Post on 22-Sep-2020
Multislice Perfusion Imaging With Arterial Spin Labelling: Applications to Functional MRI
Bradford A. Giil, B. Sc.
Medical Physics Unit
McGill University
Montreal, Quebec, Canada
A thesis submitted to the Faculty of Gtaduate
Studies and Research in partial fuifilIrnent of the
requirements of the Degree of
Master of Science in Medical Physics.
@Bradford A. Gill, August 1999
National Librafy Bibliothèque nationafe du Canada
Acquisitions and Acquisitions et Bibîiiraphic Services services bibliographiques
The author has granted a non- exclusive licence ailowing the National Library of Caaada to reproduce, loan, distriibute or seil copies of this thesis in microform, paper or electronic formats.
The author retains ownership of the copyright in this thesis. Neither the thesis nor substantial extracts &om it may be printed or otherwise reproduced without the author's permission.
L'auteur a accordé une licence non exclusive pennettant é la Bibliothèque nationale du Canada de reproduire, prêter, distribuer ou vendre des copies de cette thèse sous la forme de microfiche/nlm, de reproduction sur papier ou sur formai électronique.
L'auteur conserve la propriété du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de celle-ci ne doivent être imprimes -
ou autrement reproduits sans son autorisation.
Abstract
This thesis presents the design, implementation and testing of a perfusion-weighted
magnetic resonance imaging sequence that is capable of acquiring several slices at a
time. Various methods of image acquisition and perfiision contrast generation are con-
sidered and tested.
The results of multi-slice acquisitions are compared with those from single-slice se-
quences in a variety of expenments to elucidate the effects of various imaging parameters
on the measured perfusion signai. A seven slice version was implernented, and was
found to give good results in the tests perfonned. This sequence will be useful in
perfusion-based finctional magnetic resonance imaging studies where the region of
interest can not be covered with a single image slice.
Resumé
Cette thèse présente la conception, l'implantation et l'essai d'une séquence d'imagerie
par résonance magnetique de la perfusion ayan la capacité d'acquerir plusieurs tranches
à la fois. Diverses méthodes d'acquisition d'images et de génération de contraste de
perfusion y sont considérées et testées.
Le résultat des acquisitions multi-tranches sont comparées a celles issues de séquences
à tranche simple, dans une séries d'experiences dont le but est d'élucider les effets de
divers paramètres d'imagerie sur le signal de periùsion mesuré. Une version a sept
tranches fut implantée et produisit de bons résultats lors des essais. Cette séquence sera
utile pour les études d'imagerie fonctionelle par résonance magnétique basées sur la
perfusion ou une tranche simple s'avérera insuffisante pour couvrir la zone d'interêt.
Acknowledgements
1 would like to thank my supervisor, Dr. G. Bruce Pike, for providing guidance, support
and an open and stimulating environment in the imaging lab.
1 wouId also like to thank my fellow students and CO-workers at the Neuroimaging
Lab, and in the Medical Physics program. 1 would especially like to thank Dr. Gérard
Crelier, Dr. Rick Hoge, Patrice Munger, John Sled, Dr. Jeff Atkinson, Valentina Petre
and Andreas Lazada for their technical assistance and advice.
For invaluable help in various administrative matters I would like to thank the Med-
ical Physics Graduate Secretary, Margery Knewstubb, as well as BIC Administrative
Assistant Jemifer Quinn.
My gratitude is also extended to the Medical Research Council of Canada, who
provided the financial support that made this project possible.
Contents
Abstract 1
Resumé
Acknowledgements 3
Table of Contents 4
List of Figures 6
Glossary of Terms 7
1 Introduction 9
2 Background 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 MRSignal 12
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Imaging 19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Fast Imaging 26
. . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Functional Brain Imaging 28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 BOLDContrast 33
. . . . . . . . . . . . . . . . . . . . . . . 2.6 Perfusion Weighted Imaging 35
3 Arterial Spin Labelling 36 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Introduction 36
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Theory 37 . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Pulsed ASL Sequences 42
. . . . . . . . . . . . . . . . . . . 3 .3.1 Fast Acquisition Techniques 42 3.3.2 Spin Tagging Approaches . . . . . . . . . . . . . . . . . . . . 48 3 .3.3 Considerations for Multi-Slice Acquisitions . . . . . . . . . . . 54
4 Methods 56 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Sequence Design 59
4.1.1 Low-Resolution Acquisitions . . . . . . . . . . . . . . . . . . 59
. . . . . . . . . . . . . . . . . . . . . . 4.2 Spin-Tagging implementation 60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 FAIR 61 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 STAR 61
. . . . . . . . . . . . . . . . . . 4.2.3 Post-Processing of Image Data 62 . . . . . . . . . . . . . . . . . . . . . . . 4.3 The Perfusion Measurement 63 . . . . . . . . . . . . . . . . . . . . . . 4.3.1 Subject Immobilization 64
. . . . . . . . . . . . . . . . . . . . 4.3.2 Functional Data Processing 65 . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Experimental Design 66
. . . . . . . . . . . . . . . . . . . . . 4.4.1 Single-Slice Experiments 67
. . . . . . . . . . . . . . . . . . . . . 4.4.2 Multi-Slice Experiments 67 . . . . . . . . . . . . . . . 4.4.3 The Effect of Inversion Slab Width 68 . . . . . . . . . . . . . . 4.4.4 Slice Profile Effects on Perfusion Data 68 . . . . . . . . . . . . . . 4.4.5 The effect of different inversion times 68
. . . . . 4.4.6 A Seven-SIice Perfusion-Based Functional Experiment 69 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 QUIPPS 70
. . . . . . . . . . . . . . . . . . . . . . . . 4.6 Reproducibility of Results 70
5 Results 71 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Single-Slice Results 71
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 FAIR 71
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.2 STAR 73 . . . . . . . . . . . . . . . . . . . 5.2 The Effect of Inversion Band Width 73
. . . . . . . . . . . . . . . . . . . . . . . . 5.3 The Effect of Slice Profile 74 . . . . . . . . . . . . . . . . . . . . . . . 5.4 The Effect of Inversion Time 75
. . . . . . . . . . . . . . 5.5 Comparison of Response Seen Across the ROI 77 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6 QUIPPS 78
. . . . . . . . . . . . . . . . . . . . 5.6.1 Reproducibility of Results 78
6 Discussion 80 . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Experimental Results 80
. . . . . . . . . . . . . . . . 6.2 The Seven-Slice GRASE-FAIR Sequence 84 . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 The Interleaved Sequence 84
7 Conclusions 86
Bibliography 87
List of Figures
. . . . . . . . . . . . . . . 2.1 Nuclear effect of an extemal magnetic field 14
. . . . . . . . . . . . . . . . . . . . . 2.2 Free Induction Decay schematic 15
. . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 The spin-echo sequence 19
. . . . . . . . . . . . . . . . . . . . . . . 2.4 One-dimensional localization 21
. . . . . . . . . . . . . . . . . . . . . . . . . 2 -5 Slice-selective excitation 24
. . . . . . . . . . . . . . . . . . . 2.6 Two-dimensional spin-warp imaging 25
. . . . . . . . . . . . . . . . . . 2.7 Three-dimensional spin-warp irnaging 26
. . . . . . . . . . . . . . . . . . . . . . . . . . . 2.8 Image reconstruction 27
. . . . . . . . . . . . . . . . . 2.9 Functional imaging expenmental design 32
. . . . . . . . . . . . . . . . . . . . . 2.10 BOLD contrast source schematic 34
. . . . . . . . . . . . . 3.1 Echo-planar imaging pulse sequence schematic 43
. . . . . . . . . . . . . . . . . . . . 3.2 GRASE pulse sequence schematic 45
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 The GRASE signal 46
. . . . . . . . . . . . . . . . . . . . . . . . 3.4 EPI-STAR image formation 50
. . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 FAIR image formation 51
. . . . . . . . . . . . . . . . . . . . . 4.1 Visual stimulus delivery system 58
. . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Visual stimulus pattern 58
4.3 Hyperbolic secant pulse profile . 1 0.24ms duration . . . . . . . . . . .
1.4 Hyperbolic secant inversion pulse profile . 2.56ms duration . . . . . . .
. . . . . . . . . . . . . . . 5 . 1 Cornparison of FAIR acquisition sequences
. . . . . . . . . . . . . . . . 5.2 Comparison of STAR acquisition sequences
5.3 The effect of inversion band width on measured relative perfùsion change .
. . . . . 5.4 Effect of slice separation on measured relative perfusion change
5.5 The effect of varying inversion time on the measured relative perfusion
change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.6 Seven-slice acquisitions with visual area positioned at three separate
. . . . . . . . . . . . . . . . . . . . . . . . . . . . locations intheRO1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.7 QUIPPS Results
Glossary of Terms
ADC ; ANOVA : ASL : BOLD : CBF : CBV : CMRO : EEG : EPt : FAIR : FID : FFT : fMN: FSE : GRASE : MEG : MN1 : M R : Ml21 : MT: NMR : PET : PICORE : QUIPSS : RARE : RF: ROI : SE : SPECT : SNR : STAR : T : TI : T2 : T; : TE : TIC : TR:
analogue to digital converter analysis of variance arteriai spin labeling blood oxygenation level dependent cerebral blood flow cerebral blood volume cerebral metabolic rate of oxygen elec troencep halography echo planar imaging flow altemating inversion recovery fiee induction decay fast Fourier transfonn functional magnetic resonance imaging fast spin ec ho gradient and spin echo magnetoencephalography Montreal Neurological Institute magnetic resonance magnetic resonance imaging magent ization transfer nuclear magnetic resonance positron emission tomography proximal inversion with a control for off-resonance effects quantitative imaging of perfusion using a single subtraction rapid acquisition with relaxation enhancement radio fiequency region of interest spin echo single photon emission computed tomography signal to noise ratio signal targening using altemating radiofiequency Tesla spin lattice relaxation time constant spin-spin relaxation time constant transverse relaxation time constant echo time tirne-intensity curve repetition time
Chapter 1
Introduction
The high-resolution anatomical images produced by magnetic resonance imaging (MM)
scanners have become increasingly important in modem medicine, and the rapid growth
of MRI technology in the past decade has allowed researchers to explore new apptica-
tions of this imaging modality.
In addition to imaging simple anatomy, MR imaging methods that measure dynamic
processes in the body are being actively explored. These methods, which attempt to
image some aspect of biological function, are broadly referred to as functional MRI, or
simply fMRI. The most intensly pursued area of fMRI is the mapping of brain function.
and it is this application which motivates this thesis. In fact, in the context of this thesis,
the term fMRI will explicitly refer to the imaging of brain function.
MRi depends on stimulated signals fiom certain nuclei in the body. This endoge-
nous signal source permits great flexibility in the possible imaging strategies.
fMRI brain mapping is perfonned by acquiring a series of images in the brain re-
gion of interest while sorne stimulus is presented or task is performed. The brain areas
activated during the experiment will show changes in local neuronal and metabolic ac-
tivities, and the relevant portions of the brain images involved will show signal changes
that are correlated with the stimulus or task. fMRl imaging sequences are designed to
be sensitive to some aspect of the metabolic changes associatied with neuronal activity
in order to produce and maximize these signal changes.
The most common method of producing neuronal activity-correlated signal changes
for use in fMW studies is the use of blood oxygenation ZeveZ dependent (BOLD) con-
trast technique. This contrast source relies on the different signal decay rates seen
between regions with diffenng concentrations of deoxyhaemoglobin in the local blood
supply. Activation increases the amount of blood fiowing into an area of the brain and
decreases the deoxyhaemoglobin concentration locally. Signal fiom this area will have
an altered relaxation time constant which c m be exploited to provide image contrast.
BOLD contrast-based fMEU using a single-shot echo-planar acquistion provides good
results, and is relatively simple to implement.
A more direct method of measuring neuronal activity is the measurement of cere-
bral blood flow (perfusion), and this can be performed by altering the magnetization
of blood flowing into the region of interest. Such methods are conceptually similar
to radioactive labelling techniques such as PET (positron emission tomography) and
SPECT (single photon emission computed tomography). These methods have a po-
tentially broader applicability than BOLD, and can be used for quantitative blood flow
measurernents.
The use of BOLD contrast in the vast majority of brain fMRI studies has been
partly due to the limited spatial extent of non-invasive perfusion-weighted imaging
techniques. Contrast agents can be used in MRi perfusion measurements with im-
proved spatial resolution, but problems with build-up in multiple-trial experiments, cost
and increased experimental complexity have limited their use. The goal of this thesis
was to design and implement non-invasive, multi-slice, perfusion-weighted irnaging
sequences to increase the spatial coverage of these techniques, which should greatly
improve their usefilness in research applications.
The thesis consists of seven chapters. The second begins with a brief review of ba-
sic M W theory and proceeds to outline fast imaging techniques, and some applications
of them in the pursuit of fûnctional brain imaging. The signal theory and spin tagging
approaches used in pulsed arterial spin labelling (pulsed ASL) perfusion imaging are
discussed in Chapter Three, along with considerations for the extension of pulsed ASL
to mu1 tiple slice acquisitions. The experimental procedures for both functional imaging
studies and for the individual tests performed to detennine the effect of vanous imag-
ing parameters on functional results are outlined in Chapter Four. The resul ts of the
performed experiments are presented in Chapter Five. The discussion section, Chapter
Six, reviews the results in the context of the function of a seven-slice perfusion acqui-
sition. Chapter Six concludes with the discussion of results obtained at our lab fiom a
sequence used to obtain simultaneous BOLD/perfÜsion measurements in experiments
to elucidate metabolic fùnctioning of the primary visual area of the human brain.
Original contributions made in this thesis are: 1) the development and testing of
low-resolution, single-shot Gradient and Spin Echo (GRASE), half-Fourier GRASE
and echo-planar acquisition and reconstruction sequences; 2) the development and
testing of Flow-measurement using Altemating Inversion Recovery (FAIR) and Sig-
nal Targetting with Altemating Radiofiequency (STAR) pulsed ASL sequences using
the above acquisition techniques; 3)expenmental evaluation of the effects of varying
various sequence design parameters on multi-slice pulsed ASL sequences; and 4) the
design of an interleaved BOLDEAIR echo-planar sequence and its use in numerous
basic activation physiology experiments.
Chapter 2
Background
2.1 MR Signal
Magnetic resonance imaging ( M N ) is based on the nuclear magnetic resonance (NMR)
phenomenon which arkes from certain nuclei in the presence of external magnetic
fields.
The NMR phenomenon involves nuclei that possess an odd number of protons
andor neutrons. Such nuclei necessarily have non-zero angular momentum and rnag-
netic moment, and, as shown by Stem and Gerlach in 1922 [22], they can be made to
interact with extemal magnetic fields to yield observable results. Stem and Gerlach's
experiment involved the splitting of a beam of Ag atoms by an externally-applied, in-
homogeneous magnetic field.
In 1946, Bloch and hrcel l [7, 721, working independently, were given credit for
discovering NMR; a feat for which they were awarded a Nobel Prize in Physics in 1952.
They provided the f b t theoretical insights into the physics describing the behaviour of
the excited nuclei, and it is an extension of their basic experimental setup that is used
in al1 NMR spectroscopy measurements and MRi.
