DATA QUALITY AND PREPROCESSING

1
DATA QUALITY AND PREPROCESSING
2
The Black Box

• The danger of automated processing and fancy
  images is that you can get blobs without every
  really looking at the real data
• The more steps done at once, the greater the
  chance of problems

Big Black Box of automated software
Raw Data
Pretty pictures
3
Know Thy Data

• Look at raw functional images
• Where are the artifacts and distortions?
• How well do the functionals and anatomicals
  correspond
• Look at the movies
• Is there any evidence of head motion?
• Is there any evidence of scanner artifacts (e.g.,
  spikes)
• Look at the time courses
• Is there anything unexpected (e.g., abrupt signal
  changes at the start of the run)?
• What do the time courses look like in the
  unactivatable areas (ventricles, white matter,
  outside head)?
• Look at individual subjects
• Double check effects of various transformations
• Make sure left and right didnt get reversed
• Make sure functionals line up well with
  anatomicals following all transformations
• Think as you go. Investigate suspicious
  patterns.

4
Sample Artifacts
Ghosts
Metallic Objects (e.g., hair tie)
Hardware Malfunctions
Spikes
5
Calculating SignalNoise Ratio
Pick a region of interest (ROI) outside the brain
free from artifacts (no ghosts, susceptibility
artifacts). Find mean (?) and standard deviation
(SD).
Pick an ROI inside the brain in the area you care
about. Find ? and SD.
e.g., ?4, SD2.1
SNR ?brain/ ?outside 200/4 50
Alternatively SNR ?brain/ SDoutside 200/2.1
95 (should be 1/1.91 of above because ?/SD
1.91)
When citing SNR, state which denominator you
used.
e.g., ? 200
Head coil should have SNR gt 501 Surface coil
should have SNR gt 1001
Source Joe Gati, personal communication
6
Why SNR Matters
Note This SNR level is not based on the formula
given
Huettel, Song McCarthy, 2004, Functional
Magnetic Resonance Imaging
7
Sources of Noise

• Physical noise
• Blame the magnet, the physicist, or the laws of
  physics
• Physiological noise
• Blame the subject

8
What affects SNR?Physical factors
Source Doug Nolls online tutorial
9
Field Strength
Huettel, Song McCarthy, 2004, Functional
Magnetic Resonance Imaging

• Although Raw SNR goes up with field strength, so
  does thermal and physiological noise
• Thus there are diminishing returns for increases
  in field strength

10
Coils

• Head coil
• homogenous signal
• moderate SNR

• Surface coil
• highest signal at hotspot
• high SNR at hotspot

11
Phased Array Coils

• A recent trend is the move to phased array coils
  to get the SNR of surface coils with the coverage
  of head coils or to get faster parallel imaging
• New magnets are shipping with 32 channels

Huettel, Song McCarthy, 2004, Functional
Magnetic Resonance Imaging
12
Voxel size

• Bigger is better to a point
• Increasing voxel size ? signals summate, noise
  cancels out
• Partial voluming If tissue is of different
  types, then increasing voxel size waters down
  differences
• e.g., gray and white matter in an anatomical
• e.g., activated and unactivated tissue in a
  functional

13
Sampling Time

• More samples ? More confidence effects are real

14
What affects SNR?Physiological factors
Source Doug Nolls online tutorial
15
A Map of Noise
Huettel, Song McCarthy, 2004, Functional
Magnetic Resonance Imaging

• voxels with high variability shown in white

16
WHY HEAD MOTION SUCKSAND WHAT YOU MIGHT BE
ABLE DO ABOUT IT
17
Head Motion Main Artifacts

• Head motion can lead to spurious activations or
  can hinder the ability to find real activations.
• Severity of problem depends on correlation
  between motion and paradigm
• Head motion increases residuals, making
  statistical effects weaker.
• Regions move over time
• ROI analysis ROI may shift
• Voxelwise analyses averages activated and
  nonactivated voxels
• Motion of the head (or any other large mass)
  leads to changes to field map
• Spin history effects
• Voxel may move between excitation pulse and
  readout

