MC Lec4-7 Flashcards
true or false: region-growing approach is a type of thresholding data process
false: region-based segmentation
true or false: the area of delineation is subjective and can vary up to 10% per observer
true
the process that divides an image into its constituent parts, objects, or ROIs is:
a) registration
b) segmentation
c) recognition
d) delinearization
b
region-based segmentation is based on:
a) homogenous parts being detected via grey-level thresholding
b) homogenous parts being detected via region growing
c) homogenous parts being detected via region splitting/merging
d) all of the above
d
edge-based segmentation involves abrupt changes in _____ _____ which correspond to edges
grey level
the simplest segmentation strategy is:
a) region-based segmentation
b) thresholding
c) statistical pixel classification
d) all of the above
b
true or false: when the grey levels of the ROI are known from the image histogram, we image can be thresholded to assign a value of 1 to the ROI and a value of 0 to other details. this is known as __________
binarization
bimodal histogram has the object and background pixel intensity grouped in ___ dominant modes
two
when the threshold selected is a constant applied over the entire image, the process is called:
global thresholding
when the threshold value T changes over an image, this is called:
variable thresholding
the two types of variable thresholding are:
local: the value of T is dependent on the properties of a neighbourhood
adaptive: the value of T depends on teh actual spatial coordinate
list the following steps for basic thresholding in order:
- select an initial estimate for the global threshold (T)
- segment the image using T, producing two groups of pixels G1 and G2 on either side of T
- compute a new threshold value, based off the averages of G1 and G2 intensity values. repeat until the T value is smaller than the predefined deltaT value
- compute the average intensity value of G1 and G2 respectively
1243
true or false: the predetermined deltaT value in global thresholding is used to control the number of iterations in situations where speed is an important issue. in general, a larger deltaT, the more iterations the algorithm will perform
false: larger deltaT = LESS iterations. logically, the difference between the computed T should get smaller and smaller each time
true or false: the average intensity of the image is a good initial choice for threshold (T) in global thresholding
true
list the three thresholding histograms:
unimodal, bimodal, trimodal
under what condition, if any, can we perform unimodal histogram thresholding?
if we know the intensity that the ROI is at. otherwise needs to do some pre-processing
which of these would make thresholding easier and harder?
a) large separation between intensity peaks
b) lots of noise/larger peaks
c) small object on large, non-uniform background
d) uniform illumination of image
e) uniform reflectance properties of image
a) easier
b) harder
c) harder
d) easier
e) easier
what is a pre-processing technique to reduce noise and enhance bimodal histogram shape?
median filter smoothing
which of these is not a way to threshold with a non-uniform background?
a) inversing the pattern to correct illumination
b) work around via partitioning variable thresholding techniques
c) correct the global shading pattern via top-hat transformation
d) all of the above enable thresholding with a non-uniform background
d
adaptive thresholding via image partitioning involves:
breaking up the image into sub-images that are approximately uniform and utilising a different threshold segment for each sub-image
true or false: adaptive thresholding typically fails when the ROI and background occupy different size regions
true. this is because the likelihood of regions containing only object or background is higher
true or false: region-growing based segmentation algorithms examine pixels in the neighbourhood based on a pre-defined similarity critereon
true
true or false: the basic difference between thresholding and region-growing segementation is that the latter guarantees the segmented regions of connected pixels, whereas the former may yield regions with holes and disconnected pixels
true. thresholding is pixel based and region growing joins regions
true or false: with region-growing segmentation, the stopping criterion can be based on the minimum number or percentage of neighbourhood pixels required to satisfy the similarity criterion
true
the region-growing process stops at the iteration in which there is no new neighbourhood pixels to satisfy the similarity criterion
true
true or false: image registration is done after segmentation and involves transforming the different sets of data into multiple coordinate systems
false: one coordinate system
true or false: image registration enables integration of medical data obtained from different times, modalities, or subjects
true
image fusion involves:
combination and registration of information from multiple images into a single representation
list three advantages of multi-image registration
