ELEC5622 Chap.5 Brain MRI Preprocessing and Registration Pipelines
Brain MRI Preprocessing and Registration Pipelines
This chapter follows a brain MRI from raw scan to regional measurement. The lecture's preprocessing problems, geometric distortion, intensity non-uniformity, noise and arbitrary intensity units, come first. The Lab 1 chain then extracts the brain, segments tissue, registers the MNI template affinely and then deformably, resamples the atlas with nearest-neighbour interpolation and measures grey matter in each region.
Project 1 reruns the same chain in batch for 90 regions per subject. The chapter quotes the Lab 1 marking criteria so you can see where explanation earns marks alongside correct output. The same steps reappear in every brain imaging study, so the habits built here transfer well beyond the lab.
What this chapter covers
- 01
Why preprocessing changes study outcomes
- 02
Intensity non-uniformity and the N3 algorithm
- 03
Denoising and intensity normalisation
- 04
Skull-stripping with bet and its options
- 05
Native space, MNI space and the ICBM152 template
- 06
Rigid, affine and deformable registration
- 07
Interpolation choices for images and label maps
- 08
Measuring regional volumes with fslstats
Worked example · free
Grey matter volume in an atlas region
- 1One voxel = 0.9 × 0.9 × 1.0 = 0.81 mm³.
- 1Region volume = 6,000 × 0.81 = 4,860 mm³.
- 1Grey matter volume = 3,300 × 0.81 = 2,673 mm³, which is 55% of the region.
Key terms
- Skull-stripping
- Skull-stripping is the removal of non-brain tissue so that only the brain remains for analysis.
- Affine registration
- Affine registration aligns two images with a linear transform combining rotation, translation, scaling and shear.
- Nearest-neighbour interpolation
- Nearest-neighbour interpolation assigns each new voxel the value of the closest original voxel, preserving label values.
- MNI space
- MNI space is the standard brain coordinate system defined by the Montreal Neurological Institute templates.
- Deformable registration
- Deformable registration aligns images with a nonlinear transform that can bend the grid locally.
- Brain parcellation
- Brain parcellation divides the brain into labelled regions, usually by transferring an atlas onto a scan.
Brain MRI Preprocessing and Registration Pipelines FAQ
Why register affinely before deformably?
Affine registration removes the large differences in position, size and orientation with few parameters. Deformable registration then refines local shape. Starting the nonlinear step from a rough alignment makes it faster and less likely to settle on a wrong match.
Why must atlas labels use nearest-neighbour interpolation?
Labels are integers naming regions. Trilinear or spline interpolation averages neighbours, creating values such as 85.5 that belong to no region. Nearest-neighbour copies an existing label, so every resampled voxel stays valid.
What does the bet -f option change?
It sets the fractional intensity threshold between 0 and 1, default 0.5. Smaller values give a larger brain outline estimate, which helps when the default removes brain tissue; larger values shrink the outline when skull remains.
What is intensity non-uniformity?
It is variation in brightness across an image that has nothing to do with anatomy, so the same tissue reads differently in different places. Its correction is the most common MRI preprocessing step, and the lecture names the N3 algorithm.
Why are MRI intensities normalised?
Conventional MR intensities are in arbitrary units, so scans from different scanners, subjects or visits are not directly comparable, even with one protocol. Normalisation puts tissue intensities on a common scale before thresholding or classification.
What is the ICBM152 template?
It is the current MNI standard brain, the average of 152 normal MRI scans matched to the earlier MNI305 template with a 9-parameter affine transform. Registering subjects to it lets atlases such as AAL label the same regions in every brain.
Exam move
Write the Lab 1 pipeline from memory as a numbered list of commands, then add one sentence on why each step comes where it does. Practise converting voxel counts to mm³ with unequal voxel sizes. Rehearse the interpolation explanation until you can give it in three sentences, since the lab marks it explicitly.
Finally, sketch native space and MNI space side by side with an arrow showing which way the template moves in Lab 1, and name the file that stores each transform. Rehearse the difference between rigid, affine and deformable transforms by listing what each one can and cannot change about a brain's shape.
Working through Brain MRI Preprocessing and Registration Pipelines in ELEC5622? Sia is AskSia’s AI Electrical Engineering tutor — ask any ELEC5622 Brain MRI Preprocessing and Registration Pipelines question and get a clear, step-by-step explanation grounded in how ELEC5622 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
