w6 Flashcards
what is Image registration
the process of transforming the different sets of data into one coordinate system. Image registration is trying to maximise the amount of shared information in two images. Can be thought as reducing the information in the combined image. necessary to compare or integrate the medical data obtained from different
time, modalities or subjects.
what is image fusion
registration + combination into a single representation
e.g. pet + ct
what are the advantages of image fusion
Improved system performance, detection, tracking, situation assessment and awareness, robustness, spatial and temporal coverage
what is Monomodal image registration
to register serial scans from the same imaging technique in temporal studies to evaluate disease progression, assess treatment response, detect changes, and monitor growth and development.
what is Multimodal image registration
to combine complementary information from different
imaging modalities and to correlate anatomical structure with functional information, e.g., CT-PET, CT-MRI, MRI-PET
describe the process of Registration Optimisation
a searching strategy to find the best registration solution i.e. the solution that produces the most similar spacial correspondance.
what is Powell optimisation
performs a succession of one-dimensional optimisations, finding in turn the best solution along each freedom degree.
The algorithm stops when it is unable to find a new solution with a significant improvement to the current solution.
describe the process of Registration Interpolation
Interpolation is required when an image needs to undergo transformations.
projective mapping can yield noninteger values for pixels. Because the target image is digital, its pixel values are defined only at integer coordinates. Thus it is necessary to infer the gray-level values at those locations; Gray-level interpolation is the technique that determines the gray-level value at noninteger coordinates.
what are the most frequently used interpolation methods (in biomedical imaging)
nearest neighbor, linear, bilinear, trilinear, cubic, bicubic, tricubic, quadrilinear, cubic spline, sinc function.
what is registration transformation
to relate the pixels of moving/study/floating images to the corresponding pixels of a fixed/reference/target image.
pk = T(qk) where pk and qk are corresponding pixels
what are the three categories of registration transformation
Rigid transformation
Affine Transformation
Deformable/Elastic/Nonlinear Transformation
what is Rigid Transformation
rotation, reflection, and translation (sliding)
preserves lengths/distances and angle measures, is often used to correct translation and rotation displacements.
what is Affine Transformation
maps parallel lines into parallel lines, can be used to correct skewing distortion introduced by e.g., a tilted gantry in CT
what is Deformable/Elastic/Nonlinear Transformation
more complex and is mainly used to correct dramatic deformations caused by changes of tissue structures, differences of volume and shape of organs.
Relies on extracted features (point, curves, surfaces). Interpolation necessary. Transformation is mostly relevant in the neighborhood of the homologous features (local not global).
in what situations is rigid transformation appropriate and not appropriate to use in
appropriate in intra-subject registration
Fails in: matching atlas to patients, inter-subject registration, presence of deformations (tumor, craniotomy)
what are the three different techniques used to perform registration
Feature-based
Intensity-based
Hybrid
what is Feature-based registration
using features to align the two images, such as: Point/landmark
Contour
Surface
what is Intensity-based registration
The registration transformation is determined by iteratively optimizing a certain similarity measure calculated from pixel values. Similarity measures include:
Minimizing the intensity difference,
Correlation techniques,
Variance of Intensity Ratio (VIR) by R.Woods,
Information theoretic techniques (Maximization of Mutual Information)
what is Hybrid registration
Combining feature- and intensity-based registration techniques.
what is the procedure for Feature-based registration
Preprocessing (Segmentation): feature extraction
Registering: Computation of the optimal transformations to register features in B onto features in A.
Verifying: Usually through visual check, comparing against phantoms
what are phantoms
a specially designed object that is scanned or imaged in the field of medical imaging to evaluate, analyze, and tune the performance of various imaging devices.
what are the advantages and disadvantages of feature-based registration
Advantage is that the transformation can be formulated
in an analytic form with efficient computational schemes.
However, a preprocessing step is needed and the results are highly dependent on the result of the feature extraction. Complicated by the problem of local minima.
describe the process of Point-based feature-based registration
involves identifying corresponding points (homologous landmarks) in the images to be registered, registering the points (Iterative Closest Point (ICP) or Thinplate
Spline technique), and inferring the image transformation.
In Point/landmark-based registration, what are the qualities of good landmarks
Landmarks should: Have uniquely defined positions. Carry substantial image information. Be well-suited for user-interaction. Facilitate efficient registration. Have salient, prominent characteristics. Be scattered over the image.
describe the Thinplate Spline technique
Spline is referred to as a long flexible strip of metal, which takes a least bent shape along the spline.
