Instructor: Moo K. Chung [email protected] Lecture...
Transcript of Instructor: Moo K. Chung [email protected] Lecture...
![Page 1: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/1.jpg)
Neuroimage Processing
Instructor: Moo K. Chung [email protected]
Lecture 10-11. Deformation-based morphometry (DBM)
Tensor-based morphometry (TBM)
November 13, 2009
![Page 2: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/2.jpg)
Image Registration
• Process of transforming one image to another.
• Transformation types: linear (rigid-body), affine (non rigid-body), nonlinear.
• Type of data obtained: 3D vector fields (displacement, deformation).
![Page 3: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/3.jpg)
Affine Transformation matrix for (rigid, non rigid)
• R: 3 x 3 matrix of rotation, scaling and shear
• p’=Rp+c, where
€
p'=ʹ′ x ʹ′ y ʹ′ z
⎛
⎝
⎜ ⎜ ⎜
⎞
⎠
⎟ ⎟ ⎟ , p =
xyz
⎛
⎝
⎜ ⎜ ⎜
⎞
⎠
⎟ ⎟ ⎟ ,c =
cx
cy
cz
⎛
⎝
⎜ ⎜ ⎜
⎞
⎠
⎟ ⎟ ⎟
![Page 4: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/4.jpg)
Linear transform
• T is a linear transform if T(ap+bq) = aT(p)+bT(q) for all numbers a, b.
• Checking if the affine transform is linear. Let T(p)=Rp+c.
Note T(ap+bq)=R(ap+bq)+c=aT(p)+bT(q)+c(1-a-b).
This shows the affine transform is nonlinear.
![Page 5: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/5.jpg)
Rigid-body:
• Rotation and Translation only
Original Image Shape doesn’t change
A. Chowdhury, University of Georgia
![Page 6: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/6.jpg)
Scaling, Translation, Rotation, Reflection, Shear
Original Image Non rigid-body
Non rigid-body
![Page 7: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/7.jpg)
Nonlinear transform
• Anatomical variability is encoded in the nonlinear transform.
Original image Target Image Result of warping
![Page 8: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/8.jpg)
Talairach Template
http://www.talairach.org/
Talairach coordinate system has been widely used to describe the location of brain structures.
![Page 9: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/9.jpg)
Piecewise Affine Transform to Talairach • The standard Talairach normalization approach
• It uses a different matrix transformation for each of the 12 pieces of the Talairach Grid
• This is a really old technique based on a single old subject.
• It provides an easy reference for comparing different study results.
• Other than for comparing results against literature, the Talairach approach is outdated.
![Page 10: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/10.jpg)
Talairach Definition • Interhemispheric plane (3+ landmarks) ⇒ 2
rotations and 1 translation • Anterior and posterior commissure (AC, PC) ⇒ 3rd rotation, 2 translations
• Scale to anterior, posterior, left, right, inferior, superior landmarks (7 parameters)
• Each cerebral hemispheres divided into six associated blocks (interhemispheric plane, AC-PC axial plane, 2 coronal planes through AC and PC.
![Page 11: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/11.jpg)
MATLAB codes for reading/writing http://www.rotman-baycrest.on.ca/~jimmy/NIFTI/
Siemens DICOM sort and conversion to NIfTI format http://www.mathworks.com/matlabcentral/fileexchange/22508
MATLAB Demonstration
NIfTI image format
http://nifti.nimh.nih.gov/
NIH initiated effort for standardizing brain imaging file format. It supersedes the analyze and MNI file format.
![Page 12: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/12.jpg)
Affine vs. Nonlinear
Non-Rigid ~ 2000 parameters Affine – 12 parameters
Lawrence H. Staib Yale University
![Page 13: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/13.jpg)
Automated image registration (AIR) package (http://bishopw.loni.ucla.edu/AIR5/index.html) uses polynomial basis.
SPM package uses cosine basis.
