Neuroimaging 2 Flashcards
(205 cards)
1
Q
What are MRI images made up of?
A
Matrices of numbers
2
Q
What is a voxel?
A
MRI uses a 3D space, so instead of pixels (2D) we use voxels. Each voxel will have an x, y and z value
3
Q
How are MRI images generated?
A
The contrast in an MRI scanner is generated by different tissues exhibiting different magnetic properties
4
Q
What is white matter?
A
A type of brain tissue that facilitates communication between different regions
It contains axons (nerve fibres), surrounded by a myelin sheath
5
Q
What is grey matter?
A
A type of tissue comprising of the cell bodies of neurons, dendrites, unmyelinated axons, and glial cells
6
Q
What is CSF?
A
Cerebrospinal fluid - a clear fluid that surrounds the brain and spinal cord, acting as a cushion and providing nutrients
7
Q
What is the word to understand that a voxel may have multiple tissue types?
A
Partial volume
8
Q
How do we view MRI scans?
A
In anatomical planes
9
Q
What is relaxation?
A
The loss of energy in hydrogen nuclei after receiving an RF pulse
10
Q
What is T1 / spin-lattice relaxation?
A
Recovery of longitudinal magnetisation
The process by which the nuclei give up their energy to their surrounding environment
11
Q
What is T2 / spin-spin relaxation?
A
The loss or decay of the magnetic moments in hydrogen nuclei in the transverse plane
The process of the nuclei falling out of sync
12
Q
What does a T1 scan look like?
A
CSF: Dark
White Matter - Very Bright
Grey Matter - Quite Bright
13
Q
What does a T2 scan look like?
A
CSF: Very Bright
White Matter: Darker Grey
Grey Matter: Brighter Grey
14
Q
What is a FLAIR scan?
A
A T2 weighted scan, where an inversion RF pulse cancels CSF signal
So looks like T2, but CSF is black
15
Q
Advantages and Disadvantages of T1
A
- Good tissue contrast (Gold standard for anatomy)
- Modest ability to detect pathology (eg. perivascular space)
- Cannot pickup white matter lesions (does not differentiate well from CSF)
16
Q
Advantages and Disadvantages of T2
A
- Superior at detecting fluid and pathology than T1
- Less anatomical detail
- More affected by motion artifacts
17
Q
What are FLAIR scans good at?
A
Excellent at investigating a range of brain pathology
18
Q
What is Sequential Susceptibility Weighted Imaging?
A
A type of imaging that is exquisitely sensitive to venous blood, haemorrhage, and iron storage
19
Q
What is Sequential SWI good at?
A
Looking at microbleeds
20
Q
What is Diffusion Weighted Imaging?
A
A specialised MRI technique that measures the movement of water in the body
21
Q
What is DWI good at?
A
Showing stroke damage before it becomes visible on other scans
22
Q
What can cause scans of the same type look different?
A
- Acquisition parameters
- Imaging hardware
- Scanner drift (over time the same scanner will gradually lose signal intensity and change the scan)
23
Q
What are the three main tools for MRI analysis?
A
- Freesurfer
- SPM
- FSL
24
Q
How does Freesurfer work?
A
- Very hands off, just click go
- Takes 8-12 hours
- Gives a lot of information, particularly cortical thickness analysis
25
How do SPM and FSL work?
1. More modular and hands on
2. SPM - MATLAB, FSL - Linux
3. Can do more specific things
26
What must be done before we can analyse an MRI image?
1. File conversion - DICOM to NIfTI
2. Ensure correct image orientation - Run a reorienting command
3. Intensity non-uniformity correction - The coil worn and head of patient can distort the scan, making central structures miscoloured. Run a command.
27
Describe the steps in a structural MRI analysis?
1. Brain Extraction
2. Tissue segmentation (& volume)
3. Masking
4. Image registration
5. Atlas segmentation
28
What is Brain Extraction?
The removal (zeroing) of all voxels that are not the brain
Never perfect
29
What is Tissue Segmentation?