18
Motion ? Intensity Changes
A
B
C
Slide modified from Duke course
19
Motion ? Spurious Activation at Edges
lateral motion in x direction
motion in z direction (e.g., padding sinks)
brain position
stat map
20
Spurious Activation at Edges

• spurious activation is a problem for head motion
  during a run but not for motion between runs

21
Motion ? Increased Residuals
?1



?2

fMRI Signal
Residuals
Design Matrix

Betas
x
what we CAN explain
what we CANNOT explain
how much of it we CAN explain


x
our data
Statistical significance is basically a ratio of
explained to unexplained variance
22
Regions Shift Over Time

• A time course from a selected region will sample
  a different part of the brain over time if the
  head shifts
• For example, if we define a ROI in run 1 but the
  head moves between runs 1 and 2, our defined ROI
  is now sampling less of the area we wanted and
  more of adjacent space
• This is a problem for motion between runs as well
  as within runs

?
time1
time2
23
Motion Correction Algorithms
pitch
roll
yaw
z translation
y translation
x translation

• Most algorithms assume a rigid body (i.e., that
  brain doesnt deform with movement)
• Align each volume of the brain to a target volume
  using six parameters three translations and
  three rotations
• Target volume the functional volume that is
  closest in time to the anatomical image

24
BVQX Motion Correction Options
Analysis/fMRI 2D data preprocessing menu

• Motion correct .fmr file (2D) before any other
  preprocessing
• Why?
• Align each volume to the volume closest to the
  anatomical
• Why?

25
Head Motion Good, Bad,
Slide from Duke course
26
and catastrophically bad
Slide from Duke course
27
Problems with Motion Correction

• lose information from top and bottom of image
• possible solution prospective motion correction
• calculate motion prior to volume collection and
  change slice plan accordingly

were missing data here
we have extra data here
Time 1
Time 2
28
Why Motion Correction Can Be Suboptimal

• Parts of brain (top or bottom slices) may move
  out of scanned volume (with z-direction motion or
  rotations)
• Motion correction requires spatial interpolation,
  leads to blurring
• fast algorithms (trilinear interpolation) arent
  as good as slow ones (sinc interpolation)
• Motion correction

29
Why Motion Correction Algorithms Can Fail

• Activation can be misinterpreted as motion
• particularly problematic for least squares
  algorithms (Friere Mangin, 2001)
• Field distortions associated with moving mass
  (including mass of the head) can be
  misinterpreted as motion

Spurious activation created by motion correction
in SPM (least squares)
Mutual information algorithm in SPM has fewer
problems
Friere Mangin, 2001
Simulated activation
30
Mass Motion Artifacts

• motion of any mass in the magnetic field,
  including the head, is a problem

31
Head Motion Field Map Artifacts
Phantom

• Bag of Saline on a Stick
• experimenter moves saline left and right every
  30 sec without touching subject or phantom

Data from Jody Culham
32
A. Pre-corrected Statistical Map 1
B. Time Course 1
7
1.0
Left
Right
Left
Right
Left
.60
Signal Change
0
-.60
-4
-1.0
r value
C. Pre-corrected Statistical Map 2
D. Time Course 2
900
Signal Change
0
0
F. Motion Correction Parameters
E. Post-corrected Statistical Map 1
Data from Jody Culham
33
Head Motion Solution to Susceptibility

• Solution
• one trial every 10 or 20 sec
• fMRI signal is delayed 5 sec
• distinguish true activity from artifacts
• IMPORTANT Subject must remain in constant
  configuration between trials

34
Different motions different effects
35
The Fridge Rule

• When it doubt, throw it out!

