fMRI_Basics_files/fMRI Presentation Web

Report
fMRI
Sonia Poltoratski
Vanderbilt University
unbridled
joy
brain
pictur
e
data
intro
psych
analysis
crippling
depression
what is
BOLD?
...is the
wild wild
Outline:
• MR Physics
• BOLD signal
• Basics of Analysis
• Evolution
• Good & Bad Practices
MR Physics
• MR in humans = proton nuclear
magnetic resonance, which detects the
presence of hydrogen nuclei
electron
-
• Since the single proton of hydrogen in
unbalanced, normal thermal energy
causes it to spin about itself
+
proton
Spins
• The proton’s positive charge
generates an electrical current
• In a magnetic field, this loop current
induces torque, called the magnetic
moment (μ)
• The proton’s odd-numbered atomic
mass gives it an angular momentum
(J)
μ
J
+ +
+
++ ++
+ ++
proton
Net magnetization (M)
Negligible under normal conditions
B0
magnetic field
Proton Precession
• Spinning objects respond to
applied forces by moving their
axes perpendicular to the
applied force
Proton Precession
magnetic field
• Spinning objects respond to
precession
axis
applied forces by moving their
axes perpendicular to the
applied force
magnetic field
Proton Precession
parallel state
(low energy level)
anti-parallel state
(high energy level)
M
transverse
longitudinal
magnetic field
Net Magnetization (M)
Net Magnetization (M)
Increasing magnetic field  increase in net magnetization
energy
The Zeeman Effect
ΔE
magnetic field strength
magnetic field B0
Signal Generation
excitation B1
photons: electromagnetic
fields oscillating at the
resonate (Larmor) frequency
of hydrogen
magnetic field B0
Signal Generation: Net M
excitation B1
M
θ
flip angle
Signal Reception
magnetic field B0
reception
decaying, time-varying signal
that depends on the
molecular environment of
the spins
Signal Reception
T1 recovery (longitudinal relaxation):
Individual spins return to their low-energy state, and net M
becomes again parallel to the main field
T2 decay (transverse relaxation):
Immediately after excitation, spins precess in phase
This coherence is gradually lost
Images depict the spatial distribution of these properties
T1 Relaxation Times
Fat
White Matter
Grey Matter
CSF
T2 Decay Times
Fat
White Matter
CSF
Grey Matter
Image Formation
• Magnetic gradient: spatially varying
magnetic field
• Adding a second gradient field
causes spins at different locations to
precess at different frequencies in a
predictable manner
Paul C. Lauterbur and Sir Peter Mansfield
at the 2003 Nobel Prize Ceremony
Image Formation
longitudinal
magnetization
slice excitation
transverse
magnetization
2D spatial encoding
acquired MR
signal in kspace
2D inverse Fourier
transform
2D MR
image
Slice Excitation
ƒ
resonant frequency
vs. position
slice direction
Slice Excitation
ƒ
resonant frequency
vs. position
when gradient is applied
frequency range of
RF pulse
slice direction
excited slice
2D Spatial Encoding
A gradient field that differs along two dimensions results in a unique
frequency assigned to each location in the space, influencing the
location’s spin phase
• Phase encoding gradient: turned on before data acquisition so that spins
accumulate differential phase offset over space
• Frequency encoding gradient: turned on during data acquisition so that the
frequency of spin precession changes over space
Resulting data is in units of spatial frequency, which can be converted
into units of distance via inverse Fourier transform
Echo Planar Imaging (EPI) allows us to collect an entire imagine in
milliseconds, either following 1 excitation (single-shot) or several
(multi-shot)
T1-Weighted Image
T2-Weighted Image
Pop Quiz!
MRI data acquisition
The experimental data were collected at the Vanderbilt
University Institute for Imaging Science using a 3T Philips Intera
Achieva MRI scanner with an eight-channel head coil. The
functional data were acquired using standard gradient-echo
echoplanar T2*-weighted imaging with 28 slices, aligned
approximately perpendicular to the calcarine sulcus and covering
the entire occipital lobe as well as the posterior parietal and
posterior temporal cortex (TR, 2 s; TE, 35 ms; flip angle, 80°;
FOV, 192 x 192; slice thickness 3 mm with no gap; in-plane
resolution, 3 x 3 mm). In addition to the functional images, we
collected a T1-weighted anatomical image for every subject (1
mm isotropic voxels). A custom bite bar system was used to
minimize the subject’s head motion.
