Perfusion-Based Functional
Magnetic Resonance
Imaging
AFONSO C. SILVA,1 SEONG-GI KIM2
1
Laboratory of Functional and Molecular Imaging, National Institute of Neurological Disorders and Stroke, 10 Center
Drive, Building 10, Room B1D118, Bethesda, Maryland 20892-1065
2
Department of Neurobiology, University of Pittsburgh, Pittsburgh, Pennsylvania 15261
ABSTRACT: The measurement of cerebral blood flow (CBF) is a very important way of
assessing tissue viability, metabolism, and function. CBF can be measured noninvasively with
magnetic resonance imaging (MRI) by using arterial water as a perfusion tracer. Because of the
tight coupling between neural activity and CBF, functional MRI (fMRI) techniques are having a
large impact in defining regions of the brain that are activated due to specific stimuli. Among the
different fMRI techniques, CBF-based fMRI has the advantages of being specific to tissue signal
change, a critical feature for quantitative measurements within and across subjects, and for
high-resolution functional mapping. Unlike the conventional blood oxygenation level dependent (BOLD) technique, the CBF change is an excellent index of the magnitude of neural activity
change. Thus, CBF-based fMRI is the tool of choice for longitudinal functional imaging studies.
A review of the principles and theoretical backgrounds of both continuous and pulsed arterial
spin labeling methods for measuring CBF is presented, and a general overview of their current
applications in the field of functional brain mapping is provided. In particular, examples of the
use of CBF-based fMRI to investigate the fundamental hemodynamic responses induced by
neural activity and to determine the signal source of the most commonly used BOLD functional
imaging are reviewed.
© 2003 Wiley Periodicals, Inc. Concepts Magn Reson Part A 16A: 16 –27,
2003*
KEY WORDS: arterial spin labeling; brain; cerebral blood flow; magnetic resonance imaging
INTRODUCTION
Measuring cerebral blood flow (CBF) is a very important method to assess tissue viability, metabolism,
Received 19 September 2002; accepted 1 October 2002
Correspondence to: Dr. A.C. Silva; E-mail: silvaa@ninds.nih.gov.
Contract grant sponsor: National Institutes of Health.
Contract grant number: RR08079.
Contract grant sponsor: University of Minnesota.
Contract grant number: NS38295.
Concepts in Magnetic Resonance Part A, Vol. 16A(1) 16 –27 (2003)
Published online in Wiley InterScience (www.interscience.wiley.
com). DOI 10.1002/cmr.a.10050
© 2003 Wiley Periodicals, Inc. *This article is a US Government work and,
as such, is in the public domain in the United States of America.
16
and function. CBF can be measured noninvasively
with magnetic resonance imaging (MRI) by using
arterial water as a perfusion tracer (1– 8). The general
principle behind arterial spin labeling (ASL) techniques is to differentiate the net magnetization of
endogenous arterial water flowing proximally to the
organ of interest from the net magnetization of tissue.
As arterial blood perfuses the tissue, water exchange
occurs, effectively changing the net magnetization of
tissue proportionally to the blood flow rate. Therefore,
CBF can be quantitatively related to the difference of
two images acquired consecutively: one with spin
labeling and another as a control. ASL MRI techniques can be implemented with either pulsed labeling
PERFUSION-BASED fMRI
(3–5, 7) or continuous labeling (2, 6, 8). The pulsed
ASL (PASL) methods use single or multiple RF
pulses to label arterial blood water spins. The continuous ASL (CASL) technique (2) uses a long RF pulse
in the presence of a longitudinal field gradient to label
the arterial spins according to the principles of adiabatic fast passage (9). ASL techniques have major
advantages over other techniques based on the administration of exogenous tracers. It is a completely noninvasive technique for measuring CBF. Further, because of the very short half-life of the labeled spins,
repeated measurements of CBF can be performed as
often as desired. Moreover, the techniques preserve
the high spatial resolution of 1H MRI, enabling a
direct and precise anatomical localization of CBF
changes. On the other hand, proper perfusion contrast
is only achieved when enough time is allowed for the
labeled arterial spins to travel into the region of interest and exchange with tissue spins. This makes it
difficult to detect changes in CBF with a temporal
resolution greater than the decay time of the label.
Due to the tight coupling between neural activity
and CBF (10), the basis for modern functional neuroimaging methods (11), functional MRI (fMRI) techniques based on either changes in blood oxygenation
level dependency (BOLD) (12–14), regional CBF
(rCBF) (5, 15), or regional cerebral blood volume
(rCBV) (16), are having a large impact in defining
regions of the brain that are activated because of
specific stimuli. With the continued improvement of
MRI hardware, the spatial and temporal resolutions of
fMRI have both improved. Higher spatial resolution
means better spatial localization of fMRI signal
changes. Better temporal resolution also means better
spatial localization, as early hemodynamic events are
likely to occur close to the site of increased neuronal
activity. Nevertheless, the spatiotemporal relationship
between hemodynamic changes probed by functional
neuroimaging techniques and electrical neuronal
events is still poorly understood and needs to be better
characterized.
In particular, the BOLD contrast mechanism can
be modeled as a complex interplay between relative
CBF, CBV, and oxygen consumption changes
(CMRO2) (17, 18). However, the relationship between these variables is not fully understood. Positive
BOLD signal changes are presumably caused by fractional increases in CBF that are unmatched by fractional increases in CMRO2 (17, 19). This mismatch
between CBF and CMRO2 changes may not be
present in all activated regions at all times (20). Furthermore, there is no clear model to quantify the
changes in CBF from the measured BOLD signal
changes. Therefore, to date, BOLD remains a quali-
17
Figure 1 A schematic representation of an MR image
voxel for the purposes of establishing the arterial spin
labeling model. Blood flowing through the arterial vessels
with flow rate CBF (mL g⫺1 min⫺1) reach the capillary bed,
where a fraction E of water exchanges with tissue water in
the extravascular space. The remaining fraction (1 ⫺ E) of
arterial water flows to the venous side of the capillary bed
without exchanging with tissue water. Also represented in
the model is the exchange of tissue water with tissue macromolecules. Adapted from (31).
tative technique. On the other hand, perfusion-based
fMRI provides absolute quantification of CBF and
provides potentially better spatial localization than
BOLD because it is sensitive to the arterial side of the
vascular tree, in particular to capillaries.
