Segmentation of Brain MRI in Young Children

1
Segmentation of Brain MRI in Young Children

• Maria Murgasova1, Leigh Dyet2,
• David Edwards2, Mary Rutherford2, Jo Hajnal2 and
  Daniel Rueckert1
• 1Department of Computing
• 2Department of Imaging Sciences
• Imperial College London

2
Motivation

• The effect of premature birth
• impaired brain development
• neurological, behavioural, learning difficulties
• To understand and treat the changes we need to
  measure
• volumes of different brain structures
• growth of different brain structures
• This requires
• segmentation of anatomical
• structures at different time points

3
Registration-based segmentation

• Non-rigid registration of atlas to subject
• Segmentation is warped from atlas to subject
• Advantage
• Does not assume any tissue intensity model
• gt successful in central brain structures
• Disadvantage
• Does not deal well with complex cortical folding

4
EM-based segmentation

• Classical model for brain MRI
• 3 basic tissue classes (WM, GM, CSF)
• Tissue intensity distributions approximately
  Gaussian
• Advantage
• Can capture complexity of cortex based on
  intensity
• Disadvantage
• Cant deal well with overlaps in tissue intensity
  distributions

Real tissue distributions of 2-year-old subject
based on manual segmentation
5
EM-based segmentation

• Sub cortical structures brighter than cortical
• Overlaps cause significant difficulties
• Classical WM, GM, CSF model not sufficient for
  correct segmentation

The histograms were normalised to unit height
6
EM-based segmentation

• Probabilistic atlas
• Aligned with the image
• Spatially constrains the segmentation process
• Helps to overcome misclassification due to
  overlaps in tissue intensity distributions

7
Application of EM to young children

• During early childhood the shape of central brain
  structures differs significantly from those
  during adulthood
• WM overestimated in sub-cortical area
• Requires a specific atlas for young children

Segmentation of 1-year-old
Adult
Adult atlas
1-year-old
8
Creating a population-specific atlas
Manual segmentation
Reference subject
Average subjects
Non-rigid registration
Population specific atlas
Affine registration
New subject
Registration-based segmentation
9
Atlas comparison

• Atlas for
• 2-year-olds
• (from
• 37 subjects)
• Adult atlas

10
Segmentation results

• Improvement in thalamus
• 1.0T brain MRI of a 2-year-old child

Image
Manual segmentation
EM with adult atlas
EM with new atlas
11
Segmentation results

• Improvement overall
• 1.0T brain MRI of a 2-year-old child

Image
Adult atlas
Our atlas
12
Validation

• 4 subjects
• manual segmentation of 6-8 slices
• manual segmentation of thalamus
• Measuring the agreement between manual and
  automatic segmentation using Dice metric

13
Atlases at different time-points

• Atlas can be generated dynamically for different
  populations

Age 1
Age 2
14
Current work

• WM, GM, CSF model does not sufficiently describe
  the properties of the brain tissues
• To obtain more precise segmentation we need to
  include more brain structures in the model
• To do
• Identify which structures should be in the atlas
  depending on their intensities
• Obtain at least one good manual segmentation of
  those structures
• Create a probabilistic atlas for all those
  structures

15
Identifying the structures

• Segmentation into 83 structures
• Transferred from adults to 2-year-olds by Ioannis

16
Identifying the structures

• New histogram tool in rview
• Visualise the intensity distributions of 83
  structures
• We identified partition of the brain into 10
  structures based on intensity

17
Combining the segmentations

• Available segmentations

Ioannis 83 structures
Merged into 10 structures
Leigh WM, GM, CSF
18
Combining the segmentations

• Ioannis 83 structures, no partition into WM and
  GM
• Leigh WM, GM, CSF cerebellum, caudate,
  thalami, corpus callosum and lateral ventricles
• 10 structures we need CSF, cortical GM, caudate,
  putamen, substantial nigra, cerebellum, thalamus,
  pallidum, brainstem, WM

19
Combining the segmentations

• PROBLEM!!!
• Which segmentation is correct?
• As I will show later the errors in manual
  segmentation are transferred to automatic
  segmentation as well

Leigh
Ioannis
20
Combining the segmentations

• Combined manual segmentation into 10 structures
• Contains some significant errors
• We will need to improve it later
• BETTER MANUAL SEGMENTATION NEEDED!!!

21
Probabilistic atlas
WM
Cortical GM
CSF
Caudate
Putamen
Subst. nigra
Cerebellum
Brainstem
Pallidum
Thalamus
22
EM segmentation

• Results of EM segmentation using probabilistic
  atlas for 10 structures
• Errors in atlas transferred to the EM results
• Otherwise LOOKS VERY PROMISING!!!

23
Manual segmentation

• New segmentation tool in rview

24
Manual segmentation

• New segmentation tool in rview

25
Manual segmentation

• New segmentation tool in rview

26
Future work

• Finish the development of the new tools in rview
• Good manual segmentation into 10 structures
• A more sofisticated method for WM-GM segmentation
  in cortical area because
• Central WM is much brighter than WM in cortex
• Small parts of WM close to brain boundary are
  left out
• Intensity varies continuouslygtpartition into
  more parts not possible

27
Acknowledgements

• This work is funded by