Visualising different types of tissue in the brain
Results in a tissue probability map for white matter, grey matter etc
30
What can we do with our Tissue Probability Map?
Good software lets us overlay map onto image, meaning the voxel values are no longer arbitrary - they represent the % of the voxel that belongs to a tissue type (an expression of partial volume)
31
How can you determine tissue volume using a Tissue Probability Map?
Total volume of voxels with some GM / WM
x
Mean value of these voxels
32
When is it appropriate to just use tissue volume?
For one patient comparison later in time - neurodegeneration leads to atrophy, so it is a straightforward comparion
33
Why might it not be appropriate to use raw volume?
Between different people - we have to generate a normalising factor and normalise the volume to their skull size
34
What is Masking?
Thresholding the TPM and make it binary, to look for white or grey matter
35
How does masking work?
You set a threshold eg 0.1, 0.5 eg. for 10% GM, 50% GM
The image then only has 0s or 1s depending on whether or not they reached the threshold
This is a way of specifying voxels of interest for further analysis
36
What is Registration?
Getting different images into the same image space through a variety of techniques
37
What is an image space?
Images are in the same space when the same co-ordinates in each image refer to the same bit of anatomy
38
Are two scans ever in the same image space by default?
No. Always assume different scans are in different image spaces, even when they are from the same person
39
What two things are needed for image registration?
1. Target image
2. Image to be registered
40
What are the two types of image registration?
Linear and non linear
41
Describe Linear registration
1. More conservative, less aggressive
2. Generally only used to compare scans of the same person, or before non linear registration
3. Up to 12 degrees of freedom
4. Linear - applies to whole image
42
How is linear registration often done?
With 6 DOF, not deforming shape
This is called a rigid body registration
12 DOF often done before non linear registration
43
What is Non-Linear registration?
1. More aggressive
2. Image parts can be differently warped in certain ways
3. Generally used across participants
44
How does registration work, for both types?
It uses a cost function to determine how similar or dissimilar different images are and therefore when a good registration has been achieved
45
What must be done after the image has been registered?
It needs a method of interpolation to generate the newly registered image, as it may not nicely line up with grid space
46
What is nearest neighbour interpolation?
Preserves the original image voxel values and applies them to the nearest voxel - good for number, not as good for anatomical shape
47
What is bi-linear interpolation?
Each voxel becomes a weighted average of what is around it. Not so good for numerical nature, better for anatomical shape
48
What can we use as a target image?
A template such as the MNI 152 - this allows us to view participants' anatomy in standard space
49
What is Atlas based segmentation?
1. Image registered from patient to standard space
2. Reverse - superimpose the atlas back to the patient's native space
3. We can now identify ROIs
50
What can affect brain volume?
1. Aging
2. Disease processes of major neurological conditions
51
What is the usual tissue interest in volume studies?
Grey matter, because it is home to computationally specific regions of interest
52
What is a region of interest study?
Focusing on specific areas within the brain for detailed investigation
53
What is the biggest downside to a region of interest (ROI) study?
You need an a priori hypothesis - lots of situations where we don't know what to focus on and don't want to run lots of comparisons as it will increase the risk of false positives
54
What is the main solution to the a priori hypothesis problem with ROIs
The VBM Pipiline
55
What is Voxel Based Morphometry (VBM)
A type of global brain analysis
A 'simple experiment' which lets you identify local grey matter changes and other associations across the brain
56
When should you use a VBM pipeline?
If you really cannot think of an a priori hypothesis (it doesn't need one)
But if you can do an a priori hypothesis, use that instead
57
What are the steps in the VBM pipeline?
