Transcript Statistical Parametric Mapping

Statistical Parametric Mapping
Lecture 4a - Chapter 7
Spatial and temporal resolution of fMRI
Textbook: Functional MRI an introduction to methods, Peter Jezzard, Paul
Matthews, and Stephen Smith
Many thanks to those that share their MRI slides online
Spatial and Temporal Resolution
Issues
• Spatial Resolution
– Spatial sampling and alaising
– Partial volume averaging alters strength of response
based on voxel size and size of responding region
• Temporal Resolution
– Temporal sampling and averaging
– Would like to sample electrical activity which happens
earlier than BOLD
– Order and timing of events would improve modeling
capabilities
Spatial Resolution Issues
• Excitatory and Inhibitory neural activity are both
energy consuming, but upstream inhibited
neurons produce less neuronal activity.
• Need to cover all regions of brain involved in the
tested brain tasks (whole brain preferred).
– Activity could be weaker due to partial volume effects
at smaller nodes in a system level activated brain
network.
– Need to improve task induced change and reduce
partial volume averaging.
• Position errors due to veins, macroscopic
susceptibility, etc.
Impact of Spatial Resolution
• Extent of BOLD response (rb) is related to the extent of neurovascular response (rv) and the imaging spatial resolution extent (rs).
• General relationship
• rb2 = rv2 + rs2
• BOLD signal is variable due to partial volume averaging
• When rv < rs (voxel larger than signal region)
• rb ~ rs
• Bold signal is reduced by partial volume averaging
• When rv > rs (voxel smaller than signal region)
• rb ~ rv
• BOLD signal minimally affected by rs
Based on classical linear system where
output(x,y,z) = input(x,y,z)  PSF(x,y,z)
But?
Figure 8.1. from textbook.
Figure 7.3 from textbook.
3.6
3
2
initial
dip
overshoot
post stimulus
undershoot
1
0
fMRI response ratio
BOLD response, %
positive
BOLD response
3.2
2.8
2.4
2.0
stimulus
time
1.6
0
4
8
12
16
Stimulus duration (s)
20
Response extent
• Initial dip – localized response (low signal)
• Overshoot next in extent (high signal)
• Plateau has greatest extent (high signal)
•
•
fMRI response ratio drops off with
stimulus duration
Dilution of signal into larger extent
seems to be dominant effect
Two Main Focus Points
• Responding well to changing hemodynamics
– Initial dip in BOLD response more spatially specific to activated
brain area than later rise in response, but later phase response is
larger and needed for fMRI.
– Hyperoxic response more broadly distributed spatially.
• Techniques to eliminate unwanted contributions to signal
(increase CNR).
– Short duration stimuli seem to be more narrowly distributed spatially
than long duration stimuli in BOLD studies.
– Higher B0 appears to improve microvascular signals more than
interfering signals
– Better RF coils improve SNR
– Improved motion correction improves CNR
– Multi-shot EPI to reduce T2* blurring supports smaller voxels
Statistical Parametric Mapping
Lecture 4b - Chapter 6
Selection of the optimal pulse
sequence for fMRI
Textbook: Functional MRI an introduction to methods, Peter Jezzard, Paul
Matthews, and Stephen Smith
Many thanks to those that share their MRI slides online
Advantages
BOLD
Disadvantages
Highest activation contrast 2x-4x over
perfusion (SPMs less noisy)
complicated non-quantitative signal
easiest to implement
no baseline information
multislice trivial
susceptibility artifacts
can use very short TR
Perfusion
unique and quantitative information
low activation contrast (need more
temporal averaging)
baseline information
longer TR required
easy control over observed vasculature
multislice is difficult
non-invasive
slow mapping of baseline
information
no susceptibility artifacts
Table 6.1a. Summary of practical advantages and disadvantages of pulse
sequences (derived from textbook)
Advantages
Volume
Disadvantages
unique information
invasive
baseline information
susceptibility artifacts
multislice trivial
requires separate run for each task
rapid mapping of baseline information
CMRO2
unique and quantitative information
semi-invasive
extremely low activation contrast
susceptibility artifacts
processing intensive
multislice is difficult
longer TR required
Table 6.1b. Continued summary of practical advantages and disadvantages of
pulse sequences (derived from textbook)
Time/secs
0
1
2
3
4
Perfusion
Venous outflow
No
Velocity
Nulling
Velocity
Nulling
Arteries
Arterioles Capillaries Venules
Veins
TI
ASL
Figure 6.1a Signal is detected from water spins in the arterial-capillary region of the
vasculature and from water in tissues surrounding the capillaries. Relative sensitivity
controlled by adjusting TI and by incorporating velocity nulling gradients (also known as
diffusion weighting). Nulling and TI~1 sec makes ASL sensitive to capillaries and surrounds.
