MRS

Magnetic Resonance
Spectroscopy (MRS)
and Its Application in
Alzheimer’s Disease
PRAVAT K. MANDAL1,2,3
1
Department of Psychiatry, Western Psychiatric Institute and Clinic, University of Pittsburgh Medical School,
Pittsburgh, Pennsylvania
2
Center for Neuroscience, University of Pittsburgh Medical School, Pittsburgh, Pennsylvania
3
Department of Bioengineering, University of Pittsburgh, Pittsburgh, Pennsylvania

ABSTRACT: Magnetic resonance spectroscopy (MRS) is a noninvasive tool to measure
the chemical composition of tissues (in vivo) and characterize functional metabolic proc-
esses in different parts of the human organs. It provides vital biological information at
the molecular level. Combined with magnetic resonance imaging (MRI), an integrated
MRI/MRS examination provides anatomical structure, pathological function, and biochemi-
cal information about a living system. MRS provides a link between the biochemical
alterations and the pathophysiology of disease. This article provides a comprehensive
description of the MRS technique and its application in Alzheimer’s disease (AD)
research. This review is a primer for students and researchers seeking a firm theoretical
understanding of MRS physics as well as its application in clinical AD research. Ó 2007
Wiley Periodicals, Inc. Concepts Magn Reson Part A 30A: 40–64, 2007

KEY WORDS: MRS; MRI; PRESS; STEAM; 2D MRS; Alzheimer’s disease

I. INTRODUCTION selectively excites a small volume of tissue (voxel)
using gradients, then records the free induction
Magnetic resonance spectroscopy (MRS) is a rap- decay (FID) and produces a spectrum from the FID
idly developing field of neuroimaging that allows originating from that voxel. In the 1980s the first
noninvasive in vivo analysis of metabolites. It MR spectrum from living brain was published, and
studies were performed on patients with stroke or
brain tumors (1–3). Over the past two decades,
Received 1 August 2006; revised 12 October 2006;
MRS has been performed on patients with a wide
accepted 12 October 2006
range of neurological and psychiatric disorders so
Correspondence to: Dr. Pravat K. Mandal; E-mail: mandalp@upmc.
edu as to increase the understanding of the pathological
Concepts in Magnetic Resonance Part A, Vol. 30A(1) 40–64 (2007)
mechanisms of these disorders. MRS is also applied
Published online in Wiley InterScience (www.interscience.wiley.
to monitor long-term changes with or without drug
com). DOI 10.1002/cmr.a.20072 therapy and to identify differences between diag-
Ó 2007 Wiley Periodicals, Inc. nostic groups.
40

MRS AND ITS APPLICATION IN ALZHEIMER’S DISEASE 41

Table 1 Nuclei Used for MRS In Vivo
Spin Frequency Inherent Sensitivity Natural
Nucleus Name Number v at B0 ¼ 1.5 Tesla at Const. Field (1H ¼ 1) Abundance (%)
1 1
H Hydrogen (protons) 2 63.87 1 99.985
13 1
C Carbon 2 16.06 0.0159 1.108
19 1
F Fluorine 2 60.08 0.833 100
23 3
Na Sodium 2 16.89 0.0925 100
31 1
P Phosphorus 2 25.85 0.0663 100
35 3
Cl Chlorine 2 6.26 0.0047 75.53
39 3
K Potassium 2 2.98 0.00051 93.08

MRS is a nondestructive technique, which does not implementing them in a hospital environment for
require any ionizing radiation. It provides a wealth of diagnostic purposes (7, 8).
information (in vivo) on various neurometabolites
from a single experiment. It does not require metabo- The Basics of MRS
lite isolation or sample treatment, as required by mass The fundamental basis of MRS is governed by the
spectrometry or other analytical methods. In recent
same principles of nuclear magnetic resonance (NMR)
years, there have been a number of technical advances
(9–21). MRS requires a magnetic field and a radio fre-
concerning both the implementation of different MRS
quency (RF) transmit pulse at a particular resonant fre-
pulse sequences, data processing, and commercial
quency to observe the signal of a specific nuclei (e.g.,
availability of more sophisticated high-field scanners. 1
H, 31P, 13C etc.) in the region of interest (Table 1).
MRS techniques have been developed and applied
The product of MRS is a ‘‘spectrum’’ with a frequency
extensively in brain research (4). The brain has mul-
axis in parts per million (ppm) and a signal amplitude
tiple levels of compartmentation ranging from the
axis (22–28). The signal amplitude (area) is a measure
type of cellular compartment (neuron versus astro-
of a particular metabolite concentration. Specific
cyte) to the type of tissue compartment (the gray
nuclei (e.g., 1H, 31P, 13C, and so on) from the metabo-
matter vs. the white matter) to distinct central nerv-
lite, depending on their characteristic signature, give
ous systems and brain functions. These compart-
ments are highly integrated and work together to rise to either a single peak or multiple peaks that are
attain various brain functions. MRS is useful in uniquely positioned along the frequency axis (X axis),
understanding the neurochemical changes in the known as the chemical shift. The dispersion of chemi-
brain due to different physiological processes. The cal shift (along the X axis) increases with magnetic
extensive numbers of MRS applications have been field strength. The peak amplitude (area) that is
reported exclusively in the brain due to the lack of directly related to the concentration of that assigned
motion artifacts in the brain. In addition, the brain is metabolite is displaced along the Y axis. In vivo 1H-
more or less spherical; hence, it is easier to adjust the MRS and 31P-MRS are the most widely used applica-
high degree of homogeneous magnetic field by shim- tions of MRS, but other atoms that are used for MRS
ming for MRS studies. However, there are suscepti- studies include 13C, 15N, 19F, and 23Na. Major metabo-
bility differences in the brain between the intracellu- lites detected by 1H MRS are as follows:
lar and extracellular space.
The unique applications of MRS in brain research N-acetyl aspartate (NAA) is a neuronal marker
are (1) quantification of oxidative state of the brain seen only in nervous tissue.
and defining neuronal death; (2) accessing and map- Glutamate (Glu) and glutamine (Gln) complex
ping neuronal damage; (3) evaluating membrane is a mixture of peaks that helps to monitor
alteration and characterizing encephalopathies (dis- glutamate metabolism in the brain for chronic
turbances in brain functioning, particularly in intel- epileptic activities.
lectual activity or higher cortical functioning). MR Lactate is a highly specific marker of cell
spectroscopy enables detection of abnormalities in death as well as tissue necrosis.
several neurodegenerative diseases, such as Alzhei- Creatine (Cr) is thought to be a marker of
mer’s disease (AD), and plays an important role in energetic status of cells.
research studies of dementia (5, 6). However, despite Choline (Cho), an indicator of membrane ac-
these advances, there is still a large gap between the tivity, is often elevated in the presence of ma-
MRS techniques development and the challenge of lignant processes.

Concepts in Magnetic Resonance Part A (Bridging Education and Research) DOI 10.1002/cmr.a

42 MANDAL

Figure 1 (A) 1H MRS spectrum (97) and (B) 31P MR spectrum (32) from parietal white matter
at 7 T in normal human brain using STEAM pulse sequence. Inset indicates voxel location.

