A longitudinal study of brain development in autism

1
A longitudinal study of brain development in
autism

• Heather Cody Hazlett, PhD
• Neurodevelopmental Disorders Research Center
• UNC-CH Dept of Psychiatry
• NA-MIC AHM Salt Lake City, UT Jan 8, 2009

2

• UNC DBP-2 Team
• DBP-2
• Co-PI Heather Cody Hazlett, PhD
• Co-PI Joseph Piven, MD
• CS Programmers Clement Vachet MS, Cedric
  Matthieu BA
• Core 1 Martin Styner, UNC Chapel Hill
• UNC Algorithm Ipek Oguz, Nicolas Augier, Marc
  Niethammer
• Utah Algorithm Marcel Prastawa
• Core 2 Jim Miller, GE Research

3
Project Cortical thickness analysis
of pediatric brain

• Project Goals
• Individual and group analysis of regional and
  local cortical thickness
• Creation of an end-to-end application within
  Slicer3
• Workflow applied to our large pediatric dataset
• Why is this needed?
• - Existing tools (e.g. FreeSurfer) are tailored
  to work with adult brain
• - Pediatric brain shows more variability in
  brain shape and maturation (esp. white matter)
  than adult brain

4
Regional cortical thickness
5
Regional Cortical Thickness - Pipeline Overview
A Slicer3 high-level module for individual
cortical thickness analysis has been developed
ARCTIC (Automatic Regional Cortical ThICkness)
Input raw data (T1-weighted, T2-weighted,
PD-weighted images) Three steps in the
pipeline 1. Tissue segmentation 2. Regional
atlas deformable registration 3. Cortical
Thickness
6

• Regional cortical thickness
• (ARCTIC) pipeline
• Step 1 Tissue segmentation
• Probabilistic atlas-based automatic tissue
  segmentation via an Expectation-Maximization
  scheme
• Tool itkEMS (UNC Slicer3 external module)

7

• Regional cortical thickness (ARCTIC) pipeline
• Step 2. Regional atlas deformable registration
• 2.1 Skull stripping using previously computed
  tissue
• segmentation label image
• Tool SegPostProcess (UNC Slicer3 external
  module)
• 2.2 T1-weighted atlas deformable registration
  using a B-spline pipeline registration
• Tool RegisterImages (Slicer3 module)
• 2.3 Applying transformation to the parcellation
  map
• Tool ResampleVolume2 (Slicer3 module)

8

• Regional cortical thickness (ARCTIC) pipeline
• Step 3. Cortical Thickness
• Sparse asymmetric local cortical thickness
• Tool CortThick (UNC Slicer3 module)
• Note All the tools used in the current pipeline
  are Slicer3 modules, some of them being UNC
  external modules.
• The user can thus compute an individual regional
  cortical thickness analysis by running the
  'RegionalCortThickPipeline' module, either
  within Slicer3 or as a command line.

9
ARCTIC Pipeline Validation Analysis on a small
pediatric dataset Initial tests have been
computed on a small pediatric dataset which
includes 2 year-old and 4 year-old cases. N 16
with Autism, 1 with Dev Delay, 3 Typ
Developing Comparison to state of the art
ARCTIC vs. Freesurfer We are currently doing a
regional statistical analysis using Pearson's
correlation coefficient on a dataset that
includes 90 cases and for two comparison groups
(2 yr-old cases and 4 yr-old cases)
10

• Project Workload Timeline
• Completed
• Workflow for individual analysis (Slicer3
  external module using BatchMake)
• 2 Tutorials "How to use the UNC modules to
  compute the regional cortical thickness" and "How
  to use ARCTIC"
• In progress
• Pediatric atlases available to the community
  through MIDAS
• Comparison to FreeSurfer pearson correlation
  analysis
• ARCTIC available to the community through NITRC
  executables (UNC external modules for Slicer3),
  source code (SVN), and Tutorial dataset
• Future work
• Workflow for group analysis (KWWidgets
  application using BatchMake)

