Framework for segmentation of MRI brain image

1
Framework for segmentation of MRI brain image

• NeuroImageing
• Instructor Dr. Klemer
• Presenter Mohammad Habibi

2
Automated brain tissue segmentation

• Classify tissue in the brain (GM,WM,CSF)
• Accurate segmentation of MRI image to different
  tissue type (GM,WM, CSF) and Regional volume
  calculation ? useful diagnostics information
• Quantization of GM/WM volume ? neurodegenerative
  disorders
• White matter metabolic or inflammatory disease
• Congenital brain malformation
• Prenatal brain damage
• Post traumatic syndrome

3
(No Transcript)
4
The segmentation framework

• Framework

Original Image
Skull Strip
Non-uniformity Correct
Extract Brain Surfaces
Classify Brain Tissues
5

• Segmentation framework

Spatially Normalised MRI
Original MRI
Template
Affine register
Spatial Normalisation - writing
Grey Matter
Segment
Affine Transform
Spatial Normalisation - estimation
Priors
Deformation
6
(No Transcript)
7
Very hard to define a one-to-one mappingof
cortical foldingUse only approximate
registration.
8
Removal of non-brain tissue

• Manual
• Automated

9
Bias correction

• MRI images are corrupted by inhemogeneity in
  magnetic field (bias).
• Image intensity non-uniformity artefact has a
  negative impact on segmentation
• Much better if this artefact is corrected
• We can formulate the effect of inhomogeneity as

10

Estimation of the Tissue type
Estimation of the inhomogeneity
11
Gaussian Probability Density

• If intensities are assumed to be Gaussian of mean
  mk and variance s2k, then the probability of a
  value yi is

12
Tissue classification

• Some definitions
• Prior probability of pixel ij belong
  to cluster k
• Probability of pixel ij belong to
  cluster k (objective)
• Number of pixel in
  cluster k, mean and variance of cluster
• pixel value in image
• Correction function
• Example

13
Algorithm

• First step

• Second step

14
Tissue Classification - Algorithm
Starting estimates for belonging probabilities
based on prior probability images
Compute cluster parameters from belonging
probabilities and bias field
Compute belonging probabilities from cluster
parameters and bias field
Converged? No Yes
Compute bias field from belonging probabilities
and cluster parameters
Done
15
Non-Gaussian Intensity Distributions

• Multiple Gaussians per tissue class allow
  non-Gaussian intensity distributions to be
  modelled.

16
Tissue Classification - Mixture Model

• Intensities are modelled by a mixture of K
  gaussian distributions, parameterised by
• Means
• Variances
• Mixing proportions

17
Modelling a Bias Field

• A bias field is included, such that the required
  scaling at voxel i, parameterised by b, is ri(b).

y r(b)
r(b)
y
18
The Extended MOG

• By combining the modified P(kq) and P(yik,q),
  the objective function becomes
• Bias and nonlinear deformations are parameterised
  by a linear combination of cosine transform
  bases, where a and b are the estimated
  coefficients.

19
Schematic of optimisation

• Repeat until convergence
• Hold a and b constant, and minimise E w.r.t. g, m
  and s2
• - Differentiate E w.r.t. each parameter, and
  solve.
• - Requires substitution of current belonging
  probabilities at each iteration.
• Hold g, m, s2 and a constant, and minimise E
  w.r.t. b
• - Levenberg-Marquardt strategy, using dE/db and
  d2E/db2
• Hold g, m, s2 and b constant, and minimise E
  w.r.t. a
• - Levenberg-Marquardt strategy, using dE/da and
  d2E/da2
• end

20
Spatially normalised BrainWeb phantoms (T1, T2
and PD)
Tissue probability maps of GM and WM
21
Segmentation of Neonatal Brain MRI

• Segmentation of brain tissues of newborn infants
  from multimodal MRI
• Particular interest in the developing white
  matter structure
• Motivation
• Analysis of growth patterns, study of neuro
  -developmental disorders starting at a very early
  age

22
Challenges

• Smaller head size
• Low contrast-to-noise ratio
• Intensity inhomogeneity
• Motion artifacts
• Division of white matter into
• myelinated and non-myelinated
• regions

T1
T2
Labels
csf
nonm. WM
gm
myel. WM
23
Challenges
Neonate 2 weeks
Adult
CNR 2.9
CNR 6.9
24
Imaging the Developing Brain
age
35 weeks
adult
44 weeks
15 months
2 years
6yrs
25
Approach

• Non-optimal input data, rely on high level prior
  knowledge
• Intensity ordering (e.g. in T2W)
• wm-myelinated lt gm lt wm-non-myelinated lt csf
• Aligned spatial priors (brain atlas)
• White matter is considered as one entity

26
Method Overview
Segmentation - Bias correction
Initialization
Refinement
27
Conclusion

• Segmentation and regional volume calculation of
  the brain give us useful diagnostic information
• Classification is based on a Mixture of Gaussians
  model (MOG), which represents the intensity
  probability density by a number of Gaussian
  distributions.
• Special cases (newborn infant)

28
Prior probability image

• Provided by Montreal Neurological Institute
• Project ICBM,NIH P-20
• Derived from accurate scans of 152 young healthy
  subjects
• Value zero to one
• Website http//www.loni.ucla.edu/

29
References

• Pan Lin Yong Yang Chong-Xun Zheng Jian-Wen
  GuAn efficient automatic framework for
  segmentation of MRI brain image Computer and
  Information Technology, 2004. CIT '04. The Fourth
  International Conference on14-16 Sept. 2004
• Suri, J.S. Two-dimensional fast magnetic
  resonance brain segmentation Engineering in
  Medicine and Biology Magazine, IEEEVolume 20, 
  Issue 4,  July-Aug. 2001 Page(s)84 - 95
• Niessen, W.J. Vincken, K.L. Weickert, J.
  Viergever, M.A.Three-dimensional MR brain
  segmentationComputer Vision, 1998. Sixth
  International Conference on4-7 Jan. 1998
  Page(s)53 - 58
• http//www.cma.mgh.harvard.edu/seg/typical_data.ht
  ml
• http//www.fmrib.ox.ac.uk/analysis/research/fast/