Stieltjes · Brunner

Fritzsche · Laun

Diffusion Tensor
Imaging Introduction
and Atlas

3-D
dataset
on CD
Diffusion Tensor Imaging
Bram Stieltjes
Romuald M. Brunner
Klaus H. Fritzsche
Frederik B. Laun

Diffusion Tensor Imaging

Introduction and Atlas
With 340 figures and CD-ROM

123
Dr. Bram Stieltjes Dr. Klaus H. Fritzsche
Deutsches Krebsforschungszentrum (DKFZ) Deutsches Krebsforschungszentrum (DKFZ)
Section Quantitative imaging-based disease characterization, Department of medical and biological informatics
department of radiology Im Neuenheimer Feld 580
Im Neuenheimer Feld 280 69120 Heidelberg
69120 Heidelberg Germany
Germany
Dr. Frederik B. Laun
Prof. Dr. Romuald M. Brunner Deutsches Krebsforschungszentrum (DKFZ)
Heidelberg University Hospital, Department of psychiartry, Department of medical physics in radiology
Division child- and adolescent psychiatry Im Neuenheimer Feld 280
Blumenstr. 8 69120 Heidelberg
69115 Heidelberg Germany
Germany

Additional material to this book can be downloaded from http://extras.springer.com

ISBN-13 978-3-642-20455-5 ISBN 978-3-642-20456-2 (eBook)

DOI 10.1007/978-3-642-20456-2

Springer Medizin
© Springer-Verlag Berlin Heidelberg 2013

Library of Congress Control Number: 2012944834

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any
other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered and executed on a computer
system, for exclusive use by the purchaser of the work. Duplication of this publication or parts thereof is permitted only under the
provisions of the Copyright Law of the Publisher’s location, in its current version, and permission for use must always be obtained
from Springer. Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations are liable to
prosecution under the respective Copyright Law.

The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in
the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore
free for general use.

While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors
nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher
makes no warranty, express or implied, with respect to the material contained herein.

Editor: Dr. Christine Lerche

Project Management: Claudia Bauer
Copyediting: Isabella Athanassiou, Heidelberg
Project Coordination: Barbara Karg
Cover Illustration: © Dr. Bram Stieltjes, Deutsches Krebsforschungszentrum, Heidelberg
Cover Design: deblik Berlin
Typesetting and Reproduction of the figures: Fotosatz-Service Köhler GmbH – Reinhold Schöberl, Würzburg
The German Cancer Research Center (Deutsches Krebsforschungszentrum) holds the rights to the software on the CD.

Printed on acid-free paper

Springer Medizin is brand of Springer

Springer is part of Springer Science+Business Media (www.springer.com)
 V

Preface

Since the advent of diffusion tensor imaging in the mid-1990s, the field of in vivo representation of human
neuronal connectivity has experienced a dramatic development in terms of technological advances as well
as of applications in both neuroscience and clinical research. Although the main focus of this book is on the
depiction of neuroanatomy derived from diffusion imaging, we hope that with the »Introduction to Diffusion
Imaging,« the book will also be a good first point of reference for navigation of the currently overwhelmingly
extensive literature on diffusion imaging.
The making of this atlas was a far more interactive process than initially anticipated. For instance, to be able
to create reproducible image views and constant size representations, we had to redesign our fiber tracking
software and in many ways, over time, the final result evolved far from the original concept. We hope that the
resulting combination of two-dimensional T1-weighted and color map-based anatomy enriched with the part
on three-dimensional white matter tract representation will help readers to quickly find their way in the most
fascinating of all mazes, the human brain.
Finally, we want to thank T. Kuder for corrections and graph preparation and Springer Heidelberg for
embarking with us on this adventurous journey; Mrs. R. Scheddin for the enthusiasm when we first presented
the idea of the atlas and Dr. C. Lerche and Mrs. C. Bauer for unrelenting support during the numerous iterations
and corrections.

Bram Stieltjes
Romuald M. Brunner
Klaus H. Fritzsche
Frederik B. Laun
Spring 2012
 VII

Contents

I Introduction
How to Use this Atlas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1 Introduction to Diffusion Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.1 Diffusion: A Primer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Timeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Theoretical Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Advanced Techniques: Fiber Tracking and Deviations from Mono-exponential Signal Decay . . . . . . . 23
1.5 Practical Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.6 Selected Applications in Neuroscience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

II Atlas
2 Two-dimensional Brain Slices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.1 Coronal View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2 Sagittal View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
2.3 Transversal View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

3 Three-dimensional Fiber Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281

3.1 Fiber Tracking of the Cerebral Hemispheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
3.2 Fiber Tracking of the Brain Stem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351

III Appendix

Index Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379

Index Atlas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
 1 I

Section I
Introduction

How to Use this Atlas –3

Chapter 1 Introduction to Diffusion Imaging –5

 3

How to Use this Atlas

4 How to Use this Atlas

jThe Timeline and Introduction

The timeline and introduction sections can be seen as a primer
in diffusion imaging. Although the atlas focuses on directional
diffusion, we chose to give an outline of the complete diffusion
literature to come to a comprehensive overview. To make the
introduction accessible to readers with a different academic
background, we designed it so that the main body of the text is
understandable without specialist knowledge in physics or math-
ematics. More detailed information on the diffusion process and
calculations can be found in separate textboxes for further read-
ing. Thus, as your knowledge of diffusion increases, so may your
interest in the content of the textboxes.

jThe Atlas
The atlas consists of two major parts: a two- and a three-dimen-
sional representation of fiber tracts.
In the two-dimensional part, we compare the conventional
T1-weighted anatomy with a diffusion tensor imaging-derived
color map. In these maps, the directional orientation of fiber
tracts is color coded in the following fashion: tracts moving left-
right are coded red (e.g., the corpus callosum), anterior-posterior
tracts are coded green (e.g., the cingulum), and craniocaudal
tracts are coded blue (e.g., the corticospinal tract). The intensity
or hue indicates the fractional anisotropy, a measure of fiber den-
sity. The whole brain is covered in the three main radiological
planes: axial, coronal, and sagittal.
The three-dimensional part covers the most prominent white
matter connections in the human brain. The complete recon-
struction process is presented in a consistent, step-by-step fash-
ion. First, the relevance and anatomy of the tract are discussed
and an initial region of interest (ROI) is shown. This ROI is cho-
sen to yield an optimal final result and is shown in white. By using
inclusion (green) and exclusion (red) ROIs, the result is further
refined. The final result is represented without the ROIs in three
different planes as well as in an oblique view for optimal appre-
ciation of the anatomical location. The color coding of these
tracts is identical to the two-dimensional color maps. The intri-
cate anatomy of important adjacent tracts is further illustrated in
combined overviews. Here, each individual tract is represented
in monochrome to enhance the visualization of the complex in-
terwoven anatomy. Again, these results are represented in three
standard planes and an oblique view.
All tracts in the three-dimensional part are also indicated in
the two-dimensional part of the atlas, and flipping between these
parts will help readers to understand the projection of the three-
dimensional tract on the two-dimensional slices. Using the
T1-weighted images, adjacency to important gray matter struc-
tures is captured. Thus, by interactively using the atlas, your
understanding of the complex neuroanatomical correlations
should grow continuously. To aid this process, the accompanying
CD contains all the tracts as illustrated in the atlas and allows you
to scroll through the data in an interactive fashion.
 5 1

Introduction to Diffusion Imaging

1.1 Diffusion: A Primer –6

1.2 Timeline –7

1.3 Theoretical Aspects – 10

1.4 Advanced Techniques: Fiber Tracking and Deviations

from Mono-exponential Signal Decay – 23

1.5 Practical Aspects – 29

1.6 Selected Applications in Neuroscience – 33

1.7 References – 37

B. Stieltjes et al., Diffusion Tensor Imaging,

DOI 10.1007/978-3-642-20456-2_1,
© Springer-Verlag Berlin Heidelberg 2013
 6 Chapter 1 · Introduction to Diffusion Imaging
1.1 Diffusion: A Primer
1 membrane gets broken! Or: Diffusion is largest in left–right
Diffusion is – for the purpose of this book – the random motion of
water molecules in fluid water. When water is frozen, the particles
stand still, but when water is liquid, the water molecules move
around owing to their thermal energy. They do so rather quickly, at
about 1,000 m/s. This above-racing-car speed, however, does not
convert into a high-speed, long-distance distribution of the mole-
cules, because they bounce at each other very frequently: The time
between bounces is about 1/1,000,000,000,000 s only. Despite
this high-speed crashing – which, of course, prevents the particle
from effectively translating like a racing car – the particles do move
away from their original position. Since this process depends so
crucially on the exact initial positions and velocities of all the
10,000,000,000,000,000,000,000 water molecules present in each
drop of water (which is information that is totally elusive to us), it
appears to us that the water molecules move in a random fashion
making it impossible to predict where they end up. Thus, because
the system is so incredibly complex, the human investigator does not
care about the individual water molecule, but only about Joe Water.
How far does Joe Water translate? Of course, that depends on the
time we let him bounce around with his fellows, but let us say: How
far does he translate in the time (approximately 0.1 s) we need to
make a magnetic resonance (MR) image of the head – where Joe is
supposed to stay momentarily. The answer, which was verified in a
variety of experiments, is: he moves about 10 μm. Great! The average
cell has approximately the same size. Joe will most likely bounce
against the cell membrane.
Thus, already slowed down by his fellows, Joe has found
another obstacle curbing his freedom. And this obstacle, although
annoying for the pert and nimble spirit of Joe, means a lot to us.
For it is thought (by many and with some entitlement) that if we
have information about the diffusion-restricting structure of the
cell membrane, then we will know more about the tissue, about
its diseases, and about its course, if it is an axonal tract, thereby
enlightening us about its orientation, be it left–right, up–down,
or top–bottom.
Thus, if we could measure diffusion, it would open up a new
world for us, providing a glance at microcellular structures,
allowing apparently straightforward interpretations to be made.
Take a drug to fight the tumor. Diffusion goes up? Good, the cell
direction? There must be a nerve bundle!
If only we were able to measure diffusion in a pain-free,
convenient fashion... And, indeed, we can! This might come as
no surprise, for otherwise you would probably not be reading this
primer of an atlas on MR diffusion tensor imaging. This might
also come as no surprise since one knows that MR imaging yields
images of the spatial distribution of water molecules; so the step
of measuring displacements does not seem to be too far off. In
fact, diffusion is in many regards measured like an MR image.
Apart from acquiring one image, one designs an experiment that
intrinsically acquires two images – one at the beginning of the
experiment, and one at the end – that are subtracted from each
other. If Joe was lazy and did not move, the two images would
cancel, but if he moved from A to B, then the images would not
cancel, and from the difference image, we could learn about the
separation of A and B. In practice, of course, things are slightly
more complicated: it is not really an image that is acquired, but a
phase image. Moreover, it is not possible to measure the differ-
ence image of one particle, but only that of the whole ensemble,
which renders the process more complex.
This complicated topic is explained in the first part of this
atlas. After a short historical overview in 7 Sect. 1.2, the required
sophisticated techniques are explained in 7 Sect. 1.3 to 7 Sect. 1.6.
We put much effort into rendering the manuscript accessible to
a wide readership. Since most readers might be happy to get
the picture without being bothered by forests of formulas, we
designed the main text such that it should be understandable
without a PhD in physics. Nonetheless, since diffusion is not an
easy topic, it might require duty and desire to get through it. The
text is written assuming that the reader is familiar with basic
mathematical tools (addition, subtraction, multiplication, divi-
sion, exponentials, and logarithms) and basic MR imaging con-
cepts (T1, T2, Larmor frequency, rotating reference frame, spin
echo, k-space), although much of the text should be comprehen-
sible without these prerequisites. Current or future PhD physi-
cists should also enjoy 7 Sect. 1.3 to 7 Sect. 1.6, since there are
special textboxes, full of formulas, presenting an in-depth treat-
ment of the topic.
1.2 · Timeline
7 1
1.2 Timeline

(. Fig. 1.1, . Fig. 1.2, . Fig. 1.3)

. Fig. 1.1 Main developments in diffusion NMR imaging: 1950–1989.

 8 Chapter 1 · Introduction to Diffusion Imaging

. Fig. 1.2 Main developments in diffusion NMR imaging: 1990–2001.

1.2 · Timeline
9 1

. Fig. 1.3 Main developments in diffusion NMR imaging: 2002–2012.

 10 Chapter 1 · Introduction to Diffusion Imaging
a b
1
. Fig. 1.4 a Particles start at the origin. b Each particle performs a random walk in space. c This leads to the distribution of the particles in space
1.3 Theoretical Aspects
1.3.1 Free Diffusion
Molecules in a fluid do not stand still but perform a random walk
due to thermal motion and repeated interactions with surround-
ing molecules [18][22][25][38][72][84][89][108]. As a conse-
quence of this random motion, particles initially residing at the
same position distribute in space (. Fig. 1.4).
Since the process is statistical, it is necessary to use statistical
measures to describe, for example, how far a particle typically
translates. The most straightforward statistical measure, the expec-
tation value ¢x² of the translation distance x, is inappropriate (see
the box entitled »The Expectation Values ¢x² and ¢x2²«) since par-
ticles translate in all directions with equal probability, such that ¢x²
is always zero. Therefore, the expectation value of the squared dis-
placement ¢x2² is usually employed because positive and negative
displacements do not cancel. For free diffusion, it can be shown
(see »The 2D0t Law and the Drunken Man«), that
x 2 = 2D0t
holds true, where t is the time and D0 is the so-called free diffu-
sion coefficient. In order to understand the meaning of this equa-
tion, it is useful to compare it with a linear translation (e.g., of
a moving car). For a moving car, the relation between the traveled
distance scar, the velocity v, and the time t is
scar = vt
Concretely, if a car moves at a velocity of v=100 km/h for 1 h, then
it translates the distance
km
scar = vt = 100 ⋅1 h = 100 km.
h propagator. Inspecting the Gaussian distribution plotted in
Traveling 2 h, the car translates scar = 200 km. Thus, the traveled
distance increases linearly with time.
For a random walk, this relationship is fundamentally differ-
ent! The deeper reason for this is that particles do not always
c
move in the same direction, but also in the opposite direction.
In a certain sense, the particle (if it were a car) drives backward
at times, thus decreasing the translation distance. For further
discussion, the square root of Eq. 1 is taken
s= x 2 = 2D0t (3)
The typical translation distance s does not increase linearly in
time, but only with its square root, which is much slower. The
quadruple time is needed for the particle to translate double the
distance. A typical free diffusion coefficient of water molecules
in water is D0 = 2 μm²/ms. Thus, during a time of 100 ms (which
is the usual time of a diffusion experiment on clinical MR scan-
ners), the typical translation distance is
μm 2
s = 2⋅2 ⋅100 ms = 400 μm2 = 20 μm
ms
It takes a diffusion time of 400 ms until s increases to 40 μm. One
consequence of Eq. 1 is that the units of D0 are somewhat non-
(1) intuitive: units are not m/s or km/h as for velocity, but μm²/ms.
This is necessary; if length and time were in the same power, the
square root time law could not hold true (the units on both sides
of Eq. 1 must be equal).
An important property of free diffusion is that the distribu-
tion of particles P0 (x, t) that start their random walk at the origin
is always known, it is a Gaussian function (. Fig. 1.5).
e − x / 4 D0t
2
(2) P0 (x ,t ) = (4)
4 πD0t
This is a consequence of the so-called central limit theorem [25].
P0 (x, t) describes the probability that a particle translates the
distance x in the time t, and is also often called the free diffusion
. Fig. 1.5, it becomes apparent that small diffusion distances
are more probable than larger ones. Large diffusion distances
are extremely improbable. At longer times t, P0 (x, t) broadens,
meaning that the particles tend to translate farther away,
while it becomes more unlikely that the particle ends up where
1.3 · Theoretical Aspects
11 1

The Expectation Values ¢x² and ¢x2²

A random walk is a statistical process whose properties must be de-
scribed by statistical measures. For diffusion, the considered measures
are usually the expectation values ¢·² of the displacements, or of pow-
ers of the displacements. Let x be the distance a particle is displaced
during its random walk. The table below shows an example of five
particles having performed five random walks.
Particle number 1 2
x 2 –3
The first particle moved 2 μm, the second particle –3 μm, the third par-
ticle ended up where it started, and so on (units are omitted in the ta-
ble). The expectation value ¢x² is the sum of all x divided by the number
of particles. In this example – although most particles did translate –
¢x² is equal to zero, since positive and negative values cancel. The same
happens for a general random walk when no drift is present: ¢x² is
always zero and is thus not a good measure of the typical particle
displacement due to diffusion. The expectation value ¢x2² is a better
3 4 5
measure since negative values do not cancel. Usually, ¢x2² is employed
0 5 –4 to characterize the random walk process.

x2 4 9 0 25 16

¢x² = (2 – 3 + 0 + 5 – 4)/5 = 0

¢x2² = (4 + 9 + 0 + 25 + 16)/5 = 10.8

The 2D0t Law and the drunken Man

is equal to N. The second term

∑ i ≠ j si s j
N

is equal to zero, since it is assumed that the individual steps are not
correlated. Therefore,
The 2D0 t law can most easily be derived for a one-dimensional
discrete random walk (of man, mouse, or molecule). The discrete x 2 = L20 N
random walk model assumes the following: The particles are located at
discrete positions. At every discrete time step of duration W, the particle Introducing the total time of the random walk t = NW, ¢x2² becomes
moves with equal probability one step to the left or to the right, thus
L20 N
the i-th step direction is si = ±1. Let L0 be the distance between the dis- x2 = t
τ
crete particle positions. The quantity that we are interested in is ¢x2²
or s = x 2 . The displacement is L0 and W are constants, which can be combined to a new constant, the
free diffusion coefficient
∑ i si ⋅ L0
N
x=
L20
D0 = ,
2τ
Consequently, when N is the number of steps,
yielding
(∑ ) (∑ s )
2 2
∑ i , j =1si s j
N N N
x2 = si ⋅ L0 = L20 i = L20
i =1 i =1 x 2 = 2D0 t
Now, a little trick is used. The last sum is split into two parts, one con- s= x 2 = 2D0 t
taining the summands for which i and j are equal, the other containing
those terms for which i and j are unequal. Thus, the random walk model directly yields the square root of time
relationship. Although this model may appear oversimplified, the
∑ i =1si2 + ∑ i ≠ j si s j ∑ i =11 + L20 ∑ i ≠ j si s j
N N N N
x 2 = L20 = L20 obtained relation s = 2D0 t is valid under very general conditions,
which is due to the central limit theorem.
The term