The purpose of this article is to describe the current
state of the art in CBF-based fMRI techniques and
review the current understanding of the relationship
between BOLD and CBF signal changes occurring
during functional brain activation.
ASL TECHNIQUES
Basic Model of ASL
The theoretical modeling of MR measurements of
perfusion with ASL has been extensively described
(2, 5, 25, 21–23) and reviewed in great detail (24 –26).
Therefore, only a general description is offered here.
Figure 1 shows a schematic representation of a
18
SILVA AND KIM
typical fraction of tissue and its associated vasculature
for the purposes of establishing the arterial spin-labeling model. Blood flowing through the arterial vessels with a CBF flow rate (mL g⫺1 min⫺1) reaches the
capillary bed, where a fraction E of water exchanges
with tissue water in the extravascular (EV) space. The
remaining fraction (1 ⫺ E) of arterial water flows to
the venous side of the capillary bed without exchanging with tissue water. Also represented in the model is
the exchange of tissue water with tissue macromolecules. According to this model, the Bloch equations
for the longitudinal magnetization of brain tissue water and macromolecular spins can be written as
dM t共t兲 M0t ⫺ Mt共t兲
⫽
⫺ kforMt共t兲
dt
T1t
⫹ krevMm共t兲⫹ CBF关Ma共t兲 ⫺ Mv共t兲兴
⫻
dMm共t兲 Mm0 ⫺ Mm共t兲
⫹ kforMt共t兲 ⫺ krevMm共t兲
⫽
dt
T1m
[1]
where Mt(t) and Mm(t) are the tissue and macromolecular magnetization per gram of brain tissue, respectively; Ma(t) and Mv(t) are the arterial and venous
water magnetization per milliliter of arterial and venous blood, respectively; T1t and T1m are the tissue
water and macromolecular longitudinal relaxation
time constants, respectively (s); is the tissue/blood
partition coefficient for water; E is the water extraction fraction; and kfor and krev are the magnetization
transfer (MT) rate constants between tissue water and
macromolecular protons, respectively (s⫺1).
As reviewed elsewhere (25, 26), Eq. [1] can be
solved for the CBF rate according to any of many
different ASL approaches. In particular, ASL can be
implemented according to two main labeling strategies: CASL or PASL.
CASL Technique
The CASL strategy consists of continuously labeling
the arterial spins proximally to the brain. The continuous in-flow of labeled water leads to the development of a steady state of tissue magnetization. Labeling of arterial water can be accomplished either by a
train of saturation pulses, as originally proposed by
Detre et al. (21), or by flow-driven adiabatic fast
passage, as proposed by Williams et al. (2). The latter
method is used more because it puts the arterial water
in antiphase with tissue water, producing the largest
net change in tissue magnetization, and therefore
maximizing CBF contrast. The state of labeled water
when it exchanges with tissue is an important param-
Figure 2 The RF coil arrangement for CASL experiments.
(A) A single RF coil is used for labeling the arterial spins at
a plane proximal to the slice of interest. In this case, offresonance labeling of arterial blood induces magnetization
transfer (MT) effects, which causes a strong decrease in
tissue signal and T1 but can be controlled for in a separate
experiment where the off-resonance RF is applied in a plane
located symmetrically distally from the slice of interest. (B)
MT effects can be avoided with a two-coil approach, which
uses a small surface coil to label the carotid arteries. This
labeling coil is decoupled from the imaging coil, thus avoiding MT effects. Therefore, with the two-coil system, multislice or 3-D acquisition can be performed without subtraction artifacts and the control experiment is simply acquired
without RF power applied to the labeling coil.
eter in the quantification of CBF. The degree of labeling can be defined as
␣共兲 ⫽ ␣ 0 䡠 exp共⫺ /T1a兲
␣0 ⫽
M a0 ⫺ Ma
2Ma0
[2]
where T1a is the T1 of arterial blood. Equation [2]
shows the degree of labeling ␣() depends on the
efficiency of labeling ␣0 and on the transit time from
the labeling plane to the imaging plane. The efficiency
of labeling depends on the method used. For continuous saturation of arterial spins, Ma ⫽ 0 and ␣0 ⫽ 0.5.
For continuous inversion, ␣0 ⫽ 1. The adiabatic flowdriven inversion process is very efficient, so that in
practice the degree of labeling is dominated by the
transit time to the imaging site. In rats, where labeling
of arterial water is performed in the neck at a plane
that cuts the common carotid arteries, the ␣ value has
been measured to be better than 0.8 (6, 27). In humans, Maccotta et al. showed an ␣ value greater than
0.9 over a broad range of blood velocities (28).
CASL can be simply implemented with a single
volume RF coil that covers not only the region to be
imaged in the brain, but also the proximal area containing the feeding arteries. This scheme is shown in
Figure 2(A). In this experiment, a labeling plane is
defined to contain the main arterial supply to the
PERFUSION-BASED fMRI
brain. For example, in rodents the labeling plane is
situated in the neck, but in humans it is situated below
the circle of Willis to include both the internal carotid
and the vertebral arteries. The off-resonance RF radiation used to label the arterial spins also saturates
tissue macromolecules, which causes a strong decrease in tissue signal and T1 because of MT effects.
Fortunately, MT effects are highly symmetric in frequency, so that an easy control can be achieved if the
same off-resonance RF radiation is applied to a plane
placed distally from the imaging site at an equal
distance with respect to the labeling plane [Fig. 2(A)].