Although it is a quantum-mechanical process, a classical description of the physics
behind NMR is adequate in an explanation of MN. In the simplest case - that of an ex-
ternal magnetic field interacting with the magnetic moments of spin-possessing nuclei
in a sample - the magnetic moment of each spin 1/2 nucleus behaves as a small bar
magnet, and the extemal field will attempt to align these nuclear moments with itself.
individual nuclei cannot be treated classically, and are forbidden from aligning entirely
with the external field. Instead, these nuclei show a majority alignment with the exter-
na1 field, with a defined projection of the magnetic moment along the direction of the
applied magnetic field. In addition to the aligning of the nuclear magnetic moments, the
nuclei in the external field are rapidly precessing. The basic precessional behavior of
nuclei possessing intrinsic angular momentum and a magnetic moment in an external
magnetic field is described by the Larmor Equation:
where B is the magnetic field strength, w the angular precessional frequency of the
atomic nucleus or 'Larmor' frequency and y the gyromagnetic ratio of the nucleus. The
gyromagnetic ratio is a constant unique to eacb type of particle with spin. 'H nuclei
are used in most MRI applications due to their prevalence (by far the most numerous
nucleus in biology) and their high sensitivity. The gyromagnetic ratio for LH nuclei is
42.58 MHdTesla Cl].
Conventionally a Cartesian coordinate system is used to describe orientation in
Figure 2.1 : Formation of the net magnetization vector by the alignment of the majority o f the nuclear spins with the external field
MM, wherein the extemal field direction is along the z mis. ' The physics goveming the behavior of the individual nuclei are quantum mechani-
cal in nature, and a quantum mechanical restriction exists on the possible orientations
of 'H nuclei in extemal magnetic fields. More specifically, because 'H nuclei are spin
1/2 nuclei, two orientations are permitted, one with the projection of the magnetic mo-
ment of the nucleus parallel and the other anti-parallel to the extemal field [22]. The
potential energy possessed by a spin depends on its orientation in the field. The two
orientations can therefore be referred to as separate energy levels.
The population of the spins in the sample is split between the two energy levels with
the lower-energy @araIlel orientation) being favoured by a ratio given by the Boltzmann
factor [l]. At room temperature, the factor is very close to unity because the energy
level difference of the spins is small compared to the thermal energy of the molecules.
The presence of a majority of spins aligning with the extemal field has the effect of
polarizing the sample and creating a net magnetization vector (M) aligned parallel to
the extemal field (Fig 2.1 ). It is the manipulation of this vector that forms the basis of
al1 NMR-based measurements including M M [ 1 1.
'Vector quantities are indicated by bold font, while normal font will represent scalar quantities.
14
Figure 2.2: Free Induction Decay (FID) schematic: the experimental setup (lefi) mea- sures the signal fiom the magnetization vector precessing in the transverse plane. The received signal is dependent on the magnetization vector in the transverse plane, and decays away exponentially with a time constant Tg. At the same time M, recovers ex- ponentially with a different time constant. The Fourier transfonn of the decaying signal yields its fiequency components (spectrum) shown in the lower right.
The magnetization of a sample can be manipdated through the application of a
correctly-tuned radio-fiequency (RF) pulse whose orientation of propagation is perpen-
dicular to the direction of the extemal magnetic field. Such an RF pulse should be tuned
to the Larmor fiequency of the sample spins in order to efficiently transfer energy to
them. The classical description of the interaction between the pulse and magnetization
vector features the magnetic field component of the RF pulse (BI) rotating M away
fiom the extemal field direction. The resonance condition in the classical description
exists in that, for the efficient rotation of M to occur. BI must be synchronized with
the rotating frame of the precessing spins - a condition only met if the RF and Larmor
fiequencies match. By controlling the duration and amplitude of the RF pulse used,
any rotation angle for M can be chosen.
The rotation of the magnetization vector by BI changes its orientation with respect
to the Bo field. The result then is that a component of the magnetization vector will
now be in the plane transverse to the Bo direction - except in special cases of rotations
that are integer multiples of 180".
An experiment in NMR consists of three basic steps: 1) the alignment of the nuclei
in an external Bo field to form a magnetization vector M; 2) the rotation of the vector
via RF fields; and 3) the detection of the resultant signal by coils surrounding the sam-
ple. The component of M rotating in the transverse plane changes the flux in a receive
coi1 and induces an electro-motive force. The detected signal fiom such an event is the
basis of al1 NMR experiments including MRI. A graphical depiction is given in Fig 2.2.
The signal fiom a simple excitation and detection routine is called a free induction
decay (FID). The strongest received signal strength occurs immediately following an
RF pulse that tilts !'II by 90" so that it lies entirely in the transverse plane. Analysis
of the FID signal fiom chemicai compounds fonns the basis of a simple experiment
in NMR spectroscopy. Mucb about the structure of molecules can be deduced from
the amount of magnetic shielding they provide to their constituent nuclei. The amount
of shielding given by chemical environment results in a change in the precessional
fiequency of nuclei. This change in precessional fiequency is called a chemical sh@.
The concept of chemical shifi is important in MRI in that species with a significant
chemical shift and abundance (eg. fat) can be incorrectly
The FiD signal strength decays away exponentially
localized.
fiom its initial value foliow-
ing excitation. The re-growth of M to its steady-state strength and orientation is an
exponential process as well, but this re-growth occurs with a different tirne constant.
The time constant Tl describes Al's re-growth to the steady state M0z. Tl is called
the spin-lotlice relaxation time constant - so called by early NMR researchers [25] .
The basic mechanisms of a system's renim to steady state energy level populations are
the quantum interactions between excited spins and the surrounding molecular environ-
ment. More specifically, an excited spin can give up a quantum of energy to the lattice,
and to do this there needs to exist a magnetic field generated by nearby molecules that is
flucniating at the spin's precessional (Larmor) fiequency. These environmental or lat-
tice fiequencies originate fiom the thermal motion of nearby atoms. The time constant
is a statistical manifestation of the likelihood of these energy transitions fiom the popu-
lation of excited spins. The steady-state populations of energy levels are themselves the
result of dynarnic processes of excitation and energy exchange, and the retum to steady
state afier excitation depends on these same processes. Tl time constants depend on
temperatures and the molecular reonentation fiequencies - characterized by correlation
times, as these in part determine the lattice magnetic field fluctuation fiequencies. and
Bo field strengths as this determines the size of the energy quanta required to cause
transitions [25] .
The exponential decay of the magnetization in the transverse plane (FID) follows a
different time constant, T;. This relaxation behaviour results fiom contributions from
two separate sources: nuclear interactions and the effects of field inhomogeneity. The
contribution from field inhomogeneity can be nulled by experimental techniques. If
such techniques are implemented, the signal decays with T2, the true spin-spin relax-
ation time constant. T2 results fiom nuclear interactions including the transitions that
give rise to the Tl constant, but also Uiclude interactions that degrade the magnetization
vector strength in the transverse plane by de-phasing the spins of the constituent nuclei.
The effect of Bo inhomogeneity on the decay of the transverse magnetization of
an excited sample comes about due to the different precessional fiequencies of the
nuclei in the different regions of the field. These effects can be removed by altering
the RF pulse sequence to inchde a 180" pulse following the initial excitation. Such a
pulse sequence (Carr-Purcell) [12] refocuses the signal that was de-phased by the field
inhomogeneity to form a spin echo. The process of spin-echo formation is shown in
Fig 2.3.
The peak of the spin echo signal is the point at which the spins refocus, eliminating
the effect of Bo inhomogeneity. The decay of the peak value of the spin echo is an
exponential process characterized by the T2 time constant. Subsequent 180" pulses can
be inserted at regular intervals to form an echo train of spin echos. This is comrnonly
done in clinical imaging procedures (e-g. Fast Spin Echo) to lengthen the time for
which signals can be acquired following the 90" excitation pulse [30].
The general behaviour of the magnetization vector can be described phenomeno-
logically by the Bloch equafion [68]:
where M, il&, My and A, are the magnetization vector and its Cartesian components
along the unit vector directions i, j and k.
Figure 2.3: Schematic of spin echo sequence. Afier a 180" excitation is applied at T E / & the spins that had dephased due to field inhomogeneity begin to re-orient them- selves to form an echo at TE.
2.2 Imaging
The extension of NMR fiom simple excitation and reception to MRI involves the use of
additional hardware to mod ie the basic experiment. More specifically, the addition of
Iinear magnetic field gradients allows spatial encoding of the received signal to occur.
The gradient fields are oriented in the sarne direction as the Bo field, but their field
strengths Vary in a linear fashion within the magnet. This gives a spatial dependence to
the magnetic field strength within the field:
or more simply:
where G,, G, and Gz are the field gradients along x, y and z respectively, and B( r) is
the external magnetic field strength at r.
The effect of these gradients is to change the precessional frequencies of the spins
at different locations within the sarnple:
A single gradient field allows simple localization of the spins. If two samples in a
Bo field are spatially separated in the presence of a linear gradient (say G,). and the
gradient direction is the same as that of the samples' separation, the received signal post
excitation will contain al1 of the precessional fiequencies of both samples. The preces-
sional fiequency of the spins in the sarnples will depend on their spatial location. A
Fourier Transform of the data set will then yield a result representative of the samples'
projection (Fig 2.4).
X-ray radiography is based entirely on projection imaging, and, x-ray Computed
Tomography (CT) reconstruction techniques can be applied to MR projection data sets.
The data for such a measurement is acquired at many projection angles - corresponding
to several gradient directions, and reconstructed using cornputer algorithms to yield the
desired image. A more elegant way of acquiring and reconstmcting MR signals uses
multi-dimensional Fourier Transforms of the received MR signals to reconstruct the
image.
The MR signal can be manipulated by the linear gradient fields such that the re-
f requency - Figure 2.4: 1D Localization: in the presence of a gradient field the precessional fie- quencies of the nuclei in the samples depend on their spatial location. The received sig- nal contains contributions from al1 of the nuclei in the sarnples. The Fourier Transform of the received data will quanti@ the amounts of nuclei precessing at each frequency, which is equivalent to the projection image of the sample.
ceived signal provides a Cartesian sampling of k-space (spatial fiequency Fourier space)
that will yield an image of the excited magnetization distribution after reconstruction
with a Fast Fourier Transform (FFT) algonthm. This can be illustrated by consider-
ing the signal equation for an excited two-dimensional sample of spins (given by the
solution to the Bloch Equation) in the presence of linear field gradients:
where kz ( t ) and kg ( t ) are given by:
kx ( t ) = 1 GZ ( ~ ) d r . 27r O
and
or rather, the wclvenumber modulating the received signal is given by the time integral
of the field gradient as they are applied to the excited spins.
The two-dimensional Fourier Transform of the desired image is given by:
Equations 2.5 and 2.5 are identical in form, and imply that a point in the k-space
interpretation of an image can be inferred directly fiom the received signal strength
of the object if the excited spins are correctly manipulated by the gradient waveforms.
Furthemore, the k-space position of the received signal is given by the time integral of
the gradient wavefonns (Equations 2.6,and 2.7).
Slice-selective excitation utilizes short RF pulses that possess a discreet bandwidth
of frequencies - al1 pulses of finite duration must be composed of a range of frequencies
depending on the pulse shape. The pulses used are typically tnincated sinc or Gaussian
in shape and centred around the central carrier fiequency which is the Larmor fiequen-
cy for the spins to be excited [68]. Sinc and Gaussian RF pulse shapes are prefemed
in most applications because they result in the most usefûl slice profiles for imaging
applications.
The process of selective excitation is based on the requirement that, for the spins
of the sample to receive energy fiom the RF field, there must exist resonance between
the precessing spins and the RF field, ie. the RF field must at least contain the Lar-
mor fiequency of the spins to be excited. If the nuclei experience a gradient field in
addition to Bo such that spins are precessing at fiequencies outside the bandwidth of
the excitation pulse, only those spins that precess within the region corresponding to
the RF bandwidth will be excited, leaving those spins precessing outside the RF range
unaffected (Fig 2.5).
After excitation, the received signal is demodulated (to remove the camer fie-
quency) and sampled by an analogue-to-digital converter (or ADC) before being stored
at the appropriate position in the raw data (k-space) matrix. The matrix is fiIled by
rapidly sarnpling data with the ADC as a gradient is applied to move the signal Iinearty
through k-space. The particular gradient used in this manner is retèrred to as the read-
out gradient. The k-space trajectory is moved in the direction perpendicular ta the
readout direction by the application of short gradient pulses. These carefully chosen
phase encode gradient pulses allow readout lines to be shifted so that a full two dimen-
sional sampiing of the desired range of k-space can be obtained. This strategy can be
directly extended to three dimensions by phase encoding in the slice selection direc-
tions. Simple two and three dimensional imaging sequences along with their k-space
trajectories are shown in Fig. 2.6 and 2.7; such sequences are exarnples of spin-waa,p
imaging.
time 4 distance j +
profile
G-slice
Figure 2.5: Slice Selective Excitation Schematic: the RF pulse contains a well-defined range of fiequencies (shown by the Fourier Transform of the wavefom) which corre- sponds to a select range of precessional fiequencies in a sarnple of spins in the presence of a gradient field. The spins in this siice of the sample will be the only ones excited by the RF pulse.
The signal received by the ADC as the readout Iine is acquired (concomitant with
the passage between negative and positive k-space values) is called a gradient-wcczlled
echo. The centre position of k-space is weighted by the signal strength of the sample
spins precessing coherently. Positions off-centre are samples of the spins that have
a phase modulation in the directions of the gradients that were applied to the spins.
The echo is formed as the phase of the spins is untwisted from an edge of k-space
to the central (coherent) position and then re-modulated as the far edge of k-space is
approac hed.
Care must be taken in the acquisition of the k-space data for an image. The sample
spacing must be sufficiently close to ensure that the reconstructed image is fiee from
N D I I
time - Figure 2.6: Two-dimensional spin-warp imaging: The sequence of RF and gradient pulses move the signal around in k-space. Each excitation is followed by a G, gradient pulse (the phase-encode gradient) that moves the signal along k,. and a shon negative and longer positive G, gradient pulse (the readout gradient) moving the signal left and then nght along k,. The Gy pulse is shown as a box as many repetitions of this sequence using different Gy gradient pulse magnitudes of both signs are used in the filling of k- space.
the effects of aliasing. These effects are avoided by ensuring that the image data is
sarnpled to fulfill the Nvquist criterion which dictates that the sampling frequency must
be at least double that of the highest frequency found within the received signal [13].
There is always a tradeoff in MRI between the achievable spatial resolution, acqui-
sition time and signal-to-noise ratio (SNR) possible in different scanning procedures.
The time taken by conventional s c a ~ i n g techniques is given by the product of the num-
ber of lines that need to be acquired and the interval between excitations (as one line is
acquired for each excitation in the simplest case).
One of the great strengths of MRI is the ability to provide images weighted with
several different types of contrast simply by varying some of the parameters of the
Figure 2.7: Three-Dimensional Spin-Warp Imaging: The sequence in 2.6 is extended so that kz lines are now acquired following G , gradient pulses (phase-encode gradient). which move the signal along k:.
acquisition. CT scans provide image contrast based on subject x-ray attenuation co-
efficients. MRI acquisitions are commody weighted to provide image contrast based
upon proton density, Tl , and G. Newer, more intncate acquisition sequences can pro-
vide images weighted by other physiologically relevant parameters, such as difision,
perfision and flow velocity 1131.
2.3 Fast Imaging
Fast techniques have been developed which greatly reduce the time required to acquire
an image. Short imaging times are clinically useful in that the motion of an organ or
body part can be effectivelyfrozen if the image is taken quickly enough. There are also
physiological phenornena that would be impossible to measure with MR if the image
times were not appreciably short.
magnitude raw data magnitude reconstructed image
Figure 2.8: An image is formed fiom its k-space data.
Fast imaging techniques reduce the time required to fil1 k-space through the use
of several different modifications of the basic sequence. The reduction of the repeti-
tion time between excitations (TR), the acquisition of multiple phase encode lines per
excitation and the reduction of the total number of phase encode lines are al1 used.
