1
Image Registration I
Comp 254Spring 2002Guido Gerig
Motivation for Image Registration
• Combine images from different modalities (multi-modality registration), e.g. CT&MRI, MRI&PET, MRI&fMRI, post-mortem&MRI, structural and functional images.
• Definition of a standard coordinate system (stereotaxic coordinates for neurosurgery, comparative analysis of corresponding regions in in neurosciences).
• Construction of atlases and normative databases.• Atlas matching for segmentation and interpretation.• Arithmetic and/or statistical operations on images
(averaging, statistical parametric mapping, deformations).• Register serial scans in temporal studies (development,
follow-up, tracking).
Image Registration: Motivation
2
MRA / PET FDGMRA / PET H2O
Motivation: Anatomo-functional correlation
Image Registration: Motivation
Courtesy of D. Vandermeulen, KUL
Motivation: Merge of PET and MRI
Three-dimensional data visualization of MRI and PET datasets for two patients: pre-symptomatic SCA1 (upper row) and sporadic OPCA (lower row). http://www.pet.med.va.gov:8080/demos.html
Copyright 1997PET CenterMinneapolis Veterans Affairs Medical Center
Image Registration: Motivation
3
Motivation: Construction of Atlases, Normative Databases
Image Registration: Motivation
SPM Software Package (K. Friston): canonical MRI images (probability maps) obtained by averaging 152 registered individual MRI.
Brain Bench: Virtual tools forstereotactic frame surgery
Image Registration: Motivation
http://www.krdl.org.sg/RND/biomed/publications/dextroscope/papers/MIA/BrainBench-1.html
Luis Serra, Wieslaw L. Nowinski, Tim Poston, et al.
3D Talairach-Tournoux atlas (left) andSchaltenbrand-Wahren atlas (right).
4
1 2 3
4 5 6
MS, MR T2, time points 1-6
Intra-patient registration over time for studying MS lesion evolution
Image Registration: Temporal Studies
Courtesy of D. Vandermeulen, KUL
Time 1
Time 2
Time series images: MS lesion development
(European BIOMORPH Project)
Image Registration: Temporal Studies
Courtesy of D. Vandermeulen, KUL
5
Challenges for Registration
• Multi-modal image data carrying different information (morphological and functional).
• Combination of images with different resolution and non-isotropic voxel dimensions, sometimes distortions.
• Different “photometric” properties of multiple images (intensity differences for tissue, bone, fluid, lesion).
• Intra-patient registration (often rigid), inter-patient registration (normative databases, rigid and elastic), and atlas to patient registration (rigid with prob.atlas, elastic).
• From rigid towards elastic registration (dealing with natural and pathological variability).
Image Registration: Challenges
Combining information from multiple images requires the geometric relationship between them to be known...
T ?
Concept of Registration
T = affine transformation(3D rotation, translation, scale, skew)
Image Registration: Concept
Courtesy of D. Vandermeulen, KUL
6
T !aligned
Concept of Registration ctd.Image Registration: Concept
misaligned
Courtesy of D. Vandermeulen, KUL
Registration Strategies
Points:anatomical orgeometrical features
– interactive– correspondence
Surfaces:objects
– segmentation– correspondence
Voxels:difference imageintensity correlationhistogram dispersion
– unimodal– linear relationship– mutual information
External markers: – non retrospective
Image Registration: Strategies
Courtesy of D. Vandermeulen, KUL
7
Types of 3D TransformationsImage Registration: Transformation Types
Examples of transformations to a regular mesh (top left): rigid (top right) for position orientation differences, affine (bottom left) for scaling differences andspline (elastic) (bottom right) for local and complex differences (illustration Ph.Thirion, INRIA)
3D Affine Transformation (non-elastic)
Image Registration: Affine Transformation
8
Short Excurse: Homogeneous Coordinates: A general view
• Acknowledgement: Greg Welch, Gary Bishop, Siggraph 2001 Course Notes (Tracking).
Series of Transformations
2D Object: Translate, scale, rotate, translate again
E Problem: Rotation, scaling, shear are multiplicative transforms, but translation is additive.
9
Solution: Homogeneous Coordinates
• In 2D: add a third coordinate, w
• Point [x,y]T expanded to [x,y,w]T
• Scaling: force w to 1 by [x,y,w]T/w → [x/w,y/w,1]T
• Any two sets of points [x1,y1,w1]T and [x2,y2,w2]T
represent the same point if one is multiple of the other.
2D homogeneous coordinate space
• Every 2D point [x,y]T in 3D space represents point along a line that passes through [x,y,w]T, where we want [x1,y1,1]T.
11
Affine Transformation
Image Registration: Affine Transformation
General case: 12 parameters define 3D affine transformation(translation (3), rotation (3), scaling (3), shear (3)).
Affine Transformation ctd.Affine Transformations can be decomposed:
3D Translation
3D Rotation about z axis,similarly about y and x
Combined 3D Transformation (here only rigid body: Translation and Rotation ⇒ 6 parameters)
Image Registration: Affine Transformation
12
Procedure:• user selects pairs of corresponding landmarks in two 3D
images (minimum number of landmarks appropriate to degree of freedom)
• choose type of transformation (rigid body, RTS, affine)• calculation of transformation parameters by least squares
fit (normal equation system)• transform images using interpolation (nearest neighbor, tri-
linear, cubic)
Affine Transformation ctd.Image Registration: Affine Transformation