what is used to analyse MRI?
a number of packages such as FSL are available
outline pre-processing MRI image
it must be ensured the image is in correct orientation
intensity non-uniformity correction- this is because the coil worn on MRI scanner and the head itself can distort magnetic field giving resulting images brighter/darker centres- this can be corrected by FSL pipeline
what are the 5 steps of structural MRI analysis methods
1- brain extraction
2- tissue segmentation
3- masks
4- image registration
5- pulling tools together
outline brain extraction
- process of removing everything that’s not brain (skull etc) from the image by giving these voxels a value of zero
what is a common error in brain extraction
there are often errors where parts of brain may be clipped or parts of skull left in image
outline tissue segmentation
- used to distinguish between GM and WM and CSF
- FSL reads a structural brain scan (ideally T1) and produces additional images (tissue probability maps) in the same space that show where the major tissue types are - “FAST pipeline”
- tissue probability maps (TPM) contain only data where relevant tissue was present in main scan- not totally binary (partial value)
- multiple TOMs can be overlaid to see overlap - voxels in TPM are no longer arbitrary- they represent % voxel belonging to each tissue type
how can tissue segmentation allow one to calculate the volume of a tissue type e.g. GM in someone’s brain?
- tissue segmentation has all the information needed to calculate someone’s overall GM and WM volumes e.g. volume of voxels identified as having some GM x mean value of GM in voxels (e.g. 71%= 0.71)
why is knowing brain volume useful?
used as a measure of brain atrophy
how do we compare brain atrophy between different people?
harder as there are natural differences between people, it will require normalising factor to take account of natural size differences
- software such as FSL has a pipeline such as sienax that will normalise WM and GM volumes to size of person’s skull (looking at cavity within skull- as skull doesn’t change volume with atrophy) so atrophy can be compared to the individual’s baseline
outline MRI masks
where the brain image is divided by itself to give binary values of 1 where brain is (e.g. 5/5) and 0 where it is not (0/0 = 0)
or could remove all voxels unless they have a minimum threshold amount of GM or WM e.g. at least 50% (can vary percentage depending on what you’re looking at)
outline image space
if you take a scan of one person and use TPM to separate out tissue types and overlay these images are all in the same image space, whereas comparing 2+ brains by directly overlapping MRIs these would not share image space completely.
Hence images are in the same space when they are in alignment, where the same coordinates line up to the same brain anatomy region
The areas which don’t overlap between scans are called native space
what is image registration?
whereas we manipulate MRI scans so that they share the same image space
it always involves a target image and the image you are wanting to register to the target
what are the 2 types of image registration?
linear registration (FLIRT)
non-linear registration (FNIRT)
outline linear registration
- used to register the images of the same person (or as first step in non-linear registration)
- up to 12 degrees of separation (the more dofs used the less rigid the registration is)
- aims to be conservative as each manoeuvre introduces noise
outline non-linear registration
- used to register different people’s brain MRIs together
- it’s called non-linear as different parts of the image can be warped in uneven ways
what is a cost function
used in software for registration to work out how to change one image to match the target image
what is a pipeline
the combination of techniques (brain extraction, tissue segmentation, masking and registration) being pulled together to complete a complex goal is a pipeline
what is a template?
an average brain scan of multiple people which has been registered into the same space
