Diffusion MRI

• Diffusion MRI: a marker of tissue microstructure

• Sensitizing the MRI signal to diffusion
• Measuring diffusion rates with MRI
• Diffusion tensor imaging
• Tractography
• Acquisition Considerations.
• Bonus Material: Less conventional diffusion MRI
methods.

1 [Link] - RAD 229

 Diffusion MRI in the Brain
Clinical Applications Neuroscience
Stroke White Matter Pathways

Diffusion

Perfusion
Structural Connectivity
Zaharachuk G, et. al. 2012.

Intracranial Infections
Brain Tumours
Trauma
Edema

2 [Link] - RAD 229

 Diffusion MRI in the Body
Breast Prostate

Park M-J et. al. 2007 Bonekamp S, et. al. 2012

Liver Musculoskeletal

Bonekamp S, et. al. 2012 Staroswiecki E, et. al. 2012

3 [Link] - RAD 229
 What is diffusion?

• The result of random collisions between molecules in

liquids and gases.

• A form of passive transport that causes mixing but no

bulk motion.

• A spontaneous, random process.

[Link] - RAD 229

 Water Diffusion in Tissue

Types of Obstacles:
• membranes
• macromolecules

Effect on Water Diffusion:

• decreased diffusion rates
• limited displacement Darwin, M. et. al. 1995.
distances (restrictions)
• distinct geometric patterns of
water displacements

5 [Link] - RAD 229

 Hindered Diffusion
• molecules bump into obstacles (e.g. cell membranes)
• decreased diffusion rate

More Obstacles Fewer Obstacles

Slower diffusion. Faster diffusion.

Shorter displacements. Longer displacements.
[Link] - RAD 229
 Restricted Diffusion

Impermeable boundaries limit the displacement distance.

Time

[Link] - RAD 229

 Diffusion Anisotropy
Isotropic Diffusion Anisotropic Diffusion

Glass of Water White Matter Tissue

x100

[Link]
[Link] - RAD 229
Diffusion MRI: a marker of tissue microstructure
Patterns of water diffusion in tissue reflect the tissue
microstructure.

• membranes
• permeability of membranes
• macromolecules
• packing density
• compartment sizes

Darwin, M. et. al. 1995.

Sensitizing the MRI signal to water diffusion is a way to

indirectly get information about tissue microstructure.
9 [Link] - RAD 229
Sensitizing the MRI Signal to Diffusion
Hahn 1950.
Carr and Purcell 1954.

Case #1: Without Diffusion

Diffusion-Weighted Spin Echo

Echo
G x encode G x decode

90° 180°
Δ

x Stejskal-Tanner, 1965.
[Link] - RAD 229
Sensitizing the MRI Signal to Diffusion

Case #1: Without Diffusion

Echo
G x encode G x decode

90° 180°

x Stejskal-Tanner, 1965.
Low Gx High Gx
[Link] - RAD 229
Sensitizing the MRI Signal to Diffusion

Case #1: Without Diffusion

Echo
G x encode G x decode

90° 180°

x Stejskal-Tanner, 1965.
Low Gx High Gx
[Link] - RAD 229
Sensitizing the MRI Signal to Diffusion

Case #1: Without Diffusion

Echo
G x encode G x decode

90° 180°

x Stejskal-Tanner, 1965.
[Link] - RAD 229
Sensitizing the MRI Signal to Diffusion

Case #1: Without Diffusion

Echo
G x encode G x decode

90° 180°

x Stejskal-Tanner, 1965.
[Link] - RAD 229
Sensitizing the MRI Signal to Diffusion

Case #1: Without Diffusion

Echo
G x encode G x decode

90° 180°

x
Low Gx High Gx
[Link] - RAD 229
Sensitizing the MRI Signal to Diffusion

Case #1: With Diffusion

Echo
G x encode G x decode

90° 180°
spins change position

x
Low Gx High Gx
[Link] - RAD 229
 The Bloch Equation
M(t) = Mx(t) + My(t) + Mz(t)
where Mx(t) = Mxo*exp(-t/T2) ;
My(t) = Myo*exp(-t/T2) ;
Mz = Mzo(1-exp(-t/T1))
dMx(t)/dt = -Mx(t)/T2 self comment

torque due to B0

relaxation

[Link] - RAD 229

 The Bloch-Torrey Equation
Torrey H.C. Physical Review 1956.