- improved system performance
- improved detection, tracking, and identification
- improved situational assessment
- improved robustness
- extended spatial and temporal coverage
true or false: monomodal image registration is used to register serial scans from in temporal studies to assess treatment response, detect change, and monitor disease progression
true
image registration involves a _______ image and a ________ image
reference, study
the aim of multimodal imaging is:
to correlate anatomical structure with functional information
list the four key components of the registration framework:
- registration transformation
- interpolation
- registration criterion
- optimization procedure
true or false: optimization is required by most registration procedures
true
interpolation is required when an image needs to undergo ________
transformation
the more complex the interpolation methods, the more surrounded points concerned, and the faster/slower the registration speed
slower
the main aim of medical image registration is to determine a transformation to relate the pixels of the _____ image to the corresponding pixels of the _______ image
study, reference
the three types of registration transformation are:
rigid, affine, deformable/elastic
rigid transformation:
a) preserves lengths and angles
b) corrects translation and rotation displacements
c) can be an initial step for a more complicated registration procedure
d) all of the above
d
affine transformation:
a) maps parallel lines into parallel lines
b) correct skewing distortion
c) all of the above
c
deformable/elastic transformation:
a) corrects complex, dramatic deformations
b) corrects skewing distortion
c) corrects translation and rotation
d) all of the above
a
rigid transformation is appropriate for:
a) matching atlas to patient
b) presence of deformations
c) intra-subject registration
d) all of the above
c
when would you use elastic/deformable transformation?
a) inter-subject registration
b) matching atlas to patient
c) presence of deformations
d) all of the above
d
the three types of registration techniques are:
feature-based
registration-based
hybrid-based
which of these is NOT feature-based registration:
a) point/landmark
b) correlation technique
c) contour
d) surface
b
which of these is NOT an intensity-based registration:
a) minimizing intensity difference
b) correlation techniques
c) contour
d) variance of intensity ratio
e) information theoretical techniques
c
what may hold back feature-based registration techniques?
adequate extraction of the features in the segmentation (preprocessing) stage
true or false: landmarks represent the same feature in different images
true
once the corresponding landmarks have been decided, _____ _____ _______ can be used to register the images
thin-plate spline technique
which of these is a feature of thin-plate spine technique?
a) produce smooth spline interpolation
b) high computation speed
c) can correct local elastic deformations by mapping to study image to reference image
d) all of the above
d
true or false: fiducial markers can be invasive and non-invasive
true
fiducial markers are widely used in:
a) mammograms
b) cancer staging
c) image-guided surgery
d) none of the above
c
surface registration uses the ___ and ____ method
hat, head
in surface registration, the hat surface is a skin surface from a ____ scan with ___ resolution. the head surface is a stack of skin contour from a ___ or ___ scan with ___ resolution. the two surfaces are aligned by minimising the mean distance between them
PET, low, CT/MRI, high
true or false: surface feature-based registration tends to fail if symmetry is not shown upon surface rotation
false: will fail if surfaces show symmetries to rotation
minimising the intensity difference in intensity-based registration involves SSD and SAD which exhibit the minimum in the case of perfect matching . what do they stand for?
SSD = Sum of Squared Differences
SAD = Sum of Absolute Differences
true or false: minimising the intensity difference in intensity-based registration is limited to monomodality image registration only
true
the correlation coefficient intensity-based registration method depends on a ____ dependence between the intensity of the images
linear
true or false: correlation based-registration techniques are aimed at monomodality image registration
false: multimodality
true or false: the correlation coefficient intensity-based registration technique can also be used for rigid motion correction of SPECT cardiac images because it requires a linear dependence
true
how does image registration play an important role in breast cancer mammography?
the rate of uptake of contrast in healthy and cancerous tissue is different
which of these is not a challenge of image registration?
a) rigid and affine registration in multimodality imaging
b) automatic approaches for the registration of heart, lung, and liver images
c) efficiency to be implemented in a clinical practice
d) objective validation of registration performance
e) elastic registration in multimodality imaging
a