Analysis and visualization of brain connectivity using diffusion tensor MR imaging

1
Analysis and visualization of brain connectivity
using diffusion tensor MR imaging

• Supervisor Daniel Rueckert
• Members Caroline Baroukh,
• Rouslan Dimitrov,
• Przemyslaw Korzeniowski,
• Danial Sheikh

2
Introduction

• Multiple MRI scans provide 3D voxel grid
• Changing MRI polarizations yields slightly
  different results based on tissue orientation
• Diffusion tensor imaging (DTI) exploits this to
  assign anisotropy tensors to voxels

3
Introduction

• After processing,
• where
• e - major axis of anisotropy ellipsoid
• fa - fractional anisotropy 01
• Pv - voxel center
• Idea use ellipsoids to trace curves (connecting
  fibers) in the brain!

Pv
4
Outline

• Aim of the project
• Overview of the application
• Tracing algorithm
• GUI
• Demo
• Group organization

5
Aim of the project
6
Aim of the project

• Application that interactively traces and
  visualizes fibers
• Regions of interest (ROIs) used as starting point
  of fibers
• ROIs loaded from a segmentation file or defined
  manually
• Provide typical medical imaging visualization
  (eg. cutting planes, glyphs, etc)

7
(Technical) Overview of the application

• Application in JAVA
• Developed from scratch
• Various toolboxes interactive 3D display
• Tracer in C
• Extends VTK
• 2 new filters
• vtkFiberTracer and vtkInterpolatedDifTensors

8
Tracer

• Input
• DTI data (eigenvector and fractional anisotropy)
• Fiber Seed points
• Output
• Fibers as polylines (connected series of line
  segments).

9
Tracer

• for each seedpoint
• f new fiber polyline
• do
• move in direction of anisotropy to P
• f.AddPoint(P)
• until numberOfSteps exceeded
• store f

Use custom second order curvature-preserving
integrator
?
Seed
10
Tracer

• for each seedpoint
• f new fiber polyline
• do
• move in direction of anisotropy to P
• f.AddPoint(P)
• if(fractional anisotrpy lt fThresh)
• break
• until numberOfSteps exceeded
• store f

Empirically fThresh 0.2
11
Tracer

• MRI provides low resolution scans 1283
• Need to use steps much smaller than the voxel
  size
• Use bilinear interpolation from 8 closest voxel
  centers
• Danger 3000 fibers 100 steps 8 samples n,
• where n is the order of the integrator, can be
  high!

12
Tracer

• Problem 30 of the voxels contain at least two
  different neural tracts traveling in different
  directions
• Solution
• Add inertia to the fiber, so that low anisotropy
  regions cannot change its direction

13
vtkInterpolatedDifTensors

• Problem Which way to go?
• Anisotropy ellipsoids have no direction!

14
vtkInterpolatedDifTensors

• Solution
• Flip vectors on the fly
• This needs sense of direction
• Recover it from previous look-up
• Implemented in vtkInterpolatedDifTensors,
  inherits interface from vtkInterpolatedVectorField

15
Tracer

• Parameters exposed in the GUI from the Streamline
  Panel
• Tracing from anatomical ROIs is done by randomly
  scattering points within them
• Tracing from a user selected sphere distributes
  the seeds on the surface

16
GUI overview Demo
17
Questions?