In the spatial dimension the location where

M(t) has reached peak, the 'Laplacian of M(t)'
will be -ve in those points. This indicates M(t)
will decrease at those points with time. If M(t) is
minimum at some point then the laplacian of
M(t) is +ve which menas M(t) will increase with
time. self comment

Diffusion equation/Heat Equation:

[Link] - RAD 229

 The Bloch-Torrey Equation
Torrey H.C. Physical Review 1956.

with
[Link] - RAD 229
 b-value

Echo
G G

90° 180°

[Link] - RAD 229

 Measuring the Diffusion Coefficient

b-value:

Non-Diffusion-Weighted Signal:

Diffusion-Weighted Signal:

Diffusion Coefficient:
[Link] - RAD 229
 Apparent Diffusion Coefficient

• For Diffusion MRI, the diffusion coefficient is referred

to as the “Apparent” Diffusion Coefficient (ADC).
Why?

[Link] - RAD 229

 Diffusion in Tissue
slower
faster diffusion
diffusion

• In tissue the self-diffusion of water molecules is hindered

and restricted by membranes and macromolecules.

• In brain tissue, water can diffusion more freely along

white matter fibers than across them.
[Link] - RAD 229
 Diffusion Anisotropy
Isotropic Diffusion Anisotropic Diffusion

Glass of Water White Matter Tissue

x100

[Link]
[Link] - RAD 229
Diffusion Contrast in the Brain

Gx Gy Gz

[Link] - RAD 229

Measuring Anisotropy: Diffusion Tensor Imaging

diffusion tensor

• Eigenvectors = axes of ellipsoid (e1 orientation of fibers)

• Eigenvalues = length of axes (rate of diffusion)

Basser et. al. 1994. Pierpaoli and Basser 1996.

[Link] - RAD 229
Measuring Anisotropy: Diffusion Tensor Imaging

diffusion tensor

• Eigenvectors = axes of ellipsoid (e1 orientation of fibers)

• Eigenvalues = length of axes (rate of diffusion)

Basser et. al. 1994. Pierpaoli and Basser 1996.

[Link] - RAD 229
Measuring Anisotropy: Diffusion Tensor Imaging

Directionally
Encoded Line Representation of
Colour Map Principal Eigenvector

Basser et. al. 1994. Pierpaoli and Basser 1996.

[Link] - RAD 229
Measuring Anisotropy: Diffusion Tensor Imaging

Fractional Anisotropy:

Mean Diffusivity:

[Link] - RAD 229

 Diffusion Tractography

• Principal diffusion direction relates to direction of fiber

orientations and can be used for fiber tracking
(tractography).
[Link] - RAD 229
 Diffusion Tractography
• Piece together discrete (voxel-wise) estimates of the underlying
continuous fibre orientation field to infer on the 3-D trajectories
of anisotropic tissue structures.
• Starting (“seed”) voxel is selected in an anatomical region of
interest and voxel-wise estimates of the principal diffusion
orientation are followed from one voxel to the next.

[Link] - RAD 229

Two Classes of Tractography Algorithms
Deterministic Probabilistic

uncertainty

(θ0, ϕ0) θ pdf ϕ pdf

θ0 ϕ0
Behrens et al., MRM 2003; 50:1077–1088

[Link] - RAD 229

Two Classes of Tractography Algorithms

Deterministic Probabilistic

Iterative Monte Carlo approach

[Link] - RAD 229

Two Classes of Tractography Algorithms

Deterministic Probabilistic

Streamlines Multiple “sample” streamlines

[Link] - RAD 229

 Tractography Stopping Criteria
When to stop?
Heuristics to avoid error propagation.
+ Knowledge of the anatomy.

Curvature Threshold: To avoid crossing boundaries and very bended

trajectories, impose a smoothness criterion.

Anisotropy Threshold: To avoid propagating in regions where the

vector field is meaningless.

Anatomical Criteria: e.g. reach grey matter.

[Link] - RAD 229

 Interpretation of Tractography
What does
Quantitative Measures
probabilistic
of Connectivity?
tractography measure?
• Number of axons connecting 2 areas? • The probability that the
dominant path through the
• Proportion of axons from a seed that diffusion field passes
reach a target? through this region.

• “Integrity” of the connecting white

matter…
• degree of myelination
• packing density
• fiber size or shape

[Link] - RAD 229

Probabilistic Tractography Measures

• They may reflect anatomical connectivity strength

• But they also reflect ….

• Connection Length: Longer connections have smaller probability

than shorter ones.

• Geometric complexity: Probabilities of connections that go

through regions of complex structure will be smaller than
connections that go through more coherent regions.

• Resolution of the spatial grid: Probabilities change if we change

the size of “bins” for displaying the spatial histogram.

[Link] - RAD 229

 Diffusion Tractography
Robust mapping of
large white matter
fiber bundles.

In Vivo Diffusion MRI

Resolution: ~2 mm

WM Bundles: 2-6 mm
(106-107 axons)

Axon Diameter ~ 2-10 µm

Schuz A and Miller R. Cortical

Lebel C. et. al. ISMRM 2007.
Areas: Unity and Diversity, 2002.
[Link] - RAD 229
 Intra-Voxel Crossing Fibers

Diffusion tensor fails to model crossing fibers.