An effective method to reduce the time per scan is to acquire oniy partial k-space
data sets for individual images - there is a fundamental redundancy (Hermitian sym-
metry) in the k-space representation of a real image that allows reconstmction to occur
with as little as one half of the full data set.
In recent years, modifications to MR scanner hardware have enabled entire k-space
data sets to be sampled following a single excitation. Such modifications require pow-
e h 1 gradient drivers and finely-tuned coils to function comectly. So-called single-shot
techniques have permitted imaging of physiological phenornena in times that are typi-
cally in the tens of milliseconds [63].
2.4 Functional Brain Imaging
The functioning of the human brain is an important topic in science, with far-reaching
implications for modern medicine. The gross anatomy of the brain has been more or
iess known for some time, but only in the past century has technology progressed to the
point where accurate fwictional importance can be ascribed to the myriad structures to
be found.
Brain function involves a complex physiological interrelationship benveen electri-
cal and chemically-mediated events. The basic unit of brain fùnction is the neuron.
which is an excitable ce11 that conducts electrical signals. or action potentials through
the use of voltage-gated ion channels in the cellular membrane.
The electrical activity of functioning neurons can be detected at short distances
by measurements of the electrical field. The technique of electroencephalography, or
EEG was originally performed directly on the surface of the brain in animals, but soon
afier the technique was extended to the placing of electrodes on the scalp. Simple
localization of the measured activity could be inferred by keeping track of the signal
amplitude measured at various locations on the scalp surface. The use of multiple
electrode arrays allowed these measurements to be done simultaneously [14].
Magnetoencephalography, or MEG, is a relatively new innovation that provides
higher-resolution activity maps than EEG (within a few millimeters as opposed to tens
of mm's in EEG) [27]. MEG is based on the detection of magnetic fields that are in-
duced by the neuronal electrical currents. MEG and EEG stand alone in methodologies
that non-invasively and directly measure electrical events in the brain, but these tech-
niques suffer the rigorous calculation routines required to do the source localizations
and the prohibitive cost of MEG machinery.
Auto-radiographic studies of animal brains begm in the 1950's. These techniques
were centred around the injection of a radioactive dye while the animal was being
exposed to a stimulus. Immediately following the stimulus, the animal would be sacri-
ficed, and a thin section of the animal's brain would be taken and placed in contact with
a radiographie plate [5 11. Functionaily important areas in the brain associated with the
stimulus would show preferential uptake of the radioactive tracer and expose the plate
to a larger degree. Development of the plate would allow these areas to be localized
with high resolution. Although modem scaming techniques have reduced the need for
animal autoradiograph studies in recent years, the technique still sees use due to the
high resolution of the data acquired.
The use of radioactive substances to measure brain fknction was also extended to
in vivo studies in humans. The passage of a radioactive bolus in the blood could be
quantitatively measured by the use of a pulse counter sensitive only to a small section
of the brain. The numbers of nuclear events recorded would provide an estimate of
the amount of radioactive blood in the small area of the sensor's field. Single photon
emission tomography (SPECT) and positron emission tornography (PET) were intro-
duced in the 1970's to provide images of radioactive tracers in the body [2]. PET in
particular was used to study several physiological parameters in the brain, and these
measurements could then be used as the basis for localizing brain function changes
brought about by various stimuli. PET had the advantage of localizing the radioactive
tracers within the human brain to a higher resolution than SPECT (about l5mrn) , with
an acceptably low radiation dose to the subject.
PET functional studies typically measure brain perfusion as the correlate of brain
activity in functional studies. Perfusion as a parameter quantifies the passage of blood
water into the tissue, and is an important physiological parameter, as it provides in-
formation about possible brain pathology and it is the basis for calculations of other
brain physiology parameters [80, 171. For Our purposes, perfusion will be assumed to
be equivalent to measurements of cerebral blood flow (CBF)[2].
Problems with PET-based functional imaging include a limited spatial resolution
and a low, but significant radiation dose to the subject. Cost is also a factor; the isotopes
used in al1 human nuclear medicine-based imaging are very short lived, and do not store
well. PET imaging facilities require that a cyclotron or some other related equipment
be located nearby to generate the necessary tracers for the studies.
The explosion of MRI technology in the eighties provided a rneans of obtaining
high resolution anatornical images of the body without exposing the subject to a dose
of ionizing radiation. The extension of PET-based functional imaging techniques to
MRI was first studied in the late eighties [6. 5, 77, 761. In these studies, the radioactive
tracers were replaced with substances that locally disrupt the magnetic field seen by
blood water. Typically images of the region of interest would be rapidly acquired while
a bolus of contrast agent passed through. Fluctuations in quantitative measurements
of the signal fiom the region could then be correlated with the stimulus given to the
subject, and functional localization of the resultant brain activity could be obtained.
Problems with contrast agent-based functional MRI (fMRi) are in large part due
to the limited reproducibility of experiments due to the accumulation of contrast agent
in the blood of the subject. The situation changed with the discovery that endogenous
contrast sources related to correlates of neuronal activation existed and could be utilized
in functional studies by correct pulse sequence designs. Images could be weighted to
show contrast based on blood oxygcnation or perfusion of tissues - through the use
of inversion pulses to prepare in-flowing blood so that it would act as its own contrast
agent [ 1 81.
MR-based fûnctional studies follow several different paradigrns, with the block-
design study being the most common, In a block study, the experiment consists of a
series of altemating t h e penods (blocks) of stimulus presentation or task performance
and control. Throughout the study, images are acquired of the region of interest in the
brain (Fig. 2.9). More recent experimental design modifications involve changing the
time course of stimulus presentation. Event-related functional studies allow researchers
to investigate brain responses to stimuli of very short durations - a more realistic repre-
sentation of many normal fünctions [78].
A functional imaging study can produce results of various forrns. The raw data set
afier reconstruction, is composed of a series of images that show the same region of
interest throughout the time course of the study. From the raw data, activation maps,
time-intensity curves and quantitative information c m potentially be calculated (Fig
2.9).
Activation maps are images that represent the statistical significance of a voxel-by-
voxel companson of the intensity values in the raw data set with the applied stimulus
in the study. A wide variety of statistical methods are used in the analysis of functional
data sets [26,90].
Time course data, or time-intensity curves, are plots of the pixel values in a region
of interest across the time scale of the raw data set acquisition. Modulations in pix-
el values averaged or summed across the region of interest show the degree of signal
change brought about by the stimulus. Regions of interest centred on areas with fiinc-
tional significance to the applied stimulus (ofien defined via an activation map) should
show changes in pixel values closely correlated with the stimulus. Other regions not
functionally involved with the stimulus should show no modulations in pixel values
outside of normal noise values.
Figure 2.9: fMRI: the block design schematic (lefi) outlines a comrnon procedure used in fMRi. Many images are acquired of the same slice or slices (top) while a stimulus is applied for discrete time periods. A tirne-intensity curve (right) shows modulations of the pixel values in a region of interest across the time course of the experiment which correspond to the times of stimulus presentation in the block design.
Quantitative data fiom activation studies attempt to relate the degree of pixel val-
ue change seen between activated and baseline States through the course of the time-
intensity curve to a physiologically relevant scale. Periùsion-based functional studies
are good candidates for quantitative results, as perfusion is itself one of the most phys-
iologically important parameters used to signie the degree of brain activation.
2.5 BOLD Contrast
The basis of blood oxygenation level dependent (BOLD) contrast techniques is that
the magnetic nature of the haemoglobin molecule within red blood cells is dependent
on whether or not it is carrying oxygen. Deoxyhaemoglobin, possesses an unpaired
electron and hence is paramagnetic as this electron will align its magnetic moment
with the applied field. The magnitude of the magnetic moment of an unpaired electron
is many times that of a 'H nucleus and will have a strong disruptive effect on the signal
seen From nearby spins due to its effect on the local B field . Oxyhaemoglobin is
slightly diamagnetic as paired electrons will align in opposite orientations in an extemal
field and cancel one another's effect on the local field environment. Such molecules
contain electronic currents inducing magnetic fields that weakly oppose the applied
magnetic field - hence the diamagnetism. At 1.5T, T1 of blood varies from 30ms to
25Oms for the range of oxyhaemoglobin concentration from 30% to 96% [91]. The
susceptibility difference between the two oxygenation States is [82]:
A x ~ ~ ~ ~ ~ - l x o a = +0.18 x 106
(2.9)
The phase change in spins caused by deoxyhaemoglobin's field disruptions shorten
the T,* relaxation time constant locally. BOLD-weighted images are acquired with
timing parameters chosen to maximize these T.'. differences in the generation of image
contrast.
The exact relationship between the BOLD signal change and functional activation
is complex, and still a matter of some debate [45]. What is observed is that areas or"
increased neuronal activation show an increase in cerebral blood flow and a decrease in
the concentration of deoxygenated blood. This results in a higher signal fiom these acti-
Figure 2.10: (top) BOLD Baseline: The blood passing through the capillary delivers a set supply of oxygen to the tissue. A concentration of deoxyhaemoglobin is present that shortens the T,* tirne constant of the local signal. (bottom) BOLD Activation: Brain activation stimulates an increase in the local blood flow resulting in an infiux of oxyhaemoglobin which decreases the concentration of deoxyhaemoglobin in the capillary and lengthens the local T; time constant.
vated areas during image acquisition. Since the images used in BOLD studies are heavi-
ly T;-weighted, they are also susceptible to various artifacts. In particular, T;-weighted
images are degraded by the effects of magnetic field inhomogeneities brought about by
magnetic susceptibility differences. Such di fferences in susceptibility are prevalent at
the interfaces between tissue and air; limiting the use of BOLD contrast in brain studies
where the region of interest lies near the air sinuses or brain stem, for example.
2.6 Perfusion Weighted Imaging
Measurements of brain perfûsion are of use in the diagnosis of disease and brain injury,
and c m also form the basis of functional brain mapping.
MN-based measurements of perfusion were first performed in the late 1980's using
exogenous contrast agents [62]. The subject in these studies would be injected with a
paramagnetic contrast agent which would be imaged as it passed through the region
of interest [77]. Ln the early 19901s, methods were developed in which the contrast
agents were replaced by magnetic labeling of blood water spins immediately proximal
to the region of interest. These methods. generally referred to as artenal spin labelling
(ASL), effectively transformed blood water into an endogenous contrast source and
allow increased repeatability of perfusion experiments in a single session.
Problems with MRI-based perfusion measurements have stemmed fiom low SNR.
the difficulty of producing quantitative results and the limited spatial extent that can be
imaged at once - early methods could only be performed in a single slice. The physical
basis of ASL techniques are expanded upon in the next chapter.
Chapter 3
Arterial Spin Labelling
3.1 Introduction
Arterial spin labelling (ASL) perfusion imaging involves the acquisition of two sepa-
rately prepared images. These images are then subtracted to yield a d~rerence Nnage
that will have pixel values dependent on the amount of perfusion occurring in the im-
aged region. These images are generally called tag and conrrol images - depending on
the sequence of pulses that were used to prepare them.
The original arterial spin labelling approaches used continuous saturation or inver-
sion of the spins of a defined volume proximal to the imaged slice [18,83]. Continuous-
wave experiments of cerebral perfusion often use a separate excitation coi1 placed
around the neck of the subject for the inversion of inflowing spins. The inverted spins
flow through the brain and exchange with the tissue in the imaging regions. An image
thusly acquired, the tag image, will show some contribution fiom the inverted water
spins. The effect of inverted spins on the image can be isolated by subtracting fiom
the tag image a similar image that was acquired without the spin inversion, the control
image.
Although a higher signal-to-noise is theoretically possible with continuous-inversion
methods [87], the technique suffers fiom problems associated with excessive energy
deposition and rnagnetization transfer (MT) artifacts when the separate inversion coi1
is not used. Magnetization transfer contrast comes about due to interactions between
excited and non-excited ' H nuclei [84]. The effect in continuous-wave inversion comes
from the off-resonance inversion pulse which doesn't excite water spins, but will excite
' H nuclei in macromolecular environrnents. Macromolecular spins have very broad
spectral linewidths and can be excited well away fiom water proton resonance - which
is very narrow [a]. The macromolecular spins that have been excited can then transfer
their magnetization to nearby water spins and change the image contrast.
h l s e d ASL techniques provide a means of controlling subtraction errors due to
magnetization effects, while providing reasonable - al though lower - SNR wi thout the
need of additional hardware to invert the inflowing spins.
3.2 Theory
A comprehensive model of the ASL signal in tissue is necessary to interpret images
in physiological terms. Viirious models specific to individual experimental techniques
have been proposed 159, 1 1, 83, 55, 54, 73, 18,961 , but recent work by Buxton et.al.
[ 101 has unified the theories into general and standard models of ASL signal for both
continuous and pulsed techniques.
The general model consists of a signal function that depends on three separate and
independent func tions representing assumptions made about the introduction, washout
and decay of inverted spins in the image slice. The standard model is the solution to
the general model with a specific set of ideal assumptions about the three independent
functions. Other solutions to the general model based on more physiologically real-
istic assumptions can be obtained, but the mathematics involved quickly become very
complicated and numencal methods ofien need to be employed to solve the equations.
The general model is based on considerations of the relative concentrations of
tagged and untagged spins in the tissue; compounded with the relaxation of the tagged
spins with the time constants of their surroundings: first with the Tl of blood - Tib - and
then with TI of tissue (after the spins have been extracted). The model represents the
magnetization signal in the voxels of the difference image and has the form:
where c ( t l ) represents the normalized concentration of magnetization amving at the
voxel at time t' (delivery function), r ( t - t l ) , the residne fincrion, gives the fraction of
the tagged spins that arrived at time t' and remain in the voxel at time t - accounting
for the effects of spin washout, and m ( t - t'), the magnetization relaxationfuncrion,
represents the fraction of the magnetization of the spins that arrived at t' and remains at
t -accounting for the reduction of magnetization by the decay process.
From the general model, the direc! relationship between pixel intensity values and
the amount of flow f in the tissue is shown. A set of ideal assumptions about the nature
of c( t ) , r ( t ) and m(t) are made to provide a specific solution to Equation 3.1 which is
applicable to most ASL applications.
The idealized representations of the component parts of Equation 3.1 that comprise
the standard ASL signal model are:
1. The delivery fiinction used in the standard model cornes about fiom the assump-
tion that the tagged spins arrive in the imaging slice via uniform plug flow. The
plug is created at a distance - typically about a centimeter - fiom the imaging
slice, and therefore a short. but unknown time period bt will pass before any of
the tagged spins reach the tissue of interest. The plug is created with a defined
width and will hence flow through the slice for a defined period of time Ï. The
short, well-defined duration of the plug fiow model allows the delivery function
to be represented as:
during the time period that the plug is passing through the image slice, and c ( t ) =
O before and afler.
2. The residue function is based on assumptions about the nature of water ex-
change between blood and tissue in the image voxels. Single-compartment ki-
netics are assumed, and this implies that instantaneous exchange between sub-
compartrnents that may exist within the voxel occurs, and therefore the tissue
concentration of inverted spins is related to the vascular concentration by a con-
stant ratio equal to the tissuehlood partition coefficient of water, A. The function
has the form:
r ( t - t') = exp[- f ( t - t ' ) /A ] .
