INVESTIGACIÓN REVISTA MEXICANA DE FÍSICA 50 (3) 272–286 JUNIO 2004

Principles of magnetic resonance imaging

A.O. Rodrı́guez
Centro de Inestigación en Imagenologı́a e Instrumentación Médica,
Universidad Autónoma Metropolitana Iztapalapa,
Av. San Rafael Atlixco 186, México, D. F., 09340. México,
Telephone No.: 85 02 45 69, Fax No.: (5255) 5804-4631,
e-mail: arog@xanum.uam.mx
Recibido el 25 de agosto de 2003; aceptado el 8 de diciembre de 2003

The concepts of magnetic resonance imaging are reviewed and its application to medical and biological systems is described. The magnetic
resonance phenomenon can be described by both classical and quantum mechanical approaches. Magnetic resonance imaging is based on the
techniques of nuclear magnetic resonance. The scanner first aligns the nuclear spins of hydrogen atoms in the patient and starts rotating them
in a perfect concert. The nuclei emit maximum-strength electromagnetic waves at the start, but over time the rotating spins get out of synch,
simply due to small differences in local magnetic fields. The unsynchronized spins cause the combined electromagnetic signal to decay with
time, a phenomenon called relaxation. A slice is selected applying a gradient in a particular direction (X, Y or Z). Magnetic resonance signals
are then formed by means of the application of magnetic field gradients along three different directions. Finally, the signals are acquired and
Fourier transformed to form a two-dimensional or three-dimensional image. Important parameters determining the image quality such as
signal-to-noise ratio, contrast and resolution are discussed too. A review of the most widely utilised imaging techniques is given including
ultra-fast sequences.

Keywords: Magnetic resonance imaging; pulse sequences; ultra-fast imaging.

Los conceptos de la imagenologı́a por resonancia magnética son revisados y se describen algunas de sus aplicaciones a sistemas biológicos
y médicos. El fenómeno de resonancia magnética puede describirse tanto con un enfoque mecánico cuántico como clásico. El escaner
primero alinea los núcleos de los espines de los átomos de hidrógeno que se encuentran dentro del paciente, y luego comienza a rotarlos de
acuerdo a un concierto perfecto. Los núcleos emiten ondas electromagnéticas al inicio, pero a medida que transcurre el tiempo los espines
pierden la sincronización, debido simplemente a un decaimiento que representa el denominado fenómeno de relajación. Posteriormente se
selecciona una rebanada por medio de la aplicación de un gradiente de campo magnético en un dirección particular (X, Y o Z). A las señales
de resonancia magnética que se generan se les aplica la transformada de Fourier para formar una imagen bidimensional o tridimensional.
También se estudian los parámetros que determinan la calidad de la imagen como el cociente señal a ruido, el contraste y la resolución.
Además, se presenta un breve resumen de las secuencias imagenológicas más usadas incluyendo las secuencias ultra rápidas.

Descriptores: Imagenologı́a por resonancia magnética; secuencias de pulsos; imagenologı́a ultra rápida.

PACS: 42.30.Va; 76.60.Lz; 76.60.Pc; 87.57.-s; 87.61.-c; 87.61.Cd; 87.63.-d

1. Introduction the most common imaging sequences, including the ultra-

fast modalities are presented. The magnetic resonance phe-
The first successfull nuclear magnetic resonance experiment nomenon can be described by both classical and quantum me-
in condensed matter (as opposed to those using beams of par- chanical approaches. In this paper, the classical approach is
ticles in high vacuum) was carried out in the laboratories of used for the sake of simplicity, although NMR can be more
Bloch and Purcell about 58 years ago [1-5]. For this work, accurately treated by quantum mechanics.
they shared the Nobel prize in physics in 1952. These ex-
periments layed the foundations of the magnetic resonance
imaging (MRI) and spectroscopy (MRS) applied to biomed- 2. Elementary resonance theory
ical sciences. Around 21 years later, two pioneering papers
by Lauterbur [6], and Mansfield, and Granell [7], published Nuclei of atoms exhibit a proportionality between their total
within a few months of one another, independently proposed magnetic moment µ and total angular momentum J. These
to use the NMR to form an image. This imaging modality two parallel vectors are related by
was named NMRI, however, due to the widespread concern
over any phrase containing the word nuclear, the acronym
was changed to MRI. This imaging modality is a powerful µ = γJ (1)
tool because of its flexibility and sensitivity to a broad range
of tissue properties. Its noninvasive nature makes it an highly where γ is the gyromagnetic ratio of the nucleus, which is
demanded technique to diagnose a wide variety of diseases. a constant, characteristic of a given nucleus. Note that the
A brief description of the principles governing the gen- value of γ is nucleus-dependent. The γ values of some diag-
eration of magnetic resonance imaging, and a review of nostically relevant nuclei are listed in Table I.
 PRINCIPLES OF MAGNETIC RESONANCE IMAGING 273

TABLE I. Properties of some NMR-active nuclei.