The solution of the Bloch equations (Eq. [1]) when
tissue macromolecules are saturated [Mm(t) ⫽ 0] is
given by
proposed (6, 23) and implemented for imaging of the
rat (6, 30 –34) and humans (35). The schematics of the
two-coil system is shown in Figure 2(B). A small
surface coil is placed over the neck region to label the
carotid arteries. Because the RF field generated by the
small labeling coil does not reach the brain, tissue
macromolecules are not saturated, eliminating the reduction in signal intensity and shortening of T1 due to
MT effects. In addition, multislice or 3-dimensional
(3-D) acquisition can be performed without subtraction artifacts. Using the two-coil system, CBF can be
calculated from two images acquired with and without
labeling of the arterial spins as (23, 31)
CBF ⫽
再
冋
册
CBF
⫺ kfor ⫹ 2␣
E
M t共t兲 ⫽ T1app 䡠 M0t
T1app
1
⫻共1 ⫺ exp共⫺t/T1app兲兲
␦⫽
冎
1
CBF
1
⫽
⫹ kfor ⫹
E
T 1app T1t
[3]
Equation [3] shows the reduction in tissue magnetization at steady state, together with the reduction in
tissue T1 to a shorter apparent constant T1app, which is
due to saturation of tissue macromolecules. For example, at 4.7 T the tissue signal is reduced by 75%
from its equilibrium value and T1app ⫽ 0.45 s, compared to T1t ⫽ 1.7 s (22). CBF can be measured from
two MR images, one in which RF is applied to the
control plane and another in which RF is applied to
the labeling plane:
CBF ⫽
M ct ⫺ M lt
1
l
E T1t M t ⫹ 共2␣ ⫺ 1兲 M ct
冋
册
19
M0t ⫺ M lt
1
⫹␦
E T1t
M lt ⫹ 共2␣ ⫺ 1兲 M0t
冉
冊
k for
1 ⫹ krevT1m
[5]
where ␦ is a term that accounts for exchange between
tissue water and macromolecules. This term accounts
for 17% of the CBF quantification at 4.7 T (23).
In addition to allowing measurements of CBF with
higher SNR and more extensive coverage than the
one-coil implementation of CASL, the two-coil approach had a key role in understanding the basic
principles of water exchange in the brain (30, 31).
Further, the use of a separate labeling coil significantly reduces RF power deposition, which is a critical consideration for use of CASL in humans (35).
Unfortunately, however, the human anatomy forces
placement of the labeling coil too far upstream from
the region of interest in the brain, causing a significant
attenuation of the degree of labeling due to transittime decay.
[4]
where M ct and M lt are the tissue magnetization in the
control and labeled states, respectively.
As mentioned above, the one-coil implementation
of CASL inconveniently saturates tissue macromolecules, causing a reduction in both in the signal amplitude an the T1, effectively reducing the signal to
noise ratio (SNR) of the CBF measurement. Furthermore, the control plane illustrated in Figure 2(A)
works only for a single slice parallel to the labeling
plane. Clever approaches to allow multislice imaging
have been proposed (8, 29), but they do not avoid
saturation of tissue macromolecules and therefore still
suffer from poor SNR. A hardware approach to eliminate saturation of tissue macromolecules has been
PASL Approach
The pulsed approach to ASL consists of labeling a
thick slab of blood upstream from the region of interest with a short RF pulse and waiting a certain time to
allow the labeled blood to mix with tissue prior to
acquiring the image. This idea has been implemented
in a number of different ways (3–5, 36). The approach
used by Kim (5) was named flow-sensitive alternating
inversion recovery (FAIR), which is illustrated in
Figure 3. In FAIR, the labeled image is acquired
inside a slice-selective inversion slab. The uninverted
blood that flows from outside the inversion slab into
the inversion slab creates the desired CBF contrast,
because uninverted arterial water speeds up the longitudinal relaxation of tissue water upon mixing due
20
SILVA AND KIM
CURRENT APPLICATIONS OF CBFBASED fMRI
Figure 3 A scheme for FAIR pulsed arterial spin labeling
(PASL). The labeled image is acquired inside a slice-selective inversion slab. The uninverted blood that flows from
outside the inversion slab into the inversion slab creates the
desired CBF contrast, because uninverted arterial water
speeds up the longitudinal relaxation of tissue water upon
mixing that is due to perfusion. The control situation is
acquired after a nonselective inversion pulse, when both
arterial water and tissue are inverted.
to perfusion. The control situation is acquired after a
nonselective inversion pulse, when both arterial water
and tissue are inverted, so that tissue relaxes back to
equilibrium with the normal relaxation constant T1t.
Using FAIR the CBF can be quantified according to
(25, 37, 38)
CBF ⫽
⌬Mt
1/T1a ⫺ 1/T1app
0
2␣0 Mt exp共⫺TI/T1app兲 ⫺ exp共⫺TI/T1a兲
冋
册
[6]
where ⌬Mt is the signal difference between the control and labeled images, TI is the inversion time, and
T1a is the T1 of arterial blood. Multislice acquisition
can also be performed with FAIR (1). In this case the
slice-selective inversion slab must contain all imaging
slices, and the different transit time to different slices
could compromise quantification of CBF. However, it
has been reported that when relative CBF changes are
determined in multislice FAIR experiments, errors
induced by different transit times are not significant
(39). Relative CBF changes measured by PASL are in
very good agreement with those measured by H15
2 O
positron emission tomography (PET) in the same region and subject during the identical stimulation task
(39). Thus, FAIR is also an excellent perfusion technique for the measurement of relative CBF changes
induced by neural activity or other external perturbations.
The coupling between neural activity and CBF makes
ASL-based MRI techniques excellent tools for monitoring brain function in both normal and pathological
states. Many pathophysiological disorders of the brain
are associated with alterations in normal perfusion
values, which can be mapped and quantified or followed in time with ASL techniques. In particular,
ASL has the potential advantages of BOLD because it
can quantify resting CBF, as well as CBF changes,
making results useful in longitudinal studies or in
comparisons among clinical populations. The applications of ASL-based techniques in assessing brain
disorders are currently quite broad and beyond the
scope of this article. The reader is referred to other
reviews (40 – 42) of current clinical applications of
ASL.
The CBF changes induced by external perturbations will be, in theory, the same whether measurements are performed on high or low field systems.
Therefore, relative CBF measurements using the perfusion-based fMRI technique can be performed even
at low magnetic fields, provided adequate SNR can be
achieved, although high fields provide a higher SNR
and a longer T1 of water.
The BOLD signal is sensitive to a large static
susceptibility effect around the tissue to sinus boundaries, making it difficult to obtain high quality images
in the frontal and temporal areas. Thus, it is difficult
to obtain fMRI in these areas. Because the perfusionbased technique does not rely on the susceptibility
effect for its contrast, a spin– echo (SE) data collection scheme can be used for recovering signals in high
susceptibility regions. Hence, perfusion-based fMRI
is better than the conventional BOLD method for
mapping highly susceptible areas.