1. Brain Extraction
2. GM Tissue Extraction
3. Templates and Registrations
4. Smoothing
5. Statistical Imaging
58
VBM 1: Describe Brain Extraction
1.Run using software like FSL
2. Check outputs for major errors - better to have a bit extra information than irretrievably lose data - subsequent stages can sometimes correct
59
VBM 2: Describe GM Tissue Segmentation
1. Generate the quantitative image (TPM) - Overlay the TPM onto the BET brain
60
VBM 3: Describe Templates & Registration
1. Take the participants' T1 scan, 12DOF linear and non linear registration to MNI 152
2. Apply same maths to their GM TPMs (co registration) - leads to a set of GM TPMs in standard space
3. Create a mathematical average of the GM TPMs - leads to average GM TPM in standard space
4. Go back to original GM TPMs, and reregister to our new average template
5. These shared space GM TPMs are our basis for voxelwise analysis
6. Multiply the deformation map by the raw transformation to preserve tissue thickness
This results in GM TPMs, in template space, where cortex thickness has been spatially matched, but values modulated according to the degree that was needed to achieve this
61
Why do we make our own template for VBMs?
Every time we register we generate noise, so this is done to try and minimise the error and noise
62
What should be the weightings for the new template?
Equal amounts of participants from each subject group
63
Why do we need the deformation map?
If we register everyone's GM TPM to the same target space, it will make thick cortexes thinner and thin ones thicker
We save the deformation map when we perform registration (Jacobian of the transformations)
64
VBM 4: Describe Smoothing
1. Average voxel values with their neighbours using software
We do this to boost signal to noise - aim to preserve the effect of atrophy whilst smoothing out random noise
FSL runs pilot statistics with sigma = 2,3,4, pick the best statistical one
65
What is signal in the context of a VBM?
Patterns in our grey matter that underpin a statistical relationship
66
What is noise?
Random numerical change on top of the true GM values
From transformations, scanners, motion etc
67
VBM 5: Describe statistical analysis
1. A voxelwise analysis takes every single voxel wit data and performs statistical analysis
2. Software can be used to change the statistical design / do permutation testing
3. End up with a final image, in template space, where each voxel is a p value
4. We perform cluster enhancement
68
What is cluster enhancement?
Where we suppress lone statistically significant values, and enhance areas where voxels are consistently significant
69
What are the advantages and disadvantages of an ROI analysis?
+ Higher measurement accuracy, and therefore higher statistical sensitivity
- Very restricted in regional scope (needs a good a-priori hypothesis)
70
What are the advantages and disadvantages of a global analysis?
+ Considers the whole brain so nothing is missed
- Requires elaborate processing and creates noise - so less statistical power
- Multiple comparisons still unideal even after cluster based correction
71
What is fMRI?
Functional Magnetic Resonance Imaging - detects changes in Blood Oxygenation Dependent Level (BOLD)
72
What is the basic working of fMRI?
fMRI BOLD signal is altered due to the increase of blood flow in response to brain activity
73
Describe the BOLD effect
The BOLD sequence does not directly measure blood flow - it is sensitive to the different magnetic properties of oxygenated and deoxygenated blood
74
What is the magnetic property of oxygenated blood?
Weakly diamagnetic (slightly lower the strength of the magnetic field - can be ignored)
75
What is the magnetic property of deoxygenated blood?
Paramagnetic (increase the strength of the magnetic field)
76
What does deoxygenated blood mean in terms of spin?
Protons near areas of deoxygenation will spin slightly faster, and get out of phase
Reduced signal because of dephasing
77
How does the magnetic properties of blood affect the BOLD signal?
1. During activation (eg, in a task) , an excess of oxygenated blood flows to the activated region, swamping the deoxygenated blood
2. The excess of oxygenated blood reduces dephasing, boosting the signal
3. The change in signal due to the different T2* of the rest and task condition can be expressed mathematically - an optimum echo time can be chosen to maximise BOLD effect and minimise noise
78
How does imaging work in normal MRI?
1. Move through k space as specified by the negative phase encode and readout gradients
2. The first line of k space is acquired during the first TR after applying the largest negative gradient
3. During second TR, gradient is slightly less negative to read 2nd line of k space
4. We repeat the process and continually move up through k space
79
What is the issue with normal imaging that means it cannot be used in fMRI?
It takes a while (each TR about 1 second, 1 TR for each line of k space)
Will take too long for fMRI as we need fast scans to measure blood changes
80
What is the solution to speed in fMRI imaging?