Time/secs
Arterial inflow
(BOLD TR < 500 ms)
0
1
2
3
4
GE-BOLD
No
Velocity
Nulling
Velocity
Nulling
Arteries
Arterioles Capillaries Venules
Veins
Figure 6.1b Gradient Echo BOLD is sensitive to susceptibility perturbers of all sizes, and is
therefore sensitive to all intravasculature and extravascular effects in the capillary-venous
portions of the vasculature. If a very short TR is used may show signal from arterial inflow,
which can be removed by using a longer TR and/or outer volume saturation.
Time/secs
Arterial inflow
(BOLD TR < 500 ms)
0
1
2
3
4
SE-BOLD
No
Velocity
Nulling
Velocity
Nulling
Arteries
Arterioles Capillaries Venules
Veins
Figure 6.1c Spin Echo BOLD is sensitive to susceptibility perturbers about the size of a red
blood cell or capillary, making it predominantly sensitive to intravascular water spins in vessels
of all sizes and to extravascular (tissue) water surrounding capillaries. Velocity nulling reduces
the signals from larger vessesl.
Maximizing Signal
• Field Strength and sequence parameters
– Higher B means higher SNR but more susceptibility issues
– TE ~ T2* (30-40 msec @ 3T) for best activation contrast
– TR large enough to cover volume of interest, sampling time
consistent with experiment, >500 msec recommended, T1
increases with increasing B
• RF coils
– Larger coil for transmit
– Smaller coil for receive
– RF inhomogeneity increases with B
• Voxel size
– Match to volume of smallest desired functional area
– 1.5x1.5x1.5 suggested as optimal (Hyde et al., 2000)
– T2* increase and activation signal increase with small voxels if
shim is poor
Maximizing Signal
• Reducing physiological fluctuations
– Cardiac and breathing artifacts (sampling issues)
– Filtering to remove artifactual frequencies from time
signal, breathing easier to manage by filtering
– Pulse sequence strategies
• Snap shot (EPI) each image in 30-40 msec reduces
impact of artifacts
• Multi-shot ghosting (spiral imaging, navigator pulses,
retrospective correction)
– Gating
• Acquiring image at consistent phase of cardiac cycle or
respiration
• Problems (changing heart rate, wasted time)
Minimizing Temporal Artifacts
• Brain activation paradigm timing
– On-off cycles usually > 8 seconds
– Maximum number of cycles and maximum
contrast between
– Cycling activations no longer than 3-4 minutes
• Post processing
– Motion correction
• Real time fMRI
– Monitoring immediately and repeat if artifacts
are excessive
– Tuning of slice location
Minimizing Temporal Artifacts
• Physical restraint
– Limited success
– Cooperative subject helps
• Pulse sequence strategies
– Clustered acquisition (auditory stimulation 4-6
seconds before acquisition)
– Set phase encode direction to minimize overlap with
brain areas of interest
– Select image plane with most motion to minimize
between plane motion artifacts
– Crusher gradients to minimize inflow artifacts
Statistical Parametric Mapping
Lecture 4c - Chapter 4
More fMRI