3. The rnagnetization relaxation function assumes that the transition fiorn the vas-
cular to the tissue environment for the water spins occurs instantaneously. The
water entering the image voxel does so at a time t' and the magnetic tag decays
fiom that point forward with the relaxation time of tissue T l . This results in a
fiinction of the fonn:
The expression for the signal strength in the difference image is then given by the
solution to Equation 3.1, with the assumptions given by Equations 3.2,3.3 and 3.4:
where the expression q, is dimensionless and represents the processes of relaxation and
clearance that are occumng in each time domain; it has the fonn:
40
where:
and
The standard model solution suggests that the received difference image signal de-
pends strongly on the parameters At and r. This is a probiem from an experimental
point of view as these values are typically not known beforehand (as they depend di-
rectly on the rate of blood flow), may change with activation during a functional ex-
periment, and require that measurements be made at several different inversion times
(TI) before they can be estimated. Relative quantitative perfusion measurements - as-
sessing the fractional change of pixel values between activated and baseline States - are
less sensitive overall, but changes in the transit delay At can affect these calculations
as well. Quantitation of perfusion using the standard model requires that the technique
used should be done in conjunction with measurements of At and T l , or be modified to
minimize the effect of these parameters on the acquisition.
3.3 Pulsed ASL Sequences
Several single-shot image acquisition sequences are applicable to pulsed spin tagging
approaches. Signal to noise ratio, image artifacts, resolution, imaging time, and energy
deposition in tissue Vary between these sequences, and represent factors which must be
weighed when selecting an optimal sequence for a particular application.
3.3.1 Fast Acquisition Techniques
Perfusion measurements with pulsed ASL depend on the rapid measurement of the
signal from prepared spins in the slice or volume of interest. The dynamic nature of
these signals requires that the image be measured quickly.
The fastest MRI acquisition techniques are based on the acquisition of a complete
set of k-space data with a single excitation. Such techniques, called single-shot ac-
quisitions, require high-performance gradient systems to allow the necessary k-space
coverage in the short period of time that the signal exists following an excitation.
Problems exist in single-shot techniques due to several effects: chemical shift; the
phase errors that accrue due to error in the signal sampling; the signal difference be-
tween readout lines acquired from different sections of the signal decay curve; and
several others. These need to be corrected either through reconstruction algorithms or
modifications to the acquisition technique.
The conceptually simplest single-shot technique, echo-planar imaging (EPI), was
the first to be developed [63]. EPI is essentially a single-shot version of a standard
spin-warp acquisition as s h o w in Fig. 2.6 (previous chapter).
Post-excitation, the signal phase is manipulated by applied gradient waveforms so
that, in the k-space representation, it sits at a corner of the space to be filled. Raster lines
of k-space data are then acquired as an oscillating hi&-intensity gradient is applied in
the readout direction. The received signal is moved in the phase encode direction by
the application of a continuous, low-intensity gradient field or the application of higher-
intensity, short-duration gradient blips between the readout lines (Fig. 3.1 ).
Figure 3.1 : (top) Gradient echo (GE) echo-planar imaging (EPI) sequence schematic: n i e sequence features a 90" slice-selective excitation pulse (applied in the presence of the slice-select gradient Gs) followed by an oscillating gradient in the readout direc- tion (GR) and short, phase-encoding gradient blips (Gp) applied between readout line acquisitions. The gradient echoes formed by the oscillating readout gradient are shown on the RF line. (bonom) The k-space trajectory of a blipped echo-planar acquisition.
The SNR of an EPI acquisition can be improved, and the Bo inhomogeneities re-
duced, by inserting a 180" pulse between the excitation pulse and readout window such
that the readout window is centred on the resulting spin echo peak. The time window
over which strong signal can be detected is extended, and the signal strength at the cen-
tre of k-space - which determines the global intensity of the reconstructed image - is
significantly increased. Such a sequence is called spin-echo EPI (SE-EPI) [68].
By inserting 180" refocusing pulses between each readout iine and acquinng each
line at the centre of the resultant spin echo, the signal strength at each line of data
c m be increased although the time between hnes increases. Such a sequence is titled
RARE for t-apid acquisition with relaxation enhancetnent or FSE for fasr spin echo
[30]. RARE sequences ofien have a fürther modification in that the order in which
the readout lines of k-space is changed so that the centre lines of k-space are acquired
before the outlying ones. This modification, temed centric phase encoding, improves
SNR and decreases the blur in the reconstmcted image due to T2 decay between readout
Iines, resulting in less T2 weighting in the reconstructed image.
Because the lines of k-space in single-shot techniques have to be acquired from a
single decay curve, the signal strength o f lines read early in the acquisition will typically
be higher (excepting SE-EPI). Consecutive lines of k-space acquired with a gradient-
echo EPI technique will show signal differences based on the T; time constant; consec-
utively acquired RARE lines are different by a factor based on T2. The simple, linear
manner in which k-space lines are acquired in EPI results in a blur artifact in the recon-
structed image. This blur c m be calculated by considering the T; exponential decay
fbnction as it applies across the k-space trajectory. In addition, the central k-space line
is acquired after half of the acquisition window has passed. This results in a reduction
in SNR for the reconstructed image. The reordering of a RARE sequence places the
first - and strongest in signal - echo at the centre of k-space, and consecutive echoes
near to the origin on either side. The resultant image is of higher SNR than one ac-
Figure 3.2: (top) GRASE imaging sequence schematic; The pulse sequence consists of slice-selective excitation (applied in the presence of a slice-select gradient Gs) fotlowed by a senes of 180" inversion pulses that f o m an echo train. Through the formation of each echo, oscillating readout gradients (GR) are applied that form three gradient echoes per spin echo (five and seven GREISE are also commonly used). Between each echo, phase encode pulses (along Gp) are applied as short blips, and the spin phase is re-set to zero between inversion pulses. (bottom) The centre portion of a k-banded GRASE k-space trajectory. The numbers at the left hand side outline the sequence in which the lines were acquired (e.g. the third line acquired is numbered as 3). The numbers shown outline the centre of k-space in the GRASE sequences developed here, which used 5GREISE.
quired via EPI, and suffers less fiom the effects of T2 decay as consecutive lines differ
by a smaller factor. Problems with RARE centre around the longer imaging time re-
quired and the increased energy deposition (due to the large number of RF pulses) in
the imaged tissues.
The Gradient And Spin Echo or GRASE sequence is a hybnd of the EPI and RARE
Phase Encoding Lim
Figure 3.3: Schematic of the signal strength and phase modulation in k-banded GRASE imaging [53]. The three bands (centre, left and right) are acquired sequentially. With- in each band, the phase-encoding pulses are arranged such that readout lines obtained at the same position relative to the spin echo are piaced together. The pattern of the readout lines is chosen to minimize coherent patterns in the signal strength modula- tions across the phase-encodes in k-space. Such coherent patterns will result in image artifacts after reconstruction. The phase Iine in the figure shows the position relative to the spin echo at which the readout Iine was sampled. A sequence with 75 readout lines is shown.
techniques [71]. In this rnethod a series of 180" pulses is used to generate an RF-
refocused echo train, and several gradient echoes are acquired over eac h spin echo. This
results in a net decrease in the acquisition time when compared to RARE imaging be-
cause the number of time-consuming 180" pulses is significantly reduced. The GRASE
acquisition typically takes longer to acquire than EPI measurements, but, due to the
refocusing pulses, the signal strength rernains high enough to measure for a longer pe-
riod of time. A GRASE sequence with only one 180" pulse is a (reordered) SE-EPI
sequence, and one with only one gradient-recalled echo per spin echo is a RARE ac-
quisition. The GRASE and RARE acquisitions' refocusing of the signal also results
in the calculated image being less prone to the artifacts created by static field inhomo-
geneities due, for example, to magnetic susceptibility di fferences in the imaged object.
interfaces between air and tissue or tissue and bone do not have uniform magnetic field-
s. This is due to the change in magnetic susceptibility at the interface of the two. The
change in susceptibility tends to dismpt the magnetic field locally and can drastically
change the value of T; in this region. The refocusing pulses in the echo train result in
the tissue signal decaying with T2 instead of T;, and the resulting image will be free
from these artifacts.
The ordering of k-space readout lines in GRASE acquisitions can be done in a
number of ways including linear, 'standard', partially randomized and k-banded (kb)
GRASE [53]. The short total acquisition time of funciional brain images makes the kb-
GRASE ordering scheme the logical choice, as this method produces the least amount
of artifacts for large field of view images in a short penod of time [53].
The speed of an acquisition is greatly increased for sequences that require only a
portion of k-space for image reconstruction. Full images c m be reconstructed from ac-
quisitions where as linle as half of the required k-space data is measured. The missing
data fiom a partial acquisition can be accounted for by changing the relative weighting
the existing data prior to reconstruction. This is possible because the magnetization
distribution of the object being imaged may be considered to be pure real, and this im-
plies that the k-space representation of the object must be Hermitian. A Hennitian data
set is symmetncal about a diagonal line through the origin of k-space. For incomplete
acquisitions, the rnissing data points c m be extrapolated fiom the existing data by a
point-by-point calculation of complex conjugates about the diagonal.
Problems with the reconstruction of partial-Fowier acquisitions corne about fiom
the presence of low-fiequency-phase modulations in the magnetization distribution that
cause artifacts in the resultant image - that is, the magnetization distribution being re-
constructed is not purely real [16, 28, 521. There exist several correction methods for
removing this phase error, including the use of homodyne detection [70]. In the homo-
dyne technique, k-space is filled such that the high fiequency data beyond a few lines
on one side of the origin are left blank, and the remaining k-space is read normally
(approximately 65% of the fi111 k-space coverage was acquired in the partial-Fourier
sequences developed here). Then two reconstructions are done: one on the fùll data set
with weighting of the high frequency terms to account for the missing data, and one
which reconstmcts only the low-frequency terms that exist on either side of the origin
in the partial acquisition. The low-fiequency reconstruction results in a reconstmcted
image that is of a very low spatiaI resolution, but contains the tow-frequency phase er-
rors that reduce the resolution of the image formed by the reconstruction of the entire
partial data set. By subtracting the phase of the low-resolution image from the phase
of the entire-set reconstruction, and then taking the real part of the resultant complex
matrix, an image is generated that is fiee fiom the phase error caused by the reduced
k-space acquisition.
33.2 Spin Tagging Approaches
Two spin tagging approaches, STAR [21] and FAIR [59, 541, are currently the most
widely-used pulsed ASL techniques. STAR is the older of the two, and is the most like
continous-wave ASL which preceded it. A third imaging technique, PICORE [89, 861,
is the most recent addition to the ASL family. This sequence is very similar to the STAR
technique, and differs only in the way in which the effects of magnetization transfer are
controlled.
STAR - s ipal fargeting using altemafing radiofiequency, is a spin labelling ap-
proach used to produce perfusion-weighted images [2 11. This technique most-often
uses an EPI sequence to acquire the images. and the term EPI-STAR is thus used to
reference such acquisitions. The technique involves the acquisition of two separately
prepared images of the same slices in the brain. One of the images is preceded by a
slab-selective inversion pulse applied proximally to the region of interest. Image acqui-
sition follows afier a time penod has elapsed to allow inflowing spins to perfuse tissue
in the region of interest. The images thus acquired are now flow-sensitive and are re-
ferred to as tag images. The second image set is acquired in a similar fashion with the
difference being that the slab-selective inversion pulse is applied distally to the region
of interest. These images should show no flow-sensitivity, and are referred to as being
the conîrol image. A subtraction of the control fkom the tag images yields difference
images that are weighted such that pixel values are directly proportional to flow. STAR
is shown schematically in Fig. 3.4.
lnstead of offset slab-selective inversion pulses, the FAIR - Jlorv-sensitive alrer-nat-
hg inversion recover y - technique employs slab-selec tive and non-selec tive invers ion
pulses centred on the slice(s) of interest in the generation of the tag and control images.
In FAIR, the control image is acquired following a non-selective inversion of the
spins surrounding the image slice, and the tag is obtained by inverting only a small
volume of spins immediately surrounding the image slice. Flow-sensitivity only exists
in the selective-inversion images as inflowing spins have a different magnetization his-
tory than those within the image slice - providing image contrast. The non-selective
inversion is insensitive to the effects of flow as there is almost no difference in the
magnetization histories of infiowing and image slice spins [55, 591. The magnetization
transfer effects are equal in both images, and hence should not appear in the difference
image. FAR is s b schematically in Fig. 3.5.
Figure 3.4: (top) EPI-STAR tag image formation. The tag image is forrned by inverting the spins in the tag region with a slab-selective inversion pulse. The inverted spins then flow through the region of interest and perfùse the tissue there, where they provide contrast to the acquired images at Tl . (bottom) EPI-STAR control image formation. The control image is formed with the slab-selective spin inversion pulse applied distally to the region of interest so no tagged spins flow through the image slices pnor to imaging.
Inc luded for completeness, the PICORE sequence -proximal inversion with a con-
trol for ofiresonance effects - is identical to the STAR technique, with the exception
that the control image is acquired with the distal slab-selective inversion pulse replaced
with a repetition of the proximal inversion pulse, but now in the absence of the selection
gradient. The off-resonançe pulse will then invert no water spins, but the macromolec-
ular spins will be excited identically in both images (resulting in no magnetization
transfer contrast in the difference image). The PICORE modification has the advantage
that only spins proximal to the image slice are inverted, and therefore blood flow into
Figure 3.5: (top) FAIR selective inversion pulse (tag image formation). The inversion band is centred around the region of interest and only a small volume of spins is invert- ed. The inflowing spins have not been inverted and hence behave as the tag spins in EPI-STAR. (bottom) FAIR non-selective inversion pulse (control image): here the spins are non- selectively inverted. Tagged blood flowing into the region of interest wiIl now be in- verted and act as the spins in the control image in EPI-STAR.
the slice fiom areas distal to the image slice contributes nothing to the difference image.
The dependence of the difference image signal strength on the parameters At and
r (Equation 3.5) imply that, for absolute or relative quantitative measurements of brain
perfusion, these parameters need to be measured or ehminated fiom the signal equation
by experimental design.
The development of simple pulse sequences that allow quantitative, perfusion-weighted
difference images to be generated with just two image acquisitions was performed by
Wong et a1.1871. Two modifications to the basic pulsed ASL methods were created and
titled QUIPSS, for quantifutive imaging of perfusion using a single subtrac fion, and
QUIPSS 11. These modifications are applicable to al1 of the pulsed ASL techniques,
and allow quantitative results to be directly obtained by eliminating the unknown pa-
rameters fiom the signal equation.
The original modification - QUIPSS - eliminates the dependence of the signal on
the transit delay At by saturating the image slice after a time delay TI1 chosen such
that:
T I L > At.
The image is then acquired at TJ2, where:
The tagged blood then enters the slice for a time period AT1 = T l 2 - TlI, and the
standard mode! solution of the signal equation for the difference image becomes:
Under these exact circumstances, the signal equation has lost al1 of its dependence
on the transit delay At. By accurately estimating the physiological constants in the
equation, a quantitative difference image can be produced. The importance of this
modification to perfusion-based functional imaging is that any variability of the transit
delay due to activation-induced changes in blood flow is eliminated. This enables the
direct cornparison of difference image pixel values to obtain quantitative relative blood
flow changes between activation and baseline conditions without further measurement.
The QUIPSS method has the limitation that only a single slice at a time can be
measured. The saturation of the imaging slice after inverted spins have begun perfusing
the imaged tissue means that slices distal to the first slice to be perfised will experience
variable amounts of transit delay that will affect the quantitative nature of the di fference
image in a manner similar to the transit delay existing between the inversion band and
the imaging of the slices.
A second acquisition sequence was developed that provides multi-slice quantitative
difference images based on simple subtraction of tag and control images. The second
modification, called QUIPSS II, is similar to its precursor, but applies the saturation
pulse differently. In QUIPSS II, the saturation pulse is applied to the region of the
inversion band at time T I I . If TII is shorter than the time width of the inversion band
r, the application of the saturation pulse truncates the inversion band so that it wi Il be
of well-defined time width T I I . So if:
Ti, < r
and
then the tagged bolus of blood will have passed entirely through the region of interest
before the slices are imaged at TI2. This results in the signal equation for the difference
image being given by:
Again, by choosing the saturation T I 1 and imaging TI:, times correctly, the subtncted
images will be independent of the effects of transit delay and hence will reflect quanti-
tative perfusion values.