∑ i =11
N

it started. Examining Eq. 4 closer, one observes that P0 (x, t) is
fully determined by one single free parameter: the free diffu-
sion coefficient D0. Thus, measuring D0, for instance by measur-
ing ¢x2² at different times using Eq. 1 (e.g., as described in 7 Sect.
1.3.3), yields the full information contained in the distribution
P0 (x, t). Further measurements thus do not yield any additional
information.
1.3.2 Restricted Diffusion
A typical cell radius is 10 μm [72], which is of the same order of
magnitude as the typical diffusion distance s. For this reason,
the diffusing water molecules »see« the cell membranes and
become restricted in their motion (cell membranes are the dom-
inating factor for diffusion restrictions [15][16][17][136]). Be-
 12 Chapter 1 · Introduction to Diffusion Imaging

cause of these restrictions, the average translation of particles is

1 reduced:

x2 tissue < x2 free (5)

For restricted diffusion, Eq. 1 is modified in order to define a

time-dependent diffusion coefficient D (t) in analogy to the free
diffusion coefficient D0:

x 2 tissue
D(t ) = (6)
2t

In the literature, D (t) is often also called the apparent diffusion

. Fig. 1.5 Distribution of particles that started at the origin and performed
free random walks of different durations (D0 = 1 μm²/ms). The shape of the
distribution is a Gaussian function. At longer diffusion time, the distribution
becomes broader and the particle is less likely to end up at its starting posi-
tion. The free diffusion constant D0 is the only free parameter and it com-
pletely determines the width of the Gaussian function at every time point.
The distribution of the particles is also called the free diffusion propagator
coefficient, a label that we reserve for the diffusion coefficient
measured with long diffusion gradients. . Fig. 1.6 illustrates the
time dependence of D (t) and shows the case of a particle starting
its random walk at x1 = 2.5 μm at time t = 0, with two perfectly
reflecting plates restricting its motion at
xplate = 0 μm and xplate = 10 μm

a b

c d
. Fig. 1.6 Diffusion propagator of a diffusing particle starting at x1 = 2.5 μm for diffusion times ranging from 0.1 ms to 100 ms. The motion is restricted
by two plates at xplate = 0 μm and xplate = 10 μm. The time-dependent diffusion coefficient D (t) decreases owing to the restrictions. It is therefore an indica-
tor of diffusion restrictions, and measuring D (t) permits inferences to be made about the structure of the restrictions. Unlike for free diffusion, the diffusion
coefficient alone does not contain the full information contained in propagator: the propagator can take too many shapes (it is almost a Gaussian function
in a and a flat box function in d to be describable by one single parameter
1.3 · Theoretical Aspects
At very short diffusion time (. Fig. 1.6a), it is very improbable
that the particle touches the plates, thus the diffusion is almost
free and D (t) ≈ D0. After a certain time (. Fig. 1.6b, c, d), the
plates heavily restrict the particle motion. The particle is
»trapped« between the plates, and it cannot translate farther
than the distance a between the plates, so that D (t) becomes
smaller than D0. It is exactly this effect that renders diffusion-
weighted imaging so interesting for diagnostics. Increased cell
restrictions (e.g., due to solid tumor growth) reduce D (t), while
reduced cell restrictions (e.g., due to necrosis) increase D (t).
Therefore, measuring D (t) allows inferences to be made about
the state of the tissue.
A second glance at . Fig. 1.6 reveals a further important
property of restricted diffusion. The diffusion propagator
P (x1, x2, t) – which describes the probability that the particle
traveled from x1 to x2 during the time t – is not shaped like
a Gaussian function anymore; it is in general an arbitrary, com-
plicated function [52]. This function is not fully determined by a
single parameter D0, but by many parameters that are influenced
by cell geometry and cell membrane permeability. Thus, measur-
ing the full function P (x1, x2, t) yields more information than
only measuring ¢x2²tissue or D (t).
The parameter
x2 tissue = (x2 − x1 )2 tissue
= 2D(t )t
is nonetheless helpful: it can be used to estimate the width of
P (x1, x2, t). It is an intuitive parameter describing how far the
particles typically translate.
1.3.3 Measurement of Diffusion Using Nuclear
Magnetic Resonance
Narrow diffusion-weighting Gradients
A special property of the nuclear magnetic resonance (NMR)
method is that the voxel-averaged quantity D (t) and even
–
the complete voxel-averaged propagator P (x, t) are detectable
in NMR diffusion experiments [5][9][11][12][14][22][23]
[24][31][32][52][91][101][129][130]. Unfortunately, for this
statement to hold true, a perfect gradient system, providing
infinitely large, temporally extremely narrow magnetic field
gradients, must be available. In practice, there is no such gra-
dient system, and if there was one, it could hardly be applied
in patient measurements because of peripheral nerve stimula-
tions [124].
Nonetheless, we will first examine the limiting case of narrow
diffusion gradients since it provides a direct link of the measured
signal to the diffusion process. In the second part of 7 Sect. 1.3.3,
we will focus on the case of elongated gradients and discuss the
occurring differences.
To examine the diffusion process, a classic NMR experiment
(. Fig. 1.7) must be modified by adding two additional bipolar
magnetic field gradients between the exciting radio frequency
pulse and signal readout [129][130]. As mentioned before, we
first assume that these gradients are very short, with infinite gra-
dient amplitude. The time between the gradients is labeled '.
13 1
These gradients shall be regarded as the limiting case for which
the gradient amplitude G increases and the gradient duration G
decreases such that their product, or the q-value q = JGG, re-
mains constant. J is the gyromagnetic ratio and is a constant.
. Fig. 1.7 describes the basic principle of the diffusion measure-
ment, while a mathematical description can be found in the
boxes entitled »Short Diffusion Gradients and Free Diffusion,«
»Short Gradients and restricted Diffusion: How to measure D (t)«
and »Q-space Imaging.« From these derivations, the following
results can be inferred:
4 The diffusion weighting due to the additional gradients
leads to a signal drop. The larger the diffusion coefficient,
and the larger the q-value and gradient separation ' are, the
stronger the signal drops.
4 If the diffusion is free, the signal decays exponentially:
S = S0 exp (–q2'D0), where S0 is the signal without diffusion
gradients and S is the signal with diffusion-weighting
gradients (. Fig. 1.8).
4 This means that D0 can be determined by performing two
measurements: one with and one without diffusion gradi-
ents. This yields S and S0. The q-value and the diffusion
time ' are determined by the used sequence. The only free
parameter is the free diffusion coefficient D0, which can be
calculated from the measured signals.
4 If the diffusion is restricted (. Fig. 1.9), the signal
decreases exponentially at small q-values only:
S = S0 exp (–q2'D(')). This means that D (') = D (t) can
be determined (exactly as the free diffusion coefficient)
by performing two measurements: one with and one
without diffusion gradients. This yields S and S0. The
q-value and the diffusion time ' are determined by the
used sequence. The only free parameter is the time-depend-
ent diffusion coefficient D (t), which can be determined
by a fit.
4 In clinical scanners, it is usually not the q-value, but the so-
called b-value that is used, and which can be adjusted on
the user interface. These two values are related by b = q2'
for short gradients. Consequently, the measured signal is
S = S0 exp (–bD(')). Thus, the time-dependent diffusion
coefficient D (') can be determined by a fit of the signal to
b. Usually, b-values of about 1,000 s/mm² are used.
4 At larger q-values or b-values (. Fig. 1.9 and 7 Sect. 1.4.4 on
diffusion kurtosis imaging), the signal decays as follows:
⎛ 1 ⎞
S = S0 exp ⎜ −bD(Δ ) + b2 (D(Δ ))2 K (Δ ) + !⎟ (for details see
⎝ 6 ⎠
[65][66][67][98]). If b is small, the b2-term is not of impor-
tance; however, for large b-values, it becomes dominant.
The kurtosis K (') is an additional measurable parameter
which is zero for free diffusion. If diffusion restrictions are
present, K (') does in general deviate from zero, and can be
either larger or smaller than zero, depending on the geome-
try of the restrictions. K (') can be regarded as a parameter
describing the deviation from Gaussian diffusion. To deter-
mine the kurtosis, the signal must be measured with several
b-values ranging typically from 0 s/mm² to more than
b = 2,000 s/mm².
 14 Chapter 1 · Introduction to Diffusion Imaging

. Fig. 1.7 Basic principle of diffusion weighting in NMR. The red arrows in circles symbolize the magnetic moment in the rotating reference frame at dif-
ferent spatial positions (lower row of images). On top of that, the spatial profile of the magnetic field is depicted, which consists of the main magnetic field
B0 and the spatially varying gradient field. The 90° radio frequency pulse generates a transversal magnetization that is dephased by the first diffusion gradi-
ent. If no diffusion is present, the particles do not move, and the second diffusion gradient rephases the magnetization. However, if diffusion is present,
the particles change their spatial position, and the second gradient does not rephase the magnetization. The signal can in some sense be regarded as the
sum of the red arrows. If diffusion is present, the arrows do not point coherently along the same direction, and the signal (or the sum of arrows) drops. This
signal drop is the stronger, the stronger and longer the magnetic field gradients are applied, and the farther the particles translate. Since the gradients are
controlled by the measurement sequence, information about the diffusion process can be inferred by measuring the signal with and without diffusion
gradients: A strong signal drop indicates strong diffusion

a b
. Fig. 1.8 a Linear and b logarithmic plot of signal drop due to diffusion weighting (free diffusion). The larger the free diffusion coefficient is, the stronger
the signal decays. Since the diffusion is assumed to be free in this example, the signal decays exponentially. The negative slope in the logarithmic plot of
signal versus b = q2' is the free diffusion coefficient D0. If the diffusion is restricted, the signal decays approximately exponentially for small b-values only. It
is experimentally observed that b-values smaller than 1,000 s/mm² can usually be considered as small in biological tissue
1.3 · Theoretical Aspects
15 1

a b c
. Fig. 1.9 a, b For free diffusion, the propagator is a Gaussian function and the diffusion-weighted signal decays exponentially. Different free diffusion co-
efficients due to different viscosities result in different slopes of the signal decay. c If diffusion restrictions are present, then the propagator deviates from a
Gaussian function and the signal does not decay exponentially. The occurring deviation can be quantified using the kurtosis parameter, which can be inter-
preted as a measure for the deviation from Gaussianity of the diffusion propagator

4 By measuring with many and large q-values, the hologram
–
of the voxel-averaged propagator P (x, t) can be determined
[22][23][31]. By Fourier transformation, the voxel-averaged
–
propagator P (x, t) can then be calculated. This technique is
called q-space imaging. In phantom experiments, it could
be applied with impressive results, for example, yielding
sharp peaks for q-values corresponding to the length scale
of hollow capillaries. In practice, the q-space technique only
plays a minor role in medical imaging for two reasons. First,
a long measurement time is required, since the signal must
be acquired for many q-values, especially in three dimen-
sions (e.g., 50 × 50 × 50 = 125,000 images must be measured
in three dimensions). Second, the condition that the
gradient duration is short against the time scale of the diffu-
sional translation is technically often hardly realizable.

Short Diffusion Gradients and free Diffusion

It is assumed that a pair of bipolar gradients of duration G and amplitude Thus, we find
G are applied. The time interval between the gradients shall be of dura-
∫
2Δ
tion ': e − iqx = dxP( x ,Δ )e − iqx = e − D0 q = e − bD0

with b = q2' = J2G2G2'.

For D0 = 1 μm²/ms, the signal drops as follows:

In the time interval between the gradients, the particle performs a

random walk and moves from the initial position x1 to the final position
x2. In the rotating reference frame, a particle obtains the phase
³
M = J x (t) G (t) dt on its path x (t) due to the time-dependent magnetic
field gradient G (t). If the gradients are short, then the particle does not
move considerably during the gradients. Therefore, the first gradient im-
prints the phase M1 = JGGx1, and the second gradient imprints the phase
M2 = – JGGx2 = – qx2, where q is defined by q = JGG. Thus the total phase
is M M1 + M2 = – qx, with x = x2 – x1. The measured signal is

e iϕ = e iγ ∫ x ( t )G ( t )dt = e − iqx

where the brackets ¢·² denote the average over all possible final
positions. Since the diffusion is free, the final particle distribution is a Hence, the free diffusion coefficient D0 can be measured using at least two
Gaussian function b-values and fitting the logarithmic signal. In the above graph, S denotes
e− x
2 /4D t
0 the measured signal at b >0 and S0 is the signal at b = 0.
P( x , t ) =
4πD0 t
 16 Chapter 1 · Introduction to Diffusion Imaging

1 Short Gradients and restricted Diffusion: How to measure D (')

In this box, we show that it is possible to detect the time-dependent The term that is linear in q is set to zero, since no net drift is present and
diffusion coefficient ³
since g (t) dt = 0. Performing the double integral one finds

x 2 tissue
D( t ) = ∫∫dt dt ¢g(t )g(t ¢) x(t ) x(t ¢) = ∫dt g(t )( x(t ) x(0) − x (t ) x ( Δ ) )
2t
= x (0 ) x (0 ) − x ( Δ ) x (0 ) − x (0 ) x ( Δ ) + x ( Δ ) x ( Δ )
by using short diffusion gradients. The phase that a random walker
= ( x ( Δ ) − x (0 ))2 = x 2 tissue
acquires on his path x (t) due to the diffusion gradient G (t) is M =
J³x (t) G (t) dt. The measured signal is proportional to ¢eiM², where the Thus, if the terms O (q3) are neglected, ¢eiM² becomes
expectation value brackets ¢·² denote the average over all possible
1
initial and final positions. e iϕ = 1 − q 2 x 2 tissue = 1− q2 ΔD( Δ ) = 1− bD( Δ )
2
Here, we assume that the gradients are very short, thus
G (t) = q/J · (G(t) – G(t – ')) = q/J · g (t) with g (t) = G(t) – G(t – '). with b = J2 G2 ' = q2'. It is convenient to reformulate the final expres-
We aim at a series expansion of sion as

e iϕ = e iq ⋅ Ú x ( t )g( t )dt 1− bD( Δ ) ≈ e − bD ( Δ )

in powers of q. For small q, we find The right-hand side of this equation is often used since it is valid for
any b if the diffusion is Gaussian.
(∫ )
1 2 2
∫
e iϕ = 1 + iq x (t )g(t )dt +
2
q i x (t )g(t )dt + O( q 3 ) The time-dependent diffusion coefficient can be determined by
measuring once with b = 0 and once with a small b-value (in practice,
1
= 1+ 0 − q2
2 ∫∫dt dt ¢g(t )g(t ¢) x(t ) x(t ¢) + O(q3 ) b<1,000 s/mm² is usually considered as small in biological tissue).

Q-Space Imaging

It is assumed that a pair of bipolar gradients of duration G and ampli-
tude G are applied. The time interval between the gradients shall be of
duration '. During this time interval, the particle performs a random
walk and moves from the initial position x1 to the final position x2.
If the gradients are short, then the particle motion during the gradients
can be neglected, and a particle obtains the phases M1 = JGGx1 = qx1
and M2 = –JGGx2 = – qx2, where q is defined by q = JGG. The propagator
P (x1, x2, ') describes the probability that the particle travels from x1 to
x2 in the time '. The measured signal attenuation E(q) is

E (q ) = ∫∫dx1 dx2 ρ( x1) P( x1, x2 , Δ )e −iq( x − x )

2 1

where U(x1) is the initial density of particles, which is usually assumed

to be homogeneous. Introducing the variable x = x2 – x1 , the double
integral can be stated as

E (q ) = ∫∫dx1 dx ρ( x1) P( x1, x1 + x , Δ )e −iqx

The integration limits were not stated explicitly, but the integration is
understood to be performed over all accessible particle positions for
–
both integrals. Introducing a mean propagator P (x, ') = dx1U(x1) P (x1, ³
x1 + x, '), one finds that the signal attenuation is

∫
E (q ) = dx P ( x , Δ ) e − iqx

Thus, the mean propagator can be determined by measuring E (q) with

a range of q-values and subsequent Fourier transformation.

Long Diffusion Gradients
In real MR systems, it is impossible to apply diffusion gradients
of infinitesimal duration; therefore temporally elongated diffu-
sion gradients are applied (. Fig. 1.10). In that case, it is not pos-
sible to detect the time-dependent diffusion coefficient D (t) in a
similar straightforward fashion as it is possible with short diffu-
sion gradients. Instead, the measured signal is in general a com-
plicated function of the temporal profile of the diffusion gradi-
ents and of the geometry of the restrictions.
Nonetheless, the following statements can be made about dif-
fusion measurements with long diffusion gradients:
4 The free diffusion constant D0 can be determined exactly
(see the boxes entitled »Free Diffusion and long Gradients:
The b-Value« and »The b-Value for Stejskal-Tanner
Gradients«). If the diffusion is free, the signal decays as
S = S0 exp(–bD0) with b = q2 (' – G/3) = J2 G2 G2 (' – G/3).
As with narrow diffusion gradients, D0 can be determined
by a fit, when having measured the signals S and S0 in one
1.3 · Theoretical Aspects
17 1

. Fig. 1.10 a A gradient echo sequence without diffusion weighting. b Two very short diffusion gradients generate a diffusion weighting and permit one
to measure the time-dependent diffusion coefficient D (t). c In real MR scanners, the gradient amplitude is a limiting factor. Therefore, long diffusion gradi-
ents (of about 30 ms) are applied. The thereby measured apparent diffusion coefficient Dapp(t) is only approximately identical to D (t). d The long diffusion
gradients result in long echo times (often >60 ms). Therefore, spin echo sequences are usually employed. Note that the diffusion weighting in d is identical
to that in c. e In practice, more complicated diffusion gradients are usually applied to minimize, for example, the effect of eddy currents [120]

measurement with and one measurement without diffusion
gradients. The parameters G, ' and G, are determined by
the applied sequence (. Fig. 1.10).
4 If restrictions are present, the so-called apparent diffusion
coefficient Dapp (t) is measured. The apparent diffusion
coefficient deviates in general from the time-dependent
diffusion coefficient D (t), but is often a good approxima-
tion to D (t) [152] (. Fig. 1.11). In this context, t denotes
the duration of the diffusion weighting, thus t = ' + G.
4 Dapp (t) is defined by Dapp (t) = –ln(S/S0)/b, using
b = J2G2G2 (' – G/3). Figuratively speaking: the measure-
ment is interpreted as if the diffusion were free, although
this is not the case, and the number that emerges is simply
renamed from D (t) to Dapp (t).
4 The apparent diffusion coefficient is a parameter that is of-
ten used in medical literature. However, it should be noted
that it depends strongly on the temporal profile of the diffu-
sion gradients [53]. For example, the b-value can be adjust-
ed either by adapting the gradient amplitude G or by chang-
ing ' or G. Two very different temporal gradient profiles
can have the same b-value but yield very different values of
Dapp (t). Therefore, it is favorable to state the parameters of
the temporal gradient profile such as, for example, G, ', and
G in publications, and not solely the b-value. Strictly speak-
ing, the b-value is a valid concept only for free diffusion.
4 As with narrow diffusion gradients, the signal decays at
large b-values like
⎛ 1 ⎞
S = S0 exp ⎜ −bDapp (t ) + b2(Dapp (t )) K app (t ) + !⎟ .
2
⎝ 6 ⎠
Thus, for the kurtosis, a very similar procedure to the one
for the diffusion coefficient is performed. To distinguish
the kurtosis measured with long diffusion gradients from
that measured with narrow gradients, it is merely relabeled
from K (t) to Kapp (t). As with narrow diffusion gradients,
the b2-term is not of importance if b is small, but it becomes
dominant at large b-values. The apparent kurtosis Kapp (t)
behaves like the kurtosis K (t): If diffusion restrictions are
present, Kapp (t) is in general unequal to zero and can be
interpreted as a parameter describing the deviation from
Gaussian diffusion [67][68][98].
4 With narrow diffusion gradients, the hologram of the voxel-
–
averaged distribution P (x, t) can be determined. This is
not possible with long diffusion gradients. Instead, it is the
mean propagator of the centers-of-mass of the trajectories
that a particle moves along during the first and second
gradient that can be determined (see the box entitled »Long
Gradients yield apparent Results«, which reproduces the
findings of [106]).
 18 Chapter 1 · Introduction to Diffusion Imaging

a b
. Fig. 1.11 The apparent diffusion coefficient Dapp (t) for the case ⌬ = ␦ (. Fig. 1.10) and the time-dependent diffusion coefficient D (t) for (a) diffusion
inside a sphere and between parallel plates, and (b) diffusion between hexagonally packed cylinders. This figure shows that the apparent diffusion coeffi-
cient Dapp (t), which is obtained with long diffusion gradients, is often a good approximation to the »true« time-dependent diffusion coefficient D (t). The
radius of the sphere and the distance of the plates was 10 μm, the free diffusion coefficient D0 was 2 μm²/ms. The data were generated using (a) matrix cal-
culations [6][10][52] and (b) Monte Carlo simulations [85]