This bent spline can be used to model the surface of deformed objects. can produce a smooth spline interpolation, has high computation speed, and can correct the local elastic deformations by mapping the study landmark points to the corresponding points in
the reference image.
what are Fiducial Markers (point based feature)
an object placed in the field of view of an imaging system which appears in the image produced, for use as a point of reference or a measure.
Non-invasive: mould, frame, dental adapter, skin markers
Invasive: screw markers, stereotactic frames.
describe the process of Surface feature-based registration (“Hat and Head” method)
Two surface models are generated: “hat” and “head.” The “hat” surface is a skin surface from PET scan with low resolution. The “head” surface is a stack of skin contour from CT or MR scans with high resolution.
The two segmented surfaces are visualised in 3D computer graphic systems and aligned by minimising the mean square distance between them.
what is the disadvantage of surface registration
Unfortunately, this algorithm may be prone to finding the wrong solution and tend to fail when the surfaces show symmetries when rotated.
what is the Simplest scheme for Intensity-Based Registration
minimising the intensity difference.
methods include: Sum of Squared Differences (SSD) and the Sum of Absolute Differences (SAD) (check equations). These are efficient to calculate.
What is the drawback of using Sum of Squared Differences (SSD) and Sum of Absolute Differences (SAD) for registration
These methods are sensitive to intensity changes. If there are lots of intensity changes throughout the image then it may not work well. This limits their application to monomodality image registration only.
what is Correlation coefficient-based registration techique (Intensity-Based Registration)
Estimates the relative translative offset between two similar images. Correlation based registration techniques were proposed to aim at multimodal biomedical image registration
what is the drawback of correlation coefficient-based registration
Because the geometric deformations of the image modalities are not likely to be linear, these correlation methods, which require a linear dependence between the intensity of the images, cannot achieve reliable registration results.
what is entropy
Entropy measures the homogeneity (purity). describes the average information supplied by a set of values, given their probability.
The lower the entropy, the greater the purity of the set. When all the variables have the same probability, the entropy achieves its maximum value, and when the probability of one variable is equal to one and the probabilities of other variables are equal to zero, then the entropy is minimum (equal to zero). In registration, the goal is to get minimum entropy.
What is the Information Theoretic definition of a histogram
Gives the distribution of probabilities. A way of viewing the entropy of an image.
Suppose image I has N pixels, then the histogram is the probability that a pixel has intensity i
what is joint entropy
Joint entropy measures the amount of information in a combined image.
If the two images are totally unrelated, the joint entropy will arrive at its maximum value and be the sum of the entropies of the individual images. the more similar the images are, the lower the joint entropy.
what is a joint histogram
The Joint Histogram is a measure of the correlation between the intensities in the two images. The values for each pixel represents the probability of the value pairs occurring together.
what does an ideal registration histogram look like
the joint histogram appears sharper and has tight clusters. i.e. the distribution is low.
what is Mutual Information
Measures degree of statistical dependence between two random variables. A measure of how well one image explains the other one.
what is mutual information registration
Involves finding the transformation that makes one image the best possible predictor for the second image. Mutual information will be maximum at the optimal registration. Maximises the clustering of the joint histogram.
suitable for multimodal image registration. It is mainly used in rigid registration schemes.
describe the hybrid registration process for MRI brain registration
Step 1 – landmark-based global registration.
Step 2 – intensity-based registration for regions away from the landmarks.
This is iterated until convergence criteria met.
describe the hybrid registration process for retinal image registration
Step 1 – binary vascular trees are extracted based on local entropy-based thresholds.
Step 2 – zero-order translation is estimated by using Mutual information technique.
Step 3 – feature points are used to estimate higher-order transformation. (landmark based)
how can image registration help with early diagnosis of neurodegenerative disorders
not all brains are the same.
Build statistical models that capture patterns / characteristics of normal brains and/or specific brain diseases. Atlas’s can be used to automatically segment and label. Registration is used to combine the test image with the atlas before segmentation.
what are the current challenges of image registration
Multimodal elastic registration of deformable organs.
Automatic approaches for the registration of heart, lung, and liver. (Hybrid methods, combining similarity measures with morphological information may provide
possibilities for elastic registration)
Efficiency, especially for multiple dimensional whole-body images.
Validation of the registration performance. Although a wide variety of registration approaches have been proposed, objective validation of these methods is not
well established. (Image databases may in the future provide a source for the objective comparison of different registration methods)
