Image Registration Flashcards

1
Q

What is image registration?

A

The process of transforming different sets of data into one coordinate system

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

What does image registration involve?

A

Estimating a smooth, continuous mapping between the points in one image and those in another

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

What can be determined from The parameters that encode the mapping?

A

The relative shapes of the images

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

What is the objective of image registration?

A

Determine the single “best” set of values for these parameters

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What do many registration approaches use?

A

Small deformation model

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What are the two broad categories of parameterisation?

A
  1. The small deformation framework does not necessilary preserve topology - although if the deformations are relatively small, then it may still be preserved
  2. The largest-deformation framework generates deformations that have a number of elegant mathematical properties such as enforcing the preservation of topology
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What is the aim of image registration?

A

Estimate optimal geometric mapping between different images

Transform images so that they match in terms of:

  • orientation (position)
  • shape (anatomy)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Why is image registration necessary?

A
  1. Time series: motion correction
    - different images are acquired at different time points
    - motion: breathing, wiggle
  2. Longitudinal:
    - disease progression - in neurodegenerative diseases - look at inflammation and atrophy
    - developmental changes
    - plasticity changes
    - clinical trials endpoints
  3. Multi-modal
    - diagnosis
    - surgical planning
  4. Template space
    - group analyses
    - replication
    - atlas-based fusion
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

What is the basics of image transformation?

A
  1. When you get an image, read it into the matrix
  2. Have 8 by 8 matrix
  3. Each matrix/pixel in the matrix has a value
    - white is 0
    - black is 1
  4. Brain images are not binary, but intensity is a gradient
  5. Each value has an index- coordinates of the value of the matrix
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

What is translation?

A

Translations are movements in the x and y direction

Move forward, back, left right up and down

For rotation the origin is the same

For translation you shift the origin

B value is the transformation - you move by 2 points in the y direction

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

What is affine transformation?

A

It is a function mapping an affine space onto itself that preserves the dimension of any affine is spaces and also preserved the ratio of the lengths of parallel line segments

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

What are examples of types of transformation?

A
  1. Rigid: 6 DoF
    - rotations (roll, pitch, yaw)
    - translations (x,y,z)
  2. Affine: 12 DoF
    - 6 rigid
    - scaling (width,height, depth)
    - shearing (x,y,z)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

What does affine transformation not necessarily preserve?

A

Angles between lines or distances between points, though it does preserve ratios of distances between points lying on a straight line

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

What is affine transformation?

A

It is the matrix you apply to an image in order to be able to transform it

The information that the matrix includes is the rotation and translation

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

What is the differences between a rigid and affine transformation?

A
  1. Rigid transformation is only 6 degrees of freedom - 3 rotations and 3 translations
  2. Affine transformation has 12 degrees of freedom - you have scaling and shearing
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

What is the principle of image registration?

A
  1. You have a target image and reference image
  2. You want to move the target image to the reference image
  3. Read it into a matrix
  4. Apply transformation matrix
17
Q

What is image registration the process of?

A
  1. Estimating the transformation matrix that will match the target image to reference image
  2. Transforming (warping/reorienting) the target image so that it matches the reference image geometrically as close as possible
18
Q

How do we match images?

A
  1. Manually
    - overlay images
    - visually guided realignment
    - check alignment
    - repeat until perfect match

But

  • time consuming
  • potential bias
  • how do we handle the complex deformations?
  1. Read images in the same coordinate system
    - estimate some transformation
    - evaluate results of transformation
    - repeat until you reach a point when there is no more improvement
19
Q

What is the image registration framework?

A
  1. Deformation model
  2. Objective function
  3. Optimisation method
20
Q

Now do you measure similarity between two images?

A
  1. Sum of square differences (SSD)
  2. Cross-correlation (CC)
  3. Mutual information (MI)
21
Q

Affine transformation type

A

3 translation
3 rotations
3 scaling
3 shearing

22
Q

What is diffeomorphic registration?

A
  1. Preserves topology
    - maintains anatomical neighbourhood relationship and connectivity
  2. Transformations are
    - one to one
    - smooth
    - continuous
    - invertible i.e. the compositions of the inverse and forward transformation results in approximately the identity transform
  3. Flow fields (u) are estimated for each image to describe the transformation
23
Q

What is the aim of feature-based registration?

A

Match geometrical features:

  • landmarks
  • surface patterns

Surface-based: better for cortical folds than volumetric

24
Q

What are the different types of transformation?

A
  1. Rigid transformation
  2. Affine transformation
  3. Non-rigid transformation
  4. Landmark transformation
  5. Diffeomorphic transformation
25
Q

When would you use the transformation

A
  1. Same person scanned different time
    Points but in close proximity
  2. We could apply rigid transformation
  3. Same person but different types of scan that have different distortions between them
  4. Apply affine transformation
  5. Align scans from different participants - non-rigid - deomorphic/landmark based/ surface based
26
Q

What is a key property of diffeomorphic registration?

A

Maintaining topology so connectivity and neighbourhood

diffeomorphic registration is one to one smooth continuous and invertible mapping

27
Q

What is mutual information?

A

The amount of information that two images have in common

28
Q

What are the different types of similarity measures ?

A
  1. Sum squared differences
  2. Mutual information
  3. Cross correlation
29
Q

Which measure is better for Insta-modal registration?

A

SSD - simple and suitable

30
Q

What does mutual information describe?

A

The shared information between two images

31
Q

What is the solution for intermediate voxel positions?

A

Interpolation

32
Q

What are the types of interpolation?

A
  1. Nearest neighbour
  2. Linear, Bilinear, Trilinear
  3. Cubic and high order splines
33
Q

What is nearest neighbour interpolation?

A

Take the value of the closest point

34
Q

What is linear interpolation?

A

Weighted average of intermediate neigjbours in one direction

35
Q

What are bi- and tri-linear interpolation?

A

Linear interpolation in 2D and 3D

36
Q

What are widely used software packages?

A
  1. SPM
  2. FSL
  3. ANTs
  4. Free surfer