Nucleus Spin Relative sensitivity Gyromagnetic ratio
γ
2
[MHz/T]
1 1
H 2
1.000 42.58
13 1
C 2
0.016 10.71
19 1
F 2
0.870 40.05
31 1
P 2
0.093 11.26

Regarding the spin I, as a quantum operator, then J can

be defined by
hI
J= (2)
2π
where h is Planck’s constant. The quantum number m of I z
(its measurable component) can only take on integer or half-
integer values in intervals of 1 from -I to I, so it follows that F IGURE 1. The allowed energy levels of a spin 1/2 in a magnetic
the allowed values of M z are: field.
−γhI −γh(I − 1) γh(I − 1) γhI
, ,··· , , (3) The spin population difference in the two spin states is re-
2π 2π 2π 2π
lated to the energy difference. According to the well-known
where I is the nuclear spin number. The quantity I charac-
Boltzmann relationship,
terises the nuclei. In particular, the proton, the electron and
µ ¶
the neutron all have I =1/2. Consider a nucleus within a con- n2 ∆E
stant magnetic field B◦ applied parallel to the Z-axis. The = exp (9)
n1 κT
interaction energy of the nucleus is described by a very sim-
ple Hamiltonian: In practice,
γB0 hIz ∆E
Hz = (4) ∆E ¿ (10)
2π κT
This operator is called the Zeeman Hamiltonian. Since
Consequently, by first-order approximation
Hz and Iz are proportional to each other, then they have the
µ ¶
same eigenfunctions. The eigenvalues of Hz are: ∆E γB0 h
exp ≈1+ (11)
γB0 hm kT 2πκT
E(m) = (5)
2π therefore
which are the allowed values of energy for a free nucleus
n2 γB0 h
with spin quantum number I, and gyromagnetic ratio γ, in =1+ (12)
the magnetic field B◦ . There are 2I+1 such Zeeman levels. n1 2πκT
The Zeeman energy is also called the energy of spin interac- and for a spin 1/2 nucleus the expression for the fractional
tion with the magnetic field. The energy levels are equally population difference is
spaced because the consecutive values of m differ by 1. The
µ ¶
energy difference between neighboring levels is; n2 − n1 γB0 h
= tanh (13)
γB0 h n 2πκT
∆E = (6)
2π where n=n1 +n2 is the total number of nuclei, κ is Boltz-
Transitions between adjacent energy levels (see Fig. 1), mann’s constant, and T is the absolute temperature, and n1 ,
can be induced by applying an alternating magnetic field n2 are the populations of the two Zeeman levels which obey
B’= B1 cos(ωt). The energy of photons of this frequency is the Boltzmann distribution. M ◦ (bulk magnetization) is pro-
portional to n1 -n2 . In thermal equilibrium there is a net popu-
ωh
E= (7) lation difference between the energy states. The small size of
2π γh/2π B◦ means that M ◦ is small and consequently the NMR
so that the resonance occurs when: sensitivity is very low at room temperature, even at high mag-
netic fields, such as 1 Tesla. We have an equalization (all
ω = ω0 = γB0 (8) components have the same phase at the start of the process)