Myo-inositol (mI), a sugar alcohol, is a marker well as other physiological parameters detected by in
of astrocytic activity and is often higher in vivo 31P MRS have been used extensively in clinical
conditions such as AD and malignant tumors. studies and linked to numerous diseases such as AD
(33), epilepsy (34, 35), migraine, brain ischemia, and
In recent years, there has been more interest in 1H seizure (36). Figure 1 represents a typical 1H and 31P
MRS, particularly after it was demonstrated that it was MRS spectrum of the brain at 7 T magnetic ﬁeld.
possible to obtain high-resolution spectra from small, To enhance the signal-to-noise ratio (SNR) of the
well-deﬁned regions in reasonably short scan times. MR spectrum, the pulse sequence and the parameters
The higher sensitivity of the proton is due to several fac- are adjusted to minimize signal intensity loss due to
tors, including higher gyromagetic ratio, higher metab- T2 (transverse) and T1 (longitudinal) relaxation of the
olite concentrations, and favorable 1H relaxation times. nuclei (e.g., 1H, 31P, and 13C). As mentioned previ-
Although the sensitivity of 31P MRS is less than 1H ously, the MRS technique is applied in conjunction
MRS, 31P MRS provides insights into the biochemistry with MRI, and both techniques share similarities and
not available by 1H MRS (29, 30). 31P MRS detects differences as outlined below.
high-energy metabolites: adenosine triphosphate
(ATP), phosphocreatine (PCr), and inorganic phosphate Similarities with MRI
(Pi). 31P MRS allows noninvasive assessment of vari-
ous fundamental biochemical, physiological, and The same scanner is used for both MRI and
energy intensive metabolic events occurring inside the MRS studies. A schematic diagram of a scan-
brain (31, 32). The steady-state phosphate signals as ner is shown in Fig. 2.



Figure 2 (A) Scanner. (B) The components of a scanner used for MRS and MRI studies.

Both techniques are based on the same physi- are orders of magnitude less concentrated com-
cal principles (i.e., the detection of energy ex- pared to the concentration of hydrogen (from
change between external magnetic fields and water) generally involved in MRI.
specific nuclei within the tissue). MRI provides information on the physical-chem-
Both techniques use a magnetic field instead ical state of tissues, flow diffusion, and motion.
of radiation. Generally, the patient is placed MRS provides chemical composition of tissues
supine (face up) inside the scanner. A special from the particular region of interest.
RF coil is placed around the patient’s head
and MRI/MRS experiments are performed.
Both techniques extensively use gradients for
spatial localization and dephasing the unwanted
magnetization.
MRI and multiple-voxel MRS experiments both
use phase-encoding gradients (Fig. 3).

Differences with MRI

In MRI, the magnetic field is used to create
images based on proton signals from water con-
tent among tissues and organs. MRI images con-
tain anatomical information based on the distri-
bution of protons (from water) as well as the rel-
ative proton relaxation rates in various tissues.
In MRS, magnetic field is used for creating a
graph. This graph consists of various peaks, each Figure 3 (A) Normal magnetic field gradient and (B)
of which represents a specific metabolite in the phase-coding magnetic field gradients that allow the
encoding of the spatial signal location along a second
specific region of interest. The presence or ab-
dimension by different spin phases. Amplitude is kept
sence, as well as increase or decrease in peak area,
fixed in a normal magnetic field gradient. In phase-encod-
provide insight into various neurochemical proc- ing gradient, amplitude is typically varied from a mini-
esses occurring in the tissue. mum value of ÀKPE to maximum value of þKPE in N
MRS is generally less sensitive than MRI because steps, where KPE refers to the amplitude of the phase-
the concentrations of nuclei (1H, 31P, and so on encoding gradient. The spatial resolution is directly
from the neurometabolites) as measured by MRS related to the number of phase-encoding steps.


44 MANDAL

Common uses of MRI include the detection of important exception is 1H-decoupled MRS studies,
AD, stroke versus tumor, recurrent or residual tu- which are particularly RF intensive and may be lim-
mor following therapy versus successfully treated ited by RF heating concerns.
tumor, infection or abscess, and many others.
MRS does not diagnose a given condition, but Gradient Coils
rather provides additional data to aid in diagnosis,
Gradient coils are used to apply gradients to the main
and it must be interpreted along with clinical his-
Bo field in X, Y, and Z directions. The gradient GZ is
tory and other imaging studies, such as MRI.
applied along the long axis of the patient to select a
In MRI, readout gradient (frequency encoding) is
slice (transverse section). This GZ gradient is usually
turned on during data acquisition time. In MRS,
supplied by a pair of Helmholz coils and has a typical
no frequency-encoding gradients are necessary
value of %1 mT mÀ1. The change in Bo from one
during data collection due to inherent chemical
end of the patient to the other will be of the order
shift difference of the nuclei (e.g., 1H, 31P, 13C,
1:1000. The coils for GY and GZ gradients are usu-
etc.) in a given tissue volume.
ally saddle shaped similar to the RF coils. These gra-
In a single-voxel MRS experiment there is no
dients allow the creation of a two-dimensional (2D)
application of phase-encoding gradient, whereas
image of a particular slice. In practice, gradients can
phase encoding gradient is necessary for MRI to
be applied in any desired direction by software con-
record spatial map.
trol of the electronics. The gradients GX, GY, and GZ
are generally switched on and off for a certain length
II. COMPONENTS OF MRS TECHNIQUE of time in the complex pulse sequences of operations
used for MRS studies. The mechanical stress pro-
Some of the integral components of MRS technology duced on the various gradient coils by rapidly chang-
are radio-frequency source, gradients, slice selection, ing magnetic fields in MRS pulse sequences accounts
and phase encoding. for the strange noises often reported by patients
undergoing MRS studies.
Radio Frequency Source
Gradient Methodology
The RF coil is responsible for generating and broad-
casting the RF energy. Specialized coils are used to The flow of electrical current through the gradient
provide improved resolution in the surface regions of coils produces gradient fields. These gradient fields are
the patient. It contains four main components: a fre- applied in short bursts of pulses. The number, duration,
quency synthesizer, a digital envelope of RF frequen- and amplitude of the gradient pulse are determined by
cies, a high power amplifier, and a coil or antenna. the particular pulse sequence and measurement param-
The final component of the RF system is the trans- eters in the protocol. Continuous linear field homoge-
mitter coil. Most MRS systems use a saddle coil to neity is made using gradient offset currents.
produce uniform RF fields over large volumes (e.g., There are four characteristics to describe gradient
body or head). This design is useful to produce uni- system performance: maximum gradient strength;
form RF penetration and to generate an effective B1 duty cycle; rise time and slew rate; and techniques
field perpendicular to Bo even though the coil open- for eddy current compensation. The major complica-
ing is parallel to Bo. Two types of coil polarity are tion of gradient pulses for spectroscopic studies is
used, linearly polarized (LP) and circularly polar- eddy currents. Eddy currents are produced in
ized (CP). response to a changing magnetic field (gradient
RF coils have two categories: volume and surface pulse). Most eddy currents decay with shorter time
coil. Volume coils are typically cylindrical shaped, a constants compared with the time between the end of
popular example being a birdcage coil. Surface coils the gradient pulse and the beginning of data collec-
are subdivided into a single-loop coil or an array coil. tion. Spectroscopic studies are particularly sensitive
Volume coils transmit and receive radio-frequency to eddy currents. In some instances, additional post-
pulses and are called ‘‘trans-receivers.’’ Surface coils acquisition corrections are necessary to obtain well-
generally receive signals only and are traditionally resolved resonances.
used to improve signal-to-noise ratio. Unlike MRI Larger gradient strength allows for better spatial
studies, most spectroscopic measurements deposit lit- resolution. The duty cycle of the gradient amplifier is
tle RF power to the patient, and specific absorption another important measure of gradient performance.
rate (SAR) limitations are infrequent in MRS due to The duty cycle determines how fast an amplifier can
long TR (repetition time) used in MR protocol. One respond to the demands of a pulse sequence. Large



tance from the gradient isocenter. gH is the nuclear
gyromagnetic ratio of proton.

