Experimental Design & Data Acquisition for Mapping Brain Activity (Block Design) Flashcards
What does the experimental setup consist of?
- Stimulus
- Projector
- Headphones
- Tactile stimulation, mirrors, human presence - Subject response and monitoring
- Response box, controls
- Physiological monitoring-pulse oximetry, respiration belt, skin conductance
Why is there increased BOLD sensitivity with high field?
- The magnitude of the dipole that is created – inhomogeneity increases with the magnetic field
- The susceptibility of the tissue will create a high dipole – higher inhomogeneity of the magnetic field
What is the choice of field strength for the experimental set-up?
Uusually 1.5T or 3T (but fMRI at 7T becoming established)
What are the advantages of higher field strength?
- Image SNR and BOLD CNR increase
- increased functional sensitivity - Can use higher spatial resolution
- increased functional specifity
What are the disadvantages of higher field strength?
- Increases in image artefacts (distortions, dropout)
- Increase in image noise (thermal or physiological noise)
- May restrict scanning of certain subjects/patients with metal in their body
What are the 3 different types of coils for the experimental set up?
• Volume/birdcage coils
- Standard on most MRI scanners
• Surface coils:
- Increase in sensitivity for specific regions near coil (e.g. occipital cortex)
• Phased-array/multi-channel coils
- 32 channel coil
- Multiple surface coils (e.g. currently maximum = 64), increases sensitivity throughout head
- Can use parallel imaging to speed up acquisition if available
- Increased temporal resolution
What is block design well suited to localise?
Functional areas and study steady state processes (e.g. attention)
What is block design powerful in terms of?
Detection i.e. to determine which voxels are activated
- signal strength varies with the length of the blocks
What does haemodynamic response (HR) allow ?
The rapid delivery of blood to active neuronal tissues
The brain consumes large amounts of energy but does not have a reservoir of stored energy substrates
What is cerebral blood flow essential for?
Maintenance of neurons, astrocytes and other cells of the brain
What does signal strength vary with?
Length of blocks
When does the signal not return to baseline during null-blocks decreasing the strength of the signal?
With short blocks (less than 10s)
Why does the detection power increase with high frequency alteration?
- It depends on the number of events/blocks
- The noise in the BOLD time course which occurs mainly at low frequencies
- Blocks with duration longer than the haemodynamic response reach a compromise between signal strength and noise
What is the advantage of having short blocks (<10s)
Time to have a lot of blocks
- Statistical power increases
What is the disadvantages of short blocks (<10s)?
The signal doesn’t have time to go back to the baseline
- The BOLD effect is smaller
What is the advantage of long blocks?
Large response during the task and has time to go back to baseline during rest
- High BOLD response
What is the disadvantage of having long blocks?
No time to have a lot of blocks - Statistical power decreases
- BOLD response and noise at similar frequency -> detection power smaller
What are the typical fMRI acquisition parameters
• Length of an fMRI run:
- 10-20 minutes, up to ~3 runs in a session (e.g. 200-500 volts)
• Dummy scans:
- Acquire at beginning (~4-5 volumes)
- Allow the longitudinal magnetisation to reach equilibrium
- Allow for subject to adjust to environment and noise
• Other:
- Time for training, behavioural testing, anatomical scans
What can the spatial and temporal resolution of fMRI be ultimately be limited by?
Characteristics of the physiological changes accompanying neuronal activation
Where can the BOLD signal originate?
Within and surrounding the smallest capillaries and the largest venous vessels
Where is the oxygenation changes detected?
Several mm downstream in the venous system from the site of neuronal activity
Why is mapping of the microvasculature required?
To remove signals in large vessels to optimise the spatial accuracy of fMRI
What has been performed to investigate the physical mechanism of BOLD contrast?
Theoretical modelling of the magnetic susceptibility gradients and Monte-Carlo stimulations
What is the consequence of having very low volume of blood within the brain?
Extravascular spin contribute heavily to BOLD contrast because local magnetic field gradient exist around vessels
- Remove large vessels from activation images
What is the ultimate temporal resulotion of fMRI likely to be determined by?
Physiological dynamics than the rate at which the data can be acquired
What does the BOLD signal respond to ?
neuronal activity which operates on a time scale of 1ms (synaptic transmission) to 100s of ms (information transport) is heavily lagged and damped by the haemodynamic response
What does the peak BOLD signal in activated primary cortex have?
measured latency of 5-8 seconds from stimulus onset and a similar time is required for the signal to return to a baseline upon stimulus cessation
What are the requirement for spatio-temporal?
- Volume TR [volume acquisition time]< 5s
- TE = 30 ms (at 3T)
- Spatial coverage of up to 15cm
What are main characteristic of EPI sequences?
- Rapid switching of gradients
* One excitation, one full k-space
What are limits of EPI?
- Fast k-space acquisition requires strong gradients
- Grading switching limited by peripheral nerve stimulation and gradient system (caveat heating)
- Readout times limited approximately T2* for high fidelity point-spread functions
What are functional MRI techniques deliberately sensitised to?
differences in T2* so that magnetic properties of deoxy-haemoglobin can manifest themselves as image contrast• Therefore sensitised to other sources of magnetic field inhomogeneity
What does components of inhomogeneity within imaging plan ecause?
