Registration
Essential Concepts
Image Spaces (associated with distinct coordinate system)
Native space – coordinates unique to subject’s head placement in MRI scanner
Standard space – standard defined coordinates across all images in this space define anatomy location
Images in this space can have different resolution
Coordinate Systems
Voxel coordinates
no units; integers
voxel count in each dimension reflecting matrix size
Origin: corner of image (e.g., 0,0,0 or 1,1,1)
Origin and axes naming conventions differ in programs
World coordinates
Have units in mm; floating values
Denote anatomy
Handedness system
Right-handed (RAS) system
x+ Left-Right
y+ Posterior-Anterior
z+ Inferior-Superior
Default for: MNI152 space, NifTi
LPS system
Default for: Dicom
Origin: can differ in programs, standard space (e.g., MNI152 space origin at anterior commissure)
Standard Templates
RULE: Know demographics of cases used to generate atlas, try match template and study cohort
Talairach and Tournoux (TT space)
1 postmortem brain, female, 60 years old
MNI152/ICBM 152 (MNI152 space)
152 healthy young adults
Study-Specific Templates (Study Specific space)
Need substantial representative cases to make
Standard Atlas
RULE: Know demographics of cases used to generate atlas; MAY BE DIFFER from template in same space, try match atlas, template and study cohort
Talairach and Tournoux (MNI152 space) - discrete
1 postmortem brain, female, 60 years old
AAL (MNI152 space) - discrete
Multiple scans from 1 healthy adult
Harvard-Oxford (MNI152 space) - probabilistic
37 healthy adults
JHU WM (MNI152 space) – probabilistic
28 young healthy adults
Registration Decision Points
Prep - QC input & reference images
Artifacts
Sharpness of anatomical features
Variability of anatomy
Comparable demographics (e.g., atlas, standard template, study cohort)
FOV sampling of body similar (e.g., amount of neck included)
Prep - Reorient (optional)
Initialization
Prep - Crop (optional)
Remove non brain areas in FOV in both images so sampling similar (e.g., neck)
Prep - Brain extraction of input & reference image (optional)
RULE: input & reference image MUST match in skull sampling (BAD: 1 w/skull, 1 w/out)
Quality needs to be good, if not – better to not do
Registration - Spatial transformations
DEFINITION: Calculate the best alignment of the images by determining parameters of the spatial transformation
Linear Transformation (within-subject, same anatomy)
Rigid-Body (6-DOF: 3 rotations, 3 translations)
Within-subject registration
Initializing other methods
Affine (12-DOF: 3 rotations, 3 translations, 3 scaling, 3 sheers/skews)
Eddy-current distortion correction
Initializing nonlinear transformations
Non-linear Transformation/Warp (between-subject, differ anatomy)
12+ DOF
REQUIRE: Initialization with affine registration
Resample/Transform – Apply the spatial transformation & creating image in new space
RULE: Avoid degradation of images; resample/interpolate only 1x
Create and check quality of separate registrations before concatenating and applying to image 1x
Last updated