Removal of Unwanted Magnetization Due to an
Imperfect 1808 Pulse. This application is critical to
remove unwanted magnetization due to an imperfect
1808 pulse and it can be explained by analyzing the
Figure 4 Application of pulsed field gradient to rephase
magnetization at different points in Fig. 4.
transverse magnetization by 1808 pulse and elimination of 90 1 þ
ða ! bÞ IZ À! ÀIY ¼ À ½I À IÀ Š
x
unwanted magnetization due to imperfect 1808. The gra- 2i
dients are placed symmetrically from the 1808 pulse. Mag- g1 GrtIz 1 Â þ Àig1 GrtIz Ã
netization at various points (a–d) are explained in the text. À À
! I e À IÀ eþig1 GrtIz
2i
ð180 þ yÞIx
duty cycles allow high-amplitude gradient pulses ðb ! cÞ !
between very short interpulse delays. imperfect pulse
1 Â Ã
À Cosy Iþ eÀig1 GrtIz À IÀ eþig1 GrtIz
Selection of Magnetization by Gradients. Gradients 2i
1 Â Ã
are used extensively for two purposes, either rephas- À Iz eÀig1 GrtIz À eþig1 GrtIz
ing (selection) (37) or dephasing (elimination) (38) 2
of a particular magnetization transfer pathway (21). Àg1 GrtIz 1 Â þ Àig GrtI þig GrtI
ðc ! dÞ À ! À Cosy I e 1 z e 1 z
Whenever gradients are applied in a particular direc- 2i
À Àig1 GrtIz þig1 GrtIz
Ã 1 1
tion (for simplicity’s sake, it is assumed here that ÀI e e À Cosy ½Iþ À IÀ Š
gradients are applied along the Z direction), it gener- 2i 2i
ates a phase factor associated with the coherence 1 Â Àig1 GrtIz þig1 GrtIz
Ã 1
À Iz e Àe ¼ À ½Iþ À IÀ Š
level. It is convenient to re-express the Cartesian 2 2i
operators IX and IY in terms of raising and lowering 1 Â Àig1 GrtIz Ã
À Iz e À eþig1 GrtIz [2]
operators Iþ and IÀ, respectively, to describe the 2
effects of field gradients to rephase transverse mag- Any magnetization associated with a phase factor
netization and removal of artifacts generated due to experiences different gradient strength, and the over-
imperfect 1808 pulse (38). all integrals become zero. Hence, at the end of 2t,
longitudinal magnetization (Iz) associated with the
Rephasing of Transverse Magnetization. If two gra- phase factor will be dephased.
dients with the same strength, shape, duration, and po-
larity are applied on either side of a 1808 pulse, trans-
Spatial Encoding and Slice Selection
verse magnetization is refocused (Fig. 4). Details of
operator formalism are provided in the appendix (21). In MRS, quantification of metabolites from a particular
The rephasing of magnetization is an important appli- region of the body is the primary objective, and the
cation of gradients and it can be explained by analyz- selection of the specific region of the body is accom-
ing the magnetization at different points in Fig. 4. plished with the help of slice-selecting gradients,
90
known as spatial encoding. Slice selection is achieved
1 þ À
ða ! bÞ IZ À! ÀIY ¼ À ½I À I Š
x
by applying a one-dimensional, constant magnetic field
2i
gradient. At the same time, a selective 908 pulse is
g1 GrtIz 1 Â þ Àig1 GrtIz Ã
À ! À I e À IÀ eþig1 GrtIz applied. Application of this selective 908 pulse in con-
2i
junction with a magnetic field gradient will rotate spins
180 Ix 1 Â À Àig1 GrtIz Ã
ðb ! cÞ À À! I e À Iþ eþig1 GrtIz that are located in a slice or a plane through the object.
2i Figure 5 illustrates the slice selection using appli-
g1 GrtIz 1 Â À Àig1 GrtIz þig1 GrtIz
ðc ! dÞ À À! I e e cation of a selective 908 pulse in the presence of field
2i gradient GZ. The selective 908 pulse excites only a
Ã 1
À Iþ eÀig1 GrtIz eþig1 GrtIz ¼ þ ½Iþ À IÀ Š [1] narrow frequency range (Do), and this narrow tissue
2i slice in the Z direction (DZ) is sampled for analysis
The net phase acquired after 2t is zero and we get as indicated by the shaded area in Fig. 5.
back the same transverse magnetization (Iþ À IÀ) that The magnitude of the slice select gradient deter-
we started with, where G is the gradient strength, t is mines the difference in precession frequency between
the duration of gradient application, and r is the dis- the two points of the gradient. Steep gradient slopes


46 MANDAL

Figure 5 Selection of a slice using gradient.

generate a large difference in precession frequency
between two points of the gradients, whereas shallow
gradient slopes generate a small difference in preces-
sion frequency between the same two points (Fig. 6).
Once a certain gradient slope is applied, then the RF Figure 6 Selection of slice thickness with steepness of
pulse is transmitted to excite the slice that contains a gradients. (A) Shallow gradient and (B) steep gradient.
range of frequencies between the two points. This
frequency range is called bandwidth, and the RF decay (FID) with time contains all the necessary in-
being transmitted at this point is called the transmit formation for reconstructing the signal as a function
bandwidth. Briefly, to achieve a thick slice, a shallow of a frequency. This is accomplished with the help of
slice select gradient and/or a broad transmit band- Fourier transform (Fig. 8A).
width is applied (see Fig. 6A). To achieve a thin Suppose we apply a 908 pulse after a field gradient
slice, a steep slice select gradient and/or narrow along the X direction, GX. This has the effect of
transmit bandwidth is applied (see Fig. 6B). ‘‘labeling’’ the spins and separates them according to
For example, in a 1.5 T magnet, water protons have distance along the X axis from the isocenter (see Fig.
a resonant frequency of approximately 64 MHz. For a 8B). The resonant frequency at some point ‘‘X’’
908 pulse with a frequency width of 1.0 kHz, the mag- along the linear field gradient relative to some refer-
netic field gradient required to selectively excite a slice ence point can be written as
of tissue of 5 mm thick is calculated as follows:
o oX ¼ g X GX : [5]
Do ¼ gDBZ ¼ gGZ DZ; and g¼ ; [3]
Bz
The equation for the FID then becomes
Hence,
Z
Do Do BZ 1:0 1:5 fðtÞ ¼ rðxÞ expðÀi g GX xtÞdx: [6]
Gz ¼ ¼ ¼
gDZ o Dz 64 Â 1000 5 Â 10À3 X

1:5
¼ ¼ 0:00468TmÀ1 : [4] The Fourier transform of the FID converts from a fre-
320
quency profile of signal intensity F(o) to a spatial
Figure 7A shows the orientation of different slices in profile of signal intensity or spin density r(x).
a human brain. In general, we assume that the slice- The frequency-encoding gradient is activated dur-
selective gradient is applied along the Z direction and ing signal acquisition and is often called the readout
it generates an axial image (see Fig. 7B). However, gradient. The echo is usually centered in the middle
for the sagittal and coronal images, GX and GY gra- of the frequency-encoding gradient, so that the gradi-
dients are selected for slice selection gradients, ent is switched on during the rephasing and the
respectively (see Fig. 7B). dephasing part of the echo. The steepness of the
slope of the frequency-encoding gradient determines
Frequency Encoding. In a uniform Bo field, after the size of the anatomy covered along the frequency-
the application of a 908 pulse, the free induction encoding axis during a scan.



Figure 7 (A) The use of different physical gradients for selecting slice in the brain. (B) Three
images (axial, sagittal, and coronal) are generated due to different slice selection (GZ, GX, and
GY) gradients.