Echo planar imaging
81
What is echo planar imaging (EPI)?
The whole of k space is acquired after a single 90 degree RF pulse - also called single shot imaging
82
What is the trade off with EPI?
Lower resolution, sensitive to various artifacts
but
< 100ms for a single slice
83
What are the different ways to acquire EPI images?
Sequentially ascending / descending
Interleaved ascending / descending - which reduces the bleedout effect from RF slices (as RF cannot be infinite in time to give no bleedout)
84
What are we looking for in fMRI?
Investigate signal intensity time course voxel by voxel
Some voxels may just be noisy, some may have an activation pattern
85
How can we sort BOLD Signals?
Model based or Model free / data driven approaches
86
Name two model based approaches
1. Linear regression (GLM)
2. Non linear regression
87
Name two model free / data driven approaches
1. Independent component analysis
2. Primary component analysis
88
Describe the first stages of the General Linear Model (GLM)
1. Move the square wave in time to fit data
2. Apply a hameodynamic response function to replicate physiological time delay and shape the curve
3. Add a linear ramp to make the curve tend upwards or downwards
4. This is called the HRF convolved linear ramp model, and should provide a reasonable fit to his data to find amplitude
89
What can we do once the HRF convolved model is on the right shape of the data?
Once the model is on the data, beta (amplitude) can be changed by an algorithm to fit
90
What is the final stage in the GLM modelling process
Voxel time series data = Regression parameter x Design matrix + noise
y = beta X + e
91
Describe data analysis in the fMRI modelling process? (T statistics)
Our t-statistic is the ratio of fitted amplitude to residual noise
T statistic is calculated voxel by voxel across the image
This can then be turned into an activation map
92
What is an activation map?
Pictures that show regions of the brain that have been activated
93
What is an important thing to remember about fMRI?
It is not quantitative, it relies on signal contrast between different states
94
Describe the block design
1. Blocks closely spaced successive trials over a short interval
2. For a two condition block design, around 20 seconds is statistically optimal
95
What are the advantages of a block design?
1. Most efficient design for detecting BOLD difference and take less time
2. Statistically powerful and straightforward to analyse
3. Can do more trials in less time than other designs
4. Good starting point for newcomers to field of fMRI research
96
What are the disadvantages of the block design?
1. Stimuli can be predictable and boring, potential confounds such as anticipation or rapid habituation
2. Inflexible for more complex tasks
3. Determining an appropriate baseline can be challenging
97
Describe the Slow Event Related Design
1. Short stimuli separated by a fairly long inter-stimulus interval (ISI) (enables HRF to fall back between trials)
2. ISI around 8 seconds plus 2 x stimulus duration
98
What are the advantages and disadvantages of the Slow Event Related Design?
+ Good for estimation of the HRF
- Takes around 35% longer than the block design comparatively
99
Describe the Fast Event Related Design
1. Short stimuli separated by variable Inter stimulus intervals
2. Jittered fixation frames allow for more closely placed trials
100
What are the advantages and disadvantages of the Fast Event Related Design?
+ Events can be truly randomised as in a behavioural study
- BOLD signal change is much lower than other designs
- Cannot colour code the HRF GLM regressor as the average signal cannot be clearly separated by stimulus type
101
What are some good practices of fMRI designs?
1. Evoke the cognitive or other process of interest
2. Collect as much data as possible, from as many subjects as possible
3. Choose stimuli and timing to create maximal change during process of interest and avoid signal overlap
4. Use reliable software to optimise design efficiency for ER designs
5. Get measure of subject behaviour in the scanner
102
For a simple block design, how do we test if there is significant activation
if t statistic (beta / e) < threshold (0.05) we can reject the null hypothesis and conclude there is significant voxel activation
103
What effect does having two different event types (eg, faces and places) have on our statistical analysis?
We have two regressors so we can do more statistical analyses using contrast of parameter estimates (COPEs)
104
What are COPEs?
Contrast of Parameter Estimates
105
How do we input COPES?
Using weight vectors
106
If [1 0] represents voxels where activation occurs due to faces and [0 1] represents voxels where activation occurs due to places, what is [1 1]?