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

274 A.O. RODRÍGUEZ

of the spin populations for a 90◦ pulse and a population in- In a frame rotating at angular frequency, ωe , about the Z
version for a 180◦ pulse. Although it is very small, the pop- axis this may be written as
ulation difference between the two spin states generates an
dM
observable macroscopic magnetization vector M from a spin = γµ × (B0 γ − ωe ) k (19)
system. Such spin system is said to be magnetized. dT
The magnitude of the bulk magnetization vector points In a standard NMR experiment, a small oscillating RF
exactly along the positive direction of the Z-axis at equilib- field is applied perpendicular to the static field B◦ . In the
rium and is: laboratory frame this can be described by
γ 2 B 0 h2 n Bx = 2B1 cos(ωt) (20)
Mz0 = |M | = (14)
4π 2 kT
This may be split into two components rotating at fre-
From this, it can be said that magnetization is directly quencies ω and -ω about the Z-axis. To eliminate the time
proportional to the external field strength B◦ and n. MRI ex- dependence, we use a coordinate system that rotates around
periments are often performed with the object being at room the Z-axis at a frequency ω transforming this rotating frame,
temperature, one is limited to increasing the magnitude of and assuming γ positive, we obtain the following effective
the applied field for an increase in the bulk magnetization. field Bef f
Eq. (13) is only valid for a spin-1/2 system, so for a general µ ¶
spin-I system, magnetization becomes ω
Bef f = B0 − k + B1 i (21)
γ
γ 2 B0 h2 nI(I + 1)
Mz0 = (15) We have neglected the component rotating at -ω. Pre-
12π 2 kT
cession takes place around this rotating field (see Fig. 2),
From Eq. (14) it can be appreciated that magnetic reso- therefore, we can use Eq. (17) and replace B◦ by Bef f .
nance imaging is a low-sensitivity technique since for a mag- If ω satisfies the resonance condition
netic field strength of 1 Tesla and using protons as the source,
three in a million protons in an object can be activated to gen- ω = γB0 (22)
erate the MR signal.
Bef f =B1 and the precession of M takes place around the vec-
tor B1 . We are referring to precession as seen in the rotating
3. Larmor precession frame of reference in which B1 is static as indicated in Fig. 3.
If we apply the oscillating magnetic field for a short period,
We shall study the magnetic moment of a number of nuclei
t1 with amplitude B1 , the magnetization will precess through
contained in a sample, in which two external fields are ap-
an angle θ=γB1 t1 . If θ=π, then the pulse inverts the magneti-
plied: a strong constant field applied along the Z-axis and
zation. This pulse is called a 180◦ pulse. And, if θ=π/2 (90◦
an orthogonal Radio Frequency (RF) field. We assume that
pulse), then the magnetization is rotated from the Z-direction
the external fields, mentioned above, provide the only forces
to the -Y-direction. After a 90◦ pulse the magnetization pre-
changing the orientation of each nucleus, and that they are
cesses in the laboratory frame, pointing normal to the static
uniform throughout the sample. The resultant nuclear mo-
field. The pulses described here are usually referred to as RF
ment per unit volume is denoted by µ. We are essentially
pulses.
interested in the variation with time of this vector. The equa-
tions describing the Larmor precession of the magnetization
can be derived using a classical argument. Classical electro-
magnetism assures that Eq. (1) holds. A magnetic moment µ
in a magnetic field B, is subject to a torque T:

T =µ×B (16)

If we apply the theorem of preservation of angular mo-

mentum to Eq. (16), then T becomes
dJ
T = (17)
dt
In the presence of a magnetic field B=B◦ k we find that by
combining Eqs. (1) & (16), the variation of the polarisation
vector M is given by
dM F IGURE 2. The Larmor precession of a magnetic moment in a uni-
= γµ × B0 k (18)
dT form magnetic field.

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

 PRINCIPLES OF MAGNETIC RESONANCE IMAGING 275

TABLE II. Representative values of approximate relaxation times

for hydrogen components of different human tissues at 1.5 Tesla
and 37◦ C.
Tissue T1 (ms) T2 (ms)
gray matter 950 100
white matter 600 80
muscle 900 50
cerebrospinal fluid (CSF) 4500 2200
fat 250 60
blood 1200 100-200