The resolution of the image along the X axis field gradient along a particular dimension prior to the
depends on a number of points used for sampling acquisition of the signal is called phase encoding. If a
(typically 256 points in a field of view of 20–40 cm). gradient field is briefly switched on and then switched
In MRS, nuclei (1H, 31P, etc.) precess in different fre- off with predefined altered amplitude before acquisi-
quencies depending on chemical environments and tion of data, the magnetization of the external voxels
this is why the application of frequency encoding will either precess faster or slower relative to the am-
gradients is not necessary in MRS. This is a major plitude of the phase encoding gradient (Fig. 9). The
difference between MRI and MRS experiments. steepness of the slope of the phase-encoding gradient
determines the degree of phase shift between two
Phase Encoding. The process of locating an MR sig- points along the gradient axis. A steep phase-encoding
nal by altering the phase of spins using a magnetic gradient causes a large phase shift between two points
along the gradient, whereas a shallow phase-encoding
gradient causes a smaller phase shift between the same
two points along the gradient. Some essential concepts
of spatial encoding are

The phase-encoding gradient alters the phase
along the remaining axis of the image, which
is usually the short axis of the anatomy.
In coronal images, the short axis of the anat-
omy usually lies along the horizontal axis of
the magnet, and therefore the X gradient per-
forms the phase encoding.
In sagittal images, the short axis of the anat-
omy usually lies along the vertical axis of the
magnet, and therefore the Y gradient performs
Figure 8 (A) Conversion of free induction decay to the phase encoding.
spectra using Fourier transformation. (B) Pictorial repre- In axial images, the short axis of the anatomy
sentation of the effective magnetic field experiencing dif- usually lies along the vertical axis of the mag-
ferent spins depending on the location. net, and therefore the Y gradient performs the


48 MANDAL

Figure 9 Representation of spatial, phase, and frequency
encoding in a typical MRI sequence. The frequency-encod- Figure 11 Pulse sequences for (A) spin echo and (B)
ing gradients are not applied in MRS pulse sequences. gradient echo. Magnetization at different points is des-
cribed in the text.
phase encoding. However, when imaging the
head, the short axis of the anatomy lies along In the frequency domain (X axis) (Fig. 10), there
the horizontal axis of the magnet, and therefore is a change in frequency depending on the location of
the X gradient performs the phase encoding. the voxel. In Figure 10, we have assumed that our
area of interest is subdivided into 5 Â 5 matrices.
A field gradient is applied along the Y direction, The amplitude of the phase gradient varies systemati-
GY as a short pulse, after the RF pulse, but before the cally; as the amplitude varies, the phase of the spin
main acquisition time of the FID. This has the effect varies differently as depicted along the phase shift
of progressively phase shifting the precessing spins axis. Frequency encoding is applied in MRS, and it is
along the Y direction, but without changing the fre- provided by the inherent chemical shift differences
quency. This labels the spins in a different way, but of different spins.
to disentangle all the information, the computing pro-
cedure used to reconstruct the second dimension in Formation of an Image
the 2D slice requires that the operation be performed After getting an image of the brain, a voxel is chosen
in many steps (typically 256) where the phase encod- in the region of interest for MRS analysis. Generally,
ing amplitude is varied incrementally. This particular two MRI sequences (e.g., a spin echo or a gradient
process determines the resolution in the Y direction, echo) are applied for generating an MRI image. A
and largely accounts for the long time needed for the detailed discussion about these two sequences is
whole imaging experiment. given below.

Spin Echo (SE). A spin echo uses a 908 RF pulse
along with a slice-selective gradient (Fig. 11A). This
selective 908 excitation pulse flips the magnetization
within the slice to the transverse plane and magnet-
ization is dephased by the first gradient. A 1808 pulse
is applied at the middle of the sequence and the mag-
netization is rephased by the second gradient. The
amplitude of the spin echo is affected by T2 relaxa-
tion; the resulting images are T2 weighted. The
degree of T2 weighting is determined by the value of
TE, which may vary from few milliseconds to hun-
dreds of milliseconds.
SE sequence employs large flip angles, so it
Figure 10 A graphical representation of the frequency requires long recovery time (TR) to allow adequate
(X axis) and phase encoding (Y axis). The amplitude of recovery of longitudinal magnetization. Typically,
the phase-encoding gradient changes sequentially. TR values range from hundreds of milliseconds to



seconds. As total scan time is dependent on TR; a SE not refocused in GE, and this is the most important dif-
sequence can be lengthy. Spin echo is the least arti- ference with a SE sequence.
fact-prone sequence and generates a high signal-to- In GE (see Fig. 11B), magnetization at different
noise ratio. However, SAR is higher in SE due to points are explained as follows:
both 908 and 1808 RF pulses. Long TR in SE
90o 1 þ
ða ! bÞ IZ À ÀIY ¼ À
! ½I À IÀ Š
x
sequence times is incompatible with 3D acquisitions.
In SE (see Fig. 11A), magnetization at different 2i
Gx t 1 Â þ ÀiGx t Ã
points are explained as follows: À À
! I e À IÀ eþiGx t
2i
ÀGx t 1 Â þ ÀiGx t þiGx t
90 1 À À! I e e
À IX ¼ ½Iþ þ IÀ Š
y
ða ! bÞ Iz ! 2i
2 Ã 1
Gx t 1 Â þ ÀiGx t Ã ÀIÀ eþiGx t eÀiGx t ¼ À ½Iþ À IÀ Š
À ! I e À IÀ eþiGx t 2i
2
Gx t 1 Â þ ÀiGx t Ã
ÀGx t 1 Â þ ÀiGx t þiGx t Ã ðb ! cÞ À À! I e À IÀ eþiGx t
À ! I e e À IÀ eþiGx t eÀiGx t 2i
2
ÀGx t 1 Â þ ÀiGx t þiGx t Ã
1 À À
! I e e À IÀ eþiGx t eÀiGx t
¼ ½Iþ À IÀ Š 2i
2
1
þGx t 1 Â þ ÀiGx t À þiGx t
Ã ¼ À ½Iþ À IÀ Š ¼ Gradient echo: [8]
ðb ! cÞ À ! I e! ÀI e 2i
2
180 1 Â À ÀiG t Ã
ðc ! dÞ À! I e x À Iþ eþiGx t Gradient and spin-echo-generated images are usually
x

2 modulated by relaxation properties (T1 and T2) of 1H
Gx t 1 Â À þiGx t ÀiGx t Ã
À ! I e e þ Iþ eþiGx t eÀiGx t nuclei in the region of interest. Images are categorized
2 into two types (e.g., T1 weighted or T2 weighted).
1
¼ ½Iþ þ IÀ Š ¼ Echo [7]
2 T1-Weighted Images. T1-weighted images are pro-
duced using either the spin SE or the GE sequences.
Gradient Echo (GE). A GE pulse sequence (see For T1-weighted images, short TR and short TE are
Fig. 11B) uses a variable RF excitation pulse (gener- used to enhance the T1 differences between tissues.
ally less than 908). Hence, the magnitude of trans- T1-weighted images have excellent contrasts (e.g.,
verse magnetization is less than spin echo, where all fluids are very dark, unless they are fast moving,
the longitudinal magnetizations are converted to a water-based tissues are midgrey, and fat-based tis-
transverse plane. After the RF pulse is applied, the sues are usually very bright). T1-weighted images are
magnetization in the transverse plane is dephased by often known as anatomy scans (Fig. 12A).
the gradient and then rephased by the second gradi-
ent. The term TE is the interval between RF excita- T2-Weighted Images. T2-weighted images are pro-
tion and the center of the gradient echo. The value of duced by SE or GE sequences, but GE images are
TE is important in determining the signal contrast of affected by the magnetic field inhomogeneity. SE T2
the image. As the transverse magnetization is subject images require a long TR, a long TE, and take longer to
to T2 dephasing, regions of tissue whose T2 value is acquire than T1-weighted images (scan time depends
short compared with TE will exhibit greatly attenu- directly on the TR). In these scans, fluids have the high-
ated signals. By contrast, regions with longer T2 will est intensity, whereas water- and fat-based tissues are
have somewhat higher signals. mid-gray. T2 images are often thought of as pathology
GE sequence often employs very short TR values, scans because collections of abnormal fluid are bright
and images exhibit T1 weighting. Tissue with short T1 against the darker normal tissues (see Fig. 12B).
appears brighter because their longitudinal magnetiza-
tion is less easily saturated. The degree of T1 weight- III. TECHNICAL ISSUES
ing also increases with flip angle, because higher flip
angles cause greater saturation. The flip angle typically MRS is performed as an adjunct to MRI. An MRI
used in GE is within the range of 208–458. GE can pro- image is first generated, the voxel is selected at the
vide faster imaging using shorter TR and shorter TEs site of interest, and then MRS spectra recorded from
than spin echo. In GE, less energy deposit occurs in that voxel. The use of spatial localization is essential
the body due to use of low flip angle (908). In GE, for in vivo MRS for selection of single voxel from a
more slices per TR are generated than SE. GE is more particular region of interest. Multiple-voxel tech-
compatible with 3D acquisitions. Chemical shifts are niques, popularly known as chemical shift imaging,


