31
1.5
22
25
1.5
16
19
2
7
10
13
2
4
2.5
1
2.5
4
2. Transitions between blocks
Power in Blocked Designs
-0.5
-0.5
-1
-1
-1.5
-1.5
Addition of linear drift within run.
10
0
94
97
88
91
82
85
76
79
70
73
64
67
58
61
52
55
46
49
40
43
34
37
28
31
22
25
16
19
7
10
13
1
10
0
94
97
88
91
82
85
76
79
70
73
64
67
0
58
61
0
52
55
0.5
46
49
0.5
40
43
1
34
37
1
28
31
1.5
22
25
1.5
16
19
2
7
10
13
2
4
2.5
1
2.5
4
2. Transitions between blocks
-0.5
-0.5
-1
-1
-1.5
-1.5
Addition of noise (SNR = 0.67)
10
0
94
97
88
91
82
85
76
79
0
70
73
0
64
67
0.5
58
61
0.5
52
55
1
46
49
1
40
43
1.5
34
37
1.5
28
31
2
22
25
2
16
19
2.5
7
10
13
2.5
4
1
10
0
94
97
88
91
82
85
76
79
70
73
64
67
58
61
52
55
46
49
40
43
34
37
28
31
22
25
16
19
7
10
13
4
1
Power in Blocked Designs
2. Transitions between blocks
Limitations of Blocked Designs
• Very sensitive to signal drift
– Sensitive to head motion, especially when only a few
blocks are used.
• Poor choice of baseline may preclude
meaningful conclusions
• Many tasks cannot be conducted repeatedly
• Difficult to estimate the HDR
2. Event-Related Designs
What are Event-Related Designs?
• Event-related designs associate brain
processes with discrete events, which may
occur at any point in the scanning session.
time
Why use event-related designs?
• Some experimental tasks are naturally
event-related
• Allows studying of trial effects
• Simple analyses
– Selective averaging
– No assumptions of linearity required
Event-Related and Blocked
Designs give Similar Results
A
B
C
2a. Periodic Single Trial Designs
• Stimulus events presented infrequently
with long interstimulus intervals
500 ms
500 ms
18 s
500 ms
18 s
500 ms
18 s
Trial Spacing Effects: Periodic Designs
20sec
12sec
8sec
4sec
ISI:
Interstimulus
Interval
SD:
Stimulus
Duration
From Bandettini
and Cox, 2000
2b. Jittered Single Trial Designs
• Varying the timing of trials within a run
Randomization = Jittering
Dale & Buckner, 1997
Extracting different task
components
A
B
Effects of Jittering on Stimulus Variance
Effects of ISI on Power
Birn et al, 2002
2c. Staggered Single Trial
• By presenting stimuli at different timings, relative
to a TR, you can achieve sub-TR resolution
• Significant cost in number of trials presented
– Resulting loss in experimental power
• Very sensitive to scanner drift and other sources
of variability
+0s
Two HDR epochs
sampled at a 3s TR.
+1s
Each row is sampled at
a different phase.
+2s
+0s
Two of the phases are
normal.
+1s
But, one has a change in one trial
(e.g., head motion)
+2s
2.5
2
1.5
1
0.5
0
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
-0.5
Post-Hoc Sorting of Trials
Data from old/new episodic
memory test.
From Konishi, et al., 2000
Limitations of Event-Related Designs
• Differential effects of interstimulus interval
– Long intervals do not optimally increase
stimulus variance
– Short intervals may result in refractory effects
• Detection ability dependent on form of HDR
• Length of “event” may not be known
3. Mixed Designs
3a. Combination Blocked/Event
• Both blocked and event-related design aspects
are used (for different purposes)
– Blocked design is used to evaluate state-dependent
effects
– Event-related design is used to evaluate item-related
effects
• Analyses are conducted largely independently
between the two measures
– Cognitive processes are assumed to be independent
Mixed Blocked/Event-related Design
…
Target-related Activity (Phasic)
Blocked-related Activity (Tonic)
Task-Initiation Activity (Tonic)
Task-Offset Activity (Tonic)
…
Mixed designs
Donaldson et al., 2001
3b. Variable Stimulus Probability
• Stimulus probability is varied in a blocked
fashion
– Appears similar to the combination design
• Mixed design used to maximize experimental
power for single design
• Assumes that processes of interest do not vary
as a function of stimulus timing
– Cognitive processing
– Refractory effects
Random and Semi-Random Designs
From Liu et al., 2001
Summary of Experiment Design
• Main Issues to Consider
– What design constraints are induced by my task?
– What am I trying to measure?
– What sorts of non-task-related variability do I want to avoid?
• Rules of thumb
– Blocked Designs:
• Powerful for detecting activation
• Useful for examining state changes
– Event-Related Designs:
• Powerful for estimating time course of activity
• Allows determination of baseline activity
• Best for post hoc trial sorting
– Mixed Designs
• Best combination of detection and estimation
• Much more complicated analyses