In the functional MRI studies, baseline drift is a
problem because functional maps are calculated from
a comparison between control and task-induced signal
intensities. This drift may be due to physiological
changes induced by effects such as anxiety and anticipation, the subject’s motion (even if subpixel movements), system instabilities due to changes in eddy
current compensation and field drift, a change in gain
setting at different scanning sessions, and an approach
to steady-state magnetization if relatively fast repetition of RF pulses is employed. Slight, not abrupt,
baseline changes can be corrected by subtraction of an
image from its neighbor image (5), improving the
sensitivity of perfusion-based fMRI. Also, this property of perfusion-based fMRI can allow the use of
PERFUSION-BASED fMRI
21
perfusion-based fMRI for longitudinal studies. For
intersubject or intersession intrasubject comparisons,
the perfusion-based techniques are the method of
choice.
Most of the fMRI studies performed today are
based on the BOLD contrast, because of its high
contrast to noise ratio (CNR) and simplicity of implementation. However, the BOLD contrast mechanism
depends on a complex interplay between CBF, CBV,
and CMRO2 (17, 18). The intrinsic relationship between these variables is not fully understood. Positive
BOLD signal changes are presumably caused by increases in CBF that are unmatched by the corresponding increases in CMRO2 (17, 19). On the other hand,
negative BOLD signals can be generated by early
increases in CMRO2 prior to increases in CBF or by
increases in CBV. Therefore, to better understand the
physiological mechanisms underlying functional hemodynamic changes, it is fundamental to obtain simultaneous measurements of BOLD, CBF, and CBV
changes. For the remainder of this article, we focus on
simultaneous measurements of BOLD and CBF performed in animals and humans and attempt to provide
a current understanding of hemodynamic regulation
during functional brain activation.
Spatial Specificity
Animal studies comparing BOLD and CBF regions of
activation using fMRI have been recently reported in
the rat somatosensory cortex (33, 34, 43, 44) and in
the cat visual cortex (45). These animal studies, performed at high magnetic field strength, have established important considerations regarding the spatial
localization of BOLD with respect to CBF and to the
expected site of increased electrical activity. For example, Figure 4(A) shows typical BOLD and CBF
cross-correlation functional maps obtained at 9.4 T
during forepaw stimulation in a rat. Good agreement
in the spatial location of activation regions was observed, and the mean separation between the center of
the BOLD and the CBF active regions was determined to be less than one pixel (33). In addition, the
number of pixels in the BOLD region was strongly
correlated to the number of pixels in the CBF region.
Figure 4(B) shows a comparison of the overlap of
BOLD, CBF, and calcium influx, as probed by T1
weighted images sensitive to the calcium analog
Mn⫹⫹ (43). The highest CBF and calcium influx
change was located in layer 4 of the somatosensory
cortex. It demonstrates the excellent overlap between
the regions of activation as reported by BOLD, CBF,
and Mn-dependent T1 contrast, which is proof of the
excellent spatial localization of BOLD signal changes
Figure 4 (A) BOLD (top row) and CBF (bottom row)
activation maps of the rat brain upon electrical stimulation
of the right forepaw. The bottom graph shows the time
course of a 9 pixel ROI placed on the center of the active
areas (middle slice) of the BOLD and CBF maps. The
temporal correlation coefficient between the BOLD and the
CBF time courses was 0.92. (B) Spatial overlap of hemodynamic versus calcium-dependent activity. The first column shows the combined maps of BOLD (red), CBF
(green), and the overlap between the two (yellow). The
second column shows the combined maps of BOLD (red),
T1 (green), and the overlap between the two maps (yellow).
The third column shows the combined maps of CBF (red),
T1 (green), and the overlap between the two (yellow).
at high magnetic field strengths. However, although
CBF maintains its high spatial specificity at all magnetic field strengths, the same cannot be said of the
BOLD contrast. The functional CBF maps measured
with FAIR in response to single orientation stimulation of the cat visual cortex at 4.7 T were specific to
a submillimeter columnar structure, in contrast to
gradient– echo (GE) BOLD, which could not resolve
columnar structures (45). This indicates that perfusion-based fMRI provides higher spatial specificity
than BOLD fMRI at medium and low magnetic fields
such as 4.7 T.
CBF and BOLD studies were also performed in
humans during finger movements and visual stimula-
22
SILVA AND KIM
Figure 5 Multislice FAIR measurements of a normal volunteer during finger movement at 4 T. The panels show
FAIR contrast (top), CBF-weighted fMRI maps overlaid on
anatomic images (middle), and BOLD fMRI maps (bottom).
CBF-weighted fMRI was obtained using the FAIR technique, while BOLD maps were obtained from nonslice
selective inversion-recovery images acquired as part of
FAIR. The color bar shows from 10 to ⬎90% changes for
CBF and from 1 to ⬎9% changes for BOLD. The arrows
indicate the central sulcus. It is interesting that the CBFweighted fMRI signals localize to tissue areas, not to large
vessels. Thus, higher spatial specificity can be obtained
using CBF techniques. Adapted from (1).
tion at 4 T (1, 37, 46, 47). Generally, supramillimeter
activation sites were consistent between CBF and GE
BOLD fMRI maps and stimulation frequency-dependent activation covaried in both functional images.
However, as shown in Figure 5, GE BOLD maps
show “activation” in large draining veins, such as the
Rolendic vein in the central sulcus, which are absent
in FAIR-based fMRI.
In addition to being strongly related to the strength
of the magnetic field, BOLD is also very sensitive to
the type of MRI pulse sequence used (44, 45, 48 –51).
Inspecting the vascular contribution to BOLD fMRI
signals constitutes a fundamental step toward a better
understanding of the spatial specificity issues in
BOLD. In particular, the BOLD effect is sensitive to
the venous blood volume and vessel size and orientation (17, 18, 52). To comprehend the nature of the
vascular contribution, it is important to separate the
macrovascular from the microvascular components
(53). The microvascular or tissue component is defined as coming from capillaries and surrounding tissues, whereas the macrovascular component arises
from large venules and veins. The microvascular effect is believed to be close to the site of neuronal
activity. However, unlike capillaries, there is not a
high density of large blood vessels in the brain; thus,
functional maps based on the macrovasculature can be
significantly distant from the actual site of neural
activity. Therefore, it is desirable to minimize the
macrovascular contribution. According to the BOLD
model (18, 52), vascular contributions to the BOLD
signal are composed of EV and intravascular (IV)
effects. The EV contribution from large vessels is
linearly dependent on the magnetic field strength (B0),
whereas the EV contribution from microvessels increases quadratically with B0. This suggests that high
magnetic fields can increase the relative contribution
of the microvascular component to the BOLD signal.