Free Diffusion and long Gradients: The b-Value

In order to estimate the diffusion-weighted signal when arbitrary is zero, since

diffusion gradients G (t) are applied, the signal attenuation
T
³
¢eiM² = ¢exp (iJ G (t) x (t)dt)²must be considered. For moderate diffusion ∫0 dt G(t ) = 0.
weightings (small M), ¢eiM² can be expanded as
Only the integration of the last term is unequal to zero:
T T
1 1
e iϕ
2 2 0 0 ∫ ∫
≈ 1− ϕ2 = 1− γ 2 dt dt ¢G(t )G(t ¢ ) x (t ) x (t ¢ ) T T T T

∫ ∫ ∫ ∫
ϕ2 = γ 2 D0 dt dt ¢G(t )G(t ¢ ) t − t ¢ = 2 γ 2 D0 dt dt ¢G(t )G(t ¢ )(t ¢ − t ) = 2b ⋅D0
0 0 0 t
where T is the total duration of the diffusion gradients.
Using the identity with the famous b-value being defined by
1 1 1 T T
x (t ) x (t ¢ ) = ( x (t ))2 + ( x (t ¢ ))2 − ( x (t ) − x (t ¢ ))2
2 2 2 ∫ ∫
b = γ 2 dt dt ¢G(t )G(t ¢ )(t ¢ − t )
0 t
and the relations for free diffusion
For free diffusion, the Gaussian phase approximation
( x (t ))2 = 2D0 t ( x (t ¢ ))2 = 2D0 t ¢ ( x (t ) − x (t ¢ ))2 = 2D0 t − t ¢
1 2
− ϕ
e iϕ = e 2
we find the expression

T T is valid. Hence the signal drops as

ϕ2 = γ 2 dt dt ¢G(t )G(t ¢ ) ( ( x (t ))2 + ( x (t ¢ ))2 + ( x (t ) − x (t ¢ ))2 )
1
2 0 0 ∫ ∫ S(b ) = e − b ⋅ D0

The integration of the terms Thus, the free diffusion coefficient D0 can be evaluated by measuring
the signal S (b) with two b-values, for instance, with b = 0 s/mm² and
( x (t ))2 and ( x (t ¢ ))2 b = 1,000 s/mm².

The b-Value for Stejskal–Tanner Gradients

In this textbox, the signal drop for the pulsed gradient scheme in the case of free diffusion is evaluated. For this purpose, we first refor-
mulate the formula
⎧ 1 for 0 < t < δ
⎪ T T
G(t ) = G ⋅ ⎨−1 for Δ < t < Δ < δ
⎪⎩ 0 otherwise ∫ ∫
b = γ 2 dt dt ¢G(t )G(t ¢ )(t ¢ − t )
0 t
6
1.3 · Theoretical Aspects
19 1

of the »Free Diffusion and long Gradients: The b-Value« box. To Thus it follows by comparing the last two equations that
evaluate this expression, some rather tedious calculations are required. T
Using the expression for the wave-vector
∫
b = dt k (t )(k (t )
0

t T
In order to calculate the b-value, the wave-vectors k (t) are required.
∫
k (t ) = γ dt ¢G(t ¢ ) = − γ dt ¢G(t ¢ ) ∫ ⎧ t for 0 < t < δ
0 t
⎪
k (t ) = γG ⋅ ⎨ δ for δ < t < Δ
and the identity
⎩⎪δ − (t − Δ ) for Δ < t < Δ + δ
T T T t¢

∫dt ∫dt ¢h(t , t ¢) = ∫dt ¢∫dt h(t , t ¢) Thus,

0 t 0 0
δ Δ Δ +δ

for any function h (t, tc), we find

∫ ∫
b = γ 2G 2 dt t 2 + γ 2 G 2 dt δ 2 + γ 2 G 2 ∫ dt (δ + Δ − t )2
0 δ Δ
⎛1 1 ⎞
T T T T b = γ 2G 2 ⎜ δ 3 + δ 2 ( Δ − δ ) + δ 3⎟
⎝3 3 ⎠
∫ ∫
b = γ 2 dt dt ¢G(t )G(t ¢ )t ¢ − γ 2 dt dt ¢ G(t )G(t ¢ )t ∫ ∫ ⎛ 1 ⎞
0 t 0 t b = γ 2G 2 δ 2 ⎜ Δ − δ⎟
T t¢ T ⎝ 3 ⎠
= γ2 ∫dt ¢∫dt G(t )G(t ¢)t ¢ + γ ∫dt G(t )t k(t )
0 0 0
T T T

∫ ∫
= γ dt ¢k (t ¢ )G(t ¢ )t ¢ + γ dt G(t )t k (t ) = 2 γ dt G(t )t k (t )∫
0 0 0

Integration by parts yields

T
b = 2[k (t )t k (t )]0 − 2 dt k (t )(k (t ) + γ t G(t ))
∫
T

0
T T

∫
= 0 − 0 − 2 dt k (t ) k (t ) − 2 γ dt k (t )t G(t ) ∫
0 0

Long Gradients yield apparent Results

In order to understand the effect of elongated gradients, an approach but during the whole duration of the gradients. The signal attenuation
like the one in the box entitled »Q-space Imaging« is used. Again, the becomes
³
signal attenuation is ¢eiM² = ¢exp (iJ G (t) x (t) dt)². Consider the follow-
ing diffusion gradients
⎡
exp ⎢i γG ⋅
⎣ (∫ x(t )dt −∫
δ

0 Δ
Δ +δ
)
⎤
x (t )dt ⎥
⎦
⎧ 1 for 0 < t < δ
⎪ Introducing a generalized q-value q = JGG (usually the term q-value is
G(t ) = G ⋅ ⎨−1 for Δ < t < Δ < δ
⎪⎩ 0 otherwise used for temporally narrow gradients), the signal becomes

with the gradient amplitude G. The random walk can be depicted as follows:
⎡iq
E (q ) = exp ⎢
⎣δ
(∫ x(t )dt −∫
0
δ Δ +δ

Δ
)⎤
x (t )dt ⎥
⎦
The integral
1 δ

δ ∫0 x(t )dt
can be interpreted as the center-of-mass of the trajectory ranging from
x (0) to x (G). Introducing the centers-of-mass of the individual random
walks
1 δ 1 Δ +δ
x cm ,1 =
δ ∫0 x(t )dt and xcm,2 = δ ∫Δ x (t )dt ,

the signal becomes

S(q ) = exp[i γq ( x cm ,1 − x cm ,2 ) ] .

This expression bears a great similarity to the q-space expression in the

»Q-Space Imaging« box, as Mitra nicely pointed out [106]: »The echo
amplitude is the spatial Fourier transform of a ‘center-of-mass’ propaga-
The gradients are applied during the blue part of the trajectory. Thus, tor, which reduces to the usual diffusion propagator in the limit of zero
the diffusion process is not only sampled at initial and final position, pulse widths.«
 20 Chapter 1 · Introduction to Diffusion Imaging

1.3.4 The Diffusion Tensor
1 these directional dependencies. One important result is that the
In white matter and in muscle fibers, the restrictions are depend-
ent on the spatial orientation [11][12][14][15][27][28][37][58]
[101][112]. Parallel to the fiber direction, the particles can diffuse
more freely than perpendicular to it. Thus, a single diffusion coef-
ficient D is not sufficient to describe the diffusion process; instead,
three diffusion coefficients Dx, Dy, and Dz are required to describe
the diffusion along the three spatial directions. For instance, if the
fiber is oriented along the x-direction, then Dx is larger than Dy
and Dz. Dy and Dz can be, but do not have to be equal. These coef-
ficients are also time-dependent, but to increase the readability,
the explicit time-dependence is omitted in the notation.
. Fig. 1.12 shows a two-dimensional example illustrating
three above-mentioned diffusion coefficients are not sufficient to
describe the anisotropic diffusion process. They would be suffi-
cient if the orientation of the fibers was known, but in general,
this is not the case. Therefore, it is necessary to introduce three
additional parameters, which describe the orientation of the
fiber in space. One approach would be to introduce three angles
D, E, and T describing the spatial orientation of the fiber (exactly
three angles are necessary and sufficient to describe the orienta-
tion of a three-dimensional body, for example, a book, in space).

The Diffusion Tensor

If the diffusion is anisotropic, then the direction n of the diffusion gradi- restricted Diffusion…« box, we aim at a series expansion in G (t), thus
ent must be taken into account. Bold characters indicate vectors. Instead the normalized temporal gradient profile g (t) = G(t) – G (t – ') is used:
of the displacement x, the projection xn of the vector x along the unit ³
G (t) = nq/J ˜ g (t). With M = q x (t) ng (t) dt, the expression for ¢eiM²
vector n is the quantity of interest. Instead of ¢x2², it is ¢(xn)2² that must becomes
be considered (for simplicity, the two-dimensional case is regarded):
( )
1 2

( xn)2 = ( xnx + yny )2 = x 2 n2x + y 2 n2y + 2 xnx yny

e iϕ = 1 +
2 ∫
iq x (t )ng(t )dt + O(q3 )
1
= n2x x 2 + n2y y 2 + 2nx ny xy
2 ∫∫
= 1− q2 dt dt ¢ g(t )g(t ¢ ) ( x (t )n)( x (t ¢ )n) + (G 3 )
⎛ x2 xy ⎞ ⎛nx ⎞ 1 ⎛ x (t ) x (t ¢ ) x (t ) y (t ¢ ) ⎞
= ( nx ny ) ⎜
⎝ xy
⎟⎜ ⎟
y 2 ⎠ ⎝ ny ⎠ 2 ∫∫
= 1− q2 dt dt ¢ g(t )g(t ¢ )nT ⎜
⎝ x (t ) y (t ¢ ) y (t ) x (t ¢ ) ⎠⎟
n + O( q 3 )

⎛D xx D xy ⎞ ⎛nx ⎞
= 2t ⋅ (nx ny ) ⎜ = 2tnT Dn
⎝D xy D yy ⎟⎠ ⎜⎝ny ⎟⎠ The double integral can be performed similarly as in the »Short Gradi-
ents and restricted Diffusion…« box.
with
1 ⎛ x2 xy ⎞
x2 y2 xy e iϕ = 1− q2 nT ⎜ ⎟ n + O( q )
3
D xx = , D yy = and D xy . 2 ⎝ xy y2 ⎠
2t 2t 2t
_
The matrix D transforms like a tensor under rotations since ¢(xn)2² is a If the terms O (G3) are neglected, ¢eiM² becomes
scalar and since n transforms like a vector._
1 ⎛ x2 xy ⎞ − bnT Dn
We now derive the relationship between D and the measured signal. e iϕ = 1− q2 nT ⎜ ⎟ n = 1− q Δn Dn = 1− bn Dn = e
2 T T

The phase that a random walker acquires on his path x (t) = (x (t), y (t))T 2 ⎝ xy y2 ⎠

³
is M = J x (t) G (t) dt, with
This derivation shows that the diffusion tensor a is good concept, even
G(t ) = (G x (t ), G y (t ))T if the diffusion is not Gaussian. The description by the diffusion tensor
is always valid for small b-values, but when using large b-values, high-
Assuming that the gradient orientation does not change in time, er-order terms, described, for example, by the kurtosis tensor, must be
G (t) can be written as G (t) = nG (t). As in the »Short Gradients and taken into account.

Usually, however, a different approach is chosen. In order to
describe the orientation and strength of the diffusion process, the
so-called diffusion tensor is often used [11][12][14][101]. The
diffusion tensor is experimentally well detectable and is mathe-
matically convenient. In »The Diffusion Tensor« box, the diffu-
sion tensor is described mathematically, while an illustrative
_ of ellipsoids can be found in . Fig. 1.12. The
description by means
diffusion tensor D can be expressed by a 3×3 matrix, which is a
small table of numbers. For isotropic diffusion the diffusion
tensor is
⎛ D 0 0⎞
D = ⎜0 D 0⎟ (7)
⎜0 0 D⎟
⎝ ⎠
This is a very cumbersome notation in the case of isotropic dif-
fusion. The intent becomes clearer when the diffusion is aniso-
tropic. In the above example of a fiber aligned along the x-direc-
tion, the diffusion tensor becomes
⎛ Dx 0 0 ⎞
D = ⎜ 0 Dy 0 ⎟ (8)
⎜0 0 D ⎟
⎝ z⎠
The upper left element Dx describes the diffusion along the
x-direction, Dy and Dz describe the diffusion along the y- and
z-direction, respectively. This notation is useful, since the diffu-
sion coefficients along the different directions can be directly
read off.
1.3 · Theoretical Aspects
Finally, we consider the general case of a fiber that is oriented
arbitrarily and has different diffusion coefficients along different
directions. In this case, off-diagonal elements appear.
⎛ Dxx Dxy Dxz⎞
D = ⎜ Dxy D yy D yz⎟ (9)
⎜D D D ⎟
⎝ zx zy zz⎠
The tensor is symmetric: Dyx = Dxy, Dzx = Dxz, and Dzy = Dyz.
Consequently, the tensor has six independent elements (or six
degrees of freedom), which describe the strength of the diffusion
(three degrees of freedom) and the orientation of_ the tensor in
space (also three degrees of freedom). The matrix D itself appears
to be rather complicated, and, indeed, even experienced radiolo-
gists and physicists can hardly estimate the shape of the tensor
directly from the matrix. However, certain statements can be
made:
4 The upper left element is always the diffusion coefficient
along the x-direction. The lower right element is the diffu-
sion coefficient along the z-direction, and the central ele-
ment is the diffusion coefficient along the y-direction.
4 Imagine that a tensor (or a fiber) is not oriented along the
x-, y-, and z-direction (as in Eq. 8), but along some other
direction (as in Eq. 9). One can always choose a second
coordinate system, which is aligned like the tensor
(. Fig. 1.12). In that »tensor« coordinate system, the tensor
is always diagonal like in Eq. 8; the off-diagonal elements
are zero. Dx, Dy, Dz in that tensor coordinate system are
always the same, no matter how the tensor is oriented in
the »real-world« coordinate system. This reflects that the
diffusion perpendicular to one specific fiber (tensor coordi-
nate system) should always be identical, no matter how the
fiber is oriented in the real-world coordinate system.
4 Dx, Dy, Dz are named eigenvalues of the diffusion tensor.
They describe the diffusion parallel and perpendicular to
the fiber.
4 The directions of the coordinate axes of the tensor coordi-
nate system are named eigenvectors. The direction corre-
sponding to the largest eigenvalue is called principal eigen-
vector and is a good indicator of the fiber direction. The
corresponding eigenvalue is named D储.
4 Tensors of identical shape, but different spatial orientation,
have different elements Dxx, Dyy, Dzz, Dxy, Dxz, Dyz.
4 Knowledge of Dxx, Dyy, Dzz, Dxy, Dxz, Dyz always allows the
computation of eigenvalues and eigenvectors.
4 The mean diffusivity,
1 1
MD = (Dx + D y + Dz ) is equal to (Dxx + D yy + Dzz ).
3 3
It is a widely used parameter describing the strength of
the diffusion. It is not dependent on the tensor orientation.
In the literature, MD is often labeled as »apparent diffusion
coefficient,« which can lead to confusion.
4 A simple approach to describe the anisotropy of the diffu-
sion would be, for example, to divide the largest by the
smallest eigenvalue, for instance, the anisotropy A could be
defined by A = Dx/Dz.
21 1
4 Usually, however, more complicated expressions are used
[13][76][77][78][79]. Common parameters describing the
anisotropy are the fractional anisotropy
3 (Dx − MD)2 + (D y − MD)2 + (Dz − MD)2
FA =
2 Dx2 + D 2y + Dz2
and the relative anisotropy
(Dx − MD)2 + (D y − MD)2 + (Dz − MD)2
1
RA =
3 MD
These indices have several advantages over simpler parame-
ters like A. First, the eigenvalues do not have to be sorted. In
noisy data, erroneous sortings may occur leading to the so-
called sorting bias. Second, they are normalized. For ex-
ample, the FA is 0 for isotropic diffusion and 1 for extremely
anisotropic diffusion. Third, these indices use all three
eigenvalues. For instance, if Dz > Dx, then A is identical
for the two cases Dy = Dx and Dy = Dz, but using the FA the
two cases can be distinguished. Fourth, all three eigenvalues
are treated equally, no eigenvalue has superior importance
over the others.
4 Many other parameters derived from eigenvalues exist.
For example, the axial diffusivity AD is the average of the
two smaller eigenvalues.
4 For short diffusion gradients, the relationship between the
elements of the tensor to the displacements is similar to the
one in Eq. 1, for example,
xy tissue
x 2 tissue
Dxx = , Dxy = ,
2t 2t
etc. Strictly speaking, the tensor should be labeled »apparent
diffusion tensor« if elongated diffusion gradients are used,
but the »apparent« is usually not stated in practice.
In order to measure the diffusion tensor, at least seven measure-
ments must be performed: one measurement without diffusion-
weighting gradients and six measurements with diffusion-
weighting gradients, which are applied along six different direc-
tions. This is necessary for measuring the six independent ele-
ments of the tensor. The diagonal elements Dxx, Dyy, Dzz can be
calculated by measuring with a diffusion-weighting gradient
oriented along the x, y- or z-direction. For example,
Dxx = −ln (S / S0 ) / b (10)
if the diffusion weighting gradient is applied along the x-direc-
tion. Applying the gradient along the xy–direction (45° between
gradient direction and x-axis), a mixed term is measured:
1 1
−ln (S / S0 ) / b = Dxx + D yy + Dxy (11)
2 2
Thus, when knowing Dxx and Dyy from previous measurements,
the off-diagonal element Dxy can be calculated. In practice, more
elaborate tensor calculation techniques are often used, which
allow, for example, for the use of more than six diffusion gradient
directions (see, e.g., [26][33][59][78][101][142]).
 22 Chapter 1 · Introduction to Diffusion Imaging

plates in . Fig. 1.6. Thus, the time-dependent diffusion

1 coefficient converges toward zero:

x 2 L2
D(t ) = ≈ →0
2t 2t
a b
In open geometries, for example, in permeable cell walls or
spheres swimming in a fluid, the particles are not trapped
and it can be shown that