11
Downloads Executable and tutorial dataset
http//www.nitrc.org/projects/arctic/ Pediatric
atlas http//www.insight-journal.org/midas/item/v
iew/2277
12
Local cortical thickness
13
Local Cortical Thickness - Pipeline Overview
Input Raw T1-weighted, T2-weighted, or
PD-weighted images Eleven steps in the pipeline
7. White matter surface inflation 8. Cortical
correspondence 9. Label map creation 10.
Cortical thickness 11. Group statistical
analysis
1. Tissue segmentation 2. Atlas-based ROI
segmentation 3. White matter map creation 4.
White matter map post-processing 5. Genus zero
white matter map image surface
creation 6. Gray matter map creation
14

• Local cortical thickness pipeline
• Step 1 Tissue segmentation
• Probabilistic atlas-based automatic tissue
  segmentation via an Expectation-Maximization
  scheme
• Tool itkEMS (UNC Slicer3 external module)

15

• Local cortical thickness pipeline
• Step 2 Atlas-based ROI segmentation
  subcortical structures, lateral ventricles,
  parcellation
• 2.1 T1-weighted atlas deformable registration
• B-spline pipeline registration
• Tool RegisterImages (Slicer3 module)
• 2.2 Applying transformations to the structures
• Tool ResampleVolume2 (Slicer3 module)

16

• Local cortical thickness pipeline
• Step 3 White matter map creation
• Brainstem and cerebellum extraction
• Adding subcortical structures (except amygdala
  hippocampus)
• Tool ImageMath (NITRC module)

17

• Local cortical thickness pipeline
• Step 4 White matter map post-processing
• Largest component computation
• White matter filling
• Smoothing Level set smoothing or weighted
  average filter
• Connectivity enforcement (6-connectivity)
• Tool SegPostProcessB (Slicer3 external module)

18

• Local cortical thickness pipeline
• Step 5 Genus zero white matter map image and
  surface creation
• Tool GenusZeroImageFilter
  (UNC Slicer3 external module)
• Step 6 Gray matter map creation
• Adding genus zero white matter map to gray matter
  segmentation (without cerebellum and brainstem)
• Tool ImageMath

19

• Local cortical thickness pipeline
• Step 7 White matter surface inflation
• Iterative smoothing using relaxation operator
  (considering average vertex) and L2 norm of the
  mean curvature as a stopping criterion
• Fixing is necessary remove vertices that have
  too high curvature (extremities)
• Tool MeshInflation (UNC Slicer3 external
  module)

20

• Local cortical thickness pipeline
• Step 8 Cortical correspondence
• Correspondence on inflated surface using
  particle system
• Tool ParticleCorrespondence (UNC Slicer3
  external module)
• Step 9 Label map creation
• Label map creation for cortical thickness
  computation (WM GM "CSF")
• Tool ImageMath

21

• Local cortical thickness pipeline
• Step 10 Cortical thickness
• Asymmetric local cortical thickness or Laplacian
  cortical thickness
• Tool UNCCortThick or measureThicknessFilter
  (UNC Slicer3 external modules)
• Step 11 Group statistical analysis
• Tool QDEC Slicer module or StatNonParamPDM

22
Pipeline validation Analysis on a small
pediatric dataset (to be done) Tests will be
computed on a small pediatric dataset which
includes 2 year-old and 4 year-old cases. N 16
with Autism, 1 with Dev Delay, 3 Typ
Developing Comparison to state of the art
(ongoing) Pipeline vs. Freesurfer We are
currently doing a regional statistical analysis
using Pearson's correlation coefficient on a
dataset that includes 90 cases and for two
comparison groups (2 yr-old cases and 4 yr-old
cases)
23

• Project Workload Timeline
• In progress
• Cortical surface inflation module in progress
• Mesh needs to be fixed at some location to have a
  correct inflation
• Future work
• Workflow for individual analysis as a Slicer3
  high-level module using BatchMake
• Workflow for group analysis

24
Contributors

• Joe Piven, MD
• Guido Gerig, PhD
• Martin Styner, PhD
• Clement Vachet, MS
• Cedric Matthieu, BA
• Rachel Smith, BA
• Mike Graves, MChE
• Sarah Peterson, BA
• Matt Mosconi, PhD

NA-MIC Team Jim Miller Ipek Oguz Nicolas
Augier Marc Niethammer Brad Davis
Parent grant funded by the National Institutes of
Health