3.3.3 Considerations for Multi-Slice Acquisitions
The extension of pulsed ASL techniques fiom single to rnulti-slice acquisitions intro-
duces potential sources of error into the difference images.
Long imaging times are a cause of some concem in multi-slice perfusion imaging.
The distal-to-proximal excitation order seems most logical as the spins saturated dunng
the image acquisition of each slice would not flow into the next slice to be imaged and
affect its quality. The problem with this excitation order is that the time over which
perfùsed spins have had to disperse within the image slice varies greatly between the
most distal and proximal slices. Theoretical calculations have shown that a proximal-
to-distal excitation order helps to alleviate the problem of differential perfusion times
and does not degrade slice quality as long as the individual image times are sufficiently
short [87]. For the flow values seen in the brain, the images will not be degraded if the
image times are kept below approximately 80ms, as this implies that the progress of the
slice excitations is faster than the fastest of the flowing spins in the brain.
Careful attention must also be paid to the profiles of the slab-selective inversion
bands in a multi-slice sequence. The hyperbolic secant pulses [S 11 often used to invert
the spins in pulsed ASL show well-defined inversion margins in inhomogeneous Bo
environrnents or flowing spins, but show some rounding of the edges of the inversion
slabs. This rounding effect can alter the slice intensities depending on their proximity to
the inversion band and these modulations will differ between the tag and control images
depending on the particular tagging approach used. Expenmental approaches using
shorter (less than 20ms), lower intensity inversion pulses improve the slice profiles -
by reducing the effects of relaxation, and a saturation pulse applied to the image area
prior to the inversion pulse to nul1 the static tissue signal help co alleviate this problem
CW- Limitations exist on the spatial extent that can be irnaged during a single pulsed
ASL measurement. Using EPI-STAR as an example, the inverted tag spins that flow
through the imaging area decay with Tl tirne constants - Tib and then Tl of tissue afier
they are extracted fiom the vascular space [IO] - fiom the moment of their inversion
onwards. This limited time window for the signal sets limits on the spatial extent of the
measurernent in that a significant time penod must pass while the spins are in transit
before they can begin perfusing distal tissue slices. The proximal slices in a perfusion
measurement have a longer period of time for perfusion of extracted spins than do
distal slices. Rapid imaging sequences help to alleviate some of these problems in that
the maximum amount of time post-inversion can be allotted for the perfusion of distal
slices, but differences in the signals seen between slices due to the different amval
times of the inverted spins can affect perfusion measurement.
Chapter 4
Methods
fMRI can be performed on subjects using a wide variety of stimuli and tasks. The mag-
nitude and location of the response seen in a fMRI experiment can Vary significantly
depending upon stimulus parameters or task details.
For al1 of the experiments performed here we used a robust visual stimulus. Visual
stimuli are well-suited to fMRI methodology development due to the relatively large
signal response seen and the welI known location and fbnctional organization of the
visual areas [75].
The stimulus used here consisted of a yellow/blue radial checkerboard pattern with
30% luminance contrast (luminance variation divided by average luminance) that alter-
nated at a fiequency of 8Hz and covered 27" of the visual field of view. The pattern
was composed of 30 spokes and 4.5 rings. This was projected onto a screen at one
end of the scanner's bore, and was visible to the subject through the use of an angled
mirror added to the head imrnobilization setup. A schematic of the stimulus presenta-
tion equipment configuration is shown in Fig. 4.1. Visual stimuli were presented with a
NECMT820LCD projector operating in 640x480 mode at 60Hz, and generated using
OpenGL-based software on a Silicon Graphics O2 cornputer [32].
As illustrated in Fig. 4.2, a small triangle was projected at the centre of the stimulus
pattern. This tnangle was also presented during the control (baseline) presentation -
consisting of a unifonn grey background at the mean stimulus luminance. To ensure
that the subject's attention was focused on the centre of the presented pattern, this trian-
gle was used to prompt feedback fiom the subject between acquisitions. The subjects
were asked to report the orientation (pointing right or lefi) of this triangle by pressing
the relevant bunon on a MRI-compatible mouse.
The visual area of the human brain is highly-organized and exists at a well defined
location along the calcarine fissure in the back of the occipital cortex. Signals generated
in the retina are processed in the lateral geniculate nucleus and then passed along nerve
fibres terrninating in the primary visual area (V1 ) or srriate corfer. The organization of
the primary visual area is such that a direct correspondence (contralateral mirror image)
exists between locations in the visual field and points in VI. Other visual areas (V2 and
up) exist lateral to V1 and perform higher-order processing of visual information [75].
The selective activation of various regions of the visual area can be achieved by the
application of specific visual stimuli. Elucidation of the relationship between stimulus
location in the visual field and the location of the evoked response is titled retinotopic
mapping and has been the focus of several experiments in PET, autoradiographic studies
and fMR1 [79, 751. The stimulus used here to test the perfusion-weighted imaging
techniques was chosen because of the large and consistant response that it evokes.
NEC UT
scanner hardware
* visual field - 220
Figure 4.1 : Visual stimulus delivery schematic. The stimulus of Fig. 4.2 was delivered to the subject lying in the scanner bore. The stimulus pattern was generated by a SGI O2 computer and displayed on a screen through the use of an LCD projector. The prompt to change the stimulus during an acquisition came fiom trigger signals from the scanner hardware.
Figure 4.2: The radial checkerboard visual stimulus pattern. Experiments were per- fonned with a yellowiblue colour scheme altemating at 8Hz. Control images were acquired with a uniforni, grey screen.
4.1 Sequence Design
The sequences used were generated using PARGEN 5.0 (Parameter Generator) for the
Siemens Magnetom Vision l.5T MRI system using the NUMARIS /3 V3B3 1 A oper-
ating system (SIEMENS Medical Systems, Erlangen, GDR.). The GRASE sequences
and the echo-planar sequence were designed to have the same readout bandwidth and
sequence timings.
4.1.1 Low-Resolution Acquisitions
Previous work at our lab [15] showed that a 128 x 128 matrix GRASE sequence provid-
ed good results in the acquisition of single-slice perfusion-weighted imaging using the
FAIR technique. Images from this sequence had a rectangular field of view in which
75 phase encode lines were acquired in order'to reduce the time required per image.
The sequence was still relatively long (160ms), and we hypothesized that a reduction
in acquisition time would facilitate the extension of the technique to multi-slice.
To this end, a lower-resolution 64x64 matnx GRASE sequence was implemented
that featured the acquisition of 45 phase-encode lines in a rectangular field of view in-
stead of the original 75. The reduction of the matrix size also meant that the acquisition
time for each readout line could be reduced as well. The end result was the reduction
of the total imaging tirne for each acquisition to Zms.
A M e r time savings was gained through the use of a half-Fourier sequence. In
this acquisition, entire brain images were reconstructed fiom raw data sets that featured
incomplete (two-thirds) coverage of k-space . A homodyne detection-based reconstruc-
tion technique was implemented to reduce the phase errors associated with half-Fourier
acquisitions [70]. This implementation reduced the tirne per acquisition to 5Oms with
little reduction in image quaiity.
The GRASE sequence developed here acquired five readout lines (five gradient-
recalled echoes) for each 180" refocusing pulse. The k-space data was separated into
three regions that were filled according to the k-banding technique proposed by Fein-
berg et.al. as discussed in the previous chapter [23]. The initial five lines of data were
recorded without phase encoding and were used to correct phase error in subsequently
acquired lines. This was done by a point-by-point phase correction algorithm wherein
the phase of these original reference lines was subtracted from the corresponding data
lines.
The half-Fourier (HF) GRASE sequence reconstntcted images from incomplete k-
space data sets (two thirds of the data required by full-Fourier GRASE). The k-space
lines are organized into two bands [23] which are reversed in order with respect to one
another. Selective weighting - doubling of the intensities of the high-fiequency k-space
data, and a homodyne detection technique - to reduce the effects of low-frequency
modulations of the incomplete data set - were used to construct the full images from
the reduced data sets [69].
4.2 Spin-Tagging Implementation
Acquisitions featuring odd numbers of image slices were used in the generation of al1
perfusion-weighted images. The inversion pulses used were positioned at the central
slice location, resulting in a symmetrical positioning of the image slices within the
inversion slabs. This was done as a convenience as the slab centre is positioned along
an image slice.
4.2.1 FAIR
The FAIR technique was implemented using a hyperbolic secant inversion pulse to
invert blood water spins prior to image acquisition [8 11. An inversion slab wide enough
to encompass the image slices plus an extra margin of 25% on each side, to minimize
profile effects, was used in the generation of the selective inversion images. The sarne
inversion pulse applied in the absence of a selection gradient was employed to yield the
non-selective inversion images. The inversion pulse profile was measured by acquiting
a magnitude image in a gel phantom in a plane perpendicular to the orientation of the
inversion pulse at an inversion time of 30ms. A plot of pixel values from a vector
passing through the inversion region is shown in Fig. 4.3.
Figure 4.3: Inversion profile for hyperbolic secant inversion pulse of 10.24ms duration.
4.2.2 STAR
The STAR implementation involved the use of regional inversion bands to generate the
tag and control images needed. These bands were positioned to be parallel to the image
slice(s) with a lem separation between the closer edge of the band to the most proximal
or distal slice. The positioning of the bands was done at the scanner's interface terminal
61
in a marner similar to that of the slice positioning. Hyperbolic secant pulses were also
used to invert the spins in this case, however PARGEN lirnited the pulse duration to
approximately 3ms. A puise profile was generated for the shortened pulse and the
results are plotted in Fig. 4.4.
?au, 1
mD ' 1 P 1 9 ~ 1 1 P n Y 4 4 n
Am I*AI
Figure 4.4: Inversion profile for hyperbolic secant inversion pulse of 2.56ms duration.
The inversion profile of the shortened pulse is not noticably different than that of
the longer version.
4.2.3 Post-Processing of Image Data
Images generated on the MRI scanner were transferred to the MRI Research Lab com-
puters via ethemet. Activation maps, time-intensity curves (TIC'S) and relative per-
fusion change values were calculated fiom the data sets using routines witten in the
PERL scripting language and MATLAB ( V5.2 The Math Works Inc., Natick MA).
PERL scripts were written to convert the native data format of the scanner to MiNC
- a file format created at the Montreal Neurological Institute to facilitate easy post-
processing of images [66]. Once in MINC format, the individual image files for each
functional run were concatenated to foxm selective and non-selective inversion image
data files - or tag and control sets in the case of STAR imaging sequences. From these a
difference image set was created. Quantitative processing was then perfomed on these
files.
4.3 The Perfusion Measurement
The data sets generated in each FAIR-based functional snidy consisted of a selective
image set acquisition followed by the imaging of the same slice(s) following the non-
selective inversion pulse. The STAR sequences featured proximal followed by distal
inversions. The repetition time (TR) of the slice measurements was set to three seconds.
This choice of TR allows for fresh (steady-state) spins to flow into the inversion slab
area following the acquisition of images. TR's as low as 2.0 seconds can be used in
perfusion weighted studies as spins flowing into the inversion slab areas can be assumed
to have no prior spin history [87]. Inversion times are typically on the order of 1.0
second (time between inversion and the excitation of the first slice to be imaged). This
time period allows inverted spins to perfUse tissue in the ROI and exit the macrovascu1ar
space, while maintaining acceptable signal strength (the effects of varying the inversion
time are studied in section 5.4). The functional experirnents were conducted for four or
six minute durations. The six minute duration experiments were used initially to test the
single-slice acquisitions. Based on the results of these studies, it was decided that four-
minute scan durations were sufficient, as the time savings allowed additional studies
to be perfonned while a single subject was in the scanner (each functional expenment
takes significantly longer than its nominal duration because of sequence load delays
between runs due to the processing and storage of images fiom the previous run).
The visuai stimulus presentation was synchronized with acquisition of the function-
al data with the baseline (uniform grey background) image being presented for the fvst
minute. The activation pattern was then presented to the subject for the second minute.
This pattem was continued for the duration of the functional nin. An output trigger
pulse from the scanner hardware was used to synchronize the stimulus to the acquisi-
tion, and it was also used to change the direction of the triangular indicator at the centre
of the visual field in both activation and control images in a random fashion. The sub-
ject was asked to report the orientation of this indicator by pressing the correct button
on an MM-compatible mouse. A record of the subject's response was displayed in the
interface room to ensure that he/she was awake and paying anention to the presented
stimulus during the acquisition.
4.3.1 Subject Immobilization
Dwing the acquisition of the functional data, the subject was required to remain as still
as possible. The extremely rapid nature of the image acquisition ensures the 'freezing'
of the subject's motion during the measurement of a slice (Le. intrascan motion), but
movement between slice acquisitions (interscan motion) can greatly affect the results
of a functional expenment.
Subject immobilization during the study was facilitated by the use of an MM-
compatible headholder consisting of a foam forrn into which the subject's head was
fitted, a custom dental imprinted bite-bar that was placed into the subject's mouth, a
saddle-shaped nose-bridge piece that was pressed lightly onto the bridge of the nose,
and two ear cups that could be cinched over the the subject's ears - with the added
benefit to the subject of reducing the scanner noise dunng the acquisition.
4.3.2 Functional Data Processing
The various image sets produced by the PERL routines could be analyzed in a number
of different ways in the generation of the study results.
MATLAB routines were developed for the calculation of t-maps [3]. These were
based on the pixel values of the difference image set referenced to a vector describing
the stimulus presented to the subject during the acquisition- Spearman Rank Correlation
was used as the statistical test in t-map generation [20]. The t-map of each slice was
then supenmposed on the anatomical data set to relate the function and anatomy.
Time-lntensity Curves (TIC'S) of averaged pixel values in a region of interest (ROI)
were plotted using a separate MATLAB routine. The ROI for these studies could
be chosen based on a t-map (wherein only pixels corresponding to significantly ac-
tive brain areas would be included by eliminating those pixels below a critical t-value
threshold), or by comparison with a previously-generated retinotopic map. Retinotopic
mapping was performed on some subjects in prior experiments performed at our lab
[45]. The mapping procedure consisted of using graded stimuli in a series of functional
experiments to generate activation maps (delineating V 1 ) that excluded contributions
from large vessels. The procedure used was developed earlier by Sereno et. al. at
UCSD [79]. The retinotopic maps were used as masks in determining the ROI'S used
in the functional experiments.
Changes behveen activated and control states are demonstrated graphically by plot-
ting the averaged pixel value in the ROI (either within a slice or a volume given by con-
sidering several slices at a time) throughout the time course of the selective-inversion
images. The same averaging was applied to the non-selective images, and the average
pixel value across the time course is shown on the TIC. The average pixel values in the
ROI for activation and baseline stimulus conditions are shown on the TIC as well, and
the relative perfùsion value was calculated according to:
-4ctivation Sel. Image - BaseEine Sel. Image Relatiue Perfusion =
Baseline Sel. Image - ,Vonsel. Image (4.1)
for the FAIR images, and:
-4ctiuation Tag Image - B a d i n e T u g Image Relative Per f vs ion = -
,4ctiuation Co.ntro1 Image - Bnseline C'ontrol Image
(4.2)
for the STAR images.
The error was estimated by using the standard error of the means calculated for
the ROI pixel values in the selective image activation and control sets and in the non-
selective image data set. Such an error estimation is given by:
a Standard Error of the Mean = -
fi (4.3
where O is the standard deviation and n is the number of samples used in the calculation.