Keitzmann, Swisher, Konig, & Tong (2012)
Pop Quiz!
MRI data acquisition
The experimental data were collected at the Vanderbilt
University Institute for Imaging Science using a 3T Philips Intera
Achieva MRI scanner with an eight-channel head coil. The
functional data were acquired using standard gradient-echo
echoplanar T2*-weighted imaging with 28 slices, aligned
approximately perpendicular to the calcarine sulcus and covering
the entire occipital lobe as well as the posterior parietal and
posterior temporal cortex (TR, 2 s; TE, 35 ms; flip angle, 80°;
FOV, 192 x 192; slice thickness 3 mm with no gap; in-plane
resolution, 3 x 3 mm). In addition to the functional images, we
collected a T1-weighted anatomical image for every subject (1
mm isotropic voxels). A custom bite bar system was used to
minimize the subject’s head motion.
Keitzmann, Swisher, Konig, & Tong (2012)
Outline:
• MR Physics
• BOLD signal
• Basics of Analysis
• Evolution
• Good & Bad Practices
BOLD signal
Blood-Oxygen-Level-Dependent Contrast (Thulborn et al., 1982; Ogawa, 1990)
Oxygenated
Hemoglobin
Diamagnetic (no unpaired electrons or
magnetic moment)
Deoxygenated
Hemoglobin
Paramagnetic (significant magnetic
moment)
20% greater magnetic susceptibility,
which impacts T2 decay
BOLD signal
Blood-Oxygen-Level-Dependent Contrast (Thulborn et al., 1982; Ogawa, 1990)
Oxygenated
Hemoglobin
Deoxygenated
Hemoglobin
Diamagnetic (no unpaired electrons or
magnetic moment)
Paramagnetic (significant magnetic
moment)
20% greater magnetic susceptibility,
deoxygenated
bloodT2isdecay
present, the
which impacts
The more
shorter the T2
Difference emerges at ~ 1.5T
Ogawa (1990)
• Blood oxygen content in rodents reflected in T2-weighted
images
• Metabolic demand for oxygen (confirmed by concurrent EEG)
is necessary for BOLD contrast
“
During an MRI experiment with an
anesthetized mouse, I saw most of the
dark lines disappear when the
breathing air was switched to pure O2
in order to rescue the mouse as it
appeared to start choking. This
observation rang a bell.
log size
fMRI vs. Other Methods
brain
MEG & ERP
map
Optical
Imaging
fMRI
Natural
Lesions
TMS
column
layer
neuron
PET
Induced Lesions
Multi-unit
recording
Single Unit
dendrite
Patch Clamp
Light Microscopy
synapse
millisecond
second
minute hour
log time
day
Outline:
• MR Physics
• BOLD signal
• Basics of Analysis
• Evolution
• Good & Bad Practices
Voxels
1mm x 1mm x 1.5mm voxels
7mm x 7mm x 10mm voxels
(Smith, 2004)
Preprocessing Stages
• Slice-timing correction: correcting for differences in
acquisition times within a TR
• Motion correction: re-alignment of images across the session
• Spatial smoothing: blurring of neighboring data points, akin to
low-pass filtering.