The mean activation due to faces and places
107
If [1 0] represents voxels where activation occurs due to faces and [0 1] represents voxels where activation occurs due to places, what is [1 -1]?
Voxels where faces activate more than places
108
If [1 0] represents voxels where activation occurs due to faces and [0 1] represents voxels where activation occurs due to places, what is [-1 1]?
Voxels where places activate more than faces
109
How can we produce images of the COPEs?
By fitting the GLM to every model
110
If we plot the t statistic it is distributed under the null hypothesis beta = 0. Explain how this affects our inference.
A small p value means the null hypothesis is unlikely
A threshold should be set (eg. p = 0.05, which would lead to false positive rate of 5%)
111
What does the t distribution depend strongly on?
The degrees of freedom, so similar t values can mean very different things
112
What do we do to the t values before we inference from them?
Turn them into t values over a standard normal distribution -
N(mu = 0, sigma = 1)
This creates a Z distribution
113
What issue do we still have with our Z distribution?
5% false positives is a lot in a >100,000 voxel scan
114
How can we try and remove false positives at the end of the process
Apply a multiple comparison correction method
115
Name the four main types of multiple comparison correction methods
1. Bonferonni - strong control over false positives, least sensitive
2. False Discovery Rate - Admits false positives, more lenient
3. Gaussian Random fields - strong control over false positives, somewhat conservative
4. Cluster enhancement
116
What is 1st level analysis?
Analysing individual subjects
117
What is 2nd level analysis?
Analysing groups of data - does the group activate on average?
118
What are the steps in doing a 2nd level analysis?
1. 2nd Level GLM
2. Preprocessing
3. Presentation of results
119
How do we do group analyses on fMRI data?
1. Each subject will have a beta value and an e (noise) value
2. We group these into matrices and form another GLM
Beta matrix = Design matrix x X + error matrix
Bk = Xg x Bg + eg
120
What is this model called?
Beta matrix = Design matrix x X + error matrix
Bk = Xg x Bg + eg
The mixed effects model, considering the group variance as well as original MRI data variance.
Statistical software can be used to run GLM contrast tests across many different hypotheses
121
What are the steps in preprocessing of fMRI group analysis?
1. Brain Extraction
2. Spatial Smoothing
3. Motion correction
3.5. Slice timing correction (often not manually performed as software can account for it)
4. Temporal filtering
5. Unwarping
6. Registration
122
fMRI Analysis 2: What is spatial smoothing?
1. Average voxel values with their neighbours using software
We do this to boost signal to noise - aim to preserve the effect of atrophy whilst smoothing out random noise
123
fMRI Analysis 3: What is involved in Motion Correction?
A rigid body transform is applied between fMRI volumes to correct for motion
124
fMRI Analysis 4: What is Temporal Filtering?
1. fMRI data often exhibits a slow signal drift over time
2. This is corrected by applying a high pass filter to the data
(Filter cutoff typically set to r+A, or r+A+r+B cycle times)
125
fMRI Analysis 5: What is Unwarping?
1. EPI images can be quite distorted, particularly at tissue boundaries
2. The distortion depends on the polarity and direction of the phase encode blips
3. The difference between acquisitions with positive and negative blips can be used to estimate a field map
4. Field map used to correct distortion
126
What are some different views for fMRI data, and what are there uses?
1. Orthographic View - fine for a few activation clusters
2. Significant Cluster Presentation - useful for clearly showing ROIs
3. Lightbox View - Good overview to spatial extent of activation, but images are small
4. 3D view - looks cool but less useful scientifically
127
Summarise fMRI overall in seven points
1. BOLD MRI measures the change in oxygenated blood levels in response to a task
2. These changes induce a signal change that can be measured voxelwise over time
3. A variety of experimental designs can be implemented in order to induce a task dependent BOLD response
4. The experimental design can be modelled and fit to the imaging data voxelwise using the General Linear Model (GLM)
5. This fitting process generates a statistic that tells us how likely a voxel is to be active during a task
6. We can "threshold" this statistic to identify activated regions
7. Higher level analysis can be performed to assess activation levels across or between subject groups
128
Statistical Inference: What is a between group comparison?