50 MANDAL

Figure 12 Conventional spin echo (A) T1 and (B) T2-weighted images of brain.

allow simultaneous acquisition of in vivo MR spectra 90 1 þ
½I þ IÀ Š
y
ða ! bÞ IZ À IX ¼
!
from several voxels in one experiment. 2
Gx t1 Â þ ÀiGx t Ã
À! I e þ IÀ eþiGx t
Spatial Localization Based on 2
Single-Voxel Technique ÀGx t 1 Â þ ÀiGx t þiGx t
À ! I e e
2
The most frequently used localization methods for Ã 1
1
H MRS of the brain are PRESS (point-resolved þIÀ eþiGx t eÀiGx t ¼ ½Iþ þ IÀ Š
2
spectroscopy) (2) and STEAM (stimulated echo ac- 1 þ Chem: shift
quisition mode) (39). The basic principle underlying ðb ! cÞ ½I þ IÀ Š !
2 TE=2
1 h þ iOH TE=2 i
single-voxel technique is to use three mutually or-
thogonal slice selective pulses and design the pulse I e þ IÀ eÀiOH TE=2
sequence to collect only the echo signal from the 2
h
Gx t 1 þ ÀiGx t iOH TE=2
i
point (voxel) in space where all three slices intersect ðc ! dÞ À ! I e e þ IÀ eþiGx t eÀiOH TE=2
(Fig. 13). PRESS pulse sequence (Fig. 14) creates a 2
1 h À ÀiGx t iOH TE=2 i
double spin echo from the pulse sequence. I e e þ Iþ eþiGx t eÀiOH TE=2
2
90 ! TE=2 ! 180 ! TE=2 ! ½Echo 1Š ! TE=2 h
Gx t 1 À ÀiGx t þiGx t iOH TE=2
À ! I e e e
! 180 ! TE=2 ! ‘‘½Echo 2Š’’ [9] 2 i
þ Iþ eþiGx t eÀiGx t eÀiOH TE=2
where TE is the echo time. The magnetization at dif-
ferent points for the PRESS sequence (see Fig. 14) is 1h i
¼ IÀ eiOH TE=2 þ Iþ eÀiOH TE=2
shown in Eq. [10] as follows: 2
Chem: shift 1 h À iOH TE=2 ÀiOH TE=2
ðd ! eÞ ! I e e
TE=2 2
i
þ Iþ eÀiOH TE=2 eþiOH TE=2
1 À
¼ ½I þ Iþ Š ! Echo 1
2
Chem: shift 1 h þ iOH TE=2 i
ðe ! fÞ I e þ IÀ eÀiOH TE=2
TE=2 2
1 h À iOH TE=2 i
ðf ! gÞ I e þ Iþ eÀiOH TE=2
2
1
ðg ! hÞ ½IÀ þ Iþ Š ! Echo 2: [10]
Figure 13 A schematic illustration of selecting a voxel 2
by three orthogonal slice-selecting pulse used in STEAM
or PRESS pulse sequences. The size and position of the
voxel is controlled by the frequency and bandwidth of the
slice-selecting pulses, as well as the amplitude of the Hence, at the time of data acquisition, we get back
associated slice-selecting gradients. the same magnetization as we started with at point b.



Figure 14 A schematic diagram of PRESS pulse sequence. Magnetization at different points is
described in the text. Chemical shift selective imaging (CHESS) pulses at the beginning of the
pulse sequence are used to suppress water peak (98). [ ]3 symbol at the bottom indicates fre-
quency-selective 908 pulse to selectively excite the water, followed by application of spoiler gra-
dient (repeated thrice) to dephase the resulting magnetization.

Thus, sensitivity is not lost in a PRESS sequence. Gx t 1 h þ iOH TE=2 i
À À
! I e À IÀ eÀiOH TE=2
However, PRESS is a longer pulse sequence. 2i '
STEAM pulse sequence selects a stimulated echo 90 1 iOH TE=2 1 8 þ 9
:I þ IÀ ; þ iIZ
¼À À !
x
e
from the pulse sequence (Fig. 15). The general pulse 2i 2
sequence scheme for STEAM is given below. 8 9 '!
1: þ
ÀeÀiOH TE=2 I þ IÀ ; þ iIZ
2
90 ! TE=2 ! 90 ! TM ! 90 ! TE=2
Gx t1
1 h
1 iOH TE=2
! ‘‘½EchoŠ’’ [11] ðd ! eÞ À e fþiIZ g
Spoiler 2i
i
where TM is the mixing time. The magnetization at ÀeÀiOH TE=2 fÀiIZ g
1 1 h iOH TE=2 n8 À 9o
different points originated from STEAM sequence
90
(see Fig. 15) are given in Eq. [12] as follows: ðe ! fÞ À À
!
x
e :I À Iþ ;
2i 2 n8 9oi
ÀiOH TE=2 : À
þe I À Iþ ;
90
1 1 h iOH TE=2 n8 À 9o
1 þ Gx t
ða ! bÞ IZ À ¼ IY : À
! ½I À IÀ Š À ! GX t
x

2i ¼À e :I À Iþ ;
1 Â þ ÀiGx t Ã ÀGX t 2i 2
À I e À IÀ eþiGx t n8 9oi
2i þ eÀiOH TE=2 :IÀ À Iþ ;
ÀGx t 1 Â þ ÀiGx t þiGx t Ã
À À
! I e e À IÀ eþiGx t eÀiGx t Chem: shift 11 þ
2i ðf ! gÞ !þ ½I À IÀ Š
1 TE=2 2i 2
¼ À ½Iþ À IÀ Š 1 1 Â À ÀiOH TE Ã 11 þ
2i À I e À Iþ eÀiOH TE ¼ ½I À IÀ Š
Chem: shift 1 h þ iOH TE=2 2i 2 2i 2
ðb ! cÞ !À I e ¼ Reduced signal intensity: [12]
TE=2 2i
i
ÀIÀ eÀiOH TE=2
ÀGx t 1 h þ þiGx t iOH TE=2 Hence, at the time of data acquisition, we get half
ðc ! dÞ À À ! I e e the magnetization that we started with at point b.
2i i
Thus, in a STEAM sequence, sensitivity is reduced
À IÀ eÀiGx t eÀiOH TE=2
by half. Figure 16 shows the 1H MRS data acquired


52 MANDAL

Figure 15 A schematic diagram of STEAM pulse
sequence. Magnetization at different points is described in
the text. Chemical shift selective imaging (CHESS) pulses
at the beginning of the pulse sequence are used to sup-
press water peak (98). [ ]3 symbol at the bottom indicates
frequency-selective 908 pulse to selectively excite the
water, followed by application of spoiler gradient
Figure 16 A comparison of the brain spectra at 3T using
(repeated thrice) to dephase the resulting magnetization.
(A) PRESS and (B) STEAM pulse sequences. The in vivo
acquisition spectra shown are from the occipital region of
a 22-year-old healthy male. For all acquisitions, the band-
in the same region by PRESS and STEAM sequences width was 2.5 kHz with the collection of 2,048 data
at 3T magnetic field (40). points. A line width broadening function of $6 Hz was
applied to simulate the in vivo line width (99).
Similarities between PRESS and STEAM
sensitive to T2-relaxation throughout the local-
Both of these pulse sequences involve sequen- ization sequence. STEAM has two echo inter-
tial application of three orthogonal gradients to vals; PRESS has four echo intervals.
select slices, during which selective RF pulses With the same hardware, shorter TEs can be
are used to excite the spins in each slice. Hence, achieved with STEAM than with PRESS.
at the end of the three-slice series, the only spins STEAM may have slightly better water sup-
excited are within the chosen volumes. pression factor, because water suppression
Both PRESS and STEAM can be applied pulses can be added during the TM period
along with the phase-encoding gradients, (which does not occur in PRESS). In addition,
which allow the defined volumes to be subdi- STEAM may have less spurious water signals
vided. This yields a signal acquisition from from the 908 slice selective pulses than the
multiple volumes simultaneously. Because the 1808 pulses in PRESS.
metabolite distribution can be represented as Another factor to consider, especially at higher
maps, this approach is known as magnetic res- field strengths, is that the amount of power de-
onance spectroscopic imaging (MRSI) or posited (i.e., SAR) is approximately twice as
chemical shift imaging (CSI) (24, 41–43). high for PRESS compared with STEAM. SAR
is not a significant factor at low fields (e.g.,
1.5 T). The Federal Drug Administration has
Differences between PRESS and STEAM
approved higher fields (up to 3.0 T) for clini-
cal use. At present, 7.0 T scanners are being
In a PRESS sequence, sensitivity is higher by a
used exclusively for research purposes.
factor of two than a STEAM sequence, given the
same echo time. This is because the stimulated
Chemical Shift Imaging (CSI)
echo is formed from only half the available equi-
librium magnetization. Chemical shift imaging (CSI) or magnetic resonance
The STEAM sequence is less sensitive to T2- spectroscopic imaging (MRSI) is an efficient tech-
relaxation effects as no T2 relaxation occurs nique for noninvasive characterization and quantifi-
during the mixing time, whereas PRESS is cation of metabolites from simultaneous acquisition