The EV component of microvessels contributes to
both SE and GE fMRI as a result of dynamic signal
averaging induced by water diffusion during an echo
time (TE). However, the EV component of large
Figure 6 A comparison between spin– echo (top) and
gradient– echo (bottom) BOLD fMRI. The BOLD statistical
maps are overlaid on two-segment EPI images acquired at
9.4 T at TE ⫽ 16 ms (GE) or 40 ms (SE). The SE and GE
images are diffusion weighted (b ⫽ 100 s/mm2 for SE, 200
s/mm2 for GE). On the left, the CCC maps are color coded
between 0.5 and 0.9, while the relative BOLD signal change
maps shown on the right are color coded between 3 and
11% signal changes. The functional maps on the left present
high cross-correlation coefficient (CCC) values in the deep
layers of the somatosensory cortex for both SE and GE
images. The SE BOLD percentage of increase map (top
right) agrees with the corresponding CCC map (top left).
However, the GE BOLD percentage of signal increase map
(bottom right) presents its highest signal increases at the pial
surface, where large superficial veins are located. Adapted
from (50).
PERFUSION-BASED fMRI
23
is crucial to remove large vessel contributions. It is
important to mention that SE BOLD at 1.5 T contains
predominantly an IV component (56, 57). Therefore,
the combination of high magnetic fields with diffusion-weighted SE sequences that are insensitive to
draining veins constitutes the best approach to BOLDbased fMRI at submillimeter resolution.
Magnitude of Signal Changes
Figure 7 A comparison of the relative SE BOLD and CBF
signal changes during electrical stimulation of the rat forepaw. Data are grouped into three diffusion-weighting
ranges. There is no difference in the correlation between SE
BOLD and CBF at different diffusion weightings. The solid
curve shows the fit of the data to the function indicated on
the graph. Adapted from (44).
vessels contributes only to GE fMRI, not to SE fMRI,
because the 180° RF pulse in SE fMRI can refocus the
dephasing effect of static field inhomogeneities
around large vessels (18, 52, 54, 55). Applying flowsensitive bipolar gradients cannot reduce these EV
effects. The vascular contribution to BOLD was extensively studied at 9.4 T. In one study, GE and SE
BOLD contrast was compared in the presence of
graded diffusion weighting to reveal the contribution
from large vessels to BOLD (50). Figure 6 shows a
comparison between diffusion-weighted GE and SE
BOLD fMRI in a rat model of somatosensory stimulation. The SE image was acquired using b ⫽ 100
s/mm2, and the GE image was acquired at b ⫽ 200
s/mm2. Whereas the SE image shows the largest signal changes in the middle layers of the somatosensory
cortex, the GE image presents high signal increases at
the pial surface, where large superficial veins are
located. A combination of a long TE and diffusion
weighting helps minimize the IV component of both
the GE and SE images. However, in GE fMRI, the
highest percentage of signal changes still take place
near the edge of the brain, because the EV component
from large vessels cannot be suppressed by a GE
sequence, even in the presence of diffusion-sensitizing gradients (50). Another study compared SE
BOLD to CBF in the same rat model, leading to the
conclusion that both CBF- and SE BOLD-based fMRI
yield tissue-specific maps at high magnetic fields (44).
To obtain accurate high-resolution functional maps, it
CBF-based fMRI can provide better localized mapping of neuronal activation because it is not sensitive
to large draining vessels. The CBF contrast mostly
reflects truly perfusing spins that have permeated the
capillary walls and entered the EV space. Because
relative CBF changes are linearly correlated to metabolic changes (19, 58), CBF can play the role of a
gold standard for quantifying neuronal activity. The
quantification of absolute CBF values requires suppression of large vessel artifacts, in particular those
originating from large arteries. We have observed a
10 –20% reduction in resting CBF values with the use
of small (b ⫽ 20 –500 s/mm2) diffusion-sensitizing
gradients (30, 44). However, the use of diffusionsensitizing gradients has no effect of quantifying relative CBF changes during somatosensory stimulation
in rats (44), suggesting the arterial vasodilatation is
proportional to the CBF changes. This is consistent
with our previous finding of significant arterial CBV
changes during increased CBF (51). Taken together,
the contribution of large vessels to functional CBF
changes, as measured by the CASL technique, does
not alter tissue-level relative CBF changes.
Because the BOLD contrast is dependent on various physiological and anatomical parameters, it is
important to compare BOLD signal changes with
CBF-based fMRI. Relative CBF and BOLD signal
changes during somatosensory stimulation in rats
have been extensively compared at 9.4 T (33, 34, 43,
44). Figure 7 shows an equivalent plot of SE BOLD
versus CBF changes. Individual diffusion-sensitized
fMRI data obtained from 10 animals (12 paws) are
plotted; low (b ⫽ 0 –5 s/mm2), intermediate (b ⫽
20 –100 s/mm2), and high (b ⫽ 150 –500 s/mm2)
diffusion gradient data are shown. The relationship
between relative BOLD and CBF changes was identical in all groups, indicating that the presence of
diffusion-weighting gradients did not affect any signal. This is consistent with previous observations that
the SE BOLD contrast has its origin in EV dynamic
averaging effects around small vessels (50), and
therefore our results show excellent correlation between SE BOLD and CBF changes during functional
stimulation at high spatial resolution. When the large
24
SILVA AND KIM
vascular component is suppressed, CBF and BOLD
fMRI contrasts are closely coupled and originate from
a similar anatomical location within a single voxel.
However, the relationship between BOLD and CBF
changes is highly nonlinear, especially at high CBF
changes. Such a relationship can also be found in CBF
and GE BOLD signals obtained from a large region of
interest, but not on a pixel by pixel basis (47). This
consideration should be taken into account when trying to use BOLD-based fMRI as a quantitative
method for mapping neuronal activity.