Preprocessing Stages
• Mean intensity adjustment: normalization of signal to account
for global drifts over time
• Temporal high-pass filtering: removal of low-frequency drifts
in time course
peak
stimulus
percent MR signal change
Hemodynamic Response Function
undershoot
initial dip
-10
-5
0
5
10
time (s)
15
20
25
Modeling the Waveform
J
HRF
fit this model to
the time series
of each voxel
J
block design
J
General Linear Modeling
Y =X. β +ε
observed
data at a
single voxel
design matrix
estimated
parameters
error
test if the slope of β is
different from zero
t stat at
each voxel
anatomical
scan image
=
Outline:
• MR Physics
• BOLD signal
• Basics of Analysis
• Evolution
• Good & Bad Practices
Nature (2012)
Voxel Resolution
Kanwisher, McDermott, & Chun (1997):
3.25 x 3.25 x 6 mm
McGugin et al. (2013):
1.25 x 1.25 x 1.25 mm
TR Duration
(not my)
unpublished
data removed
for web use
(Tong Lab data)
7Tesla, TR = 200ms
Outline:
• MR Physics
• BOLD signal
• Basics of Analysis
• Evolution
• Good & Bad Practices
‘The Seductive Allure of Neuroimaging’
“
Non-experts judge explanations with neuroscience
information as more satisfying than explanations without
neuroscience, especially bad explanations.
”
(Weisberg et al., J Cog Neuro 2008)
The Nader Effect
Pitfalls in fMRI
• Study Design
• What is your contrast?
• What conclusions can we draw from fMRI activation?
• Statistical Analysis
vs
Correcting for Multiple Comparisons
(Bennett et al. 2010)
Voodoo Correlations in Social Neuroscience
Puzzlingly High Correlations in fMRI Studies of Emotion, Personality, & Social Cognition
Vul et al. (2009)
• Noticed R > 0.8 correlations, seemingly
higher than possible under constraints
of fMRI and variability of personality
measures
• Non-independence error:
• Selecting a small number of voxels
based on some trait
• Only reporting the correlation of the
trait to those voxels
• 54% of surveyed papers, including those
published in Science, Nature, and
Neuron
Pitfalls in fMRI
• Study Design
• What is your contrast?
• What conclusions can we draw from fMRI activation?
• Statistical Analysis
• Correction for Multiple Comparisons
• Independently-selected ROI’s
• Software & Human Error
Act carefully and critically at all
stages
of fMRI research!
The Finer Things in fMRI
Event-Related
Design
fMRI-A:
Adaptation
Multi-Voxel
Pattern Analysis
Event Related Design
• Allows us to mix events of
different types, avoiding
effects related to blocking
J
J
J
block design
• Events can be categorized
or defined post-hoc based
on subject’s responses
• In slow ERD, the BOLD
response is allowed to
return to baseline between
events
J
J
J
event-related design
Rapid Event Related Design
events:
individual HRFs:
summed HRFs:
(BAD)
J Q I J Q I J Q I
(GOOD
Rapid Event Related Design
events:
jittered order
& ISI
individual HRFs:
summed HRFs:
JQ
Q
IJ
I
fMRI-A: Adaptation
The resolution of fMRI
makes it difficult to
distinguish between
homogenous and
heterogenous populations:
Neuronal population is adapted by repetition of a stimulus
2. Some property of the stimulus is changed
3. Recovery from adaptation is assessed:
1.
• Signal remains adapted = neurons are invariant
• Signal recovers = neurons are sensitive to the changed property
(Grill-Spector & Malach, 2001)
Example: Face Viewpoint Invariance
Adapt to identical view
Change the property of interest
In both cases, signal is reduced
In (L) case, signal recovers
(Grill-Spector & Malach, 2001)
Multi-Voxel Pattern Analysis
(re: Kamitani & Tong, 2005)
Multi-Voxel Pattern Analysis
• AKA: fMRI decoding, MVPA, multivariate analysis
• In univariate analysis described so far, we:
• Assume independence of each voxel
• Test whether each voxel responds more to one condition than the other
• MVPA is designed to test whether 2+ conditions can be
distinguished based on activity pattern in a set of voxels
• Critically, MVPA can sometimes identify differences in
conditions when average activity is equal
(review: Pratte & Tong, 2012)
Multi-Voxel Pattern Analysis
a. Subjects view stimuli from two
categories & feature selective
voxels are selected
b. Data is divided into training and
test runs; Training voxel patterns
are decomposed and tagged by
category
c. Training runs are input to a
classifier function
d. The classifier defines a multidimensional decision boundary,
and category membership for the
test run is predicted
(review: Norman et al., 2006)
(xkcd.com)

similar documents