36
Head Restraint
Vacuum Pack
Head Vise (more comfortable than it sounds!)
Bite Bar
Thermoplastic mask
Often a whack of foam padding works as well as
anything
37
Prevention is the Best Remedy

• Tell your subjects how to be good subjects
• Dont move is too vague
• Make sure the subject is comfy going in
• avoid princess and the pea phenomenon
• Emphasize importance of not moving at all during
  beeping
• do not change posture
• if possible, do not swallow
• do not change posture
• do not change mouth position
• do not tense up at start of scan
• Discourage any movements that would displace the
  head between scans
• Do not use compressible head support
• For a summary of info to give first-time
  subjects, see
• http//defiant.ssc.uwo.ca/Jody_web/Subject_Info/fi
  rsttime_subjects.htm

38
BV Preprocessing Options
39
Spatial Smoothing

• Application of Gaussian kernel
• Usually expressed in mm FWHM
• Full Width Half Maximum
• Typically 2 times voxel size

Slide from Duke course
40
Effects of Spatial Smoothing on Activity
Unsmoothed Data
Smoothed Data (kernel width 5 voxels)
Slide from Duke course
41
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

Slide from Duke course
42
BV Preprocessing Options
43
A Brief Primer on Fourier Analysis

• Sine waves can be characterized by frequency and
  amplitude

peak high point trough low point frequency
number of cycles within a certain time or space
(e.g., cycles per sec Hz, cycles per
cm) amplitude height of wave phase starting
point
amplitude
peak
trough

• (b) has same frequency as (a) but lower
  amplitude
• (c) has lower frequency than (a) and (b)
• (d) has same frequency and amplitude as (c) but
  different phase

Source DeValois DeValois, Spatial Vision, 1990
44
Fourier Decomposition

• Any wave form can be decomposed into a series of
  sine waves

Frequency spectrum
Source DeValois DeValois, Spatial Vision, 1990
45
Time Course Filtering
46
Temporal and Spatial Analysis

• Temporal waveforms
• e.g., sound waves
• e.g., fMRI time courses
• Spatial waveforms
• can be one dimensional (e.g., sine wave gratings
  in vision) or two dimensional (e.g., a 2D image)
• e.g., image analysis
• e.g., an fMRI slice (k-space)

Source DeValois DeValois, Spatial Vision, 1990
47
Fourier Synthesis

• centre low frequencies
• periphery high frequencies
• You can see how the image quality grows as we add
  more frequency information

Source DeValois DeValois, Spatial Vision, 1990
48
K-Space
Source Travelers Guide to K-space (C.A.
Mistretta)
49
Linear Drift
Huettel, Song McCarthy, 2004, Functional
Magnetic Resonance Imaging
50
Low and High Frequency Noise
Source Smith chapter in Functional MRI An
Introduction to Methods
51
BV Preprocessing Options
Before LTR
After LTR
52
BV Preprocessing Options

• High pass filter
• pass the high frequencies, block the low
  frequencies
• a linear trend is really just a very very low
  frequency so LTR may not be strictly necessary if
  HP filtering is performed (though it doesnt hurt)

Before High-pass
After High-pass
linear drift
1/2 cycle/time course
2 cycles/time course
53
BV Preprocessing Options

• Gaussian filtering
• each time point gets averaged with adjacent time
  points
• has the effect of being a low pass filter
• passes the low frequencies, blocks the high
  frequencies
• for reasons we will discuss later, I recommend
  AGAINST doing this

After Gaussian (Low Pass) filtering
Before Gaussian (Low Pass) filtering
54
Physiological Noise

• Respiration
• every 4-10 sec (0.3 Hz)
• moving chest distorts susceptibility
• Cardiac Cycle
• every 1 sec (0.9 Hz)
• pulsing motion, blood changes
• Solutions
• gating
• avoiding paradigms at those frequencies

55
Take home Messages

• Look at your data
• Work with your physicist to minimize physical
  noise
• Design your experiments to minimize physiological
  noise
• Motion is the worst problem When in doubt, throw
  it out
• Preprocessing is not always a one size fits all
  exercise