Noise Removal Techniques for Diffusion Tensor Imaging

1
Noise Removal Techniques for Diffusion Tensor
Imaging
Bin Chen, Edward Hsu
Departments of Biomedical Engineering Duke
University
Center for In Vivo MicroscopyDuke University
Medical Center
2
Diffusion in MRI
Conventional transport
3
Diffusion in Anisotropic Tissues
examples brain white matter myocardium
cartilage
one tensor model
S So exp ( - b gT?D ? g)
4
Diffusion Tensor Imaging
1 acquire diffusion images in at least 6
non-coplanar directions 2 solve linear system
of equations for the diffusion tensor 3
diagonalize tensor into eigenvalues and
eigenvectors
5
DTI In Action
eigenvalue
conventional T2
fractional anisotropy
6
Challenges in MR-DTI
diffusion encoding via signal attenuation
large dataset size -- minimum 7 images
7
Quest for Higher SNR
Stained
Unstained
1. Active Stain
signal enhancement via T1 reduction
Challenge Noise removal without spatial blurring
8
Noise Removal in K-Space
Image space
K-Space
IFT
Fermi
X
Wiener
IFT
X
9
Image Based PDE Filtering
Nonlinear PDE Filtering
10
Data Acquisition
1 un-weighted 256 x 128 x 128 scan 2T
instrument, FOV 10 x 10 x 8 cm, TR 250 ms,
TE 27 ms, 2 averages 12 diffusion-weighted 256 x
64 x 64 scans encoded in 12 directions
b factor 900 s/mm2 6 mouse brains
11
Analysis
Tensor Construction
b0
DWI x 12
Dxx? ?xx Dxy ? ?xy Dxz ? ?xz Dxy ? ?xy Dyy ?
?yy Dyz ? ?yz Dxz ? ?xz Dyz ? ?yz Dzz ? ?zz
noise level quantification
12
Image Noise Removal Result
Original Fermi Wiener
PDE
b0 image
13
PDE filtering on tensor eigenvector?
correct modeling errors Direct benefit streamline
tracking
Potential Advantages
Challenges
Image based PDE filtering can not be implemented
on vector field directly. Constrains are needed
to conserve the eigenvector norm
14
Vector PDE Filtering
F
Analog keep the speed of a car constant
15
Vector PDE Filtering Result
Coherence Angle (RMS) ---------------------------
----------- mean ? SEM Original 0.7594 ?
0.0829 Fermi 0.7580 ? 0.0891 Wiener 0.6364 ?
0.0793 PDE(img) 0.4922 ? 0.0638
PDE(vec) 0.3111 ? 0.0298 -------------------------
--------------- ANOVA on 6 dataset
16
DTI Scalar Anisotropy Index
Contrast for tissue microstructure
conventional T2
fractional anisotropy
1.0 0.5 0.0