Professional Documents
Culture Documents
Voxel-Based Morphometry With Unified Segmentation
Voxel-Based Morphometry With Unified Segmentation
Thanks to:
John Ashburner and the FIL Methods Group.
Preprocessing in SPM
• Realignment
– With non-linear unwarping for EPI fMRI
• Slice-time correction
• Coregistration
• Normalisation SPM8b’s unified tissue
segmentation and spatial
• Segmentation normalisation procedure
“Contrast”
spm T
Preprocessing Stat. modelling Results query
Image
fMRI time-series
“Contrast”
Preprocessing Stat. modelling Results query
Image
fMRI time-series
“Contrast”
Preprocessing Stat. modelling Results query
Image
SPM for structural MRI ? Group-wise
statistics
High-res T1 MRI
High-res T1 MRI
High-res T1 MRI
?
The need for tissue segmentation
GM and WM segmentations
overlaid on original images
Lower segmentations
have been cleaned up
Limitations of the current model
• Multispectral modelling k μ k , k σ k , { s }
– (New Segment Toolbox)
• Deeper Bayesian philosophy
– E.g. priors over means and variances
– Marginalisation of nuisance variables
– Model comparison
• Groupwise model (enormous!)
• Combination with DARTEL (see later and new seg tbx)
• More tissue priors e.g. deep grey, meninges, etc.
• Imaging physics
– See Fischl et al. 2004, as cited in A&F introduction
Voxel-Based Morphometry
Segment
Normalise
VBM in pictures
Segment
Normalise
Modulate (?)
Smooth
VBM in pictures
Segment
a1xyz
a 2 xyz
Normalise Y X xyz exyz
Modulate (?)
aNxyz exyz ~ N (0, xyzV )
2
Smooth
Voxel-wise statistics 1 0
1 0
X
0 1
0 1
VBM in pictures
Segment
Normalise
Modulate (?)
Smooth
Voxel-wise statistics
VBM Subtleties
• Whether to modulate
• Adjusting for total GM or Intracranial Volume
• How much to smooth
• Limitations of linear correlation
• Statistical validity
Native
Modulation intensity =
tissue density
• Shape is really a
multivariate concept
– Dependencies among
volumes in different
regions
Above: (ii) is globally thicker, but
• SPM is mass univariate locally thinner than (i) – either of these
– Combining voxel-wise effects may be of interest to us.
information with “global”
integrated tissue volume Below: The two “cortices” on the right
provides a compromise both have equal volume…
– Using either ANCOVA or
proportional scaling
Figures from: Voxel-based morphometry
of the human brain… Mechelli, Price,
Friston and Ashburner. Current Medical
Imaging Reviews 1(2), 2005.
Total Intracranial Volume (TIV/ICV)
Deformation
Original Warped Template vector field
Determinant of
Jacobian Matrix
encodes voxel’s
Jacobian Matrix
Longitudinal VBM
Smoothed
CSF “modulated” by
Warped early Difference Relative volumes
relative volume
Nonrigid registration developments
Linear Average
(Not on Riemannian manifold)
DARTEL average
template evolution
Grey matter
average of 452
subjects – affine
Iterations
471 subjects
– DARTEL
Questioning Intersubject normalisation
• High-field high-resolution
MR may have potential to
image cytoarchitecture
• Will registration be better
or worse at higher
resolution?
– More information to use
– More severe
discrepancies?
– Need rougher
deformations
– Non-diffeomorphic?
4.7T FSE
Voxel i, class k p ( yi | ci k ) N ( k , k )
p ( yi , ci k ) k N ( k , k )
p ( yi ) k N ( k , k )
k
p (y ) k N ( yi | k , k )
Assumes independence
(but spatial priors later...)
i k
Could solve with EM
p ( y | μ, σ , γ ) (1-5)
Unified segmentation from the GMM upwards…
Spatially modify mean and variance with bias field
p (y ) k N ( yi | k , k )
i k
p (y ) k N yi k
, k
i k i ( ) i ( ) (10)
Unified segmentation from the GMM upwards…
Anatomical priors through mixing coefficients
p (y ) k N yi k
, k
i k i ( ) i ( )
bik k
k bik
bij j
Implementation prespecified:
j estimated: k
bik k
p (y ) N yi k
, k
i k bij j i ( ) i ( ) (12)
j
Unified segmentation from the GMM upwards…
Aside: MRF Priors (A&F, Gaser’s VBM5 toolbox)
bik k
k
bij j
j
bik exp km rmi rmi probable number of neighbours
j bij expm
K
jm rmi
(45)
Unified segmentation from the GMM upwards…
Spatially deformable priors (inverse of normalisation)
k bik
p(y ) N yi k
, k
i k j bij i ( ) i ( )
j
(13)
Unified segmentation from the GMM upwards…
Spatially deformable priors (inverse of normalisation)
k bik
p(y ) N yi k
, k
i k j bij i ( ) i ( )
j
bik bik ( )
k bik ( )
p (y ) N yi k
, k
i k j bij ( ) i ( ) i ( )
j
p ( y ) p ( y | α , β, μ , σ , γ )
(14, pretty much)
Unified segmentation from the GMM upwards…
Objective function so far…
k bik ( )
p (y ) N yi k
, k
p(y ) i p(ky
| α, βj b,ijμ(, σ) , γ) i ( ) i ( )
j
log p (y | α, β, μ, σ, γ )
log p ( yi | α, β, μ, σ, γ )
i
k bik ( )
log N yi k
, k
i k j bij ( ) i ( ) i ( )
j (14, I think...)
Unified segmentation from the GMM upwards…
Objective function with regularisation
p ( y ) p ( y | α , β, μ , σ , γ )
Assumes priors
p(y, α, β | μ, σ, γ ) p(y | α, β, μ, σ, γ ) p(α ) p(β) independent
p (α ) N (0, C )
log p (y | α, β, μ, σ, γ )
p (β) N (0, C )
F log p (y, α, β | μ, σ, γ )
C : α T C1α log p(α ) log p (β)
gives deformation’s
bending energy
(15,16)
Unified segmentation from the GMM upwards…
Optimisation approach
Maximising:
k bik ( )
F log N yi k
, k log p( ) log p ( )
i k j bij ( ) i ( ) i ( )
j