Advanced Normalization Tools (ANTs) http://picsl.upenn.edu/ANTS/
Nonlinear registration tools
![Page 14: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/14.jpg)
Deformation field obtained from AIR
• The deformation field d is a transformation from a subject image to a target image:
• Example: AIR 3rd degree warping
2 5 2 5 5 2 6
5 5 2 8 3 7 2 10 2 7 3
8 2 7 7 2
' 1.24 0.47 1.02 10 0.99 1.25 10 2.66 10 4.04 10 2.35 109.23 10 5.39 10 3.60 10 1.04 10 9.3 10 1.17 10
2.2 10 1.24 10 6.82 10 1.24 10
x x y z x xy y xzyz z x x y xy y
x z xyz y z
− − − − −
− − − − − −
− − − −
= − − − × + + × − × + × + ×
+ × + × + × + × − × + ×
− × + × − × − × 7 2 8 2 7 31.69 10 2.76 10 .xz yz z− −− × − ×
![Page 15: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/15.jpg)
Deformation based Morphometry (DBM)
• It uses deformation fields obtained by nonlinear registration of brain images.
Read M.K.Chung.Book… Chapter 7 upto section 7.2
![Page 16: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/16.jpg)
Visualizing Deformation
Red: tissue expansion Blue: tissue shrinking Yellow: deformation change
![Page 17: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/17.jpg)
Deformation field
visualization
SPM result
![Page 18: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/18.jpg)
Displacement vector fields
• The vector difference between the final position and the original position.
• Relation between displacement and deformation
target initial voxel position
displacement deformation
![Page 19: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/19.jpg)
Displacement vector
![Page 20: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/20.jpg)
Visualizing Displacement Vector Field
Displacement is easier to model statistically than deformation
![Page 21: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/21.jpg)
28 normal subjects, 1.5T MRI, 2 scans per subjects
First scan: 11.5±3.1 year, second scan: 17.8 ±3.2 year
Problem: localize the regions of anatomical change over time
![Page 22: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/22.jpg)
Modeling on the rate of displacement change Why? Easier than modeling on deformation itself.
: Covariance matrix
: Gaussian random vector field
![Page 23: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/23.jpg)
Estimating the rate of change
• For subject j, displacement is given by
• Finite difference estimation:
![Page 24: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/24.jpg)
• Rate of displacement
• Sample mean
• Sample covariance
• Hotelling’s T-square (related to Mahalanobis distance)
€
V = V j
j=1
n
∑€
V j
€
ˆ Σ =1n −1
(V j −V )(j=1
n
∑ V j −V )T
€
H =VT ˆ Σ −1V ≈ cF3,n−3
Hotelling’s T-square statistic
![Page 25: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/25.jpg)
€
Vij = µ j + Σ1/ 2eij
Group index j, subject index i
€
H0 :µ1=µ2 vs. H1 :µ1≠µ2
Hotelling’s T-square for two sample test
n and m samples each
![Page 26: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/26.jpg)
• Sample means: ,
• Sample covariance: pool the variance across groups
• Hotelling’s T-square statistic for two samples
€
V1
€
V2
€
ˆ Σ =1
n + m − 2(Vi1 −V1)(
i=1
n
∑ Vi1 −V1)T + (Vi2 −V2)(
i=1
m
∑ Vi2 −V2)T⎡
⎣ ⎢
⎤
⎦ ⎥
€
H = V 2 −V 1( )T ˆ Σ −1 V 2 −V 1( ) ≈ cF3,n+m−4
![Page 27: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/27.jpg)
Red: Tissue growth p < 0.025 Blue: Tissue loss p < 0.025 Yellow: Structure displacement p < 0.05
Front
Back
Right Corpus Callosum
3D SPM These 3D blobs of signal are meaningless without
superimposing them onto MRI.
![Page 28: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/28.jpg)
Red: Tissue growth Blue: Tissue loss Yellow: Structure displacement
![Page 29: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/29.jpg)
![Page 30: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/30.jpg)
Red: Tissue growth p < 0.025 Blue: Tissue loss p < 0.025 Yellow: Structure displacement p < 0.05
Front
Back
Right Corpus Callosum
3D SPM
![Page 31: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/31.jpg)
![Page 32: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/32.jpg)
How it should be done correctly? To reduce the total amount of registration errors, 1st scans should be registered to the 2nd scans first.
Registration procedures in Chung et al. (2001) is not optimal.
Limitation of Chung et al., 2001
![Page 33: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/33.jpg)
Tensor-based Morphometry (TBM)
• It uses higher order spatial derivatives of deformation fields to construct morphological tensor maps.
• From these tensor maps, 3D statistical parametric maps (SPM) are created to quantify the variations in the higher order change of deformation.
![Page 34: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/34.jpg)
Determinant out of this
![Page 35: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/35.jpg)
Jacobian determinant (JD) • The Jacobian determinant J of the deformation field is
mainly used to detect volumetric changes.