Experimental Design
Perfusion-weighted sequences were wrïtten and tested for STAR and FAIR spin-tagging
methods using both single and multi-slice acquisitions. Activation maps based on t-
values of Spearman Rank Correlation tests performed on the data sets were generated
along with time-intensity curves showing the relative perfusion values between activat-
ed and baseline stimulus presentation. The results of the single-slice studies provide a
basis of cornpanson that enabled the selection of an imaging technique as being most
suitable for multi-slice applications. Once this was accomplished, irnaging parameters
were studied to find their effects on multi-slice sequences and predict a limit for the
possible spatial extent to be covered.
4.4.1 Single-Slice Experiments
Single-slice FAIR acquisitions used a 7.5 mm inversion band centred on the 5mm image
slice while STAR sequences used a 90 mm inversion band offset fiom the image slice
by a distance of 10 mm.
Typical relative pefision changes between activation and baseline States were found
to be on the order of 20 - 60%, as expected, and were found to Vary between subjects.
The results shown are fiom multiple subjects.
4.4.2 Multi-Slice Experiments
The GRASE FAIR sequence was chosen as the method that was most suitable for ex-
tension to a multi-slice acquisition. This decision was made because the GRASE-FAIR
sequence provided good functional data in preliminary studies, and had the advantage
of eliminating much of the image distortion seen in EPI.
The quality of perfusion data acquired as part of a multi-slice acquisition can suffer
from errors arising from slight changes in the acquisition parameters. Images in a multi-
slice sequence will have flow weighting based on a much wider inversion pulse, may
show effects fiom slice profile effects due to the proximity of adjacent slices, and may
also suffer due to slight differences in inversion times that exist between slices (as each
slice acquisition takes 80 ms).
4.4-3 The Effect of Inversion Slab Width
The inversion band width's effect on perfusion data was studied using a series of
single-slice GRASE-FAIR measurements acquired at the centre of several inversion
slab widths. Single-slice acquisitions were chosen for this set of experiments as they
will not have confounding effects fiom nearby slice acquisitions. Inversion band widths
fiom 55.5m.m to 106.5mm were tested. Testing of this range of slab widths provides
an estimate of this parameter's effect on the results of functional experiments with slab
widths wide enough to incorporate up to seven 5mm slices with lmm separation - the
centre of an inversion slab of 106mm will correspond with the most distal slice of a
seven-slice acquisition.
4-4-4 Slice Profile Effects on Perfusion Data
To evaluate the effect that imaging nearby slices has on the perfusion data within a
slice, a seven slice sequence with contiguous 5 mm slices and an inversion band of
52.5 mm was used. Functional studies were performed using this sequence centred
on the visual area of the brain. Studies difEered in the distance separating the imaged
slices. Cornparisons of the perfusion values seen in the centre slice - slice 4 - were used
to assess the separation at which the perfusion values were the sarne as those seen in
single-slice acquisitions.
4-4.5 The effect of different inversion times
The inverted magnetization recovery during the acquisition of a single slice is negli-
gable, but becomes appreciable when multiple slices are acquired. The inversion time
difference between the imaging of separate slices is a possible source of error in a perfi-
sion measurement. To find the range of possible inversion tirnes over which acceptable
perfusion data can be obtained, a series of functional data sets were acquired in which
the inversion time of the image set was varied between runs. The inversion time was
defined relative to the centre slice. The ROI used to generate the TIC included the entire
masked volume defined by a retinotopic map. The different slices contributing to this
result al1 diEer slightly in inversion time across the ROI.
4.4.6 A Seven-Slice Perfusion-Based Functional Experiment
The results of the tests performed in the previous sections suggested that a multi-slice
sequence would be successful in acquiring up to seven slices at a tirne. A seven-slice
sequence was assembled, with 5 mm slice width, I mm slice separation and an inversion
band width of 55.5 mm centred on slice four in the imaged area. The spatial coverage
of the imaging sequence was now greater than the size of the primary visual cortex.
To compare the response measured by the sequence across all of the slices, functional
studies were performed with the centre of the primary visual cortex - the Calcarine
fissure - lying along slices at different locations in the excitation order. A study with
one of the more proximal slices aligned with the calcarine will have flow weighting due
to spins that have flowed a short distance from the edge of the inversion band and will
have seen a shorter inversion time than slices in more distal locations. By comparing the
activations seen fkom the visual area as it is centred on the various slice locations while
keeping al1 of the other acquisition parameters constant, a measure of the uniformity
of perfusion sensitivity across slices was obtained. The TIC'S calculated here were
averaged across the entire ROI volume covered by the slices.
4.5 QUIPPS
A single-slice QUIPPS GRASE FAIR sequence was developed. An experiment com-
paring the response seen with that fiom GRASE FAIR was performed to test the ef-
fect of the QUIPSS modification on the fiinctional results. The QUIPSS modification
(according to the Standard ASL Signal Model) eliminates the effects of the unknown
parameters r and At fiom the perfusion-weighted images.
This cornparison will yield an estimate of the accuracy of the non-QUIPPS perfu-
sion results obtained in Our studies.
4.6 Reproducibility of Results
To test the reliability of the methods used. An experiment was designed where finc-
tional data was acquired fiom a singie subject on two occasions a week apart. The
results fiom these tests were compared to ensure consistency.
Chapter 5
Results
Results of the various iÛnctional data experirnents outlined in the previous chapter are
presented here. The t-map and TIC of runs are presented. as well as the results of an
analysis of variance (ANOVA) test performed on the data sets. The ANOVA test was
used to test the probability that the difference in means of the functional results are due
to chance alone. In cases where a significant difference was found, Tukey's Honestly
Significant Difference (HSD) Test was used for post hoc-pairwise comparison of the
individual data sets [95].
5.1 Single-Slice Results
5.11 FAIR
The FAIR single-slice results are shown in Fig.5.1. The experiments were of six minute
duration, and used a retinotopic map to mask the ROI. The studies were performed
sequentially on a single subject.
The measured relative pefision change was measured as 30 5 3% for EPI-FAIR,
23 dz 3% for GRASE FAIR and 24 k 3% for Partial-Fourier GRASE FAIR. The ANOVA
test was performed on the data sets to detennine if it was reasonable to assume that the
differences in the three relative pefision values seen was actually significant, and not
due to chance alone. The results of this statistical test indicate that the means of the data
sets do not differ significantly enough to mie out chance as the cause of the differences
seen (F=2.33, dW9, p>O.OS).
The localization of function seen in the three aqusition modalities is displayed in
the upper row of Fig. 5.1. The three modalities show activation in the same area of
the brain. The EPI-FAIR sequence results showed the largest area of activation, the
GRASE FAIR results showed activation in a smaller area located within the activation
area of the EPI-FAIR results, and the Partial-Fourier GRASE FAIR results were smaller
still, but were located at the same position within the brain as the GRASE FAIR results.
Figure 5.1 : EPI-FAIR (lefi), GRASE FAIR (centre) and Partial-Fourier GRASE FAIR (right) t-maps and TIC'S.
5.1.2 STAR
The results of the single-slice STAR techniques are shown in Fig. 5.2. Each experiment
was of a six minute duration, and used retinotopic maps to mask the ROI.
The STAR results were significantly noisier than their single-slice FAIR counter-
parts - the standard deviations of the data sets were approxirnately 25% of the calcu-
lated value, as compared to 10% for the FAIR results. The ANOVA analysis found no
significant difference in the relative perfusion change values of the data sets outside of
that expected fiom chance alone (F=0.45, deS9, p>0.05).
The relative perfusion change values calculated fiom the STAR data sets were 24 * 4% for EPI-STAR, 24 -t5% for GFWSE STAR, and 28 3 ~ 6 % for Partial-Fourier GRASE
STAR.
Activation t-maps are displayed in the upper portion of Fig. 5.2. The t-maps show
activation in the same areas of the primary visual cortex with similarly-shaped areas of
activation.
5.2 The Effect of Inversion Band Width
The results of the study investigating the effects of inversion band width are s h o w in
Fig. 5.3. Al1 studies were performed sequentially on a single subject, and were four
minutes in duration.
The various inversion bands produced results of regular shape and consistant rela-
tive perfusion change values (404%).
The ANOVA test performed on the data sets showed there to be no significant dif-
ference in the calculated relative perfusion change values (F= 1 S9, df=79, p>0.05).
Figure 5.2: EPI-STAR (left), GRASE STAR (centre) and Partial-Fourier GRASE STAR (right) t-maps and TIC'S.
5.3 The Effect of Süce Profile
The results of the expenments exploring the effect of varying the slice separation are
shown in Fig. 5.4. These four-minute studies were perfomed sequentially on a single
subject. A retinotopic rnap was used to mask the ROI.
The centre slice of seven-slice acquisitions at various slice separations was used
to determine the effects of imperfect slice profiles on the measured relative perfusion
change. Increased relative perfûsion values (relative to that measured in a single-slice
experiment) were seen at slice separations below Imm.
q ml. - - . - - . - - - - . - - - . g 1 - A- ROI RXN Vau.
lm- kavabon Avg Sm I m g . . , - - - - - - - - - - - - - Casnn ~ u g 5r lrnig.
- Ncn-Sm b n ~ g e R O I A v g --
m. ; .9 .; A P 1 4 1 - - 6 - m u
4 a. 2 2 ,, --- - 4 / - ~ v g ROI mi Vatus
.a- Ccovamn Aug S.( lm- . - - I Nan-SH bnage ROI Avg
-0 ; O O n . > r-". r "
Figure 5.3: The TIC and relative perfusion change value measured in a single slice centred in inversion bands of various widths: 55.5mm (top left), ïl.5mm (top right), 90mm (bottom lefi) and 106.5mm (bottom right).
5.4 The Effect of Inversion Time
The results of measurements of the effects of differing inversion times are shown in Fig.
5.5. The four-minute experiments used here were perfomed sequentialty on a single
subject.
Seven slice acquisitions with one millimeter separation between slices were used
to test for the effects of inversion tirne on the measured relative perhision change. A
retinotopic map was used to mask the ROI in al1 slices. The response fiom al1 slices are
included in the results. inversion times stated are to the first image slice excitation.
Figure 5.4: Calculated relative perfusion changes at various slice separations. The measured relative perfusion change measured in a single slice aqusition is included as the dotted line-
The earliest inversion time study (2OOms) shows the pixel averages for the setective
image to be lower than those of the non-selective inversion. The caiculated relative
perfusion change value does not differ in sign from those of longer inversion times, but
is significantly different in the appearance of the TIC. The calculated relative perfusion
change was 24 zt: 5% for this inversion time.
The remaining inversion time experiments were of a more regular form. The ANO-
VA test perfonned on the data sets showed that a significant difference existed be-
tween the relative perfusion change values of the data sets in the study (F=7.58 d e l 19
p<0.000 1). Tukey's HSD test showed that results fiom the study with an inversion time
of 1400ms were significantly different from every other data set.
The measured relative perfusion change has a maximum value at an inversion time
of 1000ms. The response seen fkom the 1200ms inversion time study is slightly lower,
but still within the error ranges of the values.
Figure 5.5: Graph of relative perfusion change values measured with a seven slice GRASE FAIR sequence as the inversion time was varied between 600 and 1100ms.
5.5 Cornparison of Response Seen Across the ROI
The results of the measurements of the primary visual area response with different
positionings of the ROI are shown in Fig. 5.6. The three four-minute studies were
performed sequentially on a single subject. A retinotopic map was used to mask the
ROI. Only the results from the slice centred on the calcarine fissure are included. Al1
acquisitions were performed using an inversion time of 1000ms.
The relative perfusion values measured in the tùnctional runs were 30 f 6% for
the acquisition centred on the calcarine fissure; 30 & 3% for the acquisition with the
calcarine aligned along the second-most distal slice; and 33 & 4% for the acquisition
with the calcarine aligned along the second-most proximal slice. The centre slice was
acquired with an effective inversion time of 12lOms; the proximally positioned acqui-
sition had an inversion time of 1 OiOms, and the distally-positioned acquisition had an
effective inversion time of 1420ms. The TIC'S are of a regular form, with the centred
acquisition having the highest level of noise in the data set. The ANOVA test showed no
significant difference in the relative perfusion change values calculated from the studies
(F=0.29 df=59 p>0.05).
Figure 5.6: Relative perfusion changes seen in one slice (centred on the calcarine fis- sure) of seven-slice GRASE-FAIR acquisitions with calcarine fissure aligned along the second-most proximal slice (lefi side data point), the centre slice (centre data point), and the second-most distal slice (right side data point).
5.6 QUIPPS
The effect of the QUIPSS modification was studied using the GRASE FAIR sequence,
the results are dispiayed in Fig. 5.7. The four-minute runs were performed sequentially
on a single subject. A retinotopic map was used to mask the ROI.
The two TIC'S are similar in appearance with a higher level of noise seen in the
GRASE FAlR acquisition. The ANOVA test perforrned showed no significant dif-
ference in the calculated relative perfusion values of the two studies (F=0.84 d e 3 9
p>0.05).
5.6.1 Reproducibility of Results
To test the reproducibility of pefision measurements, a single subject was tested at two
separate times a week apart. Each expenment consisted of five functional runs during
each of which a single slice aligned with the calcarine fissure was imaged in four minute
runs with the usuaI stimulus presentation. The results of these measurements are shown
Figure 5.7: Cornparison of TIC'S for GRASE FAIR single-slice acquisitions. The con- ventional results are displayed at lefi, and those fiom a sequence with the QUIPPS modification are shown at right.
Table 5.1 : Table of experiinental results testing the reproducibility of perfusion results. A single subject was tested with an identical set of experiments a week apart.
below.
A two-tailed, paired t-test was applied to the results to predict the likelihood that the
difference seen between the two means was due to something other than chance [74].
The result of the test indicated that the results between the two tests are not significantly
different t ( 4 ) = - 1.16, p > 0.05.
Chapter 6
Discussion
6.1 Experimental Results
The single-slice results for the FAIR and STAR perfusion-weighting approaches show
similar results for the calculated relative perfusion change value but differ in the degree
of noise seen in the image. The results fiom both approaches are largely independent
of the imaging method used, but the FAIR techniques show cleaner TIC data (better
Sm). This improvement in signal may result fiom the decreased distance that the
leading edge of the tag bolus has to travel to get to the imaging area (a few mm in
FAIR as opposed to slightly more than a cm in STAR), and may also be in part due to
the limited size of the inversion band used in the STAR technique (a 9 cm inversion
slab was used). The selection of GRASE-FAIR as the method best-suited to extension
to multiple slices was based on the quality of the results fiom the single-slice sfudy.
The results were essentially the same as those fkom the EPI-FAIR sequence, but the
refocusing of the signal in the GRASE acquisition results in less spatial distortion than
that seen in EPI sequences [7 11.