Temporal Characteristics
The inherent temporal resolution of ASL methods of
quantifying CBF is inherently low. Proper perfusion
contrast is achieved when enough time is allowed for
the labeled spins to travel into the region of interest
and exchange with tissue spins. In addition, it is
necessary to acquire two images, usually in an interleaved manner, to determine CBF: one with spin
labeling and another as a control. Thus, the typical
temporal resolution of ASL methods is on the order of
a few seconds (3.5, 5.8, and 8). In order to obtain
dynamic CBF changes with high temporal and spatial
resolution, we have recently devised a novel MRI
technique, coined pseudocontinuous ASL (PCASL)
(32). The PCASL technique consists of using a short
ASL RF pulse in conjunction with an ultrafast imaging sequence, such as echo–planar or spiral imaging.
The ASL RF pulse is made short to allow for high
temporal resolution but long compared to the imaging
time, so that high labeling duty cycles (and thus the
efficiency) can be maintained. For example, PCASL
has been implemented using 78-ms ASL pulses in
conjunction with a 30-ms echo-planar imaging (EPI)
sequence (32, 24). Under these conditions, CBF images could be formed every 108 ms with a labeling
efficiency of 59% (␣ ⫽ 0.59). Two separate experiments are performed in PCASL: one with spin labeling and the other as a control. Once the CBF images
are formed according to Eq. [5], an analysis of the
temporal characteristics of the CBF time course is
desired. For this, a temporal deconvolution of the
CBF time course becomes necessary. This is because
instantaneous changes in CBF cause slow variations
in the MRI signal. The basic principle of the ASL
technique is the transfer of the longitudinal magneti-
Figure 8 (A) MRI-measured (gray) and true (black) CBF
curves obtained at 108-ms temporal resolution during electrical stimulation of the rat forepaw. The MRI-measure
curve was deconvolved with the tissue T1 decay curve to
produce the true CBF response. (B) Averaged onset times of
CBF and BOLD in the surface (gray bars) and deep (white
bars) regions of the somatosensory cortex. (*) The onset of
the BOLD response in the cortical surface was significantly
longer than deep in the cortex (p ⬍ 0.03). (**) CBF changes
in the deep cortex occurred earlier than the corresponding
BOLD changes (p ⬍ 0.003). (***) The onset of superficial
CBF changes was significantly delayed with respect to deep
in the cortex (p ⬍ 0.004). (C) Averaged times to peak of
CBF and BOLD. There was no significant time to peak
differences across regions for either BOLD (p ⬎ 0.28) or
CBF (p ⬎ 0.39). (*) However, the CBF peak response
occurred faster than the BOLD response in both regions
(p ⬍ 0.001). Error bars ⫽ 1 SD. Adapted from (34).
PERFUSION-BASED fMRI
zation state of the arterial water spins to the tissue
spins. This transfer is limited by T1app, repetition time
(TR), and the RF flip angle and cannot occur instantly. Therefore, step changes in perfusion (and,
consequently, in T1app) are only reflected a few seconds later in the tissue magnetization. By performing
a deconvolution of the MRI-measured CBF signal
with the initial magnetization decay curve, this latency in the MRI measured CBF response can be
removed. After this deconvolution process, the resulting CBF time course accurately reflects the dynamics
of the actual CBF changes.
Figure 8(A) shows the MRI-estimated (gray) and
the deconvolved CBF (black) time courses obtained
during somatosensory stimulation in rats using the
PCASL technique. The MRI-estimated CBF curve
was deconvolved with the initial 10 s of the control
magnetization decay, generating the deconvolved
CBF signal. It can be clearly seen how the CBF
response measured with MRI is delayed with respect
to the deconvolved curve. Note from Figure 8(A) that
the deconvolution adds oscillatory noise to the resulting curve. However, the CBF changes elicited by this
model of activation are very robust, to the point that
the results presented here are not compromised by the
additional noise introduced by the deconvolution process. Because we used GE EPI as the readout imaging
sequence in our PCASL technique, BOLD signal
changes could be measured from the control series of
images and directly compared to the corresponding
CBF changes. Figure 8(B) shows the onset time of
BOLD and CBF in the superficial and deep regions of
the somatosensory cortex following the onset of stimulation. CBF changes in the deep layers of the somatosensory cortex occurred earlier than the corresponding BOLD changes (p ⬍ 0.003). However, in
the superficial layers, the onset of the CBF response
was delayed and it was similar to the latency of the
superficial BOLD signal changes. Figure 8(C) shows
the BOLD and CBF times to peak. The CBF peak
response occurred faster than the BOLD response in
both regions (p ⬍ 0.001).
CONCLUSIONS
Perfusion-based fMRI is specific to tissue signal
changes, a critical feature for proper quantification of
the functional response and for high-resolution functional mapping. Unlike the conventional BOLD technique, the CBF change is an excellent index of the
magnitude of neural activity change. Perfusion-based
fMRI provides high spatial resolution because the
contribution of draining veins to the CBF-weighted
25
signal is minimal. The perfusion changes induced by
neural activity are faster than the BOLD response. By
combining ASL with the BOLD technique, both the
CBF and venous oxygenation level can be obtained,
which can be used for examining the sources of the
BOLD contrast. Because slow baseline changes can
be eliminated by pairwise subtraction of images,
CBF-based functional images can be obtained even
when baseline signals are modulated because of system instabilities, different gain settings, or physiological changes. Thus, perfusion-based fMRI is the tool
of choice for longitudinal functional imaging studies.
Overall, the perfusion-based fMRI technique is an
excellent complementary approach for functional
mapping of human and animal brains.
ACKNOWLEDGMENTS
The authors would like to acknowledge the financial
support from the National Institutes of Health (to the
University of Minnesota) and the Keck Foundation.
REFERENCES
1. Kim SG, Tsekos NV, Ashe J. Multi-slice perfusionbased functional MRI using the FAIR technique: Comparison of CBF and BOLD effects. NMR Biomed 1997;
10:191–196.
2. Williams DS, Detre JA, Leigh JS, Koretsky AP. Magnetic resonance imaging of perfusion using spin inversion of arterial water. Proc Natl Acad Sci USA 1992;
89:212–216.
3. Kwong KK, Belliveau JW, Chesler DA, Goldberg IE,
Weisskoff RM, Poncelet BP, Kennedy DN, Hoppel BE,
Cohen MS, Turner R. Dynamic magnetic resonance
imaging of human brain activity during primary sensory
stimulation. Proc Natl Acad Sci USA 1992; 89:5675–
5679.
