A Survey of Medical Image Registration

J.B.Maintz,M.A ViergeverMedical Image Analysis,1998

Medical Image SPECT (Single Photon Emission

Computed Tomography) PET (Positron Emission

Tomography) MRI (Magnetic Resonance Image) CT (Computed Tomography)

Image Modalities AnatomicalDepicting primarily morphology

(MRI,CT,X-ray) FunctionalDepicting primarily information on

the metabolism of the underlying anatomy (SPECT,PET)

Medical Image Integration Registration Bring the modalities involved into

spatial alignment Fusion Integrated display of the data

involved

Matching, Integration,Correlation,…

Registration procedure Problem statement Registration paradigm Optimization procedure

Pillars and criteria are heavily interwined and have many cross-influences

Classification of Registration Methods

Dimensionality

Nature of Registration basis

Nature of transformation

Domain of transformation

InteractionOptimization procedure

Modalities involved

Subject Object

Dimensionality Spatial dimensions only

2D/2D 2D/3D 3D/3D

Time series(more than two images), with spatial dimensions

2D/2D 2D/3D 3D/3D

Spatial registration methods 3D/3D registration of two images 2D/2D registration Less complex by an order of magnitude both

where the number of parameters and the volume of the data are concerned.

2D/3D registration Direct alignment of spatial data to projective

data, or the alignment of a single tomographic slice to spatial data

Registration of time seriesTime series of images are required for various

reasons Monitoring of bone growth in children (long time

interval) Monitoring of tumor growth (medium interval) Post-operative monitoring of healing (short

interval) Observing the passing of an injected bolus through

a vessel tree (ultra-short interval)

Two images need to be compared.

Nature of registration basis Image based

Extrinsicbased on foreign objects introduced into the imaged space

Intrinsicbased on the image information as generated by the patient

Non-image based (calibrated coordinate systems)

Extrinsic registration methods Advantage

registration is easy, fast, and can be automated.

no need for complex optimization algorithms.

Disadvantage Prospective character must be made in the pre-

acquisition phase. Often invasive character of the marker objects. Non-invasive markers can be used, but less accurate.

Extrinsic registration methods Invasive

Stereotactic frameFiducials (screw markers)

Non-invasiveMould,frame,dental adapter,etcFiducials (skin markers)

Extrinsic registration methods The registration transformation is

often restricted to be rigid (translations and rotations only)

Rigid transformation constraint, and various practical considerations, use of extrinsic 3D/3D methods are limited to brain and orthopedic imaging

Intrinsic registration methods

Landmark based Segmentation based Voxel property based

Landmark based registration Anatomical

salient and accurately locatable points of the morphology of the visible anatomy, usually identified by the user

Geometricalpoints at the locus of the optimum of some geometric property,e.g.,local curvature extrema,corners,etc, generally localized in an automatic fashion.

Landmark based registration The set of registration points is sparse

---fast optimization procedures

Optimize Measures Average distance between each landmark Closest counterpart (Procrustean Metric) Iterated minimal landmark distances

Algorithm Iterative closest point (ICP) Procrustean optimum Quasi-exhaustive searches, graph matching and

dynamic programming approaches

Segmentation based registration

Rigid model based Anatomically the same structures(mostly surfaces) are extracted from both images to be registered, and used as the sole input for the alignment procedure.

Deformable model based An extracted structure (also mostly surfaces, and curves) from one image is elastically deformed to fit the second image.

Rigid model based “head-hat” method

rely on the segmentation of the skin surface from CT,MR, and PET images of the head

Chamfer matchingalignment of binary structures by means of a distance transform

Deformable model based Deformable curves

Snakes, active contours,nets(3D)

Data structureLocal functions, i.e., splines

Deformable model approachTemplate model defined in one imagetemplate is deformed to match second image

segmented structure unsegmented

Voxel property based registration

Operate directly on the image grey values

Two approaches: Immediately reduce the image grey value

content to a representative set of scalars and orientations

Use the full image content throughout the registration process

Principal axes and moments based

Image center of gravity and its principal orientations (principal axes) are computed from the image zeroth and first order moment

Align the center of gravity and the principal orientations Principal axes :Easy implementation, no

high accuracy Moment based: require pre-segmentation

Full image content based Use all of the available information

throughout the registration process.

Automatic methods presented

Paradigms reported Cross-correlation Fourier domain

based .. Minimization of

variance of grey values within segmentation

Minimization of the histogram entropy of difference images

Histogram clustering and minimization of histogram dispersion

Maximization of mutual information

Minimization of the absolute or squared intensity differences

…

Non-image based registration

Calibrated coordinate system If the imaging coordinate systems of the

two scanners involved are somehow calibrated to each other, which necessitates the scanners to be brought in to he same physical location

Registering the position of surgical tools mounted on a robot arm to images

Nature of Transformation Rigid Affine Projective Curved

Domain of transformation

GlobalApply to entire

image

LocalSubsections have

their own

Rigid case equation

Rigid or affine 3D transformation equation jiji xay

Rotation matrix

rotates the image around axis i by an angle

)(iri

Transformation Many methods require a pre-

registration (initialization) using a rigid or affine transformation

Global rigid transformation is used most frequently in registration applications

Application: Human head

Interaction Interactive Semi-automatic Automatic

Minimal interaction and speed, accuracy, or robustness

Interaction Extrinsic methods

Automated Semi-automatic

Intrinsic methods Semi-automatic

Anatomical landmark Segmentation based

Automated Geometrical landmark Voxel property based

Optimization procedure

Parameters for registration transformation

Parameters computed

Parameters searched for

Optimization techniques Powell’s method Downhill simplex method Levenberg-Marquardt optimization Simulated annealing Genetic methods Quasi-exhaustive searching

Optimization techniques

Frequent additions:Multi-resolution and multi-scale

approaches

More than one techniquesFast & coarse one followed by accurate & slow one

Modalities involved Monomodal Multimodal Modality to model Patient to modality

Subject Intrasubject Intersubject Atlas

Object Different areas of the body

Related issues

How to use the registration Registration & visualization Registration & segmentation

ValidationValidation of the registrationAccuracy,…