MNTP Summer Workshop DTI module

1
MNTP Summer Workshop DTI module

• Jordan Hamm (BA, BSc)
• University of Georgia, Athens,
  Georgia
• Alexandra Reichenbach (MSc, Dipl-Ing)
• Max Planck Institute for Biological Cybernetics,
  Tuebingen, Germany

2
Outline of Program

• Technical considerations
• Mean ADC values, FA, and tract volume as
  measurements
• Application
• Rationale of the project
• Approaches
• Automatized / manual
• Tract / ROI based
• Results
• Conclusions

3
Technical Considerations
What is b-value? -higher b-values may probe
different diffusion -more sensitive to
differences in restricted (Assaf, 2004)
Do more angles provide any benefit beyond more
SNR? -i.e. are more gradient directions just
redundant? -6 dir(8 times) or 50 directions (1
time)?
Is motion correction effective? - Leemans vector
table rotation
4
Technical Considerations
What effect does b-value, angular resolution, and
motion correction have on common diffusion
metrics? - Scanned 2 subjects - Compared
parameters in -tract reconstructions -5x5mm
ROIs
for maximum sensitivity
5
Tract-based analysis Anterior portion of corpus
callosum
Qualitative analyses
Raw
6 dir, b1200
50 dir, b1200
50 dir, b2400
Motion Corrected
-Tracts produced with FACT algorithm (BF
approach) using tensors in 6 direction data and
using non-negativity constrained spherical
de-convolution in 50 direction data.
6
Evaluation of b-value, ang. res., and motion
correction in tract reconstructions
First compared average FA of a tract to overall
tract volume
As volume of a tract increases, overall average
FA of that tract decreases - so tract integrity
is not necessarily revealed in a tract based
analysis. Instead, tract volume and/or number of
tracts are best used for tract based analyses
7
Effect of b-value on tractography
Assessed number of voxels involved in each
reconstructed tract from each scan.
Initially, b-value didnt appear to affect
tractability. But.
8
Effect of b-value on tractography
Motion correction (12 parameter) with vector
table rotation reveals benefit of higher b-values
(Leemans and Jones, 2009)
9
Benefit of motion correction
Motion correction appears to improve tracking,
but differentially for different b-values. Why?
- longer scans? more movement? - b2400 scan
10 longer (2 min) -higher b-values are more
sensitive -scan artifacts
10
ROI based analyses
Manual selection of 3x3 voxel ROI Compared
between b-values, ang. res., and raw/motion
corrected data -Mean diffusivity (verified with
known values) -FA estimate
11
Diffusion coefficient estimate
Mean diffusivity variable between b1200 and
b2400 before motion correction

• -Overall variance of ADC values reduced after
  motion correction
• -also closer to prescribed 7.0 X 10-4
  (Johansen-Berg and Behrmans, 2009)
• B2400 with motion correction is best
• ROI close to CSF, to which lower b-values are
  more sensitive.
• Again, differential effects of motion correction
  seen

12
Analysis of ROI FA values
Why does FA in a voxel cluster decrease with more
resolution, but tract volume increase?
-Higher b-values yield more consistent measure of
fractional anisotropy across subjects -Some
anisotropy captured by low b-values could be
non-axonal which does not contribute to long
range tractography -lower b-values have more
hindered and less restricted
13
DTI application Project on Congenital
Prosopagnosia (CP)

• Learning aims
• Learn different DTI analysis software and their
  strengths weaknesses
• Explore a real scientific question with different
  DTI approaches
• Get to know pitfalls and possible difficulties on
  real data

Haxby et al. (2000)
14
CP project Rationale

• Familiar vs. unknown faces elicit specificBOLD
  activation in healthy controls but notin CP
  patients in
• left precuneus/posterior cingulate cortex
• anterior paracingulate cortex
• Outside the core system for face processing
• HypothesisStructural changes in white matter
  tractsbetween these regions might underlie
  thefunctional differences
• Target tract Cingulum

Avidan Behrmann (2009)
15
CP project Approaches

• Measurements (for ROIs or tracts)
• Fractional anisotrophy (FA)
• Radial diffusivity (RD)
• Transverse diffusitivity (TD)
• Number of detected fibers ( fibers)
• Number of voxels within detected tract ( voxels)
• Approaches
• Automatic fiber seeding based on fMRI group
  coordinates
• Extraction of cingulum fibers based on anatomy
  (manual seeding)
• ROI analysis of sup. cingulum with automatic
  seeding based on standard space coordinates
• (probabilistic tracking from fMRI group
  coordinates, FSL)
• Data previously acquired from 17 controls 6
  patients
• TR/TE 4900/82ms 6 directions b 850 s/mm2
  1.61.63mm voxel size
• Is this angular resolution sufficient for these
  regions (fiber crossing!)?

16
CP project Results of automatic seeding based on
fMRI data

• Transformation of fMRI MNI coordinates in native
  space (FSL FLIRT)
• Construction of spheric ROIs around these
  coordinates (MATLAB)
• Extraction of tracts traversing both ROIs
  (ExploreDTI)
• Only about 1/3 of the subjects had tractable
  fibers
• Increasing the radius of the ROI did not solve
  the problem

background FA values
ROIs 18mm diameter
anterior paracingulate cortex
precuneus / posterior cingulate cortex
17
CP project Results of cingulum tracts based on
anatomy

• Analysis with DTI Studio, manual seeding by 2
  independent investigators
• Comparison of left right cingulum in healthy
  controls and DTI patients
• Results (whole tracts as ROI)
• Inter-rater reliability gt .8
• No group differences in corpus callosum (CC)
• control tract
• FA TD larger in left than in right cingulum
• consistent with literature
• Significant differences in fibers total ?

()
18
CP project ROI cingulum analysis

• Analysis with Explore DTI, MNI coord of ROI
  transformed in native space
• Results (only ROI voxels included)
• Larger FA value left than right in controls can
  be explained by a smaller RD ? fibers more
  directed
• TD left in CP patients smaller than in controls?
  fibers more directed in controls
• ? in line with fMRI data activation of left
  precuneus/PCC in controls but not in patients

19
CP project Discussion

• Automatic seeding based on fMRI data fails
• Possibly due to large inter-individual
  differences BUT no individual fMRI available
• Possibly due to insufficient tractability with 6
  direction data higher angular resolution data
  is acquired at the moment ? ExploreDTI can model
  multiple fibers in a voxel (CSD)
• Analysis data-driven, no operator bias
• Manual cingulum tracking
• High inter-rater reliability due to
  standardized method of ROI definition? DTI
  Studio easy-to-use user-friendly GUI, ideal
  for exploration and manual interventionBUT
  supports only tensor model
• Results in controls are consistent with
  literature
• Automatic seeded ROI analysis
• No manual intervention, no operator bias
• ? Besides ILF and IFOF the left cingulum is
  another tract involved in face processing that
  seems to be compromised in CP patients

20
Achnowledgements

• Seong-Gi Kim Bill Eddy
• Kwan-Jin Jung
• Marlene Behrmann
• John Migliozzi
• Tomika Cohen
• Rebecca Clark
• NIH