4. Edelman RR, Siewert B, Darby DG, Thangaraj V,
Nobre AC, Mesulam MM, Warach S. Qualitative mapping of cerebral blood flow and functional localization
with echo–planar MR imaging and signal targeting with
alternating radio frequency. Radiology 1994; 192:513–
520.
5. Kim SG. Quantification of relative cerebral blood flow
change by flow-sensitive alternating inversion recovery
(FAIR) technique: Application to functional mapping.
Magn Reson Med 1995; 34:293–301.
6. Silva AC, Zhang W, Williams DS, Koretsky AP. Multislice MRI of rat brain perfusion during amphetamine
stimulation using arterial spin labeling. Magn Reson
Med 1995; 33:209 –214.
7. Wong EC, Buxton RB, Frank LR. Implementation of
quantitative perfusion imaging techniques for func-
26
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
SILVA AND KIM
tional brain mapping using pulsed arterial spin labeling.
NMR Biomed 1997; 10:237–249.
Alsop DC, Detre JA. Multisection cerebral blood flow
MR imaging with continuous arterial spin labeling.
Radiology 1998; 208:410 – 416.
Dixon WT, Du LN, Faul DD, Gado MH, Rossnick S.
Projection angiograms of blood labeled by adiabatic
fast passage. Magn Reson Med 1986; 3:454 – 462.
Roy CS, Sherrington CS. On the regulation of the blood
supply of the brain. J Physiol 1890; 11:85–108.
Raichle ME. Behind the scenes of functional brain
imaging: A historical and physiological perspective.
Proc Natl Acad Sci USA 1998; 95:765–772.
Bandettini PA, Wong EC, Hinks RS, Tikofsky RS,
Hyde JS. Time course EPI of human brain function
during task activation. Magn Reson Med 1992; 25:
390 –397.
Ogawa S, Tank DW, Menon RS, Ellermann JM, Kim
S-G, Merkle H, Ugurbil K. Intrinsic signal changes
accompanying sensory stimulation: Functional brain
mapping with magnetic resonance imaging. Proc Natl
Acad Sci USA 1992; 89:5951–5955.
Turner R. Magnetic resonance imaging of brain function. Am J Physiol Imaging 1992; 7(3– 4):136 –145.
Kwong KK, Chesler DA, Weisskoff RM, Donahue
KM, Davis TL, Ostergaard L, Campbell TA, Rosen BR.
MR perfusion studies with T1-weighted echo planar
imaging. Magn Reson Med 1995; 34:878 – 887.
Belliveau JW, Rosen BR, Kantor HL, Rzedzian RR,
Kennedy DN Jr, McKinstry RC, Vevea JM, Cohen MS,
Pykett IL, Brady TJ. Functional cerebral imaging by
susceptibility-contrast NMR. Magn Reson Med 1990;
14:538 –546.
Ogawa S, Menon RS, Tank DW, Kim SG, Merkle H,
Ellermann JM, Ugurbil K. Functional brain mapping by
blood oxygenation level-dependent contrast magnetic
resonance imaging. A comparison of signal characteristics with a biophysical model. Biophys J 1993; 64:
803– 812.
Weisskoff RM, Zuo CS, Boxerman JL, Rosen BR.
Microscopic susceptibility variation and transverse relaxation: Theory and experiment. Magn Reson Med
1994; 31:601– 610.
Fox PT, Raichle ME. Focal physiological uncoupling
of cerebral blood flow and oxidative metabolism during
somatosensory stimulation in human subjects. Proc
Natl Acad Sci USA 1986; 83:1140 –1144.
Roland PE, Eriksson L, Stone-Elander S, Widen L.
Does mental activity change the oxidative metabolism
of the brain? J Neurosci 1987; 7:2373–2389.
Detre JA, Leigh JS Jr, Williams DS, Koretsky AP.
Perfusion imaging. Magn Reson Med 1992; 23:37– 45.
Zhang W, Williams DS, Detre JA, Koretsky AP. Measurement of brain perfusion by volume-localized NMR
spectroscopy using inversion of arterial water spins:
Accounting for transit time and cross-relaxation. Magn
Reson Med 1992; 25:362–371.
Zhang W, Silva AC, Williams DS, Koretsky AP. NMR
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
measurement of perfusion using arterial spin labeling
without saturation of macromolecular spins. Magn Reson Med 1995; 33:370 –376.
Detre JA, Zhang W, Roberts DA, Silva AC, Williams
DS, Grandis DJ, Koretsky AP, Leigh JS. Tissue specific
perfusion imaging using arterial spin labeling. NMR
Biomed 1994; 7:75– 82.
Calamante F, Thomas DL, Pell GS, Wiersma J, Turner
R. Measuring cerebral blood flow using magnetic resonance imaging techniques. J Cereb Blood Flow Metab
1999; 19:701–735.
Barbier EL, Lamalle L, Decorps M. Methodology of
brain perfusion imaging. J Magn Reson Imaging 2001;
13:496 –520.
Zhang W, Williams DS, Koretsky AP. Measurement of
rat brain perfusion by NMR using spin labeling of
arterial water: In vivo determination of the degree of
spin labeling. Magn Reson Med 1993; 29:416 – 421.
Maccotta L, Detre JA, Alsop DC. The efficiency of
adiabatic inversion for perfusion imaging by arterial
spin labeling. NMR Biomed 1997; 10:216 –221.
Talagala SL, Barbier EL, Williams DS, Silva AC, Koretsky AP. Multi-slice perfusion MRI using continuous
arterial water labeling controlling for MT effects with
simultaneous proximal and distal RF irradiation. In
Proceedings of the 6th Annual Meeting of the ISMRM,
Sydney, Australia, 1998; Vol. 1, p 381–381.
Silva AC, Williams DS, Koretsky AP. Evidence for the
exchange of arterial spin-labeled water with tissue water in rat brain from diffusion-sensitized measurements
of perfusion. Magn Reson Med 1997; 38:232–237.
Silva AC, Zhang W, Williams DS, Koretsky AP. Estimation of water extraction fractions in rat brain using
magnetic resonance measurement of perfusion with arterial spin labeling. Magn Reson Med 1997; 37:58 – 68.