⎛1 β ⎞
D(t ) = D0 ⎜ + + O (t −3/2 )⎟
⎝ α D0t ⎠

c d Here, D is the so-called tortuosity and E is a geometry-de-

pendent constant. D is an important parameter for many
. Fig. 1.12 Visualization of the diffusion tensor. The diffusion tensor is
often visualized by ellipsoids. These ellipsoids are defined by the isosurface physical phenomena beside diffusion, like the conductivity,
of the Gaussian propagator function P0 (x, t) = c, where c is a real constant but the link to the geometry is not as straightforward as in
smaller than 1. a Isotropic diffusion. b Anisotropic diffusion, the diffusion in the short time limit. Note that recent works by Novikov
x-direction is larger than in y-direction: Dxx > Dyy. The fiber is aligned along and Kiselev indicate that the Et–1 attenuation term may
the x-axis. c The tensor has the same shape as the tensor in b, but is aligned
not be universal: This term may actually be a more general
along the primed, rotated coordinate system. In the primed coordinate sys-
tem, the tensor has the same elements as the tensor in b. In the unprimed function Et–a, where a is a real number reflecting the struc-
coordinate system, the tensor has off-diagonal element Dxy ≠ 0. d A two- tural ordering [115].
dimensional example: The two tensors are not distinguishable if the diffu- 4 Many relations also hold true for long diffusion gradients,
sion is only measured along the x- and y-direction; a third measurement for example, in the short time limit,
is required. In three dimensions, six directions must be used, since three
directions are not sufficient
⎛ I 4 S ⎞
Dapp (t ) = D0 ⎜1 − 3/2 D0t + O (t )⎟
⎝ I1 9 π V ⎠

1.3.5 On the Time Dependence of D (t), Here, I and I1 are simply real numbers that can be derived
the Link to Restrictions, and the Influence from the temporal shape of the diffusion gradients. Thus,
of the Gradient Profile the surface-to-volume ratio (and the parameters N, U and
R–1) can be measured using long diffusion gradients.
The main purpose of diffusion-weighted imaging is to gain struc- 4 Other relations are different for long diffusion gradients, for
tural information. Thus, a natural question is: What is the rela- example, in closed domains Dapp (t) does not converge with
tionship between the time-dependent diffusion coefficient and t–1, but with t–2 toward zero for long diffusion times.
the structure of the restrictions? Is it possible to infer information 4 To measure the diffusion in the short time limit is experi-
that goes beyond the intuitive arguments stated in . Fig. 1.6 mentally very challenging, since the required gradient am-
(which basically stated more restrictions, less diffusion)? And, plitudes are hardly realizable as the b-value scales with t3
indeed, there are certain statements that can be made. Here, we and thus becomes too small. One interesting feature of long
only reproduce these statements, but do not present detailed diffusion gradients is that the short time limit can be recov-
derivations, which can be found in the literature [6][34][46][52] ered, although the whole diffusion process is in the long
[60][62][73][83][107][126][127]. time limit. This can be achieved by applying oscillating
4 At short diffusion time, the time-dependent diffusion coef- cosine-shaped gradients [36][131]. Then, the characteristic
ficient is given by time scale is the frequency of the oscillations and not that of
the total duration of the diffusion gradients. The advantage
⎛ 4 1S 1S 1 S ⎞
D(t ) = D0 ⎜1 − D0t + tκ + tρ − tD0 R −1 + O(t 3/2 )⎟ is that the applicable b-value becomes larger. This technique
⎝ 9 πV 3V 6V 12 V ⎠
is applicable in animal scanners, but the available gradient
where S/V is the surface-to-volume ratio, N is the permeabi- amplitudes are usually not sufficient in human whole-body
lity of the cell membrane, U is the surface relaxation, scanners.
and R–1 is the mean curvature. Thus, by measuring D (t) 4 The shape of closed domains can be determined exactly
at several short diffusion times, these parameters can be (see the box entitled »The Link to Restrictions in closed
detected. Domains«) [22][86].
4 In the long time limit (t of), open and closed boundaries
must be distinguished. When the boundary is completely
closed, the particles are trapped and cannot translate much
farther than the typical length scale of the boundary L.
For example, L could be the distance between the reflecting
1.4 · Advanced Techniques: Fiber Tracking and Deviations from Mono-exponential Signal Decay
The Link to Restrictions in closed Domains
In this textbox, we consider the temporal gradient profiles
⎧ −1 for 0 < t < T ⋅ δ1
⎪ attenuation is
Gδ1, δ2 (t ) = G ⋅ ⎨δ1 / δ 2 for T ⋅(1− δ 2 ) < t < T
⎪⎩ 0 otherwisse
where G1 and G2 are dimensionless parameters. T is the total duration of
the gradient profile. As in the box entitled »Q-space Imaging,« Q-space
imaging gradients correspond to limG1oGoGG1,G2 (t). The other inter-
esting case is limG1oGoGG1,G2 (t).
First, we consider q-space imaging gradients. As in the »Q-space Imag-
ing« box, the measured signal attenuation E(q) is
E(q) = ∫∫dx1dx 2ρ( x1 )P( x1 ,x 2 , Δ )eiq( x − x )
1 2
Assume that the particles reside in a closed domain with homogene-
ous initial density U(x1). We now evaluate the signal attenuation in the
long time limit: Then, the initial particle position is not correlated to
the final particle position, and the particle resides at any position with
equal probability when the second gradient is applied. Thus:
E(q) = ∫∫dx1dx 2ρ( x1 )ρ( x 2 )eiq( x − x )
1 2
1.4 Advanced Techniques: Fiber Tracking and
Deviations from Mono-exponential Signal
Decay
1.4.1 Fiber Tracking
Since diffusion tensor imaging allows us to estimate the main
nerve fiber direction within a voxel, a natural extension is to re-
construct the course of whole nerve tracts [29][57][109][110]
[111][132][150]. . Fig. 1.13 shows a sketch of a brain with diffu-
sion tensor ellipsoids plotted in several brain voxels. The subject
is observing a beautiful red-leaved flower. This information is
caught by the eye, and passed to the visual brain cortex through
the optical tract. This tract is observable in the tensor ellipsoids:
they are elongated in tract voxels and reflect the main fiber direc-
tion. In the sketch, the diffusion is isotropic in voxels not contain-
ing the optic tract (the smiley has only one nerve tract, which is
of course different in human brains).
. Fig. 1.13 illustrates a straightforward approach to recon-
structing tracts from the tensor ellipsoids, the so-called FACT
algorithm (fiber assignment by continuous tracking) [109][150].
The tract is started in one voxel and it follows the orientation of
the ellipsoid in this voxel until it hits the border of the voxel.
Then, this process is iterated in the adjacent voxel. Usually sev-
eral starting points – which are also called seeds – are chosen. For
example, in . Fig. 1.13, the tract starting in the center of the initial
voxel does not reach the end of the optic nerve tract, since the
tract is interrupted when it reaches a voxel with isotropic diffu-
sion. The other tract, which starts in the upper part of the initial
voxel, reaches the visual brain cortex. Thus, by choosing several
seeds instead of only one seed, the likelihood that the actual tract
is reconstructed is increased.
23 1
The shape of the domain can be described by the pore space function
F(x), which is 1 inside the domain and 0 outside the domain
(U(x) = F(x)/V, with V being the volume of the domain). Thus the signal
1
E (q) =
V2 ∫ ∫
dx 1χ( x 1 )e iqx1 dx 2 χ( x 2 )e − iqx 2 = χ * ( q )χ ( q ) = χ 2 ( q )
Hence, it is possible to measure the power spectrum of the pore space
function, but, as in X-ray scattering experiments, for example, the
inverse Fourier transform cannot be performed since the phase infor-
mation is lost. The quantity χ( q ) is the analogue to the form factor.
Second, we consider the temporally asymmetric gradient profile. As
described in the box entitled »Long Gradients yield apparent Results,«
the first, long gradient imprints a phase that is identical to that of the
center-of-mass of the particle trajectory. In the long time limit, the
center-of-mass of the trajectory becomes identical to the center-of-
mass of the domain. Hence, E(q) becomes
∫
E ( q ) = exp(iqx cm ) dx 2 χ( x 2 )e − iqx 2 = exp(iqx cm ) χ ( q )
Thus, the pore space function can be determined directly by inverse
Fourier transformation. This experiment is essentially an imaging ex-
periment in disguise.
In . Fig. 1.13, an important property of fiber tracking results
can be appreciated: the strong dependency on the chosen algo-
rithm and its parameters. The termination criterion of the algo-
rithm depicted in . Fig. 1.13 is to stop the tract if an isotropic
voxel is reached. However, if the tract were allowed to maintain
its direction in an isotropic voxel under the condition that the
next voxel is anisotropic, then the broken fiber would make it to
the end of the optic tract. In practice, of course, voxels are usu-
ally not perfectly isotropic, so one would instead choose a certain
anisotropy threshold to terminate the fiber. The final tracking
result will in any case depend strongly on this threshold. As there
is often no unique and perfect choice of algorithm and parame-
ters, the obtained results should not be regarded as perfect
ground truth. Nonetheless, it was shown in a series of works that
the main fiber tracts can be reconstructed correctly.
Since the initial works on fiber tracking, many fiber tracking
algorithms and strategies have been developed. Because a de-
tailed review of these techniques is beyond the scope of this in-
troduction, we only present an in-depth description of the algo-
rithm used in 7 Chap. 3.
1.4.2 The Tracking Algorithm Used in this Book
In this section, the tractography algorithm that was used to gen-
erate the fiber tracking results in 7 Chap. 3 is described. It was
originally introduced by Reisert et al. [121], and is a so-called
global tracking algorithm [80][100]. This algorithm ranked first
in an evaluation study by Fillard et al. on the performance of
tractography algorithms [40].
The tracts are made up of small building blocks. The algo-
rithm runs through many iterations. At each iteration, it may give
 24 Chapter 1 · Introduction to Diffusion Imaging
a b
. Fig. 1.13 a Sketch of a person looking at a flower. The visual information travels through the optical tract to the visual cortex. The tract orientation can
be estimated by measuring the diffusion tensor ellipsoids, which are anisotropic in the nerve tract. b The tract itself can be reconstructed by following the
direction of the ellipsoids, but starting from different seed points yields different tracts. While some of the reconstructed tracts arrive at the optical cortex,
this is not necessarily true for all tracts. Major fiber tracts in the brain can usually be reconstructed correctly (7 Chap. 3), but one should be careful not to
interpret the reconstructed tracts as a perfect representation of reality
birth to a new building block, placing it at an arbitrary position
xi with a random orientation ni and endpoints x–i and x+i at
x±i = xi ± lni (i is the index of the building block). The length l is a
free parameter and was set to 3.75 mm in the datasets in this atlas.
Instead of giving birth to a new building block, the algorithm
can also shift and rotate an existing building block, or, to avoid
overcrowding, it can just put one to death. Moreover, it has a
fourth option: it can glue the building blocks together to form
a new fiber, while tearing apart some old fibers. The information
about the union of building blocks that form a fiber is stored in
tuples E = (i1, i2), where i1 and i2 are the indices of the connected
endpoints (e.g., of x1– and x+5). The fifth option of the algorithm is
to optimize the fibers by shifting one element of one fiber such
that it fits better into the fiber. All these events, birth, death, shift,
fiber generation, and fiber optimization, happen with a certain
probability at each iteration: pbirth = 0.25, pdeath = 0.05, pshift = 0.15,
pfiber = 0.45, and popt = 0.1.
In order to achieve a reasonable tracking result, the above-
mentioned steps are only tentative and the generated tentative
fiber configurations are accepted or rejected based on probabil-
istic criterions. When starting the algorithm, most tentative
fiber configurations are accepted. While the algorithm proceeds,
the acceptance rate is slowly decreased until the fiber configura-
tion is completely frozen (indeed this process is determined by
a parameter T called temperature).
The probabilistic criterion used here needs two ingredients:
One to yield beautiful, long tracts, and a second one guaranteeing
that the obtained tracts match the acquired data. The quality of
both is described by »energy« parameters Eint and Eext. A better
result is represented by a smaller energy.
First, we consider the internal energy Eint, which shall control
the connection behavior. The endpoints of connected building
blocks should not be separated too far. Thus, a penalty potential
U (i1, i2) is introduced:
U (i1, i2 ) =
1
( 2
xi + ai1 lni1 − x + xi2 + ai2 lni2 − x
l2 1
2
)−L
_
Here, x = (xi1 + xi2)/2 is the midpoint of the line connecting xi1
and xi2, and L is a free parameter of the algorithm that was set to
L = 140.625. Large L yield many fibers, while small L prevent
fibers from being connected. The quantity Di describes whether
the tip (Di = 1) or the shaft (Di = –1) of the arrow is connected.
U(i1, i2) is smallest if the building blocks lie on the same straight
line.
The internal energy Eint is the sum of the penalty potentials
of all connected fibers (unconnected fibers do not interact):
Eint = ∑ U (i1, i2 )
connections
Second, we consider the external energy Eext, which describes
how well the reconstructed fibers match the acquired diffusion-
weighted signals. To do so, we must first define the signal that a
single building block is supposed to yield. While there are many
possibilities, we choose the following prototype signal:
2
si (x , n) = e −bD& (n e − x − xi /σ2
T n )2
i
Where x is the position in the brain where the signal shall be
calculated, n the diffusion direction for which the signal shall be
calculated, and – as before – xi and ni are the position and orien-
tation of the building block. The first term describes the signal
drop in the diffusion tensor approximation (see 7 Sect. 1.3.4),
and the diffusion tensor is shaped like a »stick.« This means that
the diffusion parallel to the fiber is controlled by the parameter
D储, which is the principal eigenvalue of the tensor, while the dif-
fusion perpendicular to the building block is assumed to be zero.
D储, is another free parameter of the algorithm, it was set to 25/b.
This is a rather large value, which was, however, empirically
found to be useful [121]. While it may appear bold to model the
tensor of a building block by a highly anisotropic stick model,
a microscopic point of view can yield some justification: since not
all building blocks are aligned along the very same direction, the
signal of all building blocks can approximately reflect the truly
measured signal. Moreover, the stick model is really straight-
forward: we are left with only one free parameter. The second
term in si (x, n) basically describes the spatial profile of the signal
that the building block generates via the free parameter V, which
was set to 0.5 mm.
1.4 · Advanced Techniques: Fiber Tracking and Deviations from Mono-exponential Signal Decay
25 1
The signal of all building blocks is just the sum of the where Eold is the total energy of the actual fiber configuration,
individual signals (all building blocks have equal weight w): and Enew is the total energy of the tentative fiber configuration.
poldonew and pnewoold represent the probabilities pbirth, pdeath, etc.
S(x , n) = w ∑ si (x , n) For example, if a particle is killed, then poldonew = pdeath and
building blocks
pnewoold = pbirth. The tentative configuration is accepted with
w is a free parameter that controls the number of building blocks probability R (if R > 1, then it is automatically accepted).
that are needed to replicate the measured signal. The larger w is, The temperature T was decreased exponentially from 1 to
the fewer building blocks are required. 0.001 in approximately 5×109 iterations. The parameter w was set
This expression for S (x, n) is already a nice ansatz, but we to 0.0006. An in-depth discussion of the influence of specific
take one further step: The aim of the whole approach is to model parameter choices can be found in [121].
the anisotropic part of the diffusion (which inherently signifies
that we are not interested in the isotropic part). Therefore, the
isotropic part of S (x, n) is subtracted 1.4.3 Intravoxel Incoherent Motion
1
S ¢(x , n) = S(x , n) −
4π
∫d Ω S(x, n) The blood flowing in the capillary bed is not moving coherently,
but along different directions. This incoherent motion resembles
Where ³ d: represents the integration over all solid angles. The diffusional motion, and in fact it is observed experimentally that
same procedure is applied to the measured signals Sm (x, n) this capillary blood motion has a similar effect on the diffusion-
weighted signal as a true diffusion process. This is most visible in
1
S ¢m (x , n) = Sm (x , n) −
4π
∫d Ω Sm (x, n) organs of short tissue T2 relaxation time, like in the liver and
pancreas, because of the increased relative weight of the blood
Sc(x, n) and Scm (x, n) are called the meanless signals. It remains to compartment [94]. The typical distances that particles translate
define an expression for the external energy. Here it is: owing to the capillary flow are much larger than owing to diffu-
sion at the typical time scales of NMR diffusion experiments in
2
Eext = S ¢ − Sm′ clinical scanners. Therefore, a blood flow-related pseudo-diffu-
L2
sion coefficient D* is measured, which is larger than the apparent
where 储 · 储L2 is the L2-norm. The total energy E is diffusion coefficient by more than a factor of 10. The most wide-
ly used model, as originally proposed by le Bihan, assumes that
E = Eint + Eext two well-separated compartments – one blood compartment and
one tissue compartment – are present [90][91]:
A low value of the total energy describes a good fiber configura-
tion. A high value describes a bad fiber configuration. One S = S0 / fe −bD * + (1 − f )e −bDapp) (12)
step remains: We need a procedure to decide whether the tenta-
tive fiber configuration is accepted or rejected. Therefore, a num- Here, f is the perfusion fraction. . Fig. 1.14 illustrates the intra-
ber R is calculated voxel incoherent motion (IVIM) effect in the example of the
pold →new healthy pancreas. Note that the IVIM effect occurs at very small
E = e −( Enew − Eold )/T p
new →old b-values (<50 s/mm²). While the effect is very important when

a b
. Fig. 1.14 a Logarithmic signal decay in the pancreas of six healthy volunteers with echo time TE=50 ms. Because of the incoherent blood flow, the signal
decays bi-exponentially. After suppression of the blood signal by means of an inversion recovery pulse, the signal decays mono-exponentially. This indi-
cates that the blood compartment plays an important role. b Effect of the echo time on the signal decay. Since pancreatic tissue has a shorter transversal
relaxation time than blood, the relative weight of the blood signal increases and the IVIM effect becomes more pronounced
 26 Chapter 1 · Introduction to Diffusion Imaging

