week 2 - image processing concepts Flashcards
(29 cards)
What is intensity non-uniformity/ bias field?
What does this look like in layman’s terms?
What is it caused by?
-artificial, smooth, low-frequency variations in image intensity that are not related to actual tissue properties
-changing of brightness towards the centre of the scan -> does not reflect tissue magnetic properties
-patient head movement, hardware imperfections, RF coil sensitivity differences
What does intensity non-uniformity correction or aka bias field correction do?
a preprocessing step which removes artificial, smooth brightness variations in MRI so that the same tissue has consistent intensity across the image.
What are some structural MRI analysis methods?
(processing images)
- brain extraction
- tissue segmentation & tissue volume
- Masks
- image registration
- pulling it all together?
What does brain extraction involve (aka skull stripping)?
removing/zeroing all voxels which are not the brain
What does tissue segmentation distinguish?
distinguishes different tissue types like grey and white matter
How do tissue probability maps (tpms) analyse MRI scanned images?
tpms remove all tissue which is not of interest (eg. grey matter tpm)
What does a TPM calculate?
Are the values arbitrary?
percentage of each tissue type at a specific voxel - voxel values are now NOT arbitrary
eg.
0% CSF
43.53% white matter
56.47% grey matter
Can you calculate the overall volume of person’s grey matter/any tissue type from a single tissue segmentation image?
HOw is this calculated?
yes
for example answer would be 0.53 litres of grey matter
overall volume of GM/WM = total volume of all voxels identified having GM/WM timse mean value of those voxels
total volume x mean voxel value = overall volume of X in whole brain
Why is it useful to know the total volume of white and grey matter or total brain volume in research?
What is the issue with using raw brain volume with many different patients?
-brain atrophy cause by neurodegenerative conditions (shrinkage of W or GM)
-people have naturally different sized brains
Which preprocessing techniques can you use to combat the issue that people have different sized brains when investigating total brain volume or total WM/GM volume?
FSL has a pipeline (sienax) that will give you grey and white matter volumes, normalised to the size of a person’s skull
What is masking?
eg. setting a threshold of a certain percentage of a tissue type in all voxels AND make threshold them binary
Do two different people have the same ‘image space’?
Do two separate scans from the same person have the same image space? why?
- no, scans from different people will have different image spaces
- not usually: it’s safest to assume any two different scans are in their own unique space
-people can move around during scan etc
What is native space?
different image space unique to an individual
What is image registration and how does it affect ‘image space’?
image registration is a processing technique that moves, stretches and squishes one image to match the alignment of another
What is the ‘target image’?
the other image is registered to it, so to make both of them have the same image space as the target image
What does FLIRT stand for?
What is it used for?
What sort of transformations does it use?
What is the alternate use of FLIRT?
-FSL’s Linear Image Registration Tool
-Register images from the SAME person using linear transformations
-linear
-used as an initial step in a non-linear registration
What do we want to reduce when using FLIRT?
the number of manoeuvres/transformation used as each one introduces more noise
What degrees of freedom can you use for each FLIRT manoeuvre/transformation?
up to 12
so you can have a 6DOF etc
What manoeuvre is used with 6 degrees of freedom with FLIRT?
What type of MRI images is it useful to do this manoeuvre with?
What is good about this manoeuvre?
rigid body registration
fMRI and DTI
it can shift the brain in many ways but WITHOUT deforming the actual shape (any stretching or squishing)
What is FNIRT used for?
What does it stand for?
register different people’s brain together
FSL NON-linear Image Registration Transformation
How is FNIRT classified as non-linear?
parts of the image can be differently warped in “uneven” ways
What mathematical function does registration require? What does this function do?
cost function -> calculates how similar/dissimilar images are to each other (real & desired outcome)
Are there multiples cost functions you can use to do an image registration with FLIRT?
How do you decide which cost function to use?
yes
depends on image modalities of the experiment - type of MRI eg T1, PET
What is a pipeline?
a combination of all these techniques (mentioned) which attempt to complete a more complex goal