Silva AC, Kim SG. Pseudo-continuous arterial spin
labeling technique for measuring CBF dynamics with
high temporal resolution. Magn Reson Med 1999; 42:
425– 429.
Silva AC, Lee SP, Yang G, Iadecola C, Kim SG.
Simultaneous blood oxygenation level-dependent and
cerebral blood flow functional magnetic resonance imaging during forepaw stimulation in the rat. J Cerebr
Blood Flow Metab 1999; 19:871– 879.
Silva AC, Lee SP, Iadecola C, Kim SG. Early temporal
characteristics of cerebral blood flow and deoxyhemoglobin changes during somatosensory stimulation.
J Cerebr Blood Flow Metab 2000; 20:201–206.
Zaharchuk G, Ledden PJ, Kwong KK, Reese TG,
Rosen BR, Wald LL. Multislice perfusion and perfusion territory imaging in humans with separate label
and image coils. Magn Reson Med 1999; 41:1093–
1098.
Wong EC, Buxton RB, Frank LR. Quantitative imaging
of perfusion using a single subtraction (QUIPSS and
QUIPSS II). Magn Reson Med 1998; 39:702–708.
Kim SG, Tsekos NV. Perfusion imaging by a flowsensitive alternating inversion recovery (FAIR) tech-
PERFUSION-BASED fMRI
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
nique: Application to functional brain imaging. Magn
Reson Med 1997; 37:425– 435.
Tsekos NV, Zhang F, Merkle H, Nagayama M, Iadecola C, Kim SG. Quantitative measurements of cerebral
blood flow in rats using the FAIR technique: Correlation with previous iodoantipyrine autoradiographic
studies. Magn Reson Med 1998; 39:564 –573.
Zaini MR, Strother SC, Anderson JR, Liow JS, Kjems
U, Tegeler C, Kim SG. Comparison of matched BOLD
and FAIR 4.0T-fMRI with [15O]water PET brain volumes. Med Phys 1999; 26:1559 –1567.
Detre JA, Floyd TF. Functional MRI and its applications to the clinical neurosciences. Neuroscientist 2001;
7:64 –79.
Warach S. Use of diffusion and perfusion magnetic
resonance imaging as a tool in acute stroke clinical
trials. Curr Control Trials Cardiovasc Med 2001; 2:38 –
44.
Detre JA, Wang J. Technical aspects and utility of
fMRI using BOLD and ASL. Clin Neurophysiol 2002;
113:621– 634.
Duong TQ, Silva AC, Lee SP, Kim SG. Functional
MRI of calcium-dependent synaptic activity: Cross correlation with CBF and BOLD measurements. Magn
Reson Med 2000; 43:383–392.
Lee SP, Silva AC, Kim SG. Comparison of diffusionweighted high-resolution CBF and spin– echo BOLD
fMRI at 9.4 T. Magn Reson Med 2002; 47:736 –741.
Duong TQ, Kim DS, Ugurbil K, Kim SG. Localized
cerebral blood flow response at submillimeter columnar
resolution. Proc Natl Acad Sci USA 2001; 98:10904 –
10909.
Kim SG, Ugurbil K. Comparison of blood oxygenation
and cerebral blood flow effects in fMRI: Estimation of
relative oxygen consumption change. Magn Reson Med
1997; 38:59 – 65.
Zhu XH, Kim SG, Andersen P, Ogawa S, Ugurbil K,
Chen W. Simultaneous oxygenation and perfusion imaging study of functional activity in primary visual
cortex at different visual stimulation frequency: Quantitative correlation between BOLD and CBF changes.
Magn Reson Med 1998; 40:703–711.
Frahm J, Merboldt KD, Hanicke W, Kleinschmidt A,
Boecker H. Brain or vein—Oxygenation or flow? On
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
27
signal physiology in functional MRI of human brain
activation. NMR Biomed 1994; 7:45–53.
Kim SG, Hendrich K, Hu X, Merkle H, Ugurbil K.
Potential pitfalls of functional MRI using conventional
gradient-recalled echo techniques. NMR Biomed 1994;
7:69 –74.
Lee SP, Silva AC, Ugurbil K, Kim SG. Diffusionweighted spin– echo fMRI at 9.4 T: Microvascular/
tissue contribution to BOLD signal changes. Magn Reson Med 1999; 42:919 –928.
Lee SP, Duong TQ, Yang G, Iadecola C, Kim SG.
Relative changes of cerebral arterial and venous blood
volumes during increased cerebral blood flow: Implications for BOLD fMRI. Magn Reson Med 2001; 45:
791– 800.
Ogawa S, Menon RS, Kim SG, Ugurbil K. On the
characteristics of functional magnetic resonance imaging of the brain. Annu Rev Biophys Biomol Struct
1998; 27:447– 474.
Song AW, Wong EC, Tan SG, Hyde JS. Diffusion
weighted fMRI at 1.5 T. Magn Reson Med 1996; 35:
155–158.
Bandettini PA, Wong EC, Jesmanowicz A, Hinks RS,
Hyde JS. Spin– echo and gradient– echo EPI of human
brain activation using BOLD contrast: A comparative
study at 1.5 T. NMR Biomed 1994; 7:12–20.
van Zijl PC, Eleff SM, Ulatowski JA, Oja JM, Ulug
AM, Traystman RJ, Kauppinen RA. Quantitative assessment of blood flow, blood volume and blood oxygenation effects in functional magnetic resonance imaging. Nat Med 1998; 4:159 –167.
Zhong J, Kennan RP, Fulbright RK, Gore JC. Quantification of intravascular and extravascular contributions
to BOLD effects induced by alteration in oxygenation
or intravascular contrast agents. Magn Reson Med
1998; 40:526 –536.
Oja JM, Gillen J, Kauppinen RA, Kraut M, van Zijl PC.
Venous blood effects in spin– echo fMRI of human
brain. Magn Reson Med 1999; 42:617– 626.
Sokoloff L, Reivich M, Kennedy C, Des Rosiers MH,
Patlak CS, Pettigrew KD, Sakurada O, Shinohara M.
The [14C]deoxyglucose method for the measurement
of local cerebral glucose utilization: Theory, procedure,
and normal values in the conscious and anesthetized
albino rat. J Neurochem 1977; 28:897–916.