. Fig. 1.15 Diffusion kurtosis tensor data of the brain of a healthy subject. a Projection of the kurtosis tensor along the z-axis, which is orthogonal to the
image plane. The kurtosis can be interpreted as a measure for the deviation of the diffusion from free, Gaussian diffusion: The kurtosis is largest in the cor-
pus callosum (where the nerve fibers are densely packed) and smallest in the ventricles (where the diffusion is rather free). b Kurtosis tensors for the white
box depicted in a, which contains the corpus callosum. Because of the diffusion restrictions, the kurtosis is largest perpendicular to the fiber direction. In
the liquor, the tensors are much smaller, which reflects the reduced diffusion restrictions. (Modified from [84])

measuring diffusion in the pancreas and liver [93][94][99][149],
it can usually be neglected in brain tissue [30].
1.4.4 Kurtosis Tensor Imaging
At large b-values, the influence of the kurtosis on the measured
signal becomes relevant, since the Gaussian phase approximation
is not valid (7 Sect. 1.3.3 and . Fig. 1.9). To describe the spatial
profile of the apparent kurtosis, a kurtosis tensor can be intro-
duced in analogy to the diffusion tensor (7 Sect. 1.3.4) [65][66]
[67][98]. The kurtosis tensor is more complicated than the diffu-
sion tensor: It has 15 independent elements and is a rank-4
tensor. This means that it cannot be represented by a two-dimen-
sional array of numbers like the diffusion tensor, but by a four-
dimensional array. Moreover, the tensor cannot be represented
by simple ellipsoids, since it can take more complicated shapes.
We will not go into details here, but only present some kurtosis
tensor data acquired in the healthy human brain (. Fig. 1.15).
1.4.5 High-angular-resolution Diffusion Imaging
The diffusion tensor is well suited to describe single fiber strings.
However, the six available degrees of freedom are not sufficient
to describe complex fiber configurations like fiber crossings
(. Fig. 1.16), for which more elaborate models and techniques are
required. The most general approach
_ is to measure the complete
voxel-averaged propagator P (x, ') with q-space imaging (see
»Q-space Imaging« box) [22][23][31]; the three-dimensional
version of q-space imaging is often called diffusion spectrum
imaging (DSI) [135][143][144][145]. This voxel-averaged propa-
gator contains detailed information about the underlying fiber
structure and was shown to reveal the geometry of crossing fibers
adequately. The required q-space data points are usually acquired
by sampling q-vectors on a Cartesian grid and then applying the
fast Fourier transform (FFT). In theory, for a completely sampled
_grid and narrow diffusion gradient pulses, this technique yields
P (x, '), but in practice, the duration of the applied gradient
pulses is often not short enough. In that case, only the propagator
of the centers-of-mass of the trajectories that the particle moves
along during the gradients is accessible (see the »Long Gradients
yield apparent Results« box). Thus,_long diffusion gradients yield
a somewhat »blurred« estimate of P (x, '), but can still yield suf-
ficient information to estimate fiber crossings. The most severe
limitation of the DSI technique is the large acquisition time need-
ed to sample the three-dimensional grid. To make the acquisition
time tolerable, the resolution in q-space has to be decreased sig-
nificantly in practice.
In order to acquire only the most relevant information and
thus achieve a reasonable trade-off between acquisition time
and information content, several high-angular-resolution diffu-
sion imaging (HARDI) schemes have been introduced. These
methods usually sample the q-space on one or more circular
shells and rely on model assumption to estimate missing data in
the q-space. To reach sufficient angular resolution, more than six
diffusion directions must be acquired; often investigators use 64
or even more diffusion directions. HARDI imaging is especially
well suited to resolve complex fiber configurations like fiber
crossings. For single fiber strings, it is often more apt to use con-
ventional diffusion tensor imaging since the acquisition time is
largely reduced. Moreover, smaller b-values may be used. Using
smaller b-values is advantageous, since the signal-to-noise (SNR)
ratio in the diffusion-weighted images is larger (7 Sect. 1.5.1).
As mentioned before, the description of the signal decay
by the diffusion tensor is very appropriate for small b-values
1.4 · Advanced Techniques: Fiber Tracking and Deviations from Mono-exponential Signal Decay
. Fig. 1.16 Diffusion-weighted signal and diffusion tensors for three fiber
bundle configurations. At small b-values, the diffusion-weighted signal
permits us to estimate the fiber direction in single-fiber configurations.
However, two crossing fibers can hardly be distinguished (signal b1). If
large b-values are used, then the two fibers become clearly separable in the
diffusion-weighted signal (signal b2). In contrast, the diffusion tensor ellip-
soids represent only the single fiber bundles correctly (tensor 1 and 2), but
not the fiber crossings: An ellipsoid just cannot take a shape that represents
a crossing (tensor 3). A straightforward approach to model the crossing is
to assume that the signal is composed of the sum of signals of the individu-
al fibers; in the plotted example, both fibers are assumed to contribute
equally to the signal. In addition, it is further assumed that the single fibers
can be described properly by the corresponding diffusion tensors. Then, by
appropriate experiments and fitting procedures, tensors 1 and 2 can be re-
trieved from the crossing fiber signal (signal b2)
(in biological tissue, b-values less than 1,000 s/mm² can usually
be regarded as small; see also »The Diffusion Tensor« box and
. Fig. 1.16). However, since the information contained in smaller
b-value measurements is insufficient to describe fiber crossings
(it is too blurry) larger b-values (e.g., b = 3,500 s/mm²) must be
applied and more elaborate models must be fitted to the data
(a tensor ellipsoid cannot take the shape of a crossing, . Fig. 1.16).
The most straightforward HARDI technique is to model
fiber crossings by two or several fibers, and to assume that each
_of these fibers can be described properly by its diffusion tensor
Di, where i is the index of the fiber (. Fig. 1.16) [1][4][81][116]
[128][140]. In practice, it is often assumed that only one or two
fibers are present in a voxel. Thus, this model requires one to
determine 13 free parameters: The relative contribution of the
two fibers to the signal and two times _ six tensor elements. While
the assumption that the tensors Di describe the signal at high
b-value properly may be doubtful (7 Sect. 1.4.4), it was shown
that fiber crossings can be detected adequately with this ap-
proach.
While model-based approaches are very intuitive, the most
widely applied technique for HARDI reconstruction is a non-
parametric approach called Q-ball imaging (QBI) [47][113][117]
[119][139]. QBI does not sample the complete q-space, but only
a spherical shell of it. It aims at the reconstruction of the so-called
orientation distribution function (ODF). The ODF represents
the probability that the diffusing particle ends up on a straight
27 1
. Fig. 1.17 Examples of orientation distribution functions (ODFs). ODFs
represent the probability that a particle ends up on a straight line along a
certain direction: A direction with a long ODF symbolizes a high probability.
Since the particle in most cases ends up on a line parallel to the fiber direc-
tion, the ODF is longest along the fiber direction. The real advantage of
ODFs over diffusion tensors becomes apparent in crossing fiber regions:
Because particles mostly end up on lines parallel to one of the fiber direc-
tions (rather than in between them), both fiber directions can clearly be
appreciated in the ODF
line pointing along the unit vector n [139]. Examples of the ODFs
of single and crossing fibers are depicted in . Fig. 1.17, where a
high probability is visualized by a long ODF. Particles tend to end
up on straight lines that are parallel to the fiber direction, and
hence the fiber direction is reflected by the ODF (as by the diffu-
sion tensor). The real advantage of using ODFs is observable in
a fiber crossing: the particles mostly end up on straight lines
parallel to one of the two fiber directions (and not in between),
and hence the orientation of both fibers can clearly be appreci-
ated in the ODF.
One fascinating aspect of QBI is that the ODF along a certain
direction, say along the x-direction, is not calculated using a dif-
fusion direction parallel to the x-direction, but using the perpen-
dicular diffusion directions lying in the y–z-plane (for details see
the »Q-Ball Imaging« boxes). In practice, since the directions are
often not perfectly orthogonal, some kind of interpolation is usu-
ally employed.
While the aim of the original QBI method was to measure the
probability that a particle ends up on a straight line, several more
recent publications proposed to measure the probability that
a particle ends up within a certain solid angle [2][3][8], this
method is often named constant solid angle (CSA) reconstruc-
tion. The difference is illustrated in . Fig. 1.18: Classic QBI
focuses, for example, on the question: How probable is it that the
particle ends up on the x-axis? In contrast, CSA-QBI asks: How
probable is it that the particle ends up within a certain deviation
of the x-axis?
Why is the CSA reconstruction advantageous? There are
several reasons. One disadvantage of the original QBI technique
is that it is intrinsically not normalized. This problem is usually
tackled by some kind of artificial normalization, for instance,
to minimal and maximal ODF values. For example, the ODFs
depicted in . Fig. 1.17 are not the ODFs that emerged from the
reconstruction. These raw ODFs are much more spherical and
were min–max normalized during post-processing.
A further HARDI technique is spherical deconvolution [137]
[138]. The idea is to interpret the signal as the weighted sum of
different fiber compartments present in a voxel (. Fig. 1.19). The
signal is interpreted as a convolution of the prototype single-
fiber signal with the probability distribution of fiber directions.
The prototype single-fiber signal is typically not chosen a priori
 28 Chapter 1 · Introduction to Diffusion Imaging

. Fig. 1.18 Comparison of the original QBI technique and the QBI technique with solid angle reconstruction.

but learned from the data themselves. The main limitation of this
approach is the susceptibility to noise and the assumption that
one prototype signal can represent all fiber populations found
within the brain (which might not hold true in the periphery
or in diseased regions). Persistent angular structure (PAS) is
another closely related, less restrictive technique [64].

. Fig. 1.19 Spherical deconvolution. For this technique, it is assumed that

the signal in a voxel is composed of the signal of several single fibers. All fib-
ers are assumed to have the same single-fiber signal (left). With this assump-
tion, the fiber distribution is calculated from the voxel signal

Q-Ball Imaging

Here, we follow Tuch [139] and prove the fascinating fact that the ODF with
∞
∫−∞dz P (r , θ, z , Δ )
∞
1 Pz (r , θ, Δ ) =
ϕ(n) =
Z0 ∫
P (nr, Δ )dr
Now the Jacobi–Anger expansion
n=∞
which describes the probability that a particle ends up on the half line
[0,nr o
e iqr r cos( q−J ) = ∑ i n Jn (qr r )ein(q−J )
_ f], can be estimated by measuring the diffusion perpendicular n=−∞
to n. P (nr, ') is the voxel-averaged propagator describing the probabil-
ity that a particle translates by nr in the time interval ' (see »Q-space
is used, where Jn is the n-th order Bessel function of the first kind:
Imaging« box). Z is a normalization constant, which is necessary to en-
2π n=∞ 2π ∞ 2π
sure the unit mass of the ODF.
In the »Q-space Imaging« box, we derived the relationship ∫0 dJE (qr ,J , 0 ) = ∑ ∫0 dJ ∫0 r dr ∫0 dq Pz (r , q, Δ )i n Jn (qr r )e in( q−J )
n=−∞

∫
E(q ) = dxP ( x , Δ ) − e iqx
Performing the --integration yields
n=∞ ∞ 2π 1 − i 2 πn
between
_ signal attenuation E(q) and voxel-averaged diffusion propaga-
tor P (x, ') assuming narrow diffusion gradients. In this textbox, we will
∑ ∫0 r dr ∫0 dθ Pz (r , θ, Δ )i n Jn (qr r )e in θ
− in
(e − 1)
n=−∞
use the three-dimensional generalization using cylindrical coordinates
r, T and z:
Since limno0 (e–i2Sn – 1)/n = –2Si and (e–i2Sn – 1) = 0 for n > 0, we finally
∞ 2π ∞ find
E (qr , J, qz ) = ∫0 r dr ∫0 dq ∫−∞dz P (r , q, z , Δ )ei (q z +q r cos(q−J ))
z r

2π ∞ 2π

Without loss of generality, it is assumed that n points along the z-direc- ∫0 dJE (qr ,J , 0 ) = 2π ∫0 r dr ∫0 dq Pz (r , q, Δ ) J0 (qr r )

tion. We now investigate the following circle integral (assuming that n

Thus, the left circle integral gives a good approximation to M(n), except
points along the z-direction)
that it does not yield the projection on a thin line, but an estimate that
2π 2π ∞ 2π ∞ is blurred by the Bessel function.
∫0 dJE (qr ,J , 0 ) = ∫0 dJ ∫0 r dr ∫0 dq ∫−∞dz P (r , q, z , Δ )eiq r cos(q−J )
r

2π ∞ 2π
= ∫0 dJ ∫0 r dr ∫0 dq Pz (r , q, Δ )e iqr r cos( q−J )
1.5 · Practical Aspects
Q-Ball Imaging: continued
For a b-value of 3,500 s/mm², using the relation for narrow gradients
b = q2' and assuming ' = 100 ms, one finds q ≈ 20 μm–1. The first root
of the Bessel function J0 (qrr) is approximately at qrr = 2.4, thus the
length scale of the blurring is about ±(2.4/20) μm = ± 0.12 μm.
1.5 Practical Aspects
1.5.1 b-Value, Averaging
One important issue in diffusion-weighted imaging is the b-value
that should be applied. In the case of isotropic diffusion at high
SNR, the relationship bDapp ≈ 1.1 should be approximately ful-
filled for the detection of Dapp (. Fig. 1.20). A typical value for
Dapp in tissue is 1 μm²/m², and thus b should be approximately
1,000 s/mm². b-Values that deviate strongly from this optimal
value lead to imprecise Dapp-values. In the case of anisotropic
diffusion, the above statement (bDapp ≈ 1.1) is not so definite, but
a good approximation is b . MD ≈1.1.
It is very important to note that the above statements are
only valid if the SNR is large, which means it should be roughly
larger than three. The large SNR requirement has to be fulfilled
in the S0-image, but, even more importantly, also for all voxels
and every diffusion direction in the diffusion-weighted images.
This is of importance, since, even if the SNR is high in the
S0-image, it could be low in the diffusion-weighted image due to
a large apparent diffusion coefficient, or due to a high diffusion
anisotropy. If the SNR is too low in the diffusion-weighted image,
the mean signal is erroneously increased because of the back-
ground noise, resulting in too-small estimates of the diffusion
coefficient (. Fig. 1.21 and . Fig. 1.22). Thus, it is indispensable
to check the SNR manually or automatically in the raw data
before quantitative values are extracted. If the SNR is too small,_
then inaccurate, in fact, arbitrary values are measured for Dapp, D
or MD.
How to proceed if the SNR is too low? There are basically five
choices. First – which is the option of choice when the data have
already been acquired – skip the low SNR voxels and evaluate only
those voxels of sufficient SNR. This yields accurate values for the
29 1
Since a typical diffusion length at ' =100 ms is 10 μm, the blurring
can be regarded as rather negligible. And indeed, several investigators
proved that crossings can properly be delineated. In order to reduce
the blurring, higher q-values can be applied.
In practice, the reconstruction is performed using linear matrix algo-
rithms. Recent publications often choose spherical harmonic basis
functions for an analytic representation of the signal and M(n). Spherical
harmonics allow for a very compact, parameterized representation of
spherical functions, avoid several numerical interpolation steps in the
reconstruction, and feature straightforward regularization using the
Laplace-Beltrami operator.
evaluated voxels, but it should be kept in mind that a certain bias
might be introduced when a heterogeneous region is evaluated
where, for example, the high Dapp-voxels are skipped. Second, use
smaller b-values. This might come at the cost of a decreased preci-
sion, but it yields accurate results. Third, when the diffusion is
anisotropic, one could avoid directions parallel to the fiber direc-
tion. This is especially favorable if the structure of interest has
a uniform orientation, like the spinal cord (. Fig. 1.21). Fourth,
increase the SNR, for instance, by decreasing resolution and/or
increasing slice thickness, or avoid reducing the field of view
(FOV) in phase direction (7 Sect. 1.5.2). It should be kept in mind
that especially high-resolution imaging (e.g., less than 2.5 μm
isotropic) is prone to low SNRs. Thus, when high-resolution
images are to be acquired, it is especially necessary to check the
SNR in the diffusion-weighted images. Fifth, a range of b-values
can be acquired, and the estimation of the diffusion coefficient
can be improved by introducing a noise parameter.
The sixth choice is no choice: it should not be attempted to
increase SNR by averaging the diffusion-weighted images, at
least not if no elaborate phase correction technique is used. In
order to understand this, one must have a closer look at the effect
of diffusion-weighting gradients. The intention of diffusion-
weighting gradients is to imprint a phase on moving particles.
This microscopic water motion is superimposed by a second
motion, the tissue pulsation. The heart pumps blood intermit-
tently through the arteries into the tissue, causing a bulk tissue
pulsation. Consequently, the diffusion-weighting gradients cause
an additional phase, which depends on the pulsation. Therefore,
the phase of the diffusion-weighted images is random (ECG trig-
gering may help but does not work perfectly). This random phase
is not a problem if single-shot techniques are used and if the
magnitude of the signal is averaged instead of the complex signal,
but otherwise the averaging yields unstable results (. Fig. 1.22).
 30 Chapter 1 · Introduction to Diffusion Imaging

a b

c d
. Fig. 1.20 Fit stability of the diffusion coefficient at high signal-to-noise ratios (SNRs). a At high SNRs, the noise is approximately Gaussian and the
standard deviation is equally large for all data points. b Plotting the signal decay logarithmically, the standard deviation is larger at small signals. c When
using a two-point fit to determine the slope – and therewith the diffusion coefficient – it is unfavorable to use two small b-values, since the fit becomes im-
precise. d Using b-values of too-large separation also yields unstable fitting results. Therefore, it is optimal to use medium b-values, the first b-value should
be zero, and the second should be chosen such that b ≈ 1.1/Dapp. The high b-value image should be averaged about three times as often as the small b-val-
ue image. The dependence on Dapp is important: Tissues with large apparent diffusion coefficients should be measured with smaller b-values

