Week Five - Medical Data Registration & Fusion Flashcards
What is the difference between image registration and fusion?
Registration = image aligning Fusion = registration + combination (merging)
Define image registration.
The process of transforming different sets of data into one coordinate system. It aims to spatially match data sets that may differ in time of acquisition, imaging device, and acquisition angle.
What is the purpose of monomodal image registration?
To register scans from the same imaging technique in temporal studies to evaluate disease progression.
What is the purpose of multimodal image registration?
To combine information from different imaging modalities in order to correlate anatomical structure with functional information.
What is an atlas?
A database of medical images that can be used to automatically segment, label, or interpret different tissues; and study anatomical/functional variability.
What advantage does image guided surgery have over conventional surgery?
Conventional surgery relies on direct vision, whereas image guided surgery allows planning and navigation by the registration of images to physical space when operating.
What role does registration play in image guided surgery? (3 things)
Register:
- Anatomical images with patient
- Intra-operative images with patient
- Segmented anatomical images with intra-operative images
What are the steps involved in the registration framework?
Reference image + study image - Similarity measure - Optimisation - Transformation & Interpolation
What is interpolation?
Uses known data to estimate data at unknown points. When an image undergoes a transformation, each pixel might not map exactly to the centre of a pixel in the output image - so interpolation is used to predict the value of that pixel.
Name 3 commonly used interpolation methods.
Nearest neighbour, linear, bi-linear.
What is the purpose of an optimisation algorithm?
To search for the optimal transformation to maximise the similarity measure.
Describe the Powell algorithm.
An optimisation algorithm that performs a succession of 1D optimisations, finding the best solution for each variable; and the single variable optimisations are used to determine the new search direction.
Define transformation.
To relate the pixels of a study image to the corresponding points of a reference image.
Name three types of transformations.
Rigid - preserves lengths and angles (corrects translation and rotation)
Affine - maps parallel lines to parallel lines (corrects skewing)
Deformable - corrects dramatic deformations caused by changes of tissue structure, difference in shape of organs etc.
Name 4 applications of deformable transformations.
Matching atlas to patients.
Inter-subject registrations.
Temporal deformations.
Between pre- and intra-operative.
What are the two central types of registration strategies?
Feature based and intensity based.
Describe fiducial marker registration.
Artificial markers are placed on the subject e.g. stereo-tactic frames, skin markers.
What is the difference between feature based and intensity based registration?
Feature based - establish a correspondence between distinct points, before a geometrical transformation is used to map the target image to the reference image.
Intensity based - compares intensity patterns in images via correlation metrics.
What are the three main types of feature based registrations?
Point/landmark
Contour
Surface
What are the three steps involved in feature based registration?
- Pre-processing (segmentation)
- Registering
- Verifying
What is an advantage and a disadvantage of feature based registration?
Transformation in analytic form allows for efficient computation.
Highly dependent on success of pre-processing.
What are some advantages (2) and disadvantages (1) to intensity based registration?
No preprocessing, automatic.
Low computational efficiency.
Why is intensity based registration only used for monomodal images?
Sensitive to intensity changes, therefore works best with 2 images with similar intensities. Correlation metrics only works when there is a linear dependence between intensities - which doesn’t happen across image modes.
What are two similarity measures used in intensity based registration?
Sum of squared differences, sum of absolute differences. Minimum in case of perfect match.