Figure 17 A pictorial representation of the scheme involving one-dimensional (1D) and chemi-
cal shift imaging (CSI) MRS study. Initial steps are to place the subject into the scanner, adjust-
ing the power level and shimming and recording MRI images. Voxel is then selected in the
desired location of the brain. Depending on the nature of the study, desired MRS pulse sequen-
ces are used for one-dimensional (1D) single-voxel MRS (A) or multivoxel CSI (B) spectra of
the brain.

of spectra from multiple voxels. Phase encoding can One advantage to this technique is that there is no
be used either with a simple FID acquisition or in chemical shift artifact problem as seen in single- or
combination with volume selective methods, such as multiple-voxel localization techniques. Therefore, it
PRESS or STEAM. Because frequency encoding is is useful for high-ﬁeld in vivo MRS applications in
not possible in spectroscopy, phase encoding must be which the chemical shift dispersion is linearly
applied in each of the directions in which spatial in- increased as a function of B0.
formation is required—namely, two directions for 2D A major technical problem is the difﬁculty of
chemical shift imaging or three directions for 3D shimming an entire slice to the level necessary for
CSI. Scan time is dependent on the number of voxels good spectra from every voxel in the matrix. When
(N) in a particular direction. Given that, each phase- setting up a CSI scan, the edges of the region of in-
encoding step requires a separate TR period, and the terest should lie within the skull to avoid the suscep-
scan time increases as N2 for 2D or N3 for 3D CSI. tibility changes associated with the bone.
Because the scan time increases so rapidly with N,
this imposes a further constraint on spatial resolution. Comparison of Single-Voxel versus Multiple-Voxel
In multivoxel 1H MRS, typical in-plane resolutions Techniques. Usually, but not exclusively, in a sin-
are in the order of 1 to 2 cm. In 31P MRS, even lower gle-voxel spectroscopy (SVS) technique, scans are
resolutions are used because of the lower sensitivity recorded at short TEs (35 ms), whereas multiple
of the 31P nucleus. voxel techniques, such as MRSI studies, are per-
The information collected in a multivoxel acquisi- formed at long TEs (e.g., TE 135 ms). In SVS,
tion can be presented as an array of spectra (Fig. spectra contain signals from more compounds and
17B). The metabolite maps can be displayed in color have better SNRs, but also have worse water and lipid
and overlaid on an MR image of the same slice. contamination. In MRSI, spectra have lower SNR,


54 MANDAL

Figure 18 A schematic diagram of a two-dimensional L-COSY pulse sequence. The RF pulse
scheme consisted of three RF pulses (908, 1808, 908) that were slice-selective along three orthog-
onal axes. A pair of B0 gradient crusher pulses were symmetric with respect to the slice-refocus-
ing 1808 RF pulse. The last slice-selective 908 RF pulse with a pair of symmetric B0 gradient
crushers also served as a coherence transfer pulse for the L-COSY spectrum. The coherence
transfer pathway diagram depicts the different stages of conversion of magnetization/coherences.
Chemical shift-selective imaging (CHESS) pulses at the beginning of the pulse sequence are
used to suppress water peak (98). [ ]3 symbol at the bottom indicates frequency-selective 908
pulse to selectively excite the water, followed by application of spoiler gradient (repeated thrice)
to dephase the resulting magnetization.

fewer detectable compounds, and variable amounts of 1.5 T and 3 T MR imaging scanners (48, 49). Due to
T2 weighting but are usually better-resolved spectra an added dimension, a localized 2D MR spectrum
with flatter baselines. SVS takes less time than MRSI has better resolution (Fig. 19) than a conventional 1D
studies. Resolution in SVS is higher than MRSI tech- MR spectrum (48, 49).
niques. Quantification of metabolites is more robust A 2D L-COSY sequence is operated on a single
in SVS compared with MRSI techniques. voxel. Two major problems (49) yet to be resolved in
the localized 2D MR spectroscopy are (1) minimiz-
ing the RF pulses used for localization and coherence
Two-Dimensional Technique
transfer, taking into consideration that some of the
A major concern with one-dimensional MRS is that brain metabolites have short T2; and (2) recording
many peaks overlap, and precise quantification is not the localized 2D spectra of human organs in a rea-
possible. In particular, the dominant peaks of gluta- sonable time duration.
thione (GSH) overlap with other metabolites. Spec- Prior to localization by the 2D L-COSY sequence,
tral editing and multiple-quantum (MQ) techniques a CHESS sequence consisting of three frequency-
can be used to differentiate glutathione and Glx selective water-suppression pulses with a bandwidth
metabolites from overlapping lipids signals (44–47). of approximately 75 Hz was used, each followed by
A drawback of the spectral editing technique is that the dephasing Bo gradient pulses.
only one metabolite can be selectively detected (44, The 2D L-COSY pulse sequence had a combination
45). Reduced signal strength of metabolites is a of three slice-selective RF pulses (908–1808–908) to
major concern with MQ techniques (46, 47). Several localize a desired voxel. The desired coherence trans-
versions of localized 2D MRS sequences (Fig. 18) fer pathways selected by a pair of gradient pulses are also
have been successfully implemented on whole-body shown along with the pulse sequence (see Fig. 18).



The magnetization at different points of a 2D-L-
COSY sequence is given in Eq. 13 as follows:

