Kim review Flashcards
absolute path
the path of a file from the root location ex: /users/notcara/documents…..etc
relative path
the path of a file from current location: ex: ./MATLAB/etc.
how would you write a path that’s a folder above where you are
../jboudreau/etc.
terminal command: pwd
print current directory
terminal command: cd
change directory
terminal command: ls
list directory
terminal command: find
search files
terminal command: cp
copy files, use option ‘-r’ to copy folders
terminal command: mkdir
make directory
terminal command: mv
move/rename files
terminal command: rm
remove/delete files, use option ‘-r’ to remove folders
is there a recycle bin in Linux command line?
NO. it is possible to recover deleted files but it takes a lot of effort
terminal command: ln
create links/shortcuts to files, use option ‘-s’ to create symbolic links
files/directories are associated with three types of users:
owner (u), group (g), and others (o)
three kinds of permissions are assigned to each type of user
read (r), write (w), execute (x)
command: ls -l
view permissions
command: chmod
change permissions
command chown
change ownerships
MRI RF intensity inhomogeneities
when part of the image is shaded darker, there’s a gradient across the image going from darker to lighter
if r= distance from the origin of magnetic field, what would the graph with r as the x axis and 1/r^2 as the y axis look like?
steep drop down between 0 and 1 on the x axis tapering off for the rest of it
the slope of r by 1/r^2 graph would __ with higher field strength
increase
N3 inhomogeneity correction: end tolerance
smaller values lead to more stable results but take longer time
N3 inhomogeneity correction: field distance (mm)
smaller distance offers higher power but takes longer computationally.
the higher the magnetic field, the __ the field distance in N3 should be used
smaller
N3 inhomogeneity correction: subsampling factor:
the larger this is, the faster the algorithm runs but the coarser the results are
N3 inhomogeneity correction: Kernel fwhm
the larger the FWHM is, the faster the algorithm runs
N3 end tolerance: range
0.01 - 0.00001
N3 subsampling factor range
1.0-32.0
N3 Kernel fwhm range
0.05-0.50
normalization changes __, does not change __
the range of intensity distribution on the image using fixed min & max values; doesn’t change histogram shape
equalization requires __, changes __, and the min & max values __
a target image to match; the histogram; may not be fixed
output from brain extraction (2)
a mesh representation of brain boundary; a binary brain mask
a mesh representation of brain boundary is __ than the binary mask, because it __ but the binary is __
more accurate; can achieve sub-voxel accuracy; more convenient to use
BET option -f
“fractional intensity threshold”. the higher this local threshold, the smaller the Norm Force to expand the brain surface, the smaller the mask
BET option -R
recursively estimate the center of the brain; makes the result more stable especially when the image includes a large non-brain portion (it removes it)
rigid registration
the most strict, only between the same subject. is a type of linear registration
linear registration
use to adjust spacing difference, or for pre-registration for template construction
affine registration
has 12 parameters, takes a long time - pre-registration for deformable registration and template construction — deformation
slide 19
slide 19
linear interpolation
takes weighted average voxel intensity values - overlaps exist between interpolation coverage and actual voxels
(registration) similarity metrics (4)
sum of squared errors, correlation coefficients, mutual information, landmark matching. used for registration and transformation
sum of squared errors
same modality, value similarity (registration of images from the same scanner and same modality)
correlation coefficients
linear similarity (two images of same modality, possibly different scanners, or registration of subject image to atlas of same modality)
mutual information
different modalities, non-linear similarity
landmark matching
not intensity-based, identifies landmarks to match between images
con of sum of squares errors method
very restricted assumption
sum of squares errors method: assumption
same tissue showing with same intensity range in two images
using sum of squared errors, what could be performed as its preprocessing
histogram-matching
correlation coefficients: assumption
there is a linear relationship between voxel intensity of two images
using correlation coefficients, what could be performed as its preprocessing
histogram-matching
histogram-matching is not required for this similarity metric
mutual information
mutual information assumption:
two images share similar structures, but may be expressed with voxel intensities in different ways
mutual information pro
can handle non-linear relationships
3 methods of tissue segmentation
K-means, Gaussian-mixture models-based clustering, Markov Random Fields (MRF)
K-means clustering
takes the mean of the intensity values of clusters
K-means clustering only works when
the size of each cluster is comparable equal
“Mickey Mouse” ear issue
k-Means clustering results in segmentations that include the ear plus some of the face, whereas EM clustering is more accurate
Gaussian mixture model (GMM) clustering
each cluster is specified by two parameters: location and standard deviation
GMM clustering: location is __ standard deviation is __
the mean of intensity values in the cluster, the range of variance in intensity in the cluster
GMM clustering: __ are mixed together in the histogram. to model the mixture, __
different tissues types; each tissue type is assigned a weight
GMM clustering: weight:
mixture proportion relative to other tissue types
Markov Random Fields
uses probability to determine which tissue type a voxel is more likely to be given the neighboring voxels
HMRF (markov random field) does not mean __
protection of topology
what will happen when the weight for HMRF is large? small?
..
history of template construction (5)
Talairach atlas, MNI305 linear, MNI152 linear, MNI152 nonlinear (most popular), ADNI40 nonlinear
what is the most popular template?
MNI152 nonlinear
morphometry methods (3)
VBM, DBM, VBM + DBM (optimized VBM)
voxel-based analysis of gray matter (6 steps)
subject images, tissue classification, then you have probabilistic map of gray matte, spatial transformation to an atlas space, atlas space, voxel-wise statistical analysis
cons of VBM (3)
tissue probabilistic maps do not necessarily reflect the volume of a tissue at a voxel; relies on the accuracy of the tissue classification; mis-registration may lead to false positives
deformation-based morphometry (DBM): by-product of __
nonlinear registration
how does DMB overcome the criticism of VBM?
…
optimized VBM: involves __
Jacobian modulation
optimized VBM: steps (3)
gray matter map, times Jacobian determinant map, = modulated map
how does the optimized VBM overcome the issue of VBM?
…
Jacobian determinants:
how much a voxel changes after registration to the template
atlas-based segmentation: involves __ (2)
registration to template, transformation (segmentation)
nearest neighbor interpolation
assigns intensity value of the nearest voxel to the interpolation center
interpolation of label maps can either be
linear, or nearest neighbor method
evaluation of segmentation accuracy
…?
linear transformation
does the same thing to each voxel, global
nonlinear transformation
you can work with a specific part of the image; local
what software do we use for rigid registration?
ANTS
DBM
doesn’t rely on segmentation
Jacobian determinant
how much a specific voxel changes in size during deformation
two indices to evaluate segmentation accuracy
dice (more popular) and jaccard