When averaging the magnitude of the image (which is done by
manufacturer-provided echoplanar sequences), the precision
increases, but the accuracy remains unchanged.
1.5.2 Sequences
The workhorse of diffusion-weighted imaging is the echo planar
imaging (EPI) sequence with a spin echo diffusion weighting
(. Fig. 1.23). The EPI sequence combines several advantages that
are favorable for DWI:
4 It is fast and many images can be acquired quickly.
4 It is a single-shot technique, thus no elaborate phase correc-
tion techniques must be employed (the phase is unstable
because of tissue pulsation during the diffusion-weighting
gradients; . Fig. 1.22).
4 It is stable and reliable.
4 It is widely available on clinical MR scanners.
EPI also has some drawbacks:
4 An inhomogeneous main magnetic field causes image
distortions. This effect is, for example, very prominent close
to the eyes, or close to the spinal cord.
4 The maximal image resolution is limited by transversal T2
relaxation. Usually the achievable image resolution is not
much larger than 128×128 voxels.
4 PI/2 ghosts may be present.
There are several possibilities to reduce the image distortions:
4 50% reduction of the FOV in phase direction at constant
resolution o 50% less distortion, but 30% decreased
SNR.
1.5 · Practical Aspects
31 1

a b
. Fig. 1.21 Systematic underestimation of the diffusion at low SNRs. a If the diffusion-weighted signal becomes as small as the background noise, then
a fit to that signal (straight line) yields too-low estimates of the slope, and hence of the diffusion coefficient. b In this regard, the directional dependency
of the diffusion must be taken into account when the diffusion is anisotropic. Especially directions that lie parallel to the fiber direction are prone to sys-
tematic errors. If the investigated structure has one main fiber direction – e.g., the spinal cord – then the applied diffusion directions can be adapted in
order to avoid directions parallel to the fiber. Otherwise, the SNR must be increased, the b-value must be reduced, or the noise must be accounted for in
the fitting routine

a b c d

. Fig. 1.22 a The transverse magnetization M is composed of a component in x-direction Mx and of one in y-direction My. b The measurement of the trans-
verse magnetization in real MR scanners is in practice disturbed by noise, and the measured distribution of Mx and My is usually approximately Gaussian. In
diffusion-weighted NMR, two special aspects must be considered: c First, owing to cardiac-driven tissue pulsation, the phase of the magnetization becomes
instable, and the measured magnetization lies approximately on the plotted circle. This is understandable: the diffusion-weighted gradients are designed
to imprint a particle phase if a particle translates a microscopic displacement (. Fig. 1.7). Tissue pulsations occur on a much larger length scale, such that
the diffusion-weighted image bears a random phase. If the signal is acquired with a single-shot technique (e.g., echo planar imaging), this does not represent
a problem, since usually only the magnitude of the signal is required to calculate the diffusion coefficient. However, if multi-shot techniques are to be used,
the different k-space segments bear different phases, resulting in strong image artifacts if no elaborate correction technique is used. d Second, if the signal
becomes too small because of the diffusion weighting, then the distribution of the signal magnitude is no longer distributed like a Gaussian function.
A Gaussian function would require that negative magnitudes could occur, which is impossible for the magnitude, or the length, of M. Even if the true signal is
almost zero, the measured signal is still larger than zero (and equal to the background noise)

4 100% increase of imaging bandwidth (the bandwidth is the
inverse of the readout gradient duration) o 50% less distor-
tion, but 30% decreased SNR.
4 Parallel imaging with acceleration factor 2 o 50% less dis-
tortion, but at least 30% decreased SNR and possibly recon-
struction artifacts at high acceleration factors.
4 But: 100% increase of resolution with constant FOV and
constant imaging bandwidth → 0% less distortions.
Thus, the reduction of distortions always comes at the cost of
reduced SNR. In general,
_ reducing the distortion by a factor of 2
reduces the SNR to 1/—2.
Besides the EPI sequence, several other diffusion-weighted
imaging sequences are described in the literature. Some of these
techniques are provided by major manufacturers, but many are
only in-house developments. In . Tab. 1.1, some of these se-
quences are referenced.
 32 Chapter 1 · Introduction to Diffusion Imaging

a b
. Fig. 1.23 a Schematic timing table of a spin echo diffusion-weighted echo planar imaging sequence. In this example, the diffusion-weighting gradients
(light green) are applied with equal amplitude along all three spatial directions. During the EPI readout, a series of bipolar gradients (blue) is applied result-
ing in a series of gradient echoes (red). Adjacent k-space lines are reached by applying short gradient pulses in phase direction (orange). The dark green
gradient that is applied before the EPI readout train determines the position of the first k-space line. The corresponding k-space is depicted in b

. Tab. 1.1 Diffusion-weighted sequences

Sequence Pros Cons Selected references

HASTE readout (turbo spin – No image distortions – CPMG condition not fulfilled owing to [97][125][141]
echo signal readout) – Better than EPI if T2* is short diffusion-weighting gradients
– Image quality on clinical scanners usually
much worse than for EPI

Segmented readouts – Higher resolution possible than – Phase correction necessary [39][75][96][125][141]
(e.g., BLADE, spiral) with EPI – increased measurement time

Line scan readout – No image distortions in phase direc- – Largely increased measurement time [7][41][42][43][44][54]
tion since no phase encoding is used [55][123]

Inner volume excitations
– Only the volume of interest is excited – Reducing the FOV in phase direction [61][69][70][71][75][87]
– High resolutions possible reduces the SNR [146][147]
– Volume of interest can be shimmed – Often increased measurement time
separately
– Can be combined with segmented
readouts

STEAM diffusion weighting – T1 signal decay between signal exci- – Half of the signal is lost [20][104][105][122][133]
tation and signal acquisition allows [151]
for long diffusion times

Steady-state diffusion- – High resolution possible – Only semi-quantitative [21][35][56][92][102]

weighted sequences – Relation of signal and diffusion coefficient [103][153]
is complicated and depends on T1, T2
1.6 · Selected Applications in Neuroscience
33 1

. Fig. 1.24 Timeline of brain maturation. (Reprinted with permission from the American Journal of Psychiatry (Copyright 2006), American Psychiatric
Association [48])

1.6 Selected Applications in Neuroscience with brain maturation, which is illustrated in . Fig. 1.25. Changes
in the mean diffusivity (labeled Dav in . Fig. 1.25) are not simul-
1.6.1 Introduction taneous in all brain regions, for example, white matter RA values
increase first in the occipital lobe, but occur rapidly during the
Diffusion tensor imaging (DTI) is an important noninvasive first year of life.
imaging method for the investigation of normative white matter
development, of neurodevelopmental disorders (e.g., autism), and
neurodegenerative disorders (e.g., amyotrophic lateral sclerosis).
Moreover, DTI is a valuable tool for studying neurological dis-
orders including multiple sclerosis, ischemic stroke, tumors, or
neuropsychiatric disorders like schizophrenia or Alzheimer’s
disease. In addition, DTI has been used in basic neuroscience
research to determine the timing and mechanisms of brain devel-
opment, brain maturation and its relation to emotional and be-
havioral development. In this section, selected findings from DTI
studies in neuroscience are presented. The focus will be on studies
that aimed to identify the trajectories of brain development in
healthy individuals and in patients with schizophrenia.

1.6.2 Normative Brain Development

The timeline of brain maturation was illustrated by Giedd et al.

as shown in . Fig. 1.24. The central nervous system begins to
develop in the human fetus 2–3 weeks after conception through
the folding and fusion of the ectoderm to create the neural tube
[19]. The rate of synaptogenesis peaks after the 34th week of
gestation (40,000 new synapses per second and continue into
adulthood; [48][95]). Myelination begins in the last 2 months
of gestation. It proceeds in a posterior-to-anterior direction
and seems to follow the maturation of functional circuits, with
sensory pathways myelinating first, then motor pathways, and
finally association areas [63]. The increase in white matter relates
to a greater connectivity and integration of initially disparate
neural circuitries [49].
DTI has been used in the evaluation of the developing human . Fig. 1.25 Axial images at the level of basal ganglia from subjects of dif-
ferent ages. In the progression from 26 weeks’ gestation to 7 years, white
brain. One striking example is the study by Neil et al. [114], where
matter RA-values increase first in the occipital lobe (at 40 weeks), but by
it was shown that diffusion anisotropy values of white matter age 7 years they are increased relatively uniformly throughout the subcorti-
change strongly in early brain development. Prominent changes cal white matter. (Reprinted with permission from NMR in Biomedicine
in mean diffusivity and relative anisotropy (7 Sect. 1.3.4) go along (Copyright 2002), Wiley [114])
 34 Chapter 1 · Introduction to Diffusion Imaging
. Fig. 1.26 Top row: longitudinal T1-weighted magnetic resonance images of the same child (at 2 weeks, 1 year, and 2 years of age). Bottom row: white
matter tractography of a neonate, a 1-year-old, and an adult shows the organization of corpus callosum white matter fibers, reflected in increasing frac-
tional anisotropy (yellow, red). (Reprinted with permission from the American Journal of Psychiatry (Copyright 2006), American Psychiatric Association [50])
The progressive increase in white matter structures is further
illustrated in . Fig. 1.26, which is taken from the study by Gilmore
et al. [50]. Since most white matter in neonates is unmyelinated,
the T1-weighted image of a 2-week-old infant exhibits low inten-
sity relative to gray matter. When myelination progresses, white
matter T1-time decreases and, accordingly, signal intensity rises
over time (. Fig. 1.26, top row). Fiber tracking results are in
agreement with the gradual increase of white matter myelination
seen on the T1-weighted images (. Fig. 1.26, bottom row). As
can be shown from this image, there is a progressive increase in
the amount of fibers reconstructed. Furthermore, the average
fractional anisotropy is strongly increasing, as indicated by the
increasing amount of red and yellow areas in the bottom row of
. Fig. 1.26. This holds true for both the corpus callosum and the
corticospinal tract.
A further illustrative example of a DTI study investigating the
maturating brain is the study by Lebel et al. [88], where the pro-
gression of tract-specific maturation was measured in subjects
aged 5 to 30 years. The results of this study suggest that there is
a hierarchical pattern of maturation by which areas with frontal-
temporal connections develop more slowly than in other regions.
Rapid development can be seen in the splenium and genu of the
corpus callosum, while the cingulum and uncinate fasciculus
demonstrate more gradual maturation. The fornix shows no age-
related changes at all (. Fig. 1.27). Based on this study, the devel-
opmental time course of all investigated structures could be elu-
cidated. The maturation of brain white matter continues through-
out adolescence and in some brain structures into the twenties.
A recent, repeated whole-brain analysis by Giorgio et al. [51]
used tract-based spatial statistics analysis of DTI data in adoles-
cents (range 13.5–18.8 years, rescanned on average after
2.5 years). The analysis revealed increases in overall white matter
volume and specifically an increase of FA in two fiber tracts
that are important for speech and motor functions, respectively
(arcuate fasciculus and corticospinal tract). . Fig. 1.28 illustrates
these age-related fractional anisotropy increases in the time
course of a group of adolescents.
1.6.3 Schizophrenia
Findings from imaging studies over the last 20 years provide
evidence that abnormalities in both gray and white matter in the
brain are associated with schizophrenia. Since the advent of DTI,
studies using this method for investigating patients with schizo-
phrenia have added support to the hypothesis that white matter
pathology plays a key role in schizophrenia (see review [118]).
For example, White et al. [148] observed a lower fractional
anisotropy in the cingulate, corpus callosum, and the frontal
lobes in schizophrenic patients with a chronic course of illness
and also in patients with a first episode and/or early onset of
schizophrenia [148]. A recent study by Kanaan et al. [74] in a
large sample of patients with schizophrenia found widespread
clusters of reduced fractional anisotropy in most major white
matter tracts. The reductions were not associated with illness
duration or neuroleptic medication. These findings support the
1.6 · Selected Applications in Neuroscience
35 1

. Fig. 1.27 Age-related fractional anisotropy (FA) increases as measured by fiber tractography in 202 individuals. Considerable changes of age-related
FA are present in nine of ten white matter tracts. Tracts are shown in a 23-year-old male subject (color of tract matches color of curve in plot). (Reprinted
with permission from NeuroImage (Copyright 2008), Elsevier [88])

disconnection model of schizophrenia, since many widespread

regions (frontal, temporal, parietal, occipital, and cerebellar
regions) showed white matter abnormalities. In accordance
with these overall findings, Kyriakopoulos et al. [82] found that
patients with early-onset schizophrenia had significantly lower
fractional anisotropy in the white matter of the parietal associa-
tion cortex bilaterally and in the left middle cerebellar peduncle
than did healthy controls (. Fig. 1.29). Correlative investigations
of DTI measurements and clinical symptoms showed an associa-
tion of decreased fractional anisotropy in the inferior fronto-
occipital fasciculus with an increased severity of hallucinations
and delusions [134] and with cognitive symptoms as shown in
. Fig. 1.30 [45].
 36 Chapter 1 · Introduction to Diffusion Imaging

. Fig. 1.28 White matter tracts (in red–yellow) showing significant, age-related FA increases over time in a group of adolescents (range 13.5–18.8 years,
rescanned on average after 2.5 years). Significant clusters are thickened for better visibility. Background image is the mean FA of all subjects, green is the
white matter skeleton thresholded at FA > 0.2 (visible where the age correlation did not reach significance). (Reprinted with permission from NeuroImage
(Copyright 2010), Elsevier [51])