90 i Gy t
ða!bÞ IZ À ÀIY ¼ À ½IÀ ÀIþ ŠÀ
! !
x

2
iÂ Ã
À IÀ eþiGy t À Iþ eÀiGy t
2
ÀGy t iÂ Ã
À À Iþ eþiGy t eÀiGx t À Iþ eÀiGy t eþiGy t
!
2
i
¼ À ½IÀ ÀIþ Š
2
Chem: shift i h À ÀiOH TE=2 þ þiOH TE=2 i
ðb!eÞ !À I e ÀI e
TE=2 2
Gz t ih i Figure 19 2D L-COSY MR spectrum of a 27-year-old
ðc!dÞ À À IÀ eþiGz t eÀiOH TE=2 ÀIþ eÀiGz t eþiOH TE=2
! healthy control in the occipito-parietal gray matter region
2
180 i h þ þiGz t ÀiOH TE=2 À ÀiGz t þiOH TE=2 i at 1.5 T scanner. The 2D raw data were zero-filled to 256
À À I e
! ÀI e
x
e e and 2,048 along F1 and F2 axes and displayed in the
2
Gz t i h magnitude mode (100).
À À Iþ eþiGz t eÀiGz t eÀiOH TE=2
!
2 i The application of 2D L-COSY on a normal brain
À IÀ eÀiGz t eþiGz t eþiOH TE=2 correlating different metabolites is shown in Fig. 19.
ih i This 2D MRS technique has a great potential for
¼ À IÀ eÀiOH TE=2 À Iþ eþiOH TE=2 application to neurodegenerative diseases (i.e., AD)
2
Chem: shift ih for quantification of neurometabolites, particularly
ðd!eÞ ! À IÀ eÀiOH TE=2 eþiOH TE=2 the major antioxidant GSH that cannot be definitively
TE=2 2
i i quantified by one-dimensional MRS technique.
À Iþ eþiOH TE=2 eÀiOH TE=2 ¼ À ½IÀ ÀIþ Š
2
Chem: shift iÂ Ã
ðe!fÞ ¼ À IÀ eþiOH t1 À Iþ eÀiOH t1 IV. ALZHEIMER’S DISEASE
t1 2
Spin À Spincoupling
Alzheimer’s disease (AD), which accounts for
t1 around 70% of dementia, is a progressive neurodege-
i Â À iOH t1 Ã
À I e fcosðpJIS t1 Þ þ 2iSZ sinðpJIS t1 Þg nerative disease manifested by cognitive deteriora-
2 tion, progressive impairment of activities of daily liv-
iÂ Ã
þ Iþ eÀiOH t1 fcosðpJIS t1 ÞÀ2iSZ sinðpJIS t1 Þg ing (ADL), and a variety of neuropsychiatric symp-
2 toms and behavioral disturbances (50, 51). In normal
ÀGz t iÂ
ðf!gÞÀ À IÀ eiOH t1 eÀiGz t fcosðpJIS t1 Þ
! aging, nerve cells (neurons) in the brain are not lost
2 in large numbers. In AD, however, many nerve cells
þ 2iSZ sinðpJIS t1 ÞgŠ stop functioning, lose connections with other nerve
iÂ cells, and die. At first, AD destroys neurons in parts
þ Iþ eÀiOH t1 eþiGz t fcosðpJIS t1 Þ À 2iSZ sinðpJIS t1 ÞgŠ of the brain that control memory, including the hip-
2
þGz t iÂ pocampus (a structure deep in the brain that helps
À À IÀ eiOH t1 eÀiGz t eþiGz t fcosðpJIS t1 Þ
! encode short-term memories) and related structures
2
þ 2iSZ sinðpJIS t1 ÞgŠ (52). As nerve cells in the hippocampus stop working
properly, short-term memory fails and a person’s
iÂ
þ Iþ eÀiOH t1 eþiGz t eÀiGz t fcosðpJIS t1 ability to do easy and familiar tasks often begins to
2 decline. AD later attacks the cerebral cortex (the
þ 2iSZ sinðpJIS t1 ÞgŠ outer layer of neurons in the brain), particularly the
iÂ Ã areas responsible for language and reasoning (53). At
¼ À IÀ eiOH t1 fcosðpJIS t1 Þ þ 2iSZ sinðpJIS t1 Þg
2 this point, AD begins to take away language skills
i Â þ ÀiOH t1 Ã and changes a person’s ability to make rational judg-
þ I e fcosðpJIS t1 Þ À 2iSZ sinðpJIS t1 Þg
2 ments (54). Psychotic symptoms develop in some
¼ Data acquisition ðduring t2 Þ: [13] patients, such as depression, hallucinations, and delu-


56 MANDAL

Figure 20 The clinical pathway (normal ? pre-MCI ? MCI ? AD) of AD progression.
Postulated sequence of spread of neurofibrillary pathology in AD, showing the medial aspect of
the cerebral cortex (101). The depth of the darkness in the brain is in proportion to the density
of tangles (102).

sions (55). Eventually other parts of the brain are compared the diagnostic or prognostic performance
involved, thereby making the AD brain unresponsive. of MRS with that of established imaging techniques.
The fundamental molecular etiology, which leads to MRS has the potential to be used in research studies
neuronal loss resulting in cognitive decline in AD, is to monitor the efficacy of drug therapy in AD by
unknown. However, there are existing data to support measuring alteration of important neurochemicals
amyloid (56), tau (57), oxidative stress (58), soluble with time. These neurochemicals are associated with
oligomeric Ab (59, 60), inflammatory cascade (61, two important biophysical processes (i.e., energy me-
62), and cholinergic neuronal loss (63) hypotheses in tabolism and lipid metabolism). A brief discussion of
AD. It is not yet known which molecular event ini- these biochemical processes follows.
tiates the pathocascade of AD (Fig. 20).
Investigators are continuing to use neuroimaging Energy Metabolism. Brain energy or oxidative me-
techniques to assess whether it is possible to measure tabolism (64) is characterized by (i) high levels of
brain neurochemicals to identify people who are at phosphocreatine (PCr) and creatine (Cr); (ii) high
risk of AD even before they develop the symptoms levels ATP production; (iii) high activity of creatine
of the disease. Over the past few years, research has kinase (CK); and (iv) high steady-state mitochondrial
expanded our understanding of the potential useful- respiration (Fig. 21A). Because the sine qua non of
ness of these techniques for research and diagnostic brain metabolism is a high rate in mitrochondrial res-
purposes. piration, the evaluation of energetic balance in the
brain under physiological and nonphysiological con-
ditions is important (65).
MRS in AD 31
P MRS detects distinct signals from the most
In AD, MRS has demonstrated changes in neuro- important metabolites involved in energy transport
chemistry due to increased oxidative stress (indicated and storage (i.e., the molecules containing high-
by depletion of brain antioxidant, glutathione) and energy phosphate bonds). ATP exhibits three peaks
altered lipid and energy metabolism with the progres- in the 31P MRS spectrum corresponding to the three
sion of the disease. No studies were identified in the phosphorus atoms (a, b, and g), and PCr exhibits one
scientific literature that positively correlated these peak corresponding to the phosphorus atom (66). In
neurochemical changes with clinical findings, estab- addition, using specific experimental conditions
lished the sensitivity or specificity of MRS in AD, or (magnetization transfer), the activity of CK, which



Figure 21 A schematic presentation of energy (A) (64) and lipid metabolism (B) (103). The
abbreviations in energy metabolism are as follows: CK, choline kinage; PDH, pyruvate dehydro-
genase complex; ATP, adenosine triphosphate; ADP, adenosine diphosphate. Glucose breaks
down to pyruvate in the cytosol during glycolysis. (B) The abbreviations in lipid metabolism are
as follows: FA, fatty acid; PC, phosphorylcholine; CK, choline kinase; CDP: cytidine diphos-
phate; PLA, phospholipase A; PLC, phospholipase C; PLD, phospholipase D; LPL, lysophopholi-
pase; CPD, cholinephosphodiesterase; PD, phosphodiesterase.

catalyzes the transfer of the phosphate group of PCr lapping signals, the significance of modifications of
to ADP, can be directly measured (67). PCr is PME and PDE resonances in pathology is not com-
detected together with Cr through its major reso- pletely known. Initially it has been proposed that the
nance on a 1H MRS at 3 ppm, corresponding to N- PME-to-PDE ratio reflects phospholipid turnover
CH3 protons (68). In short, the MRS technique is (64). PME and PDE corresponds to the molecules
helpful in measuring noninvasively ATP, PCr, and involved in the anabolism and catabolism of phos-
lactate concentrations of the brain under normal and pholipids, respectively (76).
pathological conditions (69–71).
1
H MRS in AD
Lipid Metabolism. The brain has a high lipid con-
1
tent, including phospholipids, galactocerebrosides, H MRS has two great advantages: the proton is the
and gangliosides (72). 31P MRS detects the phospho- most sensitive stable nucleus, and almost every com-
rus atoms of the head groups in bilayer phospholipids pound in living tissue contains hydrogen atoms.
of neuronal membrane. These narrow PDE resonan- However, there are technical difficulties. First, the
ces of the 31P MR spectrum are primarily from GPC presence of an intense signal from tissue water and,
and GPE, which are free and mobile in the cytosol in some cases, from lipids swamp the much smaller
and involved in brain lipid metabolism (73) (see Fig. signals from metabolites of interest that are present
21B). The PME is mostly composed of signals from at much lower concentration. Another major problem
phosphoethanolamine (PE) and phosphocholine (PC) arises from the narrow chemical shift range of 1H
(74, 75). Because these resonances consist of over- signals (about 8 ppm). Thus in order to apply in vivo


58 MANDAL

in white matter, but a moderate inverse association
between frontal white matter mI levels and global
mental function has been found (78). The combined
NAA/mI ratio is robust in discriminating possible
AD cases from age-matched control subjects (78,
84). The NAA/mI ratio in patients with AD has also
been shown to significantly correlate with Mini-Men-
tal State Examination (MMSE) scores and even to
significantly predict MMSE change after 12 months
(85).
There are intriguing suggestions that 1H MRS
may have a useful role in prognosis of mental func-
tion and tracking of disease progression. A notewor-
thy finding has been the equivalence of in vivo chol-
ine estimates between pathologic groups and control
subjects.