transverse relaxation time respectively. The transverse rate of

decay may differ from the longitudinal one.
In terms of the original variables, the complete set of so-
lutions is therefore
µ ¶
−t
Mx (t) = exp (Mx (0) cos(ω0 t)
T2
+My (0) sin(ω0 t)) (26)
µ ¶
−t
My (t) = exp (My (0) cos(ω0 t)
T2
−Mx (0) sin(ω0 t)) (27)
µ ¶ µ µ ¶¶
−t −t
Mz =Mz (0) exp +Mz (0) 1− exp (28)
T1 T1
Spin-lattice relaxation, which is characterized by T 1 , the
spin-lattice relaxation time occurs as a result of an exchange
F IGURE 3. a) Effective field Bef f in the rotating frame. b) The of energy between the spin system and the lattice. The lat-
magnetisation about Bef f . tice is defined as the assembly of sample molecules treated
as a reservoir of thermal energy, determined by the motion of
molecules. The transverse magnetization is not related to the
4. The Bloch equations and relaxation times energy of the spins and its evolution is influenced by quantum
transitions, which cause the transfer of energy between spins,
So far, we have not taken into account the interactions be-
leaving the total energy unchanged. This process is called
tween the spins and their environment, since we were deal-
spin-spin relaxation and is defined by T 2 , the spin-spin relax-
ing with isolated spins. However, for a sample made up of
ation time.
matter, it is necessary to consider the internal magnetic and
Inhomogeneity in B◦ also gives rise to decay of trans-
electric fields. These fields can cause additional motion of the
verse magnetization characterized by a decay time, T 2 *. In
magnetization, known as relaxation. The problem of the evo-
this case, the precession vector proceeds at different rates in
lution of the magnetization under the influence of the sum of
different sections of the sample. This accelerates the process
a constant, and a rotating field with simultaneous relaxation
of transverse relaxation, so that T ∗2 <T 2 .
was first solved by Bloch [3]. He proposed a set of equations
In the rotating frame, Bloch’s equations can be solved for
which describe how a spin system evolves. Thus,
continuous irradiation with a weak RF field, so that, satu-
dMz M0 − Mx ration is avoided. This saturation condition means that the
= + γ (M × B)z (23)
dt T1 decay of the transverse magnetization resulting from a π/2 is
dMx Mx complete before another pulse is applied. In other words,
= γ (M × B)x − (24) after a π/2 pulse there is no magnetization along the Z-
dt T2
direction: a new π/2 pulse transmitted at the time would re-
dMy My sult in no FID signal, as there would be no longitudinal mag-
= γ (M × B)y − (25)
dt T2 netization to rotate into the transverse plane. In this case, the
where B=B◦ +B1 . Two different times T 1 and T 2 have been magnetization precesses around Bef f (see Fig. 3). Neglect-
introduced, which are the longitudinal relaxation time and the ing the transient term, substituting M ◦ =χ◦ B◦ , and defining

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

276 A.O. RODRÍGUEZ

ω ◦ =γ B◦ , then the steady state solutions can be expressed as where φ0 is the magnetization phase (the phase angle gives
M x =χ0 ◦ H 1 and M y =χ00 ◦ H 1 the direction in the X-Y plane of the two-dimensional trans-
Dispersion mode. verse magnetization vector), Ω is the demodulation reference
frequency, and field angle θB . In 1950, Hahn [8] showed
χ0 (ω0 − ω)T2
χ0 = T2 ω0 (29) that an echo of the MR signal could be forced by subjecting
2 1 + (ω0 − ω)2 T2 the sample to two RF pulses. Hahn applied a π/2 pulse (90◦
Absorption mode. pulse) to a sample to observe a FID which follows a turn-off
of the pulse. Inhomogeneities cause a spread in the preces-
χ0 1 sion frequency, so that some of the spins go out of phase with
χ00 = T2 ω0 (30)
2 1 + (ω0 − ω)2 T2 respect to the others. Because of this dephasing, the resultant
signal decays with a time of the order of 1/(γ∆B), where ∆B
χ = χ0 − iχ00 (31)
is the spread in the static field over the specimen. If how-
χ is the complex susceptibility and χ◦ is the sample’s static ever a second π/2 pulse is applied at a time τ after the first
susceptibility. Eqs. (30) and (31) describe the NMR line- pulse, another signal re-appears at a time 2τ after the initial
shapes predicted by Bloch’s equations as shown in Fig. 4. pulse. He named the signal the spin echo. Later, Hahn him-
self showed the existence of echoes from Bloch’s equations.
This solution proved that if τ is varied the echo amplitude di-
5. Free induction decay and spin echoes minishes exponentially with a time constant T 2 . A pictorial
description of the process of echo formation in a 90◦ -τ -90◦
If a specimen is left in a high magnetic field for a long enough
sequence is shown in Fig. 5.
time, protons in the sample will tend to align themselves
along the direction of the external magnetic field. Forming
a macroscopic nuclear magnetization of the sample. But, if 6. Fourier transform NMR
the magnetization is perturbed away from alignment with the
field, using a 90◦ pulse, precession of the resulting magne- The early NMR experiments used continuous wave (CW) de-
tization will occur. There will be a gradual dephasing of tection. Nowadays, systems use the so called Fourier Trans-
this magnetization and consequently a loss of coherence of formation NMR methods, which employ short intense RF
the precessing magnetization. The dephasing of the trans- pulses, to excite a wide bandwidth of frequencies simulta-
verse magnetization causes a gradual decrease of the signal neously. Recently the wavelet transform has also been used
induced in the RF coil, leading to a decaying NMR signal to generate an image from MR signals [9]. The Fourier trans-
which is called the free induction decay (FID). (That is, de- form of the resulting FID reveals the mixture of frequencies
cay free of B1 ). This represents the total time-varying coher- produced by the applied gradients, chemical shift variations,
ent magnetic field derived from the sum over all precessing etc., for on resonance spin species. The FID is the expo-
proton spin fields, which induce a small EMF in any RF coil nential T 2 decay whose Fourier transform is the lineshape of
properly oriented to detect the corresponding flux changes. It Eq. (30). Experimentally, the lineshape will be broader than
can be used to locate the resonance peak for water, and de- that predicted by T 2 relaxation only. This is because the
termine the RF amplitude and duration necessary to produce
maximum signal. The theoretical expression for the FID for
the complex form of a demodulated signal due to an RF spin
flip at t=0 is:
Z
t →−
s(t) ∼ ω0 d3 re(− T2 ( r )) |B| (→−
r ) |M (−→r , 0)|
→
− →
− →
−
×ei(t(Ω−ω( r ))+φ0 ( r )−θB ( r ) (32)