. Fig. 1.29 FA map indicating areas where FA in schizophrenia patients is reduced in comparison with healthy subjects. (Reprinted with permission from
Biological Psychiatry (Copyright 2008), Society of Biological Psychiatry [82])
References
. Fig. 1.30 Scattergrams for fornix FA and Doors and People Test (DPT) measures of visual, verbal, recall, and recognition memory for normal controls
and patients with schizophrenia. (Reprinted with permission from Schizophrenia Research (Copyright 2008), Elsevier [45])
References
[1] Alexander DC, Barker GJ (2005) Optimal imaging parameters for fiber-
orientation estimation in diffusion MRI. Neuroimage 27: 357–367
[2] Aganj I, Lenglet C, Sapiro G et al. (2009) Multiple Q-shell ODF recon-
struction in Q-ball imaging. Med Image Comput Comput Assist Interv
12 (Pt 2): 423–431
[3] Aganj I, Lenglet C, Sapiro G et al. (2010) Reconstruction of the orienta-
tion distribution function in single- and multiple-shell q-ball imaging
within constant solid angle. Magn Reson Med 64: 554–566
[4] Anderson AW (2005) Measurement of fiber orientation distributions
using high angular resolution diffusion imaging. Magn Reson Med 54:
1194–1206
[5] Avram L, Ozarslan E, Assaf Yet al. (2008) Three-dimensional water diffu-
sion in impermeable cylindrical tubes: theory versus experiments. NMR
Biomed 21: 888–898
[6] Axelrod S, Sen PN (2001) Nuclear magnetic resonance spin echoes for
restricted diffusion in an inhomogeneous field: Methods and asymp-
totic regimes. J Chem Phys 115: 6878–6895
[7] Bammer R, Herneth AM, Maier SE et al. (2003) Line scan diffusion imag-
ing of the spine. AJNR 24: 5–12
[8] Barnett A (2009) Theory of Q-ball imaging redux: Implications for fiber
tracking. Magn Reson Med 62: 910–923
[9] Bar-Shir A, Avram L, Ozarslan E, Basser PJ, Cohen Y (2008) The effect of
the diffusion time and pulse gradient duration ratio on the diffraction
pattern and the structural information estimated from q-space diffu-
sion MR: experiments and simulations. J Magn Reson 194: 230–236
[10] Barzykin AV (1999) Theory of spin echo in restricted geometries under
a step-wise gradient pulse sequence. J Magn Reson 139 :342–353
[11] Basser PJ, Mattiello J, LeBihan D (1994) Estimation of the effective self-
diffusion tensor from the NMR spin echo. J Magn Reson B 103: 247–254
[12] Basser PJ, Mattiello J, LeBihan D (1994) MR diffusion tensor spectros-
copy and imaging. Biophys J 66: 259–267
37 1
[13] Basser PJ, Pierpaoli C (1996) Microstructural and physiological features
of tissues elucidated by quantitative-diffusion-tensor MRI. J Magn Re-
son B 111: 209–219
[14] Basser PJ, Pierpaoli C (1998) A simplified method to measure the diffu-
sion tensor from seven MR images. Magn Reson Med 39: 928–934
[15] Beaulieu C (2002) The basis of anisotropic water diffusion in the nerv-
ous system – a technical review. NMR Biomed 15: 435–455
[16] Beaulieu C, Allen PS (1994) Determinants of anisotropic water diffusion
in nerves. Magn Reson Med 31: 394–400
[17] Beaulieu C, Allen PS (1994) Water diffusion in the giant axon of the
squid: implications for diffusion-weighted MRI of the nervous system.
Magn Reson Med 32: 579–583
[18] Bernstein MA, King KF, Zhou XJ (2004) Handbook of MRI Pulse Sequenc-
es. Elsevier Academic Press, Burlington MA
[19] Black IB, DiCicco-Bloom E, Dreyfus CF (1990) Nerve growth factor and
the issue of mitosis in the nervous system. Curr Top Dev Biol 24: 161–
192
[20] Boretius S, Wurfel J, Zipp F, Frahm J, Michaelis T (2007) High-field diffu-
sion tensor imaging of mouse brain in vivo using single-shot STEAM
MRI. J Neurosci Meth 161 :112–117
[21] Buxton RB (1993) The diffusion sensitivity of fast steady-state free pre-
cession imaging. Magn Reson Med 29: 235–243
[22] Callaghan PT (1991). Principles of Nuclear Magnetic Resonance Micros-
copy. Clarendon Press, Oxford
[23] Callaghan PT, Coy A, Macgowan D, Packer KJ, Zelaya FO (1991) Diffrac-
tion-like effects in NMR diffusion studies of fluids in porous solids. Na-
ture 351: 467–469
[24] Callaghan PT, MacGowan D, Packer KJ, Zelaya FO (1997) High-resolu-
tion q-space imaging in porous structures. J Magn Reson 90: 177–182
[25] Chandrasekhar S (1943) Stochastic Problems in Physics and Astronomy.
Rev Mod Phys 15: 1–89
[26] Chang LC, Jones DK, Pierpaoli C (2005) RESTORE: Robust estimation of
tensors by outlier rejection. Magn Reson Med 53: 1088–1095
 38 Chapter 1 · Introduction to Diffusion Imaging
[27] Chenevert TL, Brunberg JA, Pipe JG (1990) Anisotropic diffusion in hu-
1 man white matter: demonstration with MR techniques in vivo. Radiol-
ogy 177: 401–405
[28] Cleveland GG, Chang DC, Hazlewood CF, Rorschach HE (1976) Nuclear
magnetic resonance measurement of skeletal muscle: anisotropy of
the diffusion coefficient of the intracellular water. Biophys J 16: 1043–
1053
[29] Conturo TE, Lori NF, Cull TS et al. (1999) Tracking neuronal fiber path-
ways in the living human brain. Proc Natl Acad Sci USA 96: 10422–
10427
[30] Conturo TE, McKinstry RC, Aronovitz JA, Neil JJ (1995) Diffusion MRI:
precision, accuracy and flow effects. NMR Biomed 8: 307–332
[31] Cory DG, Garroway AN (1990) Measurement of translational displace-
ment probabilities by NMR: an indicator of compartmentation. Magn
Reson Med 14: 435–444
[32] Cotts RM (1991) NMR – Diffusion and diffraction. Nature 351: 443–444
[33] Coulon O, Alexander DC, Arridge S (2004) Diffusion tensor magnetic
resonance image regularization. Med Image Anal 8: 47–67
[34] Desjardins S, Gilkey P (1994) Heat-content asymptotics for operators of
laplace type with Neumann boundary-conditions. Mathematische
Zeitschrift 215: 251–268
[35] Ding S, Trillaud H, Yongbi M et al. (1995) High resolution renal diffusion
imaging using a modified steady-state free precession sequence. Magn
Reson Med 34: 586–595
[36] Does MD, Parsons EC, Gore JC (2003) Oscillating gradient measure-
ments of water diffusion in normal and globally ischemic rat brain.
Magn Reson Med 49: 206–215
[37] Doran M, Hajnal JV, Van Bruggen N et al. (1990) Normal and abnormal
white matter tracts shown by MR imaging using directional diffusion
weighted sequences. J Comput Assist Tomogr 14: 865–873
[38] Einstein A (1905) Über die von der molekularkinetischen Theorie der
Wärme geforderte Bewegung von in ruhenden Flüssigkeiten suspen-
dierten Teilchen. Annalen der Physik 322: 549–560
[39] Ellingson BM, Sulaiman O, Kurpad SN (2010) High-resolution in vivo
diffusion tensor imaging of the injured cat spinal cord using self-navi-
gated, interleaved, variable-density spiral acquisition (SNAILS-DTI).
Magn Reson Imaging 28: 1353–1360
[40] Fillard P, Descoteaux M, Goh A et al. (2011) Quantitative evaluation of
10 tractography algorithms on a realistic diffusion MR phantom. Neu-
roimage 56: 220–234
[41] Finsterbusch J, Frahm J (1999) Diffusion-weighted single-shot line scan
imaging of the human brain. Magn Reson Med 42: 772–778
[42] Finsterbusch J, Frahm J (1999) Single-shot line scan imaging using
stimulated echoes. J Magn Reson 137: 144–153
[43] Finsterbusch J, Frahm J (2000) Diffusion tensor mapping of the human
brain using single-shot line scan imaging. J Magn Reson Imaging 12:
388–394
[44] Finsterbusch J, Frahm J (2000) Gradient-echo line scan imaging using
2D-selective RF excitation. J Magn Reson 147: 17–25
[45] Fitzsimmons J, Kubicki M, Smith K et al. (2009) Diffusion tractography
of the fornix in schizophrenia. Schizophr Res 107: 39–46
[46] Fordham EJ, Mitra PP, Latour LL (1996) Effective diffusion times in mul-
tiple-pulse PFG diffusion measurements in porous media. J Magn Re-
son A 121: 187–192
[47] Fritzsche K, Stieltjes B, Laun FB, Meinzer HP (2009) Towards Reliable
Quantification of Fiber Integrity in Q-Ball Imaging. MICCAI, London,
pp 213–220
[48] Giedd J (1999) Brain development, IX: human brain growth. Am J Psy-
chiatry 156: 4
[49] Giedd JN, Lalonde FM, Celano MJ et al. (2009) Anatomical brain mag-
netic resonance imaging of typically developing children and adoles-
cents. J Am Acad Child Adolesc Psychiatry 48: 465–470
[50] Gilmore JH, Lin W, Gerig G (2006) Fetal and neonatal brain develop-
ment. Am J Psychiatry 163: 2046
[51] Giorgio A, Watkins KE, Chadwick M et al. (2010) Longitudinal changes
in grey and white matter during adolescence. Neuroimage 49: 94–103
[52] Grebenkov DS (2007) NMR survey of reflected Brownian motion. Rev
Mod Phys 79: 1077–1137
[53] Grebenkov DS (2010) Use, misuse, and abuse of apparent diffusion
coefficients. Concept Magn Reson A 36A: 24–35
[54] Gudbjartsson H, Maier SE, Jolesz FA (1997) Double line scan diffusion
imaging. Magn Reson Med 38: 101–109
[55] Gudbjartsson H, Maier SE, Mulkern RV et al. (1996) Line scan diffusion
imaging. Magn Reson Med 36: 509–519
[56] Gudbjartsson H, Patz S (1995) Simultaneous calculation of flow and
diffusion sensitivity in steady-state free precession imaging. Magn Re-
son Med 34: 567–579
[57] Hahn HK, Klein J, Nimsky C, Rexilius J, Peitgen HO (2006) Uncertainty in
diffusion tensor based fibre tracking. Acta Neurochir Suppl 98: 33–41
[58] Hansen JR (1971) Pulsed NMR study of water mobility in muscle and
brain tissue. Biochim Biophys Acta 230: 482–486
[59] Hasan KM, Narayana PA (2003) Computation of the fractional anisot-
ropy and mean diffusivity maps without tensor decoding and diago-
nalization: Theoretical analysis and validation. Magn Reson Med 50:
589–598
[60] Haus JW, Kehr KW (1987) Diffusion in regular and disordered lattices.
Phys Rep 150: 263–406
[61] Hiepe P, Herrmann KH, Ros C, Reichenbach JR (2011) Diffusion weight-
ed inner volume imaging of lumbar disks based on turbo-STEAM acqui-
sition. Z Med Phys 21: 216–227
[62] Hurlimann MD, Latour LL, Sotak CH (1994) Diffusion measurement in
sandstone core: NMR determination of surface-to-volume ratio and
surface relaxivity. Magn Reson Imaging 12: 325–327
[63] Huttenlocher PR (2002) Neural Plasticity: The Effects of the Environ-
ment on the Development of the Cerebral Cortex. Harvard University
Press, Cambridge
[64] Jansons KM, Alexander DC (2003) Persistent Angular Structure: new
insights from diffusion MRI data. Dummy version. Inf Process Med Im-
aging 18: 672–683
[65] Jensen JH, Falangola MF, Hu C et al. (2011) Preliminary observations of
increased diffusional kurtosis in human brain following recent cerebral
infarction. NMR Biomed 24: 452–457
[66] Jensen JH, Helpern JA (2010) MRI quantification of non-Gaussian water
diffusion by kurtosis analysis. NMR Biomed 23: 698–710
[67] Jensen JH, Helpern JA, Ramani A, Lu H, Kaczynski K (2005) Diffusional
kurtosis imaging: the quantification of non-gaussian water diffusion by
means of magnetic resonance imaging. Magn Reson Med 53: 1432–1440
[68] Jensen JH, Helpern JA (2011) Effect of gradient pulse duration on MRI
estimation of the diffusional kurtosis for a two-compartment exchange
model. J Magn Reson 210: 233–237
[69] Jeong EK, Kim SE, Guo J, Kholmovski EG, Parker DL (2005) High-resolu-
tion DTI with 2D interleaved multislice reduced FOV single-shot diffu-
sion-weighted EPI (2D ss-rFOV-DWEPI). Magn Reson Med 54: 1575–1579
[70] Jeong EK, Kim SE, Kholmovski EG, Parker DL (2006) High-resolution DTI
of a localized volume using 3D single-shot diffusion-weighted Stimu-
lated echo-planar imaging (3D ss-DWSTEPI). Magn Reson Med 56:
1173–1181
[71] Jin N, Deng J, Zhang L et al. (2011) Targeted single-shot methods for
diffusion-weighted imaging in the kidneys. J Magn Reson Imaging 33:
1517–1525
[72] Johansen-Berg H, Behrens TEJ (2009) Diffusion MRI. Elsevier, London
[73] Kac M (1966) Can one hear the shape of a drum? Am Math Monthly 73:
1–23
[74] Kanaan R, Barker G, Brammer M et al. (2009) White matter microstruc-
ture in schizophrenia: effects of disorder, duration and medication. Br
J Psychiatry 194: 236–242
[75] Karampinos DC, Van AT, Olivero WC, Georgiadis JG, Sutton BP (2009)
High-resolution diffusion tensor imaging of the human pons with a
reduced field-of-view, multishot, variable-density, spiral acquisition at
3 T. Magn Reson Med 62: 1007–1016
[76] Kingsley PB (2006) Introduction to diffusion tensor imaging mathemat-
ics: Part I. Tensors, rotations, and eigenvectors. Concept Magn Reson A
28A: 101–122
[77] Kingsley PB (2006) Introduction to diffusion tensor imaging mathemat-
ics: Part II. Anisotropy, diffusion-weighting factors, and gradient encod-
ing schemes. Concept Magn Reson A 28A: 123–154
References
[78] Kingsley PB (2006) Introduction to diffusion tensor imaging mathemat-
ics: Part III. Tensor calculation, noise, simulations, and optimization.
Concept Magn Reson A 28A: 155–179
[79] Kingsley PB, Monahan WG (2005) Contrast-to-noise ratios of diffusion
anisotropy indices. Magn Reson Med 53: 911–918
[80] Kreher BW, Mader I, Kiselev VG (2008) Gibbs tracking: a novel approach
for the reconstruction of neuronal pathways. Magn Reson Med 60:
953–963
[81] Kreher BW, Schneider JF, Mader I et al. (2005) Multitensor approach for
analysis and tracking of complex fiber configurations. Magn Reson
Med 54: 1216–1225
[82] Kyriakopoulos M, Vyas NS, Barker GJ, Chitnis XA, Frangou S (2008) A
diffusion tensor imaging study of white matter in early-onset schizo-
phrenia. Biol Psychiatry 63: 519–523
[83] Latour LL, Svoboda K, Mitra PP, Sotak CH (1994) Time-dependent diffu-
sion of water in a biological model system. Proc Natl Acad Sci USA 91:
1229–1233
[84] Laun FB, Fritzsche KH, Kuder TA, Stieltjes B (2011) Introduction to the
basic principles and techniques of diffusion-weighted imaging. Radio-
loge 51: 170–179
[85] Laun FB, Huff S, Stieltjes B (2009) On the effects of dephasing due to
local gradients in diffusion tensor imaging experiments: relevance for
diffusion tensor imaging fiber phantoms. Magn Reson Imaging 27:
541–548
[86] Laun FB, Kuder T, Semmler W, Stieltjes B (2011) Determination of the
defining boundary in nuclear magnetic resonance diffusion experi-
ments. Phys Rev Lett 107: 048102
[87] Laun FB, Stieltjes B, Schluter M, Rupp R, Schad LR (2009) Reproducible
evaluation of spinal cord DTI using an optimized inner volume se-
quence in combination with probabilistic ROI analysis. Z Med Phys 19:
11–20
[88] Lebel C, Walker L, Leemans A, Phillips L, Beaulieu C (2008) Microstruc-
tural maturation of the human brain from childhood to adulthood.
Neuroimage 40: 1044–1055
[89] Le Bihan D (1995) Diffusion and Perfusion Magnetic Resonance Imag-
ing. Raven Press, New York
[90] Le Bihan D, Breton E, Lallemand D et al. (1988) Separation of diffusion
and perfusion in intravoxel incoherent motion MR imaging. Radiology
168: 497–505
[91] Le Bihan D, Breton E, Lallemand D et al. (1986) MR imaging of intra-
voxel incoherent motions: application to diffusion and perfusion in
neurologic disorders. Radiology 161: 401–407
[92] Le Bihan D, Turner R, MacFall JR (1989) Effects of intravoxel incoherent
motions (IVIM) in steady-state free precession (SSFP) imaging: applica-
tion to molecular diffusion imaging. Magn Reson Med 10: 324–337
[93] Lemke A, Laun FB, Klauss M et al. (2009) Differentiation of pancreas
carcinoma from healthy pancreatic tissue using multiple b-values:
comparison of apparent diffusion coefficient and intravoxel incoherent
motion derived parameters. Invest Radiol 44: 769–775
[94] Lemke A, Laun FB, Simon D, Stieltjes B, Schad LR (2010) An in vivo veri-
fication of the intravoxel incoherent motion effect in diffusion-weight-
ed imaging of the abdomen. Magn Reson Med 64: 1580–1585
[95] Levitt P (2003) Structural and functional maturation of the developing
primate brain. J Pediatr 143 [4 Suppl]: S35–45
[96] Liu C, Bammer R, Kim DH, Moseley ME (2004) Self-navigated interleaved
spiral (SNAILS): application to high-resolution diffusion tensor imag-
ing. Magn Reson Med 52: 1388–1396
[97] Lovblad KO, Jakob PM, Chen Q et al. (1998) Turbo spin-echo diffusion-
weighted MR of ischemic stroke. AJNR 19: 201–208; discussion 209
[98] Lu H, Jensen JH, Ramani A, Helpern JA (2006) Three-dimensional char-
acterization of non-gaussian water diffusion in humans using diffusion
kurtosis imaging. NMR Biomed 19: 236–247
[99] Luciani A, Vignaud A, Cavet M et al. (2008) Liver cirrhosis: intravoxel in-
coherent motion MR imaging – pilot study. Radiology 249: 891–899
[100] Mangin JF, Poupon C, Cointepas Y et al. (2002) A framework based on
spin glass models for the inference of anatomical connectivity from
diffusion-weighted MR data – a technical review. NMR Biomed 15:
481–492
39 1
[101] Mattiello J, Basser PJ, Le Bihan D (1997) The b matrix in diffusion tensor
echo-planar imaging. Magn Reson Med 37: 292–300
[102] McNab JA, Jbabdi S, Deoni SC et al. (2009) High resolution diffusion-
weighted imaging in fixed human brain using diffusion-weighted
steady state free precession. Neuroimage 46: 775–785
[103] McNab JA, Miller KL (2008) Sensitivity of diffusion weighted steady
state free precession to anisotropic diffusion. Magn Reson Med 60:
405–413
[104] Merboldt KD, Hanicke W, Bruhn H, Gyngell ML, Frahm J (1992) Diffusion
imaging of the human brain in vivo using high-speed STEAM MRI.
Magn Reson Med 23: 179–192
[105] Merboldt KD, Hanicke W, Frahm J (1991) Diffusion imaging using stim-
ulated echoes. Magn Reson Med 19: 233–239
[106] Mitra PP, Halperin BI (1995) Effects of finite gradient-pulse widths in
pulsed-field-gradient diffusion measurements. J Magn Reson A 113:
94–101
[107] Mitra PP, Sen PN, Schwartz LM, Ledoussal P (1992) Diffusion propagator
as a probe of the structure of porous-media. Phys Rev Lett 68): 3555–
3558
[108] Mori S (2007) Introduction to Diffusion Tensor Imaging. Elsevier, Am-
sterdam
[109] Mori S, Crain BJ, Chacko VP, van Zijl PC (1999) Three-dimensional track-
ing of axonal projections in the brain by magnetic resonance imaging.
Ann Neurol 45: 265–269
[110] Mori S, Kaufmann WE, Davatzikos C et al. (2002) Imaging cortical asso-
ciation tracts in the human brain using diffusion-tensor-based axonal
tracking. Magn Reson Med 47: 215–223
[111] Mori S, van Zijl PC (2002) Fiber tracking: principles and strategies –
a technical review. NMR Biomed 15: 468–480
[112] Moseley ME, Cohen Y, Kucharczyk J et al. (1990) Diffusion-weighted MR
imaging of anisotropic water diffusion in cat central nervous system.
Radiology 176: 439–445
[113] Moussavi-Biugui A, Stieltjes B, Fritzsche K, Semmler W, Laun FB (2011)
Novel spherical phantoms for Q-ball imaging under in vivo conditions.
Magn Reson Med 65: 190–194
[114] Neil J, Miller J, Mukherjee P, Huppi PS (2002) Diffusion tensor imaging
of normal and injured developing human brain – a technical review.
NMR Biomed 15: 543–552
[115] Novikov DS, Fieremans E, Jensen JH, Helpern JA (2011) Time-depend-
ent diffusion and kurtosis as a probe of tissue structure. In: Proceedings
of the 19th Scientific Meeting, International Society for Magnetic Reso-
nance in Medicine, Montreal, Canada, 2011
[116] Peled S, Friman O, Jolesz F, Westin CF (2006) Geometrically constrained
two-tensor model for crossing tracts in DWI. Magn Reson Imaging 24:
1263–1270
[117] Perrin M, Poupon C, Rieul B et al. (2005) Validation of q-ball imaging
with a diffusion fibre-crossing phantom on a clinical scanner. Philos
Trans R Soc Lond 360: 881–891
[118] Peters BD, Blaas J, de Haan L (2010) Diffusion tensor imaging in the
early phase of schizophrenia: what have we learned? J Psychiatr Res 44:
993–1004
[119] Poupon C, Rieul B, Kezele I et al. (2008) New diffusion phantoms dedi-
cated to the study and validation of high-angular-resolution diffusion
imaging (HARDI) models. Magn Reson Med 60: 1276–1283
[120] Reese TG, Heid O, Weisskoff RM, Wedeen VJ (2003) Reduction of eddy-
current-induced distortion in diffusion MRI using a twice-refocused
spin echo. Magn Reson Med 49: 177–182
[121] Reisert M, Mader I, Anastasopoulos C, Weigel M, Schnell S, Kiselev V
(2011) Global fiber reconstruction becomes practical. Neuroimage 54:
955–962
[122] Rieseberg S, Merboldt KD, Kuntzel M, Frahm J (2005) Diffusion tensor
imaging using partial Fourier STEAM MRI with projection onto convex
subsets reconstruction. Magn Reson Med 54: 486–490
[123] Robertson RL, Maier SE, Mulkern RV et al. (2000) MR line-scan diffusion
imaging of the spinal cord in children. AJNR 21: 1344–1348
[124] Schaefer DJ, Bourland JD, Nyenhuis JA (2000) Review of patient
safety in time-varying gradient fields. J Magn Reson Imaging 12:
20–29
 40 Chapter 1 · Introduction to Diffusion Imaging