31
P MRS in AD
31
P is a naturally occurring nucleus, which has been
Figure 22 The application of 1H MRS in AD. Spectra
most extensively used for studying in vivo tissue
are shown for Control and AD brain in the posterior cin-
gulate region. It is a region of the brain in which there
energetic processes. The spectra are simple as the
appears to be progressive pathological change throughout MR signals are observed only from the relatively mo-
the course of AD, and is hence a suitable target for longi- bile compounds, which are in 2–10 mM concentra-
tudinal studies. NAA/Cr ratio is lower in AD patients tion. Thus, monitoring the relative concentration of
compared with Control subjects. In most cases, the ratio various 31P metabolites noninvasively helps to study
NAA/Cr shows good specificity and sensitivity for distin- the biochemistry of diseased and normal states of tis-
guishing AD patients from Control subjects (104). sues and to monitor the efficacy of several therapeu-
tic interventions. The spectrum (see Fig. 1B) shows
1
H MRS successfully, it is necessary to suppress the characteristic resonances from b-ATP at À23 ppm
intense interfering signals (i.e., water and lipids). and g-ATP signal at À6.0 ppm. The signal at À7.5
Moreover, other technical and experimental problems ppm contains contributions from the a-phosphate
related with the localization, interpretation, and quan- groups of ATP and adenosine-di phosphate (ADP).
tification of 1H MRS spectra should be taken into The resonances at 0 ppm and 5 ppm are due to PCr
consideration accurately. and inorganic phosphate (pi), respectively. The
1
H MRS has yielded a growing body of interesting chemical shift position of Pi is sensitive to pH and
and largely replicable evidence of characteristic provides a noninvasive indicator of intracellular pH.
metabolite changes in AD (Fig. 22). A consistent Besides Pi, ATP, and PCr, signals from phosphomo-
finding has been a reduction in NAA levels in AD noesters (PME; 6–8 ppm) and phosphodiesters (PDE;
brains in temporoparietal region (77), temporal lobe 2–4 ppm) are also observed. The metabolic state of
(78–80), and parietal lobe (81). Overall, NAA cells can thus be studied by monitoring the PME
decrease in AD has been shown in at least 18 reports, peak.
31
including in vitro studies showing a correlation with P MRS studies of AD have shown abnormalities
AD pathology (82, 83). in the levels of membrane phospholipids and high-
NAA depletion is higher in gray matter compared energy metabolites that appear dependent on the se-
with white matter in AD. Another striking finding in verity of the illness (22, 23). In normal aging there is
the literature has been the unforeseen elevation of mI a decrease in PMEs accompanied with a concomitant
levels by about 15% to 20% in the gray matter of increase in PDE levels, which is different from the
patients with AD. Subjects with age-associated mem- profile of biochemical changes in AD (86, 87). Pette-
ory impairment show no significant increase in mI in grew and coworkers (22, 86) have reported elevated
the temporoparietal region (77), yet one study dem- PME levels in the initial stages of AD compared with
onstrated an increased mI signal in the posterior cin- age-matched controls. As the illness progresses, PME
gulate of individuals with mild cognitive impairment levels drop. In contrast, PDE levels and high-energy
(79). No significant mI changes have been confirmed metabolites, such as PCr and Pi, appear to increase as



of the brain thought to be involved in a particular psy-
chiatric disease, as determined by other imaging
modalities, such as positron emission tomography
(PET) or single photon emission computed tomogra-
phy (SPECT). However, the real possibility exists
that the brain abnormality is located in a brain region
not sampled by the MRS voxel. As such, the abnor-
mality could be missed altogether. Many studies
attempt to sample several locations in a given sub-
ject’s brain, but in reality, it is only a small percent-
age of the entire brain volume. Recent studies are
using the technique of MRS imaging (MRSI) to sam-
ple dozens of voxels at a time, thus reducing this
potential for sampling errors. To get high-quality MR
Figure 23 MRS data phosphomonoesters (PME) in AD spectrum, voxel selection in cerebrospinal fluid
and Control subjects. Values represent percent areas cal-
(CSF) and near the skull and scalp should be avoided.
culated in terms of the ratio between PME under-peak
area to the sum of all under-peak areas within the same
spectral curve (105). Tissue Volume Correction
the dementia worsens and seem to correlate with the Most MRS techniques use cubic or rectangular vox-
number of senile plaques (22). It has been proposed els, which do not usually correspond with the curved
that the increase in PME reflects early, possibly caus- shapes of the sampled brain regions. As such, a given
ative, abnormalities in membrane metabolism, voxel often samples a combination of cerebrospinal
whereas the increase of PDE and PCr reflects neuro- fluid (CSF), gray matter, and white matter. Because
nal degeneration and death (22, 86). CSF has no measurable proton MRS metabolites, the
Abnormalities in the lipid composition (Fig. 23) presence of a large fraction of CSF within a voxel
have been identified in different regions of the brain will artifactually lower the metabolite concentrations.
of AD patients. Anisotropy studies have additionally Furthermore, metabolite concentrations are different
demonstrated abnormal membrane fluidity in hippo- in gray matter and white matter (89). New postpro-
campal synaptosomes (88). Taken together, such cessing techniques have been developed using ana-
findings suggest that aberrations in the synthesis and tomical images to take these tissue components into
degradation of membrane phospholipids are meta- account (89). It is also possible to incorporate voxel
bolic events that occur in AD brains. The intracere- tissue composition data into the statistical analysis
bral availability of phospholipid precursors and (90) or correct metabolite concentrations (91). To
metabolites, as well as the occurrence of high-energy select a voxel in a desired brain region, it is possible
phosphates, can be estimated by the analysis of the to shift the acquisition grid for MRSI studies.
31
P spectral curve within a discrete brain area with
the aid of MRS. MR Spectra Quantification
It is possible that 31P MRS findings will change
during the progression of AD. Longitudinal studies Absolute concentration measurements are the ulti-
in larger population are needed for the time course of mate goal of in vivo 1H MRS. Because the signal
31
P MRS changes in AD. Because 31P MRS yields area is proportional to the amount of nuclei in ques-
lower SNR ratio, 31P MRS requires larger voxels, tion, it is in principle possible to quantify metabolite
which limits the specificity of the findings from a concentration in vivo. In practice, however, signal
certain region of the brain. These drawbacks can be quantification present major technical problems.
overcome using higher-field scanner due to increased First, the spectrum itself can be difficult to interpret.
SNR at higher field. It may contain many overlapping peaks (especially if
acquired at a short time echo) and due to broad base-
line that could come from metabolites with short T2.
Voxel Selection
Second, there is inevitable T1 and T2 weighting of
As the exact location of the postulated biochemical the resonance peaks, which is dependent on the tim-
abnormalities is unknown in neurodegenerative disor- ing of the localization sequence, as well as signal loss
ders, the optimal location for the MRS voxel selec- and distortions of coupled peaks. Third, the quality
tion is important. Often voxels are selected on regions of the localization (i.e., suppression of signals from


MRS

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Andere mochten auch

Andere mochten auch (20)

Ähnlich wie MRS

Ähnlich wie MRS (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

MRS