Geometric distortion of image
What causes large scale inhomogeneity?
The magnetic susceptibility differences between air, bone and different tissue types
What causes a loss of image intensity due to spin dephasing?
Those that co-linear with slice selection direction
Why is geometric distortion particularly serious problem with EPI?
Very low frequency per point in the phase encoding direction
• Long TE = more dropout
• Long readout time = bigger distortions
When do human tissues exhibit changes in the magnetic field?
About 1-2ppm
What is another common problem with echo-planar images?
Presence of Nquist ghost
What are Nquist ghost?
Low intensity secondary images which appear a half FOV away from the real image
What is the presence of Nquist ghost due to?
Timing of phase differences between odd and even echoes in the echo train
They are minimised by calibration using a pre-scan under a bipolar gradient prior to data acquisition
Why may Nqyuist ghost represent a problem in real fMRI?
They can be unstable over time due to shim and gradient amplifier instabilities
When is image realignmnt affected?
if the intensity between total power of real and ghost image is too great
What happens if the centre of k-space is not acquired?
Transverse DROPOUT
What are the limits of EPI?
- Distortion
2. Dropout
What is distortion?
- Pixel displacement in phase-encoding direction and signal non-uniformity
- Problem for spatial localisation of activations
- May lead to inaccurate coregistration and reduce sensitivity of group studies
What is dropout?
- Signal disappears due to complete signal dephasing (spurious gradinet within voxel along slice-selection direction) or due to the fact that the centre of k-space is not acquired (spurious gradient along the phase-encoding direction or the readout direction added to the imaging gradient
- Cannot get a signal from certain regions of brain
What is reduced drop out?
- Use shorter TEs (but need relatively long TE for BOLD effect)
- Use thinner slices
- Optimised EPI sequences (e.g. tilted slices, compensation gradient, Z-shimming, change phase-encoding polarity
What is the benefit of combined angulation and z-shimming?
- In-plane gradient is reduced by angling slices
2. Larger through slice susceptibility compensated by z-shimming
What is reduced distortion?
Shorter acquisition times, use parallel imaging
Optimised EPI sequences (e.g. tilted slices, compensation gradients, change phase-encoding polarity)
Use B0 field map to measure field inhomogeneity and then correct for it
What is 2D EPI for high spatio-temporal resolution?
3 mm isotropic resolution Volume TR ~ 3 s, depending on coverage ~ 70 ms / slice 64 x 64 matrix No parallel imaging Possibly with some z-shim and angulation
What is increased temporal resolution used for?
For real-time imaging
To acquire more samples per unit time
To better sample physiological effects
What is increased spatial resolution?
Spatial specificity, e.g. sub-nuclei / functional units
Reduce artefacts, e.g. susceptibility effects and physiological noise
What is 2D Multiband EPI?
Simultaneous excitation of multiple slices
Volume acquisition time reduced by MB factor
What is 3D EPI for high spatio-temporal resolution?
Volume excitation
2nd phase encoding direction
Volume acquisition time reduced by through-plane acceleration factor
What are the challenges for high spatio-temporal resolution?
Increased B0 inhomogeneity
Increased B1+ inhomogeneity
Higher Specific Absorption Rate (SAR)
What are the typical metrics for the estimate of functional sensitivity?
Signal to noise ratio (SNR)
Based on one volume :
𝑆𝑁𝑅=(𝑚𝑒𝑎𝑛 𝑠𝑖𝑔𝑛𝑎𝑙 )/(𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑡ℎ𝑒 𝑛𝑜𝑖𝑠𝑒)
What are examples of 2D vs 3D in terms of SNR?
As the coverage increases:
3D EPI: The signal combines additively but the noise in quadrature giving a N increase in SNR
2D EPI: The TR increases => flip angle increases, increasing the SNR until the Ernst angle plateaus at 900
What are examples of MB vs 3D in terms of SNR?
With acceleration, the gains in SNR are reversed as the volume repetition time decreases.
In practice, other factors play a role, e.g. g-factor penalties, as well as sensitivity to motion and physiological noise
What is another typical metrics for estimate of functional sensitivity?
Temporal Signal to Noise ratio (tSNR)
Based on a time-series:
𝑡𝑆𝑁𝑅=(𝑡𝑒𝑚𝑝𝑜𝑟𝑎𝑙 𝑚𝑒𝑎𝑛 𝑠𝑖𝑔𝑛𝑎𝑙)/(𝑡𝑒𝑚𝑝𝑜𝑟𝑎𝑙 𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑡ℎ𝑒 𝑠𝑖𝑔𝑛𝑎𝑙)
What is the equation for tSNR (A)?
𝑚𝑒𝑎𝑛(𝑆)/√(𝑣𝑎𝑟(𝑆) )
What is tSNR and what does it indicate?