[125] Seifert MH, Jakob PM, Jellus V, Haase A, Hillenbrand C (2000) High- [149] Wirestam R, Borg M, Brockstedt S et al. (2001) Perfusion-related param-
1 resolution diffusion imaging using a radial turbo-spin-echo sequence: eters in intravoxel incoherent motion MR imaging compared with CBV
implementation, eddy current compensation, and self-navigation. J and CBF measured by dynamic susceptibility-contrast MR technique.
Magn Reson 144: 243–254 Acta Radiol 42: 123–128
[126] Sen PN (2004) Time-dependent diffusion coefficient as a probe of ge- [150] Xue R, van Zijl PC, Crain BJ, Solaiyappan M, Mori S (1999) In vivo three-
ometry. Concept Magn Reson A 23: 1–21 dimensional reconstruction of rat brain axonal projections by diffusion
[127] Sen PN, Schwartz LM, Mitra PP (1994) Probing the structure of porous tensor imaging. Magn Reson Med 42: 1123–1127
media using NMR spin echoes. Magn Reson Imaging 12: 227–230 [151] Yongbi MN, Ding S, Dunn JF (1996) A modified sub-second fast-STEAM
[128] Sotiropoulos SN, Bai L, Morgan PS et al. (2008) A regularized two-tensor sequence incorporating bipolar gradients for in vivo diffusion imaging.
model fit to low angular resolution diffusion images using basis direc- Magn Reson Med 35: 911–916
tions. J Magn Reson Imaging 28: 199–209 [152] Zielinski LJ, Sen PN (2003) Effects of finite-width pulses in the pulsed-
[129] Stejskal EO (1965) Use of spin echoes in a pulsed magnetic-field gradi- field gradient measurement of the diffusion coefficient in connected
ent to study anisotropic, restricted diffusion and flow. J Chem Phys 43: porous media. J Magn Reson 165: 153–161
3597–3603 [153] Zur Y, Bosak E, Kaplan N (1997) A new diffusion SSFP imaging tech-
[130] Stejskal EO, Tanner JE (1965) Spin diffusion measurements: spin echoes nique. Magn Reson Med 37: 716–722
in the presence of a time-dependent field gradient. J Chem Phys 42:
288–292
[131] Stepisnik J (1981) Analysis of NMR self-diffusion measurements by a
density matrix calculation. Physica B 104: 350–364
[132] Stieltjes B, Kaufmann WE, van Zijl PC et al. (2001) Diffusion tensor imag-
ing and axonal tracking in the human brainstem. Neuroimage 14:
723–735
[133] Stocker T, Kaffanke J, Shah NJ (2009) Whole-brain single-shot STEAM
DTI at 4 Tesla utilizing transverse coherences for enhanced SNR. Magn
Reson Med 61: 372–380
[134] Szeszko PR, Robinson DG, Ashtari M et al. (2008) Clinical and neuropsy-
chological correlates of white matter abnormalities in recent onset
schizophrenia. Neuropsychopharmacology 33: 976–984
[135] Takahashi E, Dai G, Wang R et al. (2010) Development of cerebral fiber
pathways in cats revealed by diffusion spectrum imaging. Neuroimage
49: 1231–1240
[136] Takahashi M, Hackney DB, Zhang G et al. (2002) Magnetic resonance
microimaging of intraaxonal water diffusion in live excised lamprey
spinal cord. Proc Natl Acad Sci USA 99: 16192–16196
[137] Tournier JD, Calamante F, Gadian DG, Connelly A (2004) Direct estima-
tion of the fiber orientation density function from diffusion-weighted
MRI data using spherical deconvolution. Neuroimage 23: 1176–1185
[138] Tournier JD, Yeh CH, Calamante F et al. (2008) Resolving crossing fibres
using constrained spherical deconvolution: validation using diffusion-
weighted imaging phantom data. Neuroimage 42: 617–625
[139] Tuch DS (2004) Q-ball imaging. Magn Reson Med 52: 1358–1372
[140] Tuch DS, Reese TG, Wiegell MR et al. (2002) High angular resolution
diffusion imaging reveals intravoxel white matter fiber heterogeneity.
Magn Reson Med 48: 577–582
[141] von Mengershausen M, Norris DG, Driesel W (2005) 3D diffusion tensor
imaging with 2D navigated turbo spin echo. MAGMA 18: 206–216
[142] Wang Z, Vemuri BC, Chen Y, Mareci T (2003) A constrained variational
principle for direct estimation and smoothing of the diffusion tensor
field from DWI. Inf Process Med Imaging 2732: 660–671
[143] Wedeen VJ, Hagmann P, Tseng WY, Reese TG, Weisskoff RM (2005) Map-
ping complex tissue architecture with diffusion spectrum magnetic
resonance imaging. Magn Reson Med 54: 1377–1386
[144] Wedeen VJ, Reese TG, Tuch DS et al. (1999) Mapping fiber orientation
spectra in cerebral white matter with Fourier-transform diffusion MRI.
In: Proceedings of the 8th Annual Meeting ISMRM, Denver, 2000
[145] Wedeen VJ, Wang RP, Schmahmann JD et al. (2008) Diffusion spectrum
magnetic resonance imaging (DSI) tractography of crossing fibers.
Neuroimage 41: 1267–1277
[146] Wheeler-Kingshott CA, Parker GJ, Symms MR et al. (2002) ADC mapping
of the human optic nerve: increased resolution, coverage, and reliabil-
ity with CSF-suppressed ZOOM-EPI. Magn Reson Med 47: 24–31
[147] Wheeler-Kingshott CA, Trip SA, Symms MR et al. (2006) In vivo diffusion
tensor imaging of the human optic nerve: pilot study in normal con-
trols. Magn Reson Med 56: 446–451
[148] White T, Nelson M, Lim KO (2008) Diffusion tensor imaging in psychiat-
ric disorders. Top Magn Reson Imaging 19: 97–109
 41 II

Section II
Atlas
Chapter 2 Two-dimensional Brain Slices – 43

Chapter 3 Three-dimensional Fiber Tracking – 281

 43 2

Two-dimensional Brain Slices

2.1 Coronal View – 45

2.2 Sagittal View – 167

2.3 Transversal View – 209

B. Stieltjes et al., Diffusion Tensor Imaging,

DOI 10.1007/978-3-642-20456-2_2,
© Springer-Verlag Berlin Heidelberg 2013
 44 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
45 2

2.1 Coronal View

 46 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
47 2
 48 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
49 2
 50 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
51 2
 52 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
53 2
 54 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
55 2
 56 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
57 2
 58 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
59 2
 60 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
61 2
 62 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
63 2
 64 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
65 2
 66 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
67 2
 68 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
69 2
 70 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
71 2
 72 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
73 2
 74 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
75 2
 76 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
77 2
 78 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
79 2
 80 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
81 2
 82 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
83 2
 84 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
85 2
 86 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
87 2
 88 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
89 2
 90 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
91 2
 92 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
93 2
 94 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
95 2
 96 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
97 2
 98 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
99 2
 100 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
101 2
 102 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
103 2
 104 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
105 2
 106 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
107 2
 108 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
109 2
 110 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
111 2
 112 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
113 2
 114 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
115 2
 116 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
117 2
 118 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
119 2
 120 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
121 2
 122 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
123 2
 124 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
125 2
 126 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
127 2
 128 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
129 2
 130 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
131 2
 132 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
133 2
 134 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
135 2
 136 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
137 2
 138 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
139 2
 140 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
141 2
 142 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
143 2
 144 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
145 2
 146 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
147 2
 148 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
149 2
 150 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
151 2
 152 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
153 2
 154 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
155 2
 156 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
157 2
 158 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
159 2
 160 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
161 2
 162 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
163 2
 164 Chapter 2 · Two-dimensional Brain Slices

2
2.1 · Coronal View
165 2
 166 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
167 2

2.2 Sagittal View

 168 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
169 2
 170 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
171 2
 172 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
173 2
 174 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
175 2
 176 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
177 2
 178 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
179 2
 180 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
181 2
 182 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
183 2
 184 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
185 2
 186 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
187 2
 188 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
189 2
 190 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
191 2
 192 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
193 2
 194 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
195 2
 196 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
197 2
 198 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
199 2
 200 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
201 2
 202 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
203 2
 204 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
205 2
 206 Chapter 2 · Two-dimensional Brain Slices

2
2.2 · Sagittal View
207 2
 208 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
209 2

2.3 Transversal View

 210 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
211 2
 212 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
213 2
 214 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
215 2
 216 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
217 2
 218 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
219 2
 220 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
221 2
 222 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
223 2
 224 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
225 2
 226 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
227 2
 228 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
229 2
 230 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
231 2
 232 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
233 2
 234 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
235 2
 236 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
237 2
 238 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
239 2
 240 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
241 2
 242 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
243 2
 244 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
245 2
 246 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
247 2
 248 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
249 2
 250 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
251 2
 252 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
253 2
 254 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
255 2
 256 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
257 2
 258 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
259 2
 260 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
261 2
 262 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
263 2
 264 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
265 2
 266 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
267 2
 268 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
269 2
 270 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
271 2
 272 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
273 2
 274 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
275 2
 276 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
277 2
 278 Chapter 2 · Two-dimensional Brain Slices

2
2.3 · Transversal View
279 2
 280 Chapter 2 · Two-dimensional Brain Slices

2
 281 3

Three-dimensional Fiber Tracking

3.1 Fiber Tracking of the Cerebral Hemispheres – 283
3.1.1 Fiber Tracking of the Corpus Callosum – 285
3.1.2 Fiber Tracking of the Cingulum – 292
3.1.3 Fiber Tracking of the Fornix – 296
3.1.4 Fiber Tracking of the Inferior Longitudinal Fasciculus – 302
3.1.5 Fiber Tracking of the Inferior Fronto-occipital and Uncinate Fasciculus – 309
3.1.6 Fiber Tracking of the Superior Longitudinal, the Superior Fronto-occipital
and the Arcuate Fasciculus – 319
3.1.7 Fiber Tracking of the Cerebral Peduncles – 334

3.2 Fiber Tracking of the Brain Stem – 351

3.2.1 Fiber Tracking of the Inferior Cerebellar Peduncle – 353
3.2.2 Fiber Tracking of the Middle Cerebellar Pedundle – 359
3.2.3 Fiber Tracking of the Medial Lemniscus
and the Superior Cerebellar Peduncle – 365

B. Stieltjes et al., Diffusion Tensor Imaging,

DOI 10.1007/978-3-642-20456-2_3,
© Springer-Verlag Berlin Heidelberg 2013
 282 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
283 3

3.1 Fiber Tracking of the Cerebral Hemispheres

 284 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
285 3

3.1.1 Fiber Tracking of the Corpus Callosum

 286 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
287 3
 288 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
289 3
 290 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
291 3
 292 Chapter 3 · Three-dimensional Fiber Tracking

3.1.2 Fiber Tracking of the Cingulum

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
293 3
 294 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
295 3
 296 Chapter 3 · Three-dimensional Fiber Tracking

3.1.3 Fiber Tracking of the Fornix

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
297 3
 298 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
299 3
 300 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
301 3
 302 Chapter 3 · Three-dimensional Fiber Tracking

3.1.4 Fiber Tracking of the Inferior Longitudinal Fasciculus

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
303 3
 304 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
305 3
 306 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
307 3
 308 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
309 3

3.1.5 Fiber Tracking of the Inferior Fronto-occipital

and Uncinate Fasciculus
 310 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
311 3
 312 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
313 3
 314 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
315 3
 316 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
317 3
 318 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
319 3

3.1.6 Fiber Tracking of the Superior Longitudinal,

the Superior Fronto-occipital and the Arcuate Fasciculus
 320 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
321 3
 322 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
323 3
 324 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
325 3
 326 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
327 3
 328 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
329 3
 330 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
331 3
 332 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
333 3
 334 Chapter 3 · Three-dimensional Fiber Tracking

3.1.7 Fiber Tracking of the Cerebral Peduncles

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
335 3
 336 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
337 3
 338 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
339 3
 340 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
341 3
 342 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
343 3
 344 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
345 3
 346 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
347 3
 348 Chapter 3 · Three-dimensional Fiber Tracking

3
3.1 · Fiber Tracking of the Cerebral Hemispheres
349 3
 350 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
351 3

3.2 Fiber Tracking of the Brain Stem

 352 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
353 3

3.2.1 Fiber Tracking of the Inferior Cerebellar Peduncle

 354 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
355 3
 356 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
357 3
 358 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
359 3

3.2.2 Fiber Tracking of the Middle Cerebellar Pedundle

 360 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
361 3
 362 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
363 3
 364 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
365 3

3.2.3 Fiber Tracking of the Medial Lemniscus

and the Superior Cerebellar Peduncle
 366 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
367 3
 368 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
369 3
 370 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
371 3
 372 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
373 3
 374 Chapter 3 · Three-dimensional Fiber Tracking

3
3.2 · Fiber Tracking of the Brain Stem
375 3
 376 Chapter 3 · Three-dimensional Fiber Tracking

3
 377 III

Section III
Appendix

Index Introduction – 379

Index Atlas – 380

B. Stieltjes et al., Diffusion Tensor Imaging,

DOI 10.1007/978-3-642-20456-2,
© Springer-Verlag Berlin Heidelberg 2013
 379 A–W

Index Introduction

A F N T
adolescents 34 FACT-algorithm 23 narrow diffusion weighting termination criterion 23
Alzheimers disease 33 fiber assignment by continuous gradients 13 time 10
amyotrophic lateral sclerosis 33 tracking 23 nerve tracts 23 time-dependent diffusion
anisotropy 21 fiber tracking 23 neurodegenerative disorders 33 coefficient 12
apparent diffusion coefficient 12, fractional anisotropy 21 neurodevelopmental disorders 33 tissue pulsation 29
17 free diffusion 10 neurological disorders 33 tortuosity 22
apparent kurtosis 17 free diffusion coefficient 10 neuropsychiatric disorders 33 tracking 23
applications in neuroscience 33 free diffusion propagator 10 NMR diffusion 13 tract specific maturation 34
autism 33 normative brain development 33 translation distance 10
averaging 29 turbo spin echo 32
G two-tensor approach 27
O
B Gaussian function 10

background noise 29
global tracking 23
gyromagnetic ratio 13
ODF 27
orientation distribution function
V
basic principle of diffusion meas- 27 velocity 10
urement 13 oscillating diffusion gradients 22 voxel-averaged propagator 15
beautiful red-leaved flower 23
bipolar magnetic field gradients
H
13
BLADE 32
HARDI 26
HASTE 32
P W
brain maturation 33 high-angular-resolution diffusion parallel imaging 31 white matter development 33
b-value 13, 29 imaging 26 peripheral nerve stimulations 13
permeability 22
persistent angular structure 28
C I phase correction 32
PI/2 ghosts 30
capillary flow 25 image distortions 30 practical aspects 29
cell membranes 11 inhomogeneous main magnetic prototype single-fiber signal 27
cell radius 11 field 30 pseudo-diffusion coefficient 25
central limit theorem 10 inner volume excitations 32
complex fiber configurations 26 intravoxel incoherent motion 25
conductivity 22
constant solid angle 27
ischemic stroke 33
IVIM effect 25
Q
CPMG condition 32 Q-ball Imaging 27
crossing fibers 26 QBI 27
K q-space 26
q-value 13
D kurtosis 13
kurtosis tensor imaging 26
diffusion propagator 13
diffusion spectrum imaging 26
R
diffusion tensor 20
distribution of particles 10
L random walk 10
restricted diffusion 11
linear translation 10
line scan readout 32
E long diffusion gradients 16
long time limit 22
S
ECG triggering 29 schizophrenia 33, 34
echo planar imaging 30 sequences 30
eigenvalues of the diffusion tensor
21
M short diffusion times 22
signal-to-noise 29
eigenvectors of the diffusion tensor mean curvature 22 SNR 29
21 mean diffusivity 21 spherical deconvolution 27
EPI 30 multiple sclerosis 33 spiral sequence 32
expectation value 10 myelination 34 squared displacement 10
exponential signal drop 13 statistical measures 10
STEAM diffusion weighting 32
structural information 22
surface relaxation 22
surface-to-volume ratio 22
380 Appendix
Index Atlas
For each structure, the respective page for the coronal, sagittal and transversal views are provided whenever available in all three planes.
The reference to the 3D-reconstruction whenever available is printed in bold.
A G
Amygdala 109, 187, 227
Anterior commissure 98, 170, 240
Arcuate fasciculus 146, 198, 254;
326
C Hippocampus 113, 187, 233
Callosal sulcus 87
Caudate, head 89, 179, 245
Caudate, tail 109
Cerebellar hemisphere 137, 213
Cerebellum 137, 171
Cerebellum, anterior lobe 227
Cerebellum, posterior lobe 225
Cerebellum, vermis 137, 221
Cerebral aqueduct 227
Cerebral peduncle 120, 184, 230;
334
Cingulate gyrus 87, 173
Cingulate sulcus 87, 173
Cingulum 110, 180, 250; 295
Circular sulcus 91, 205, 251
Claustrum 99
Corona radiata 112, 184, 266;
343
Corpus callosum, body 98, 170,
268; 290
Corpus callosum, genu 86, 170,
248; 290
Corpus callosum, isthmus 120
Corpus callosum, major forceps
138, 186, 250
Corpus callosum, minor forceps
74, 186, 240
Corpus callosum, splenium 132,
172, 252; 290
Cortico-spinal tract 120, 170, 212;
343
E Trigeminal nerve 216
External capsule 98, 198, 246
Extreme capsule 98
F 375
Fornix, body 108, 172, 256; 301
Fornix, column 102, 172, 250; 301
Fornix, crus 120, 172, 256; 301
Fornix, fimbria 132, 194, 236; 301
Fourth ventricle 137, 177, 217
Frontal lobe 203
Fronto-pontine tract 92, 184, 230;
342
Globus pallidus 105
Gyrus rectus 83, 171, 227
H Optic tract 104, 178, 232
I Parietal lobe 171
Inferior cerebellar peduncle 134,
182, 212; 375
Inferior frontal gyrus 103
Inferior fronto-occipital fasciculus
116, 196, 236; 316
Inferior longitudinal fasciculus
116, 196, 226; 308
Inferior parietal lobe 131,
Inferior temporal lobe 137, 239
Insula 113, 201, 245
Internal capsule, anterior limb 98,
190, 250
Internal capsule, genu 106
Internal capsule, posterior limb
116, 184, 250
Interpeduncular fossa 117
L 120, 196, 268; 331
Lateral ventricle, anterior horn 83,
179, 247
Lateral ventricle, atrium 135,
Lateral ventricle, body 115,
Lateral ventricle, posterior horn
149, 187, 247
Lateral ventricle, temporal horn
113, 199, 225
Lateral ventricle, trigone 263
Longitudinal cerebral fissure 85
M
Mamillary body 113
Medial lemniscus 130, 170, 214;
Middle cerebellar peduncle 112,
170, 216; 375
Middle frontal gyrus 95
O
Occipital lobe 161, 169, 227
Optic chiasm 98, 172, 230
Optic nerve 88, 226, 240
Optic radiation 126, 196, 240
Orbital gyrus 85, 203, 239
P
Postcentral gyrus 135
Precentral gyrus 131
Putamen 93, 187, 243
S
Sagittal stratum 136, 196, 242
Septum pellucidum 99, 171, 251
Septum pellucidum, cavum 91,
253
Superior cerebellar peduncle 132,
178, 226; 375
Superior frontal gyrus 91
Superior frontal sulcus 91
Superior fronto-occipital fasciculus
110, 188, 278; 325
Superior longitudinal fasciculus
Sylvian fissure 87, 203, 243
T
Temporal horn 113, 199, 225
Temporal pole 99, 197, 214
Temporoparieto-occipitopontine
tract 122, 186, 230; 348
Thalamus 111, 243
Third ventricle 105, 241
U
Uncinate fasciculus 98, 198, 240;
313