[Link] - NeuroRad Fellows

 Diffusion Spectrum Imaging

• Diffusion spectrum imaging (DSI) maps crossing fibers by

estimating orientation distribution functions (ODFs).
Wedeen et al. 2005.
[Link] - RAD 229
 q-space Imaging
Diffusion-weighted Spin Echo
gradient
gradient duration
strength

  diffusion time interval

 

• Spin displacement probability density function:

• Gradient wave vector: Callaghan 1991.
[Link] - RAD 229
 q-space Imaging
sample q-space

DWIs

3D radial
qz FFT projection
y y

qx x
qy x
spin displacement orientation distribution
q space probability density function (ODF)
function (PDF)
Wedeen et al. 2005.
[Link] - RAD 229
 q-ball Imaging
Funk-Radon Transform

• Approximates the dODF using the Funk-Radon Transform on

!

a spherical acquisition scheme (i.e. one q-value).

Tuch, 2004. Tuch et. al. 2003.

[Link] - RAD 229
 Spherical Deconvolution
Anderson and Ding, 2002. Tournier et. al. 2004.

⊗ -1 =

Response
DWI Signal fiber ODF
Function

Deconvolve DWI signal with a response

function to recover the fiber ODF.

[Link] - RAD 229

 Several Model-Based Approaches

• Multi-Tensor Model
+

• Ball and two stick model, Behrens, 2003.

+ +

[Link] - RAD 229

 Intra-Voxel Crossing Fibers

parallel fanning crossing kissing bending

tensor

Jbabdi S, et. al. Brain Connect, 1:169-183, 2011.

[Link] - RAD 229
False Positives in Diffusion Tractography

O’Donnell LJ and Westin C-F, Neurosurg Clin N Am, 22:185-188, 2011.

[Link] - RAD 229

False Negatives in Diffusion Tractography

O’Donnell LJ and Westin C-F, Neurosurg Clin N Am, 22:185-188, 2011.

[Link] - RAD 229

 We Need More Validation Tools
Very few methods exist (in vivo or ex vivo) to map 3D white matter pathways and they
all have limitations.

Polarized Classical Electron

Tracers Light 2D Histology Dissection Microscopy
Imaging Myelin Stains

Saleem (2008)
J. Comp. Neur.
Larsen (2007) Seung S. [Link]
Microsc Res Tech Lawes (2008) NeuroImage

Optical Coherence CLARITY 3D Histology

Tomography

Magnain (2015) Neurophotonics Tomer (2014) Nature Protocols

[Link] - RAD 229

Post-mortem Diffusion Imaging

Benefits
• Diffusion orientation preserved.
• No motion.
• Long scan times possible.
• Can be compared with histology.

McNab,..Miller, 2009. McNab, Edlow et. al. 2013.

Acquisition Considerations

[Link] - RAD 229

 The effect of b-value
Multi-Shell Diffusion Acquisitions
Why bother?
One Orientation Two Orientations Three Orientations

Signal at
different
b values
(s/mm2)
b=1000
b=2000
b=3000
b=4000
b=5000

Higher b value gives us more angular contrast!!!

[Link]

[Link] - RAD 229

 Multi-Shell Diffusion Acquisitions
The effect of b-value
Why bother?
One Orientation Two Orientations Three Orientations
Multi-Shell Diffusion Acquisitions
Why bother?
Signal at
different One Orientation Two Orientations Three Orientations
b values
(s/mm2) Signal at
b=1000 different
b=2000 b values
b=3000 (s/mm2)
b=1000
b=4000 b=2000
b=5000 b=3000
b=4000
b=5000

b=300 b=1000 b=2000 b=3000

But SNR
goes down Higher b value gives us more angular contrast!!!
very quickly
with b…

[Link]

[Link] - RAD 229

b-value: q-value vs. diffusion time

c c

q diffusion
time

Time

[Link] - RAD 229

Different Types of Resolution for Diffusion MRI
spatial resolution spin displacement angular resolution
resolution
30°

40°

50°

60°

90°
1/kmax 1/qmax
no. of diffusion
how far out area of diffusion
encoding directions
in k-space encoding gradient
[Link] - RAD 229
Benefits of Strong Gradients for Diffusion MRI

180° 90° 180°

90°

G G G G

TE TE

1) Improved SNR at high q-values.

2) Ability to obtain short diffusion times and high q-
values simultaneously.
3) Closer to the short pulse approximation.
[Link] - RAD 229
 Bonus Material

Lots of different flavors of diffusion imaging.

Variations in the acquisition and/or modeling that help to

gain sensitivity to different microstructure features.