Voxel position
Deformation
Jacobian determinant
€
J(x) = det ∂d(x)∂x
= det∂d j
∂xi
⎛
⎝ ⎜
⎞
⎠ ⎟
€
x = (x1,x2,x3)'
€
d = (d1,d2,d3)'
![Page 36: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/36.jpg)
Interpretation of Jacobian determinant (JD)
• JD measures the volume of the deformed unit-cube after registration.
• In images, a voxel can be considered a unit-cube
• JD measures how voxel volume changes after registration.
![Page 37: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/37.jpg)
JD Computation
• 1D example:
• 2D example:
€
ʹ′ x = 2x +1J(x) = 2
€
ʹ′ x = 2x + y +1ʹ′ y = x + 2y
J(x,y) = 3
0
1
1
3
(0,1) (3,1)
(2,2) (4,3)
![Page 38: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/38.jpg)
Is it possible to perform TBM using only an affine registration?
• For affine transformation p’=Ap+B, the Jacobian determinant is det(A) at every vovels.
• Every voxel will have the same scalar value.
• Affine registration based TBM only detect global size difference.
TBM requires really good high order registration technique to work properly.
![Page 39: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/39.jpg)
Statistically significant regions of local volume change JD > 1 volume increase, JD < 1 volume decrease over time
My first paper Chung et al. 1999. HBM meeting
![Page 40: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/40.jpg)
Generalization of Jacobian determinant in arbitrary manifold = determinant of Riemannian metric tensors = local volume (surface area) expansion
Surface-based Jacobian determinant
![Page 41: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/41.jpg)
Distributional assumption: Normality of JD
€
J(p) ≅1+ tr ∂U∂ ʹ′ p ⎛
⎝ ⎜
⎞
⎠ ⎟ =1+
∂U1
∂p1+∂U2
∂p2+∂U3
∂p3
€
d(p) = p +U(p)
€
∂d(p)∂ ʹ′ p
= I +∂U(p)∂ ʹ′ p
Note that we modeled displacement U to be a Gaussian random field. Any linear operation (derivative) on a Gaussian random field is again Gaussian. So J is approximately a Gaussian random field.
![Page 42: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/42.jpg)
Lognormal distribution
• Random variable X is log-normally distributed if log X is normally distributed N(mu, sigma^2)
• For E log X=0, the shape of density:
Some lognormal distribution looks normal so how do we check if data follows normal or lognormal?
![Page 43: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/43.jpg)
Properties of JD
• J(p) >0 for one-to-one mapping • J(p) > 1 volume increase; J(p) <1 volume decrease • Due to symmetry, the statistical distribution of J(p) and 1/J(p) should be identical.
Mapping d(p) JD J(p)
Inverse mapping JD 1/J(p)
€
d−1(p)Subject 1 Subject 2
![Page 44: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/44.jpg)
• Domain
• If J(p)=1,
• Symmetry:
• These 3 properties show that JD can be modeled as lognormal distribution.
€
−∞ < logJ(p) <∞
€
logJ(p) = 0
€
log J−1(p)[ ] = −logJ(p)
Distributional assumption: Lognormality of JD
![Page 45: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/45.jpg)
Testing normality of data
• How do we check if JD is normal or lognormal emphatically?
• Quantile-quantile (QQ) plot can be used. For given probability p, the p-th quantile of random variable X is the point q that satisfies
P(X < q) = p.
![Page 46: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/46.jpg)
QQ-plot compares quantiles
x y
x
y
![Page 47: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/47.jpg)
Normal probability plot showing asymmetric distribution
Longer tail
![Page 48: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/48.jpg)
Checking normality across subjects on cortical measure
Chung et al., 2003 NeuroImage
![Page 49: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/49.jpg)
Fisher’s Z transform on correlation
Tricks for increasing normality of data
![Page 50: Instructor: Moo K. Chung mkchung@wisc.edu Lecture 10-11.brainimaging.waisman.wisc.edu/~chung/neuro.processing/lectures.2009/... · determinant is det(A) at every vovels. • Every](https://reader033.fdocuments.net/reader033/viewer/2022041508/5e26a38898c49624b66ef66b/html5/thumbnails/50.jpg)
Increasing normality in heat kernel smoothing
Thickness 50 iterations 100 iterations
QQ-plot QQ-plot QQ-plot