- Key metric in fMRI
2. It indicates the minimum effect size detectable
What does the tSNR depend on?
the number of samples and temporal correlations of the data :
variants of the tSNR (√𝑁*tSNR, t- score of the mean)
How can you increase tSNR?
Increase SNR
Bigger voxels
Higher field
What limits tSNR improvement?
Physiological noise that introduces temporal variance
And the impact of physiological noise increases with the signal (voxel volume, field strength)
What are the source of noise in fMRI?
- Non-biological noise2
2. Biological noise
What is non-biological noise?
Instrumentation noise (e.g. scanner drift)
Thermal noise from subject and scanner
Other environmental noise
Increases ~ linearly with magnetic field
What is biological noise?
Head motion Non task-related neural activity Respiration (~ 0.3 Hz) Cardiac cycle (~ 0.8-1.2 Hz) Increases > linearly with magnetic field
What is the benefit of minimizing variance?
- Increases variance in acquired signal
2. Can lead to both false positive and false negative
What is the effect of motion?
Motion reduces the functional sensitivity
What is cardiac cycle and its effect on fMRI?
1.Signal intensity changes mainly in vasculature caused by cardiac pulsatility
- Signal intensity variations:
In-flow effects causing intra-voxel dephasing
Tissue movement
What is respiratory effects in fMRI?
Respiratory motion causes shifting of the brain image, and variance around edges.
Across-breath variation cause changes in arterial level of CO2
What are examples of measuring cardiac and respiratory effect?
- Pulse oximeter
2. Respiration belt
How can measured cardiac and respiratory phase be modelled?
Using a sum of periodic functions e.g. sines and cosine of increasing frequency (Fourier set)
What is the modelled effects for physiological noise?
removed from original fMRI signal
or included in fMRI statistical model
How can you prevent motion?
Constrain the volunteer’s head Give explicit instructions: Lie as still as possible Try not to talk between sessions Swallow as little as possible Make sure your subject is as comfortable as possible before you start Try not to scan for too long Mock scanner training for participants who are likely to move (e.g. children or clinical groups)
How can you correct for motion during acquistion?
Estimate the motion in real-time with a camera
Change automatically the position of the field of view based on the recorded displacement
How can you correct for motion after acquisition (more common)?
Estimate the rigid body transformation between volumes
Apply the transform to realign the data
What is the cycle length for a periodic block design?
0.01 and 0.1Hz
How can you minimise variance?
Sample more rapidly and filter out the frequencies of interest
What is co-registeration of fMRI analysis?
Match images from same subject but different image modalities:
anatomical localisation of single subject activations
achieve a more precise spatial normalisation of functional image using detailed structure of anatomical image.
What is re-alignment of fMRI analysis?
Match images from same subject and same modality:
Each transform can be applied in 3 dimensions
Therefore, if we correct for both rotation and translation, we will compute 6 parameters
What is the process of realignment ?
A reference image is chosen, to which all subsequent scans are realigned – normally the first image. These operations (translation and rotation) are performed by matrices and these matrices can then be multiplied together
Realignment
Visit each voxel in space of the transformed image.
Transform the voxel coordinate.
Find the voxel in the original image.
Calculate voxel values in transformed image
Usually this requires sampling between the centres of voxels so requires interpolation:
eg trilinear interpolation, bspline interpolation.
What is realignment and unwarping of fMRI analysis?
Even after realignment, there may be residual errors in the data need unwarping
Realignment removes rigid transformations
(i.e. purely linear transformations)
Unwarping corrects for deformations in the image that are non-rigid in nature
Correct for change in EPI distortions due to head motion
Estimate the change in distortion from time series?
Change wrt rotaton about x-axis ( )
Change wrt rotaton about y-axis ( )
How can distortions be corrected for?
Estimates can be combined with measured field map
What is spatial normalization of fMRI analys?
Increase sensitivity by averaging over subjects
Extrapolate findings to the population as a whole
Compare results from different studies by aligning them to a reference space
E.g. Talairach space or MNI space
What is the process of spatial normalization?
Find non-linear transformation that matches image to anatomical reference
Why smooth for fMRI analysis?
Increase SNR
Inter-subject averaging for multi-subject studies
Validity of statistical inference for statistical analyses (i.e. imposing normal distribution on data).
Typically performed by convolution with a Gaussian kernel.
Experimental setup and design
Consider number of conditions, block length, number of repetitions within psychological constraints
Choose fMRI acquisition parameters to optimise between sensitivity and keeping scanning times down.
Could the data be improved by using different coils or higher field strength if available?
EPI is necessary to meet spatio-temporal requirements but introduces artefacts in the image
Need to acquire a large coverage in a short time and a TE around T2* : Echo-planar –Imaging is appropriate
Dropout and distortions are introduced but can be minimized or corrected
Functional sensitivity
Can be estimated for a given TE with the tSNR and variants
tSNR is limited by physiological noise
The physiological noise is a limiting factor but several options allow to reduce its impact
Physiological monitoring, Motion correction, Increase temporal resolution and filter out physio frequencies
Once images have been acquired, several pre-processing steps are required before using statistical analysis
coregistration, realignment, normalization, smoothing