Two participant groups (independent sample t-test)
129
Statistical Inference: What is a within group comparison?
Same participants assigned to 2 conditions (paired sample t-test)
130
Statistical Inference: What is linear association?
Finding correlation between two variables
131
Statistical Inference: What is a mixed design?
Two groups and two conditions
132
Statistical Inference: What are some advanced methods?
Real time fMRI, multivariate analysis
133
What are some advantages of fMRI experiments?
1. Versatility (structure and function)
2. Simple to run statistics with software
3. In vivo methodology
4. Good for longitudinal designs
134
What are some limitations to fMRI experiments?
1. Rather devoid of biological information
2. Temporal resolution is suboptimal
3. Counterindications (metal, pregnancy)
4. Expensive to buy and set up
135
What is Resting State fMRI?
Measuring spontaneous brain activity when a person is at rest using the BOLD signal
136
What are the advantages of Resting State fMRI?
1. Helps understand functional connectivity between brain groups
2. Shows intrinsic brain organisation
3. Can be done in any population (infants, unconscious, people that can't do tasks)
4. Provides an objective clinical biomarker (eg. for Alzheimer's or Depression)
5. Much quicker and easier to set up than a task based fMRI
137
What noise can a resting state fMRI involve?
1. Physiological noise (cardiac, respiration)
2. Subject motion
3. Scanner instabilities
4. Image reconstruction errors
138
What is the Default Mode Network?
The primary network active during rest
Mainly dorsal medial prefrontal cortex, dmPFC, posterior cingulate cortex (PCC), and precuneus (PCUN).
139
What is the Default Mode Network involved in?
1. Self referential thoughts - thinking about your own identity, belief and experiences
2. Social cognition - understanding the perspectives of others
3. Autobiographical memory - recalling past experiences
4. Future planning - thinking about potential future events
140
What other well known networks can be active at rest?
8 well known networks - the Beckman 8
141
What is a key question in resting state fMRI?
Whether signal fluctuations we observe actually correspond to synchronised neural fluctuations
142
How can we isolate neural signals from resting state fMRI?
Lower frequency signals come predominantly from neural activity
Higher frequency signals tend to come from cardiac and respiration
So look at lower frequency signals
143
What are the 3 types of connectivity we can examine with resting state fMRI?
1. Functional connectivity
2. Structural connectivity
3. Effective connectivity
144
What is functional connectivity?
How different areas of the brain are related by function
145
What is structural connectivity?
Physical interconnections between brain regions
146
What is effective connectivity?
Influence of bran regions on each other
147
What is a usual time frame for resting state fMRI?
6-10 minutes
148
What is the fMRI pipeline?
1. Pre processing
2. Denoising
3. Post processing
149
Resting state fMRI 1: Describe pre processing
1. Realign image
2. Co register to T1 reference image
3. Slice timing correction
4. Normalise to standard space and smooth
150
Resting state fMRI 2: Describe denoising
Removal of head movement and physiological noise
151
Resting state fMRI 3: Describe post processing
Either
Independent Component Analysis (ICA)
or
Seed Based Connectivity
152
Resting state fMRI: Describe independent component analysis
- Multivariate voxel based approach
- Finds interesting structures in the data
- Spatial and global approach
- Simple as it is model free
153
Resting state fMRI: Describe seed based connectivity
- Calculates correlation coefficients between two regions
- Requires ROIs to do it
154
Resting state fMRI: Summarise Independent Component Analysis
- Data driven
- Finds whole resting state networks without predefined seeds
- Harder to interpret, requires group level analysis
- Best for exploring large scale network dysfunction
155
Resting state fMRI: Summarise Seed Based Connectivity
- Hypothesis driven
- Simple, interpretable, focuses on specific region
- Requires predefined ROIs, may miss wider activity
- Best to study specific circuits
156
Resting state fMRI: Summarise ROI to ROI Connectivity
- Measures direct correlation between predefined ROIs
- Good for group comparisons as more flexible than SBC
- Still relies on predefined ROIs, may miss unknown connections
- Best for comparing functional connectivity between conditions or group
157
What is diffusion?