The inversion slab width was shown to have little effect on the quality of the func-
tional data and the measured relative perfiision changes over the range of widths stud-
ied. Only the inversion widths that were relevant to the numbers of slices studied in
other investigations were explored. The limit of this slab width was not realized dur-
ing this investigation, but it must necessarily exist as extremely wide inversion bands
would be indistinguishable fiom the non-selective inversion pulse, and not be flow-
weighted at all. The results fiom the widest inversion pulse studied suggest that an
inversion pulse wide enough to incorporate seven slices would not significantly alter
the perfusion-weighting of the imaged slices, i.e. the centre slice from an inversion
band wide enough to include thirteen slices would roughly correspond to the most dis-
ta1 slice of 3 seven slice acquisition.
lmaging slice profile has a large effect on the calculated value of relative pef i s ion
change seen in the expenments. The changes seen in the contiguous slice and half-
millimeter separation experiments show relative changes that are approximately 25%
higher than that of a single-slice experiment perfomed on the same subject. Separa-
tions of one millimeter or larger show perfusion values are within experimental error
of the single-slice results. The inflated perfusion change values result from the dif-
ferent effects that imperfect slice profiles have on inflowing spins following selective
and non-selective inversion pulse. Signal from inflowing spins in selec tive-inversion
images have steady-state magnetization, but non-selective inversion spins are only par-
tially recovered at the time of image acquisition. Imperfect slice profiles result in slight
rotations of spins in nearby slices, and this will attenuate the signal from steady-state
inflowing spins to a larger degree than their non-selective inversion counterparts. The
relative perfusion change calculation relies on the change seen in activation and base-
line pixel averages in the selective inversion images. These will be attenuated in a
similar fashion by the imperfect slice profiles of the adjoining slices, not greatly af-
fecting the difference. The pixel average in the non-selective images, however, will
show less attenuation fiom slice profile effects due to the saturation of the inflowing
spins. The overestimation in relative perfusion therefore likely results from the reduced
difference between the selective-inversion baseline pixel average and the non-selective
pixel average (a reduction in the value of the denorninator).
Inversion time was found to have little efTect on the calculated relative perfusion
change and activation t-maps
Relative perfusion change measurements were found to be robust to changes in in-
version time between 600ms and 1200ms. Studies using inversion times outside of this
range showed noisier TIC'S, poorer-quality t-maps and changes in the value of the rela-
tive perfbsion change calculation. These results suggest that inversion time differences
between slices acquired as part of a seven-slice sequence would have minimal effect on
the quality of fiinctional image data.
Within the inversion band of a seven-slice GRASE-FAIR acquisition a similar re-
sponse was seen after positioning the same responding structure, the primary visual
cortex, at proximal, central and distal locations. A noisier functional data set was ob-
tained fiom the central positioning experiment, but the calculated relative perfusion
change value seen was very similar to the two more-proximal studies. The centred
slice results having an increased level of noise is most likely due to imperfect position-
ing of slices along the calcarine fissure between experiments. Slight diflerences in the
alignment of the slice of interest along the calcarine may affect the signal seen in an
experiment .
The QUIPPS modification to the single-slice GRASE FAIR sequence showed no
significant difference in the resultant data sets. Similar TIC'S and calculated relative
perfusion changes were seen. The increased level of noise seen in the non-QUIPSS
acquistion may be due to imperfect subtraction of the tissue signal between the selective
and non-selective inversion images. The signal from static tissue in the QUIPSS images
will be more efficiently nulled by the saturation pulsed applied to the image slices,
making this sequence less sensitive to this source of noise.
The QUIPSS modification is important to absolute quantitative measurements of
perfusion level. as this method eliminates the confounding factor of At fiom the signal
equation of the standard model. in FAIR measurements, the spatial gap between the
Ieading edge of the inversion slab and the imaged slice is typically a few mm, and Our
use of the body coi1 in the spin inversions results in a large bolus width (large T ) as al1
of the spins in the sensitive volume of the scanner would be inverted in the generation
of the non-selective image. The implications of this in reference to Buxton's Standard
model (outlined in Section 3.2 of this thesis) are that the second condition of equation
3.5 would apply for the signal function. The small spatial gap between the leading edge
of the inversion bolus and the first imaged slice would result in a small value of At in
equation 3.5. If we assume this At contribution to be negligable, and the value of T to
be large enough so that the second condition of equation 3.5 is valid for the inversion
time used, the relative changes in measured signal during an activation experiment will
be directly proportional to the relative flow (and hence perfusion) changes in the brain.
Questions remain about the validity of the above suggestions. The actual value of
At and T for a typical FAIR experiment have not been studied at Our lab in reference to
the standard model. The result of the single-slice QUIPPS cornparison performed here,
however, suggests that our assumptions are valid.
6.2 The Seven-Slice GRASE-FAIR Sequence
The results of the various tests performed on sequence parameters suggest that a seven-
slice GRASE-FAIR sequence with lmm slice separation is effective in extending the
spatial coverage of pefision-weighted imaging sequences fiom single slice acquisi-
tions. An increase in spatial coverage fiom 5mm to 41mm was obtained (although the
increased coverage necessarily contains 6 lmm slice gaps that do not contribute to the
functional data).
6.3 The Interleaved Sequence
Currently, BOLD contrast-based fMRI is predominmt over perfusion-weighted imag-
ing techniques. This is due to limitations in SNR and the extent of the spatial coverage
possible in the perfusion methods. Questions exist, however. pertaining to the relation-
ship between BOLD and perfusion-based contrast sources, and the changes in cerebral
metabolic rate of oxygen consumption (CMR02) (assumed to be coupled to changes in
cerebral blood flow - CBF). Several models describing this relationship have been pro-
posed, and experiments performed at Our lab have tested this relationship and compared
it to the relationship proposed by the Deoxyhaemoglobin Dilution Mode1 [45].
The interleaved sequence that was the means of acquiring this functional data fea-
tured an echo-planar readout and acquired a BOLD-weighted image after the selective
inversion image acquisition of a FAIR sequence and another afier the non-selective im-
age acquisition. The use of an interleaved sequence (developed by the author) provided
the direct cornparison of BOLD and perfusion-weighted measurements dunng a sin-
gle functional run (the two data sets are temporally interleaved and have exact spatial
correspondence). The expenments done at Our lab were performed with graded visual
stimuli, and with different levels of hypercapnia - which allowed the manipulation of
CBF independent of CMR03.
The results of these experiments supported the Deoxyhaemoglobin Dilution Model,
suggested that ABOLD signal (relative BOLD signal change) is a linear measure of tis-
sue workload in visuai activation studies, and related ACBF to 3CMRO2 in a constant
2 : 1 ratio. Furthemore, a PET-based determination of ACMRO? with a radial
checkerboard stimulus (which had been performed in a separate experiment) produced
results which were the same as those found at in out MN experiment (2.5% change).
This work formed the bulk of the PhD thesis of Rick Hoge, and was the source of three
papers [45] - [47] and numerous abstracts [34] - [43] and [48] - [SOI.
The extension of the interleaved perfusion-BOLD teclmique to multiple slices us-
ing an echo-planar sequence would allow the activation physiology experiments to be
extended to other brain structures. As well, increased spatial coverage in further visual
studies would allow the entire primary visual cortex to be studied in a single experi-
ment.
Chapter 7
Conclusions
The pursuit of a perfusion imaging technique that, at least partially, alleviates the prob-
lem of limited spatial extent has been the focus of this thesis. Candidate acquisition
sequences were designed and tested with FAIR and STAR spin-labelling modalities in
single-slice acquisitions. The GRASE FAIR sequence was chosen for extension to a
multiple-slice acquisition, and was found that up to seven slices could be effectively
imaged at a time.
The imaging of perfusion in multiple slices introduces potential problems. Vari-
ous possible sources of error in multi-slice acquisitions were addressed and tested to
determine their effect on perfusion results. The results of the study suggested that the
perfusion measures seen were robust to most of the parameters tested over the ranges
of interest.
An increase in the spatial coverage of perfusion-weighted sequences will allow the
acquisition of data fiom activation foci less easily identified than the primary visu-
al area, whose location along the calcarine fissure makes it particularly convenient to
study. Also, tùnctional structures in areas of high magnetic susceptibility change are not
good candidates for BOLD imaging methods, as the TG-weighting of the images leads
to an unacceptable level of artifacts. Perfusion-weighted GRASE imaging is especially
well-suited to these tasks as the constant refocusing of the signal dunng acquisition
greatly reduces artifacts fiom this source.
Issues remain in regard to the quantitative nature of the blood flow signal measured
in MR-based perfusion imaging. The goal of much of the current research is to provide
absolute quantitative measurements of cerebral perfusion. The pursuit of this has gen-
erated much of the theory that was used in this thesis (the Standard Model. QUIPSS,
and QUIPSS II).
Questions remain about the validity of the perfusion change results reported here.
The assumption that relative blood flow changes are directly proportional to the perfision-
weighted signal changes seen in Our images is as yet unverified. Encouraging results
such as the constancy of calculated values between single and multi-slice acquisition-
s, the agreement that was seen between Our methods and PET for the same stimulus
between different subjects, and the constancy of results with and without the QUIPSS
modification have been reported here. A more exhaustive study comparing the multi-
slice sequence reported here with a QUIPSS II version would be useful in answering
the questions that remain.
Additionally, quantitative comparisons with PET studies performed on the same
subjects would be the most reliable assay of the effectiveness of ASL perfusion mea-
surements. PET results remain the most reliable assays of brain perfusion, and the
reproduction of these using ASL methods would be of great importance to fMRI.
Bibliography
[ 1 ] A. Abragam. Principles of nuclear rnugnetisrn. Oxford Science Publications, 1961.
[2] C. J. Aine. A conceptual overview and critique of functional neuroimaging tech- niques in humans: 1. MRVfMRI and PET. Critical Reviews in Netrrobiology, 9(2-3):229-309, 1995.
(31 P.A. Bandettini and A. Jesmanowicz. Processing strategies for time-course data sets in functional MRI of the human brain. Magnetic Resonance in Medicine, 30(2): 16 1-73, Aug. 1993.
[4] J. C. Baron, R. S. Frackowiak, K. Herholz, T. Jones, A. A. Lammertsma, B. Ma- zoyer, and K. Wienhard. Use of pet methods for measurernent of cerebral energy metabolism and hemodynamics in cerebrovascular disease. Jownal of Cere61-al Blood Flow and Metabolism, 9(6): 72-2, Dec 1989.
[SI J. W. Belliveau, D. N. Kennedy Jr., R. C. McKinstry, B. R. Buchbinder, R. M. Weisskoff, M. S. Cohen, J. M. Vevea, T. J. Brady, and B. R. Rosen. Functional mapping of the human visual cortex by magnetic resonance imaging. Science, 254(5032):7 16-9, Nov 1 199 1.
[6] J. W. Belliveau, B. R. Rosen, H. L. Kantor, R. R. Rzedzian, D. N. Kennedy, R. C. McKinstry, J. M. Vevea, M. S. Cohen, 1. L. Pykett, and T. J. Brady. Func- tional cerebral imaging by susceptibility-contrast NMR. Magnetic Resonance in Medicine, 14(3):538-46, Jun 1990.
[7] F. Bloch. Nuclear induction. Physical Reviav, 70:46W73, 1946.
[SI M. J. Bronskill and P. Sprawls, editors. n e physics of MRi 1992 surnmer school proceedings. American Association of Pysicists in Medicine, 1992.
[9] R. B. Buxton and L. R. Frank. A mode1 for the coupling between cerebral blood flow and oxygen metabolism during neural stimulation. Journal of Cerebral Blood Flow and Metabolism, 17(1):6472, Jan 1997-
[ i O] R.B. Buxton, L.R. Frank, E.C. Wong, B. Siewert, S. Wanch, and R.R. Edelman. A general kinetic model for quantitative perfusion imaging with artenal spin la- belling. Magnetic Resonance in Medicine, 40:383-396, 1998.
[ I l ] F. Calamante, S. R. Williams, N. van Bruggen, K. K. Kwong, and R. Turner. A model for quantification of perfusion in pulsed labelling techniques Cpublished erratum appears in NMR biomed 1996 sep;9(6):277]. NMR in Biomedicine. 9(2):79-83, Apr 1996.
[12] H.Y. Cam and E.M. Purcell. Effects of diffusion on fkee precession in nuclear magnetic resonance experirnents. Ph-vsical Review, 94(3):63M3 8, 1 954.
[ 131 Z.H. Cho, J. P. Jones, and M. Singh. Foundations of Medical Imuging. John Wiley and Sons, hc., 1993.
[14] E. Clarke and K Dewhurst. An illusmted HrSrory of Bvain Funcrion. Norman Publishing, second edition, 1996.
[ 151 G.R. Crelier, R.D. Hoge, P. Munger, and G.B. Pike. Perfusion-based functional magnetic resonance imaging wi th single-shot RARE and GRASE acquisitions. Magnetic Resonance in Medicine, 4 1 ( 1 ): 1 32-1 36, January 1 999.
[16] J. Cuppen and Andre van Est. Reducing MRimaging time by one-sided recon- struction. In Topical Conference on Fast MM Techniques, pages 1 5-1 8, May 1987.
[17] J. A. Detre, D. C. Alsop, L. R. Vives, L. Maccotta, J. W. Teener, and E. C. Raps. Noninvasive MW evaluation of cerebral blood flow in cerebrovascular disease. Neurologv, 50(3):633-4 1, Mar 1998.
[18] J. A. Detre, J. S. Leigh, D. S. Williams, and A. P. Koretsky. Pefision imaging. Magnetic Resonance in Medicine, 23 ( 1 ) : 3 7 4 5 , Jan 1 992.
[19] J. A. Detre, W. Zhang, D. A. Roberts, A. C. Silva, D. S. Williams, D. J. Grandis, A. P. Koretsky, and J. S. Leigh. Tissue specific perfusion imaging using arteriat spin labeling. NMR in Biomedicine, 7( 1 -2):75-82, Mar 1994.
[20] J.L Devore. Probabiiiîy and Statistics for Engineering and the Sciences. Duxbury Press, fourth edition, 1995.
[2 l ] R. R. Edelman, B. Siewert, D. G. Darby, V. Thangaraj, A. C. Nobre, M. M. Mesu- larn, and S. Warach. Qualitative mapping of cerebral blood flow and functional localization with echo-planar mr imaging and signal targeting with alternating ra- dio fiequency. Radiology, 192(2):5 1 3-20, Aug 1994.
[22] R.M. Eisberg. Fundamentals of modem physics. John Wiley and Sons, Inc., 196 1.
[23] D. A. Feinberg, B. Kiefer, and G. Johnson. Grase improves spatial resolution in single shot imaging. Magnetic Resonance in Medicine, 33(4):52%3 3, Apr 1995.
[24] L. R. Frank, E. C. Wong, and R. B. Buxton. Slice profile effects in adiabatic inversion: application to multislice perfusion imaging. Magnetic Resonaiice in Medicine, 38(4):55844, Oct 1997.
[25] R. Freeman. A handbook of nuclear magnetic resonance. Longman Scientific and Technical, 1988.
[26] K.J Friston, A.P. Holmes, K.J. Worsley, J.P. Poline, C.D. Frith, and R.S.J. Frack- owiak. Statistical parametric maps in functional imaging: a gened iinear ap- proach. Human Brain Mapping, 2: 189-2 10, 1995.
[27] J.S. George, C.J. Aine, J-C. Mosher, D.M. Schmidt, D.M. Ranken. H.A. Schlirt, C.C. Wood, J.D. Lewine, J.A. Sanders, and J.W. Belliveau. Mapping function in the human brain with magnetoencephalography, anatomical magnetic resonance imaging, and functional magnetic resonance imaging. Journal of Chical Newo- ph-vsiology, 12(5):40&3 1, Sep 1995.
[28] P. M. Glover, P. F. Tokarczuk, and R. W- Bowtell. A robust single-shot partial sampling scheme. Magnetic Resonance in Medicine, 34( 1 ): 74-9, Jul 1 995.
1291 D. F. Gochberg, R. P. K e ~ a n , and J. C. Gore. Quantitative studies of magne- tization transfer by selective excitation and t l recovery. Magnetic Resonance in Medicine, 38(2):224-3 1, Aug 1997.
1301 J. Hennig, A. Nauerth, and H. Friedburg. Rare imaging: a fast imaging method for clinical IN. Magnetic Resonance in Medicine, 3(6): 823-3 3, Dec 1 986.
