Automatic Segmentation of Brain Tissues

1
Automatic Segmentation of Brain Tissues

• Maria Murgasova
• Visual Information Processing
• Department of Computing
• Imperial College London

2
Outline

• Background
• Segmentation methods
• EM segmentation
• Results
• Future work

3
Motivation

• Brain development in prematurely born children
• Finding links neurological and behavioural
  changes ltgt brain shape and volume changes
• Precise automatic segmentation needed

4
Segmentation of Brain MRI

• Why is it not straightforward?

5
Immature brain segmentation

• 1 year old less contrast

6
Immature brain segmentation

• Neonatal brain even less contrast

7
Automatic segmentation methods

• Segmentation into WM, GM, CSF
• Segmentation of structures within the adult brain
  (thalamus, cortex )

8
Automatic segmentation methods

• Atlas-based
• Deformable atlas non-rigidly registered
  information transferred
• Intensity based
• EM, kNN, neural networks
• (non-)parametric intensity distribution?
• Probabilistic atlas prior information

9
Automatic segmentation methods
10
Automatic segmentation methods

• DL Collins, AP Zijbejdos, AC Evans. Improved
  automatic gross celebral structure segmentation.
  Poster. 4th International Conference on
  Functional Mapping of the Human Brain
• J Ashburner, K Friston, W Penny Human Brain
  Function, 2nd edition, http//www.fil.ion.ucl.ac.
  uk/spm/doc/books/hbf2/
• K. Van Leemput, F. Maes, D. Vandermeulen, P.
  Suetens. Automated model-based tissue
  classification of MR images of the brain. IEEE
  Trans. Med .Imag. (Special Issue on Model-Based
  Analysis of Medical Images), vol. 18, pp.
  897-908, Oct 1999
• C. A. Cocosco, A. P. Zijdenbos, A.C. Evans. A
  fully automatic and robust brain MRI tissue
  classification method. Medical Image Analysis 7
  (2003), pp. 513-527
• M. Prastawa, J. Gilmore, W. Lin, G. Gerig.
  Automatic segmentation of neonatal brain MRI
  In C. Barillot, D.R. Haynor, P. Hellier (Eds.)
  MICCAI 2004, LNCS 3216, Springer, Berlin pp.
  10-17
• P. Anbeek, A. Koeman, M. J. van Osch, K. L.
  Vincken, J. van der Grond Automated Brain
  Tissue Segmentation in Neonatal MR Imaging.
  Proc. Intl. Soc. Mag. Reson. Med. 13 (2005)

11
Intensity inhomogeneity correction

• Low frequency intensity change
• Need for correction methods

12
Expectation Maximisation

• Tissue intensities modeled by Gaussians
• Prior information probabilistic atlas alingned
  with data
• Iteratively calculate the Gaussians and
  segmentation using EM method

13
Gaussian mixture model
14
Probabilistic Atlas

• Template to align with data
• Probability for WM, GM, csf
• The brighter the intensity the higher the
  probability of tissue

15
Partial volume effect

• Blending of intensities of two tissues
• GM and CSF
• gt not Gaussian distribution

16
Results Adult brain

• Labels
• Good results

17
Results Adult Brain

• Probability maps
• WM, GM, csf
• Finer detail

18
1-year-old brain

• Differs from adult gt need for a child atlas
• We use 5-9 year old atlas
• No atlas available for younger children and
  neonates

19
1-year-old problem

• Deep GM classified as WM in thalamus

20
Neonatal brain T1

• Myelination
• Even less contrast between tissues

21
Neonatal brain histogram T1
22
Neonatal brain histogram T2
23
Neonatal brain T2

• Better contrast gt reasonable results
• Not possible to detect myelination

24
Future Work

• Segmentation algorithm for 1-year-olds
• Segmentation algorithm for neonates
• Construct an atlas of neonatal and young child
  brain
• Combine atlas-based and EM segmentation
• Combine registration and segmentation
• Experiment with other methods end theories (kNN,
  MRF, MDL)

25
Thank you for your attention

• Special thanks to my supervisor
• Dr. Daniel Rueckert