医学图像配准 (Medical Image Registration)

目录

Classification

Transformation

 Registration Algorithms

Landmark Based

Surfaced Based

Voxel Intensity Based

Information Theory Based

Registration using basis functions

Registration using splines

Other Physics Based Registration

Optimization

Visualization

Validation

---

Classification

1. Dimensionality

2D-2D, 3D-3D, 2D-3D

2. Nature of registration basis

-Image based

        Extrinsic, Intrinsic

-Non-image based

3. Nature of the transformation

-Rigid, Affine, Projective, Curved

4. Interactive

-Interactive, Semi-automatic, Automatic

5. Modalities involved

-Monomodal, Multimodal, Modality to model (the third one can be involved in the first two categories)

6. Subject

-Intra-subject, Inter-subject, Atlas (the third one can be involved in the inter-subject)

7. Domain of transformation

-Local, Global

8. Optimization procedure

9. Object

Transformation

Relate the position of features in two images

-Rigid (Rigid-body)

        Translations and rotations

-Affine

        Also allows scaling and shearing

-Curved

        Allows the mapping of straight lines to curves 

- Perspective

        The parallelism of lines need not be preserved

(Curved and perspective can be summarized as Non-linear)

 Registration Algorithms

Methods used to find the transformation

1. Rigid & affine

-Landmark based (feature point)

-Information theory based

-Edge based

-Voxel intensity based

2. Non-rigid

-Registration using basis functions

-Registration using splines

-Physics based

-Elastic, Fluid, Optical flow, etc.

Landmark Based

Identifying corresponding points in the images and inferring the image transformation.

1. Types of landmarks

-Intrinsic

        internal anatomical structure

-Extrinsic

        artificial objects attached to the patients

2. Compute the average or centroid of each set of points -> translation

3. Rotate this point set about the new centroid until the sum of the squared distances between each corresponding point pair is minimized.

Surfaced Based

1. Method

-Extract corresponding surfaces

-Compute the transformation by minimizing some measure of distance between the two surfaces

2. Algorithms used

-The "Head and Hat" Algorithm

-The Iterative Closest Point Algorithm

-Registration using crest lines

Voxel Intensity Based

1. Method

-Calculating the registration transformation by optimizing some measures calculated directly from the voxel values in the images

2. Algorithms used

-Registration by minimizing intensity difference

-Correlation techniques

-Ratio image uniformity

-Partitioned intensity uniformity

Information Theory Based

1. Maximize the amount of shared information in two images

-Reduce the amount of information in the combined image

2. Algorithms used

-Joint entropy

        Joint entropy measures the amount of information in the two images combined (min)

-Mutual information

        A measure of how well one image explains the other, and is maximized at the optimal alignment (max)

-Normalized mutual information

Registration using basis functions

1. Represent the deformation field using a set of basis functions

-Fourier (trigonometric) basis functions or wavelet basis functions

-Implement smoothness constraint by a linear combination of basis functions

-The trigonometric basis functions correspond to a spectral representation of the deformation field where each basis function describes a particular frequency of the deformation

Registration using splines

1. Assumption

-A set of corresponding points or landmarks (control points) can be identified

2. At control points, interpolate or approximate the displacements to map the location of the control points in both images

3. Between control points, they provide a smoothly varying displacement field

Elastic Registration

1. Model the deformation as a physical process resembling the stretching of an elastic material

-The physical process is governed by the internal force & external force

-Described by the Navier linear elastic partial differential equation

2. The external force drives the registration process

-The external force can be the gradient of a similarity measure

        e.g. local correlation measure based on intensities, intensity differences, or intensity features such as edge and curvature.

-Or the distance between the curves and surfaces of corresponding anatomical structures

Other Physics Based Registration

1. Fluid registration

-The image was modeled as a highly viscous fluid

2. Registration using mechanical models

-Use a three-component model to simulate the properties of rigid, elastic, and fluid structures

3. Registration using optical flow

Optimization

Many registration algorithms require an iterative approach

-An initial estimate of the transformation is gradually refined

-In each iteration, the current estimate of the transformation is used to calculate a similarity measure

-Make another estimate of the transformation, evaluate the similarity measure again, and continue until the algorithm converges

-No transformation can be found that results in a better value of the similarity measure, to within a preset tolerance

Visualization

1. Color overlay

2. Interleaved pixel or chessboard fusion

3. Dynamic alternating display

4. Split view displays

5. Subtraction images

6. Etc.

Validation

1. Measurements using computer-generated models, images of physical phantoms of accurately known construction and dimensions, and images of patients or volunteers.

-Robustness

-Accuracy

2. Assessment of accuracy

-Estimate of some geometrical measure of alignment error

-Compare the system to be validated against a gold standard

-Visual assessment