The Brownian motion of molecules due to thermal processes
158
What is diffusion imaging?
An MRI technique that measures the mobility of diffusion to generate images
159
What is the principle of diffusion imaging?
At the same temperature, some things diffuse more than others - diffusion imaging measures the difference
160
How does diffusion imaging work?
1. Uses T2 weighted echo planar imaging (same as fMRI)
2. A spin echo MRI sequence is sensitised to diffusion by adding large diffusion weighting gradients around the 180 degree pulse
161
How do tissues with restricted diffusion work in diffusion weighted imaging?
1. Gradient applied
2. Wait
3. Reverse gradient applied
If there is not much motion after the first gradient, the reverse gradient will just resestablish the full signal
162
How do tissues with lots of diffusion work in diffusion weighted imaging?
1. Gradient applied
2. Wait
3. Reverse gradient applied
If there is more motion, the molecules will be out of place when rephasing, meaning that the reverse gradient will not be "right" to reset it
This means the original signal is not restablished and is weaker
163
What can we change in diffusion weighted imaging?
We can change the b value to balance signal to noise ratio
b0 - baseline image, not sensitive to diffusion
b1000 - sensitive to relatively high and low diffusion
164
What does a higher b value mean in diffusion weighted imaging?
A higher b value means we become more sensitive to areas with a lower diffusion rate
165
What is a b0 image?
Just a normal T2 EPI, not sensitive to diffusion
166
What can we do with the b0 image?
Combine it with the other image (b1000) to find the Apparent Diffusion Coefficient
167
What can we do with the ADCs?
1. In many human tissues, the amount of diffusion changes depending on the direction of the applied diffusion gradient and anatomy
2. We can use the b0 and combine it with images from all different gradient directions to get lots of ADCs
3. Average out all the ADCs per voxel to get a ADC map / Mean Diffusion map of the braiin
168
What is directional resolution?
If we average ADCs out over different directions, it creates "directional resolution"
This means that the more directions we measure, the more accurate a picture we can gain in 3D space
169
What is the bare minimum number of directions in diffusion imaging?
The bare minimum is 3, in the x y and z plane
It is commonly done in clinical scenarios with cost and time pressure
Resulting data is referred to as a DWI (Diffusion weighted image)
170
What is the first step of more complex diffusion imaging?
1. In human tissue, diffusion is restricted by the complex microstructural environment
2. To better characterise this environment, the diffusion process may be described by more complex mathematical models such as a diffusion tensor
3. The diffusion tensor defines an ellipsoid in each voxel that characterises the diffusion property
171
What is isotropic diffusion?
Equal diffusion in all direction - a spherical diffusion tensor ellipsoid
172
What is anisotropic diffusion?
Varying diffusion across directions - a oval shaped diffusion tensor ellipsoid
173
Where might you find anisotropic diffusion?
In the white matter tracts
174
How are diffusion tensor ellipsoid stored?
As eigenvalues and eigenvectors
175
What calculations can we do with diffusion tensor models?
Mean diffusivity
Fractional anisotropy
176
What is mean diffusivity?
Lambda 1 + Lambda 2 + ... + Lambda n / n
177
What is fractional anisotropy?
A number between 0 and 1 measuring anistropy
Closer to 1, more anisotropic
Needs at least 6 directions to calculate
178
What have studies shown regarding MD and FA?
MD and FA change in disease processes, even when they are very early on - very sensitive
The more organised the white fibres, the healthier - so in neurodegeneration, MD goes up and FA goes down
179
What can we do with FA maps?
Apply colour coding to show principal diffusion direction
180
What is an application of diffusion imaging?
In strokes
1. Diffusion initially reduced due to cytotoxic oedema
2. Over time, diffusion recovers towards more normal, slightly increased due to vasogenic oedema
Diffusion is reduced quicker in DWI than in a standard T2 image
181
What can be done to improve diffusion images?