[3 11 P. Herscovitch and M. E. Raichle. What is the correct value for the brain-blood partition coefficient for water? Journal of Cerebral Blood Flow and Metabolism. 5(1):6S9, Mar 1985.
[32] R. Hoge. Glstim: An openGL based stimulus presentation program for functional MN. http://www.bic.mni.mcgil l .ca~users/rhoge/GLstim/GLstim. html.
[33] R.D. Hoge. Enhancement of volume and temporal resolution for functional mag- netic resonance brain imaging. Master's thesis, McGill University, 1995.
1343 R.D. Hoge, J. Atkinson, B. Gi11, G.R. Crelier, S. Marrett, and G.B. Pike. Additive combination of perfusion responses to bypercapnia and visual stimulation. In Pro- ceeding of The Eighth International Society of Magnetic Resonance in Medicine Con ference.
[35] R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Additive combination of perfusion responses to hypercapnia and visual stimulation. In Pro- ceeding of The Eighth international Society of Magnetic Resonance in Medicine Con ference.
[36] R.D. Hoge, J. Atkinson, B. Gi11, G.R. Crelier, S. Marrett, and G.B. Pike. Compar- ison of bulk CBF/CMRO coupling in human Vl during monocular and binocular stimulation. In Proceeding of The Eighth International Society of Magnetic Res- onance in Medicine Conference.
[3 71 R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Compar- ison of bulk CBFKMRO coupling in human VI during monocular and binocular stimulation. in f roceeding of the Fïfth International Conference on Functional Mapping of the Hurnan Brain.
[38] R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. investi- gation of BOLDsignal dependence on CBF and CMRO?: The deoxyhemoglobin dilution model. In Proceedingof the Fifrh International Conference on Functionai Mapping of the Human Brain.
[39] R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Linear coupling between cerebral blood flow and oxygen. in Proceeding of the Fi/ih International Conference on Functional Mapping of the Human Brain.
[40] R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Metabol- ic BOLD signal attenuation demonstrated using hypercapnically matched perfu- sion levels. In Proceeding of the Fqth International Conference on Funetional Mapping of the Human Brain.
[4 11 R.D. Hoge, J. Atkinson, B. Gi11, G.R. Crelier, S. Marrett, and G.B. Pike. Non- linear BOLD and perfusion dynamics in human V 1 . in Proceeding of The Eighth International Socieiy of Magnetic Resonance in Medicine Confererzce.
(421 R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Non- linear BOLD and perfusion dynarnics in human VI. In Proceeding of the F13h International Con ference on Functional Mapping of the Hurnan Brain.
1431 R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Stimulus- dependent response waveforms in human V1: Detection of BOLD and perfusion overs hoo t. Ln Proceeding of the F'@h International Con ference on Functional Màpping of the Humun h i n .
[44] R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Stimulus- dependent response waveforms in human V1: Detection of BOLD and perfusion overs hoo t. In Proceeding of The Eighth International Society of Magnetic Reso- nance in Medicine Conference.
[45] R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. hves- tigation of BOLD signal dependence on CBF and CMRo,: the deoxyhemoglobin dilution model. Submitted to Magnetic Resonance in Medicine, Oct, 1 998.
1463 R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Linear coupling between cerebral blood flow and and oxygen consumption in activated human cortex. Submitted to Proceedings of the National Acaderny of Science, Oct. 1998.
[47] R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Non- linear BOLD and perfusion dynamics in human VI. Szibmitted to Netiroirnage, Oct. 1998.
[48] R.D. Hoge, J. Atkinson, B. Gill, G.R. Crelier, S. Marrett, and G.B. Pike. Investi- gation of BOLD signal dependence on CBF and CMRo, : The deoxyhemoglobin model. In Proceeding of The Eighth International Society of Magnetic Resonance in Medicine Conference, May 1999.
[49] R.D. Hoge, B. Gill, J. Atkinson, G.R. Crelier, S. Marrett, and G.B. Pike. Inves- tigation of CMRo,/CBF coupling in human VI using fMRi. In Proceedings of the Fourth international Conference on Functional Mapping of the Human Brain, June 1998.
[SOI R.D. Hoge, B. Gill, G.R. Crelier, and G.B. Pike. Cornparison of perîusion and (bold) responses in human visual cortex to blob-selective stimuli, inter-blob- selective stimuli, and hypercapnia. In Proceedings of the Seventh Inter-national Society of Magnetic Resonance in Medicine Con ference, A p d 1 998.
[5 11 D.H. Hubel and T.N. Wiesel. Brain mechanisms of vision. Scient@ her ican, Sept. 1979.
1521 P. M. Jakob, M. A. Griswold, K. O. Lovblad, Q. Chen, and R. R. Edelman. Half- fourier burst imaging on a clinical scanner. Magnetic Resonance in Medicine, 3 8(4):534-40, Oct 1997.
CS33 G. Johnson, D. A. Feinberg, and V. Venkataraman. A cornparison of phase en- coding ordering schemes in t2-weighted grase imaging. Magnetic Resonance in Medicine, 36(3):427-35, Sep 1996.
[54] S. G. Kim. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (fair) technique: appiication to functional mapping. Mugnetic Resonance in Medicine, 34(3):293-30 1, Sep 1995.
[55] S. G. Kim and N. V. Tsekos. Perfusion imaging by a flow-sensitive altemat- ing inversion recovery (fair) technique: application to functional brain imaging
Cpublished erratum appears in magn reson med 1997 may;37(5):675]. Magnetic Resonance in Medicine, 3 7(3):42%3 5, Mar 1 997.
[56] S.G. Kim and N.V. Tsekos. Multi-slice perfusion-based fùnctional M M using the FAIR technique. In Proceedings of the Yh Meeting of the Iniernational Sociep of Magnetic Resonance in Medicine, page 3 75, 1 997.
[57] K. K. Kwong. Functional magnetic resonance imaging with echo planar imaging. Magnetic Resonance Quarteriy, 1 1 ( 1 ): 1-20, Mar 1995.
[58] K. K. Kwong, J. W. Belliveau, D. A. Chesler, 1. E. Goldberg, R. M. Weisskoff, B. P. Poncelet, D. N. Kemedy, B. E. Hoppel, M. S. Cohen, R. Turner, and a. 1. et. Dynamic magnetic resonance imaging of human brain activity during pnmary sensory stimulation. Proceedings of the National Academ-v of Sciences of the United States of America, 89(12):5675-9, Jun 15 1992.
[59] K. K. Kwong, D. A. Chesler, R. M. Weisskoff, K. M. Donahue, T. L. Davis, L. Ostergaard, T. A. Campbell, and B. R. Rosen. Mr perfusion studies with t l- weighted echo planar imaging. Magnetic Resonance in Medicine, 34(6):878-87, Dec 1995.
[60] K.K. Kwong. Current issues in fûnctional MN. NMR in Biomedicine, 1 0: 1 57- 159, 1997.
[61] S. Lai and G. H. Glover. Three-dimensional spiral fMRI technique: a comparison with 26 spiral acquisition. Magnetic Resonance in Medicine, 39( 1):68-78, Jan 1998.
[62] D. Le Bihan, editor. Dt#iusion and Perfiusion Magnetic Resonance Inzaging, Ap- plications to fMRI. Raven Press, 1 995.
[63] P. Mansfield and I.L. Pykett. Biological and medical imaging by NMR. Jorn-na/ of Magnetic Resonance, 29:3 55-3 73, 1978.
[64] A. C. McLaughlin, F. Q. Ye, J. J. Pekar, A. K. Santha, and J. A. Frank. Ef- fect of magnetization transfer on the measurement of cerebral blood flow using steady-state arterial spin tagging approaches: a theoretical investigation. Magnet- ic Resonance in Medicine, 37(4):50 1-1 0, Apr 1997.
1651 S. Meiboom and D. Gill. Modified spin-echo method for measuring nuclear re- laxation times. Review of Scientijïc Instruments, 29(8):68849 1, 1958.
[66] P. Neelin. introduction to MINC. http://www.bic.mni.mcgill.ca/so~are/minc/ minc. htm 1.
[67] D. G. Nishimura, P. Irarrazabal, and C. H. Meyer. A velocity k-space analysis of flow effects in echo-planar and spiral imaging. Magnetic Resonance in Medicine, 33(4): 54-6, Apr 1 995.
[68] D.G. Nishimura. Principles of magnetic resonance imaging. Stanford University Course Notes, 1996.
[69] Douglas C Noll, Craig H Meyer, John M Pauly, Dwight G Nishimura, and Albert Macovski. A homogeneity correction method for magnetic resonance irnaging with time-varying gradients. IEEE Transaciions on Medical Imuging, 10(4):629- 637, Dec 199 1.
[70] Douglas C Noll, Dwight G Nishimura, and Albert Macovski. Homodyne detec- tion in magnetic resonance irnaging. IEEE Transactions on Medical irnaging, 1 O(2): 154-1 63, Jun 199 1.
1711 K. Oshio and D. A. Feinberg. Grase (gradient- and spin-echo) imaging: a novel fast M M technique. Magnetic Resonance in Medicine, 20(2):344-9, Aug 199 1.
[72] E.M. Purcell, H.C. Torrey, and R.V. Pound. Resonance absorbtion by nuclear magnetic moments in a solid. Physicul Review, 69:37-38, 1946.
1731 D. A. Roberts, J. A. Detre, L. Bolinger, E. K. Insko, and J. S. Leigh Jr. Quantitative magnetic resonance imaging of human brain perfusion at 1.5 t using steady-state inversion of artenal water. Proceedings of the National Actrdemy of Sciences of ihe UniiedStates ofAmerica, 91(1):33-7, Jan 4 1994.
[74] J. Rochford. Staiistics for the neurosciences. Department of Psychiatry McGill University, 1998.
(751 1. Rock, editor. The Perceptual World, chapter 1. pages 3-24. W. H. Freeman and Company, 1990.
[76] B. R. Rosen, J. W. Belliveau, B. R. Buchbinder, R. C. McKinstry, L. M. Porkka, D. N. Kennedy, M. S. Neuder, C. R. Fisel, H. J. Aronen, K. K. Kwong, and a. 1. et. Contrast agents and cerebral hemodynarnics. Magnetic Resonance in Medicine, l9(2):28%92, Jun 199 1.
[77] B. R. Rosen, J. W. Belliveau, J. M. Vevea, and T. J. Brady. Perfusion imaging with NMR contrast agents. Magnetic Resonance in Medicine, 14(2):249-65, May 1990.
[78] B. R. Rosen, R. L. Buckner, and A. M. Dale. Event-related functional MM: past, present, and future- Proceedings of the National Academy of Sciences of the United States of America, 95(3):773-80, Feb 3 1998.
[79] M.I. Sereno, A.M. Dale, J.B. Reppas, K.K. Kwong, J.W. Belliveau, T.J. Brady, B.R. Rosen, and R.B.H. Tootell. Borders o f multiple visual areas in humans re- vealed by functional magnetic resonance imaging. Science, 268:889-93, May 1995.
[80] T. Siegal, R. Rubinstein, T. Tzuk-Shina, and J. M. Gomori. Utility o f relative cere- brai blood volume mapping derived fiom perfùsion magnetic resonance imaging in the routine follow up of brain tumors. Journal of Neurosurgew, 86(1):22-7. Jan 1997.
[81] M.S. Silver, R.I. Joseph, C.N. Chen, VJ. Sank, and D.I. Hoult. Highly selective n/2 and T pulse generation. Journal of Magnefic Resoriance, 34:52%33, 1 984.
[823 R.M. Weisskoff and S. Kiihne. MRI susceptometry: image-based measurement of absolute susceptibility of MR contrast agents and hwnan blood. Magneric Resonance in Medicine, 24(2):3 75-83, April 1 992.
[83] D. S. Williams, J. A. Detre, J. S. Leigh, and A. P. Koretsky. Magoetic resonance imaging of perfusion using spin inversion of arterial water [published erratum appears in prac natl acad sci u s a 1992 may 1;89(9):4220]. Proceedings of the National Academy of Sciences of the United States of Arnerica, 89( 1 ):2 1 2 4 , Jan 1 1992.
1841 S D Wolff and R S Balaban. Magnetization transfer contrast (MTC) and tissue water proton relaxation in vivo. Magnetic Resonance in Medicine, 1 O( I ): 1 3 5- 1 44, 1989.
185 ] S. D. Wolff and R. S. Balaban. Magnetization transfer imaging: practical aspects and clinical applications. Radiohgy, 1 92(3): 593-9, Sep 1 994.
[86] E. C. Wong, R. B. Buxton, and L. R. Frank. Implementation of quantitative per- fusion imaging techniques for functiona1 brain mapping using pulsed artenal spin labeling [see comments] . NMR in Biomedicine, 1 O(4-5):237+9, Jun-Aug 1 997.
[87] E. C. Wong, R. B. Buxton, and L. R. Frank. Quantitative imaging of per-fusion us- ing a single subtrac tion (quipss and quipss II). Magnetic Resonance in Medicine, 39(5):702-û, May 1998.
[88] E.C. Wong, R.B. Buxton, and L.R Frank. A theoretical and experimental compar- ison of continuous and pulsed arterial spin labelling techniques for quantitative perfusion irnaging. Magnetic Resonance in Medicine, 4O:MfM 55, 1 998.
[89] E.C. Wong, L.R. Frank, and R.B. Buxton. Quantitative multislice perfusion imag- ing using QUIPSS II, EPISTAR, FAIR and PICORE. In Proceedings of the jLh
Meeting of the Intemational Socieiy of Magnetic Resonance in Medicine, page 85, 1997.
1901 K.J. Worsley, S. Marrett, P. Neelin, A.C. Vandal, K.J. Friston, and A.C. Evans. A united statistical approach for determining significant signals in images of cerebral activation. Human Bmin Mapping, 458-73, 1996-
[91] G.A. Wright, B.S. Hu, and A. Macovski. Estimating oxygen saturation of blood in vivo with MR imaging at 1 3. Journal of Magnetic Resonance Imaging, 1 (3 ):27% 84, May 199 1.
1921 Y. Yang, J. A. Frank, L. Hou, F. Q. Ye, A. C. McLaughlin, and J. H- Duyn. Multi- slice imaging of quantitative cerebral perhision with pulsed arterial spin labeling. Magnetic Resonance in Medicine, 39(5): 825-3 2, May 1 998.
1931 Y. Yang, G. H. Glover, P- van Gelderen, A. C. Patel, V. S. Mattay, J. A. Frank, and J. H. Duyn. A comparison of fast mr scan techniques for cerebral acti- vation studies at 1.5 tesla Cpublished erratum appears in magn reson med 1998 mar;3 9(3):following 505 j. Magnetic Resonance in Medicine, 3 9( 1 ):6 1-7, Jan 1998.
[94] F. Q. Ye, V. S. Mattay, P. Jezzard, J. A. Frank, D. R. Weinberger, and A. C. M- claughlin. Correction for vascular artifacts in cerebral blood flow values mea- sured by using artenal spin tagging techniques. Magnetic Resonance in Medicine, 3 7(2):226-35, Feb 1 997.
[95] J.H. Zar. Biostatistical Analysk Prentice-Hall, foürth edition, 1 999.
[96] W. Zhang, A. C. Silva, D. S. Wiiliams, and A. P. Koretsky. Nmr measurement of pefision using arterial spin labeling without saturation of macromolecuIar spins. Magnetic Resonance in Medicine, 33(3):370-6, Mar 1995.
[97] W. Zhang, D. S. Williams, J. A. Detre, and A. P. Koretsky. Measurement of brain perfusion by volume-localized nmr spectroscopy using inversion of artenal water spins: accounting for transit tirne and cross-relaxation. Magnetic Resonance in Medicine, 25(2):362-7 1, Jun 1992.