[Link] - RAD 229

Gaussian vs. Non-Gaussian Diffusion
Gaussian Diffusion:

• large displacement and small q Diffusion

Tensor
• for isotropic or a single fiber population
Non-Gaussian Diffusion:

• small displacement and large q

• intra-voxel crossing fibers
• bi-exponential diffusion
➡ restriction, multiple compartments, exchange

[Link] - RAD 229

 Diffusion Kurtosis Imaging
Cumulative Expansion of Diffusion MRI Signal

Gaussian kurtosis
DTI
• quantify the non-Gaussianity
of the diffusion signal.

Jensen J et. al. 2005.

[Link] - RAD 229
 What about fiber patterns in grey matter?

Deep Grey Matter

Lizard Brain Cerebral Cortical
Grey Matter
Cognitive Processing

White Matter
Information Pathways
Atlas of the Human Brain in Section, 2nd ed.,
Roberts et. al. 1987.

[Link] - RAD 229

 Cortical Diffusion Orientations
In Vivo 2 mm Ex Vivo 0.7 mm In Vivo 1 mm
No coherent pattern of Radial diffusion Radial diffusion
diffusion orientations orientations orientations

McNab JA et. al. NI McNab JA et. al. NI

46:775-785, 2009. 69:87-100, 2013.
[Link] - RAD 229
Cortical Diffusion Orientations

Qiyuan Tian, Christoph Leuze

[Link] - RAD 229
b-value: q-value vs. diffusion time

c c

q diffusion
time

Time

[Link] - RAD 229

 Oscillating Gradient DWI
Diffusion-Encoding Frequency
Time Integral
Waveforms Spectrum
f(t) |F(w)|2
g(t)
PGSE

OGSE
sine

OGSE
cosine

π
π/2
RF
0 TE
Does et. al. 2003. Parsons EC et. al. 2006. Gore et. al. 2010.
[Link] - RAD 229
 Oscillating Gradient DWI
Study time dependence of ADC over a series of short
diffusion times rather than looking at the mean displacement
after a single long diffusion time.
T2
Can be used to distinguish:

• disperse flow vs. restriction

PGSE
• pore sizes
• surface-to-volume ratio OGSE
• diffusion path tortuosities
tumour microstructure

Does et. al. 2003. Parsons EC et. al. 2006. Gore et. al. 2010.
[Link] - RAD 229
 Diffusion-Diffraction Minima
NMR in Porous Media
z

Signal, E(q,Δ)

• minima occur at
q=1/Lr, where Lr is
the restriction
length.
Callaghan 1991. Shemesh et. al. 2010.
[Link] - RAD 229
Mapping Compartment Sizes in Brain Tissue

Axon Diameter Mapping: Acquire a range of q-values

and diffusion times and fit the data to a model for intra
and extra-axonal diffusion.

AxCaliber Model
Stanisz GJ et. al. MRM (1997).
Assaf Y. et. al. MRM (2008).
Barazany [Link]. al. Brain (2009).
Ong HH et. al. NI (2010).
Alexander DC et. al. NI (2010).
Zhang H et. al. NI (2011). hindered restricted

[Link] - RAD 229

 Initial Work in Fixed Tissue
Assaf Y. et .al. MRM 59:1347-1354 (2008).
Optic Nerve

Signal Decay
Signal Decay

q-value (µm-1) q-value (µm-1)

Sciatic Nerve
Probability

Probability

Axon Diameter (µm) Axon Diameter (µm)

Spectroscopy: 7T, Gmax=1200mT/m,51min.
[Link] - RAD 229
Initial Work in In Vivo Humans Using 300 mT/m Gradients

0.4

µm
0.35

0.3

Genu
0.25

Probability
0.2

Splenium
0.15

0.1

0.05

0
0 2 4 6 8 10 12
a (um)

McNab, Edlow,....Wald. NeuroImage, 2013.

Huang, ....McNab. ISMRM 2014.

[Link] - RAD 229

Angular-Double Pulsed Field Gradient Imaging
Cory 1990, Mitra 1995, Koch [Link] and Cohen 2011.

Ψ = angle between dir1 and dir2

Normalized Signal

Normalized Signal

Ψ [deg] Ψ [deg]

[Link] - RAD 229

 Angular-Double Pulsed Field Gradient Imaging

Free Diffusion
Normalized Signal

Normalized Signal

Ψ [deg] Ψ [deg]
Cory 1990, Mitra 1995, Koch [Link] and Cohen 2011. [Link] - RAD 229
 Diffusion MRI

• Diffusion MRI: a marker of tissue microstructure

• Sensitizing the MRI signal to diffusion
• Measuring diffusion rates with MRI
• Diffusion tensor imaging
• Tractography
• Acquisition Considerations.
• Bonus Material: Less conventional diffusion MRI
methods.

72 [Link] - RAD 229