DTI ModuleMNTP 2011
InstructorKwan-Jin Jung, Ph.D.
(Carnegie Mellon University)
Technical AssistantNidhi Kohli
(Carnegie Mellon University)
David Schaeffer (University of Georgia)
Lauren Libero (University of Alabama at Birmingham)
Sara Levens, Ph.D.(University of Pittsburgh)
LEARNING OBJECTIVES Effects of
Segmented sampling Motion correction Fiber orientation estimation method fMRI based ROIs vs. drawing ROIs
Anatomical separation of sensorimotor cortex
TERMINOLOGY Diffusion encoding gradient direction
Vector table (x, y, z components) Angular resolution
Diffusion-weighting (b-values) Duration & amplitude s/mm²
b0 = 0 s/mm² No diffusion gradient
METHOD OF ACQUISITION Segmented sampling
Complementary diffusion encoding directions
64 (A) - 10 min 64 (B) - 10 min 128 (A + B) - 20 min
Useful for special populations
MOTION CORRECTION How to correct:
1. Estimate the motion2. Rotate image and vector table accordingly
Intended Collected Head correction WRONG
Head & vector table correction
CORRECT
MOTION CORRECTION No correction
No vector rotation Interpolation
Estimates how much you rotate vector table
Based on distributed b0 images – “real motion”
Rota
tion
(deg
rees
)
Rota
tion
(deg
rees
)
TimeBEFORE
AFTER
6
3
0
-3
-6
Time
6
3
0
-3
-6
MOTION CORRECTION Simulation method
Collect two diffusion scans1. 6 direction scan (low b-value)
Why? – Fast (little time for motion) Edges of brain are clearly defined
2. 6 or more direction scan (higher b-value)
Assume no motion on scan 1, then simulate what higher b-value volume should look like
Low b-value (b=800
s/mm²) DWI(scan 1)
Assume no motion
Co-registervolumes
(estimating motion)
High b-value (b=2000
s/mm²) DWI(scan 2)
Find D (diffusion tensor)
S=S0e-
bD
Find S (simulated
high b-value)
S=S0e-
bD
Rotate vector table
FIBER ORIENTATION ESTIMATION METHODFiber/voxel Data Acquisition Analysis
Single fiber 6 – 12 directions Tensor
Multiple fibers > 25 directions (HARDI)
CSD (Q-ball, multi-tensor)
FIBER ORIENTATION ESTIMATION METHOD Tensor
Performs well for straight tracts (like motor) Performs poorly for crossing and branching
fibers (like Genu)
Constrained Spherical Deconvolution (CSD) Better for detecting branching and crossing
fibers (Tournier et al., 2007)
CSD VS. TENSOR
CSD Tensor0
20000
40000
60000
80000
100000
120000
Average Number of Tracts in Genu
N F
iber
s
GenuTensor
GenuCSD
DRAWING ROIS Manually draw ROIs Using fMRI
Collect fMRI data – find center of activation (x, y, z) Matrix transformation
Convert from fMRI coordinates into DWI native space
SEGMENTING SENSORIMOTOR Finger closing fMRI results as ROI Separation of sensory and motor areas
Clustering – fiber end-point distribution
Central Sulcus
SUMMARY Sampling schemes can be
advantageously altered for use with special populations
Simulation is a promising method for more accurate motion correction
CSD Fiber tracking is most appropriate for resolving fiber crossings
SUMMARY fMRI-based ROIs can be
used to track fibers from areas of activation
DTI can be used as a tool to segment brain areas that are not separable based on diffuse fMRI activation maps
ACKNOWLEDGMENTS Dr. Kwan-Jin Jung Nidhi Kohli MNTP Leaders: Dr. Eddy & Dr. Kim MTNP Trainees & Participants DTI Trainees 2009 & 2010 Funding:
NIH grants: R90DA023420 and T90DA022761
DISTRIBUTED B0
SCANNING PARAMETERS
MOTION CORRECTION
No correction
Interpolation
Simulation
Motion