Samenvatting Medische Beeldanalyse (8DC00)

Samenvatting Medische Beeldanalyse – 8DC00

Index
MEDICAL IMAGE REGISTRATION......................................................................................... 2
Introduction to image registration..................................................................................2
Geometrical transformations.......................................................................................... 2
Image transformation..................................................................................................... 6
Point-based registration.................................................................................................. 7
Evaluation of the registration accuracy...........................................................................9
Intensity-based similarity metrics.................................................................................10
SEGMENTAION OF MEDICAL IMAGES................................................................................14
Evaluation of segmentation.......................................................................................... 14
Segmentation in feature space.....................................................................................16
Segmentation in feature space – clustering..................................................................18
Segmentation in feature space – classification.............................................................18
Generalization............................................................................................................... 20
Overfitting..................................................................................................................... 20
Overfitting in kNN......................................................................................................... 21
Dimensionality reduction..............................................................................................21
Atlases.......................................................................................................................... 23
Active shapes models................................................................................................... 25
COMPUTER-AIDED DIAGNOSIS.........................................................................................27
Introduction to CAD; Linear regression.........................................................................27
Logistic regression and neural networks.......................................................................28
Performing machine learning experiments...................................................................30
CAD convolutional neural networks..............................................................................31




Page 1 out of 34

,MEDICAL IMAGE REGISTRATION
Introduction to image registration

Image registration: the determination of a geometrical transformation that aligns one
view of an object with another view of that object, or another object. This can be the case
because of:
- Different positioning of the patient
- Movements of organs
- Movements of patient
- Distortions caused by imaging system
- Changes caused by interventions (e.g. surgery, chemo) in between the acquisition
of the images.




Applications of image registration:
- Combining information from different sources
- Comparison: differences in (groups of) subjects, or monitoring changes in a single
subject
- Segmentation
- Motion correction
- Image-guided treatment
- Atlas, model of average anatomy.

Review of linear algebra
Matrix transpose:

Special matrices and vectors:
- Unit vector

- Symmetric matrix

- Orthogonal matrix




Determinant: the determinant of a transformation matrix T is the signed area of a unit
square shape after transforming it with T. The sign reflects whether the orientation has
changed.
 Here no flip but rotation!



Geometrical transformations




Page 2 out of 34

,Translation:




For which the distance between two points (in 2D) can be described by the following
formula:


Rotation:




LET OP: Not every matrix can be considered as a rotation matrix. Rotation matrices must:
- Be orthogonal

- Have a determinant equal to 1




Page 3 out of 34

, Scaling




Shearing:




Reflection
Horizontal:




Vertical:




Composition of transformations: it is also possible to combine different sorts of
transformations. They can be combined by multiplying the transformation matrices.
- Rotation + translation (rigid) (links)

- Rotation, scaling + translation (rechts)




Page 4 out of 34