1. Computational Neuroanatomy for Dummies
Lucy Lee &
Jeremie Pariente
(neither of whom have ever actually done any VBM, yet…)

2. Contents Part I: The Theoretical bit: Lucy
Part II: The Practical bit: Jeremie

3. Analysing Structural Images
What questions are we asking?
What information do we need?
How can we get this information from our data?

4. Analysing Structural Images
Rigid body registration
Normalisation
Segmentation
Modulation by normalisation warps
Statistical analysis

5. Pre-processing Structural Data
One structural image per subject
May have more than one time-point
May have more than one modality of image
Aims
Register images acquired at different times
Register images from more than one modality
Normalise structural images to standard anatomical space

6. Rigid Body Registration
Assumes no change in shape of images
between sessions
between subject and template
Within modality
[Realign]
Between Modality
[Coregister]
Use different optimisation approaches

7. Affine Transforms Shear Translations by tx and ty
x1 = x0 + tx
y1 = y0 + ty
Rotation around the origin
by  radians
x1 = cos() x0
+ sin() y0
y1 = -sin() x0
+ cos() y0 
x1 = x0 + h y0
y1 = y0
Shear
x1 = sx x0
y1 = sy y0
Zooms by sx and sy

8. Spatial Normalisation
Why
Standard coordinate system
Minimise the difference in shape between images
How
Initial Linear normalisation
Non-linear Warping
Problems
Computational
No one-one mapping between anatomical landmarks
Perfect matching would remove all spatial information!
Solutions
correct gross differences & smooth normalised images
Use information from normalisation in analysis

9. Spatial Normalisation - Non-linear
Deformations consist of a linear combination of smooth basis functions
Contain information about relative position of structures
Algorithm minimises mean squared difference between template and source image

10. Information from Normalisation
Original
Warped
Template
Relative volumes
from Jacobian determinants of deformation fields

11. Segmentation Dividing MR images into different tissue classes Why?
Grey & white matter
CSF
Why?
Compare size / shape brain areas
Within subjects over time
Between subjects e.g. patients vs controls
Do proportions of different tissue types change?

12. Segmentation - Mixture Model
Intensities of T1 signal in image is modelled by a mixture of Gaussian distributions
Mean intensity
variances
mixing proportions

13. Segmentation - Priors Prior probability maps help with segmentation
Requires initial registration to standard space
Assumed to be representative of population
Prior probability
of a voxel belonging
to a particular tissue
type is derived from
segmented images
of 151 subjects

14. Segmentation - Algorithm
Compute belonging probability of voxel to cluster type:
given the priors, cluster parameters & sensitivity field
Segmented
DATA

15. Morphometry Identify & Characterize structural differences
Between populations
Within subjects over time
Correlations between brain shape & variables
Disease severity
Age
Statistical Inference
Global differences
Regionally specific differences (voxel by voxel)

16. Morphometry Deformation Based Morphometry
Macroscopic anatomic differences between subjects
Differences in relative positions of structures
Requires very exact initial co-registration
Sensitive to differences in brain volume
Multivariate statistics on the deformation fields

17. Morphometry Tensor Based Morphometry Regional structural differences
Jacobian or Gradient of deformation fields provides information about local shapes
Volume, Length & Area
Simple TBM
Comparing differences in relative volumes of structures
Remove information about position
‘Strain Tensor’ maps voxels in one brain to template
Perform statistics on the different tensor elements

18. Morphometry Voxel Based Morphometry
Voxel wise comparison in local concentration of grey (or white) matter between different subjects
Local composition of the brain tissue
Analysis performed after removing information about position & shape differences
Jeremie …