1. Motion correction
2. Susceptibility Induced Distortion Correction (Unwarping)
182
What is Susceptibility induced distortion connection (unwarping)?
1. EPI images can be heavily distorted, particular at tissue boundaries
2. The distortion depends on the polarity of the phase encode gradient blips
3. Software can reverse the blips and correct for the distortion
183
What is tractography?
A technique used in diffusion imaging to visualise and reconstruct the trajectories of white matter fibre tracts in the brain
184
How does tractography work?
Overlay the principle eigenvector onto the FA map at each voxel to give an indication of the major white fibre orientation
185
How does streamline tractography work?
Starts from a single seed point and continuously path finds through local voxels until certain criteria are met
186
What are the potential issues with streamline tractography?
1. Only finds a single pathway from each seed point
2. Branching fibres can cancel out diffusion measurements, confuses the algorithm and not true to anatomy
187
What is ODF sampling?
A technique that can be used in tractography to estimate / balance the different diffusion direction probabilities
188
How does ODF work?
1. ODF performed many (>1000) times to define the principal diffusion direction for each trial
2. Based on these, generates a probability that region A crosses to region B - called a probabilistic tractography
189
What is the advantage to probabilistic tractography?
It can be used to generate more reliable heatmaps and is better than basic tractography
190
What can tractography be used for?
Extracting and creating 3D ROI analyses, combining diffusion parameters with clinical or cognitive investigations
191
What is global tractography?
1. Propagates from a number of seed points and forces them to all join together
2. Represents how likely they are to actually do this and generates geodesic pathways
192
What are the advantages / disadvantages to global tractography?
+ Better at finding fine detail tracts that penetrate deep into the cortex
+ Becomes a lot more accurate
- Opens itself up to a lot more noise
193
What is Tract Based Spatial Statistics (TBSS)?
Essentially voxel based morphometry using fractional anisotropy
194
Why can't we just do a normal VBM on FA images?
Image registration with FA images creates a massive amount of image registration errors. We can line up images spatially, but hard to line up anatomically
195
What are the steps in TBSS?
1. After registration to standard space, an FA skeleton is made
2. The FA skeleton represents the core average total shape of the group
3. Individual subject data are then projected onto the skeleton by taking the nearest maximum FA value
4. This helps refine misregistrations and ensure that the projected skeleton FA data are comparable for voxel based analysis
196
What is the spin of a 1H atom?
A spin of 1/2
Different elements have different spins
197
What affects the magnetic field that a 1H atom feels?
The chemical structure that it is a part of
1. Electron cloud of molecule shields 1H nuclei from applied B0 field
2. Resonant frequency of 1H changes with molecule - expressed as a frequency chemical shift
198
How does 1H NMR Spectroscopy imaging work?
1. If we supress the strong water signal, we can then distinguish the difference between the weaker signals of important metabolites
2. This is done at a high field strength to move the spectra further from their neighbours
199
Give an example of 1H NMR spectroscopy imaging?
Brain tumour showing lactate due to anaerobic respiration - scan can be colour coded and superimposed on an anatomical resolution
200
How are concentration heatmaps generated in NMR spectroscopy?
Integrate the area under spectra peaks to generate heatmap
201
What do we need to image with other nuclei?
Other nuclei resonate at different frequencies, so we need separate RF coils
202
Why does imaging with other nuclei give weaker signals?
Due to less abundance and concentration
203
How can we boost low signals in other nuclei scans?
Using hyperpolarisation makes it easier to pickup weaker concentrations
204
Give some examples of other nuclei and what they are used to image?
1. Deuterium can be used to label structures - Signal resonates at different frequency to 1H but mimics it chemically
2. 3He can be used for lung imaging via hyperpolarisation
3. We can check the level of Lithium in neuropsychiatric patients' brains
4. Carbon13 can be used to look at metabolic exchange over the blood brain barrier
5. Phosphorous can be used to measure brain ATP metabolism
6. Xenon can be used to measure brain gas exchange
205