Textbook: Functional MRI an introduction to methods, Peter Jezzard, Paul
Matthews, and Stephen Smith
Many thanks to those that share their MRI slides online
Effects of Field Homogeneity
R2* = R2 + R2mi +R2ma
• R2 = transverse relaxation rate due to spin-spin
interactions and diffusion through microscopic
gradients
• R2mi = transverse relaxation rate due to
microscopic changes, i.e. deoxyhemoglobin
• R2ma = transverse relaxation rate due to
macroscopic field inhomogeneity
R2*a is relaxation rate during activation
R2*r is relaxation rate at rest
R 2*  ( R 2 *a  R 2 *r )  ( R 2  R 2 mi  R 2 ma ) a  ( R 2  R 2 mi  R 2 ma ) r
R 2*  ( R 2  R 2 mi ) a  ( R 2  R 2 mi ) r  ( R 2 a  R 2 r )  ( R 2 mi a  R 2 mi r )
R 2*  R 2  R 2 mi
Note: macroscopic
components subtract off
Approximate GM Relaxation And Activation
Induced Rexalation Rate Changes
1.5T
3T
T2
100 ms
80 ms
T2*
60 ms
50 ms
T2’
150 ms
133.3 ms
R2 = (1/T2)
-0.2 s-1
-0.4 s-1
R2* = (1/T2*)
-0.8 s-1
-1.6 s-1
R2’ = (1/T2’)
-0.6 s-1
-1.2 s-1
• T2, T2* and T2’ (from ASE) of GM decrease with increasing field strength
• During activation relaxation rates decrease (T2 increase) slightly
• Activation induced changes in relaxation rates (R2s) indicate potential for
signal production
Echo Time Optimization
0.025
T2*r=80ms
Signal, arb
0.020
70ms
TEopt = optimal TE for BOLD
contrast lies between T2*a
and T2*r
60ms
0.015
50ms
0.010
40ms
30ms
20ms
10ms
0.005
0.000
T2*a = 1/R2*a
T2*r = 1/R2*r
0
50
100
TE, ms
150
Subscripts a and r
indicate during activation
and rest.
Fig. 4.1 BOLD response as a function of TE for different values of T2*r. Note that TEopt
~ T2* and that BOLD response increases with increasing T2*r.
Effects of Field Homogeneity
number of voxels
4000
1.9mm
3.8mm
5.9mm
3000
2000
1000
0
0
50
100
T2*, ms
150
Fig. 4.2 Change in histogram of T2* for thick slab through brain with changing slice
thickness. Note broadening of distribution with increasing thickness with shift away
from T2*a toward shorter T2*r.
4x4x4 mm3
2x2x2 mm3
Spin Echo
Gradient Echo EPI
Fig. 4.3 EPI obtained with TE= 60 and TR=3000 msec and 63 and 95 ky lines. Note
recovery of signal loss in d vs c and ghosting in c.
navigator phase, degrees
Intra-scan Motion Signal
0.2
0.1
0.0
-0.1
-0.2
-0.3
0
500
1000
navigator index
1500
Fig. 4.4 Phase fluctuations at center of k-space over 42 seconds. Spikes are due to
cardiac cycles and slower periodic signal due to respiratory cycles.
Why would phase advance and retard?
Statistical Parametric Mapping
Lecture 4d
The big Picture of Brain
Many thanks to those that share their MRI slides online
Brain Lobes +
Frontal Lobe
Cerebrum Lobes
Occipital Lobe
• Frontal
Temporal Lobe
Cerebellum
• Parietal
• Temporal
• Occipital
Brainstem
Brodmann’s Functional Map
Mango and Anatomy
• Talairach Daemon (TD)
– Anatomical/functional labels
– 5 hierarchical levels
•
•
•
•
•
Hemispheres
Lobes
Gyri
Tissue
Cellular
• Spatial Normalization
– Supports x-y-z coordinate lookup of
anatomical/functional labels using the TD