F IGURE 5. The Hahn spin experiment: i) application of π/2 pulse,

ii) dephasing of spins, iii) position of spin isochromats following a
F IGURE 4. Lorentzian lines for absorption and dispersion. π pulse, and iv) position of magnetisation after refocusing.

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

 PRINCIPLES OF MAGNETIC RESONANCE IMAGING 277

magnetization is also attenuated by dephasing magnetic field 8. Image formation

inhomogeneities. This is added to the decay function to yield
the effective decay time, where To form an image is necessary to perform the spatial localiza-
tion of the MR signals which is normally a two step process.
1 1 1 First a slice of the body is selected for imaging. Second a
∗ = + (33)
T2 T2 T2in magnetic field gradient can be applied along to any or a com-
T 2in is the effective relaxation time due only to inho- binations of following directions X, Y, and Z, and according
mogeneities. To excite all the spin frequencies of interest to a preestablished imaging sequence to generate an image
the RF pulse should have a bandwidth large enough to span with a specific orientation.
the appropriate frequency range. The bandwidth can be cal-
culated from the Fourier transformation of the pulse shape. 8.1. Magnetic field gradients
The linewidth of the resonance peak can be obtained From
Eq. (30). Defining the linewidth ∆f as the width at half In order to generate an image, it is necessary to measure the
height, then spatial variation of MR parameters such as spin density, or the
spin-lattice relaxation time, T 1 . These variables are not inde-
1 pendent of the spatial coordinates of the spin system. These
∆f = (34)
πT2 measurements are made by degrading the uniformity of the
The Fourier transformation of a single FID can produce static magnetic field so that the magnetization precesses at
all the necessary information to create a spectrum. different frequencies. Therefore, there is a variation of res-
onance frequency across the sample. We may modify the
uniformity of the field B◦ by applying linear magnetic field
7. The chemical shift gradients across the sample. The Hamiltonian describes the
Other important NMR parameters that can distinguish spins in interaction of isolated spins at position r in a magnetic field
a particular environment are: the self-diffusion coefficient D, gradient as follows
the isotropic chemical shift δ, and the hyperfine splitting J. h
In a real spin system, all nuclei in atoms and molecules have H= (ω0 Iz + γI · G · r) (38)
2π
associated electrons. If a magnetic field is applied, the sur-
rounding electron clouds tend to circulate in such a direction where G is a magnetic field gradient tensor containing nine
as to produce a field which opposes that applied, causing a components. The only terms which contribute significantly
small chemical shift. The nucleus experiences a total field: are the Z-axis ones. Therefore, the effective gradient Hamil-
tonian is
Bef f = B0 (1 − d) (35) X
H1 = γ (G · ri )Izi (39)
where d is the shielding. This shielding perturbation results i
in a shift of the resonant frequency for nuclei in different en-
vironments, and this resultant effect is very useful in NMR where G has the components
spectroscopy. The chemical shift may be expressed as,
∂Bz
f − fT M S 6 Gx = (40)
δ= 10 (36) ∂x
fT M S ∂Bz
Gy = (41)
where δ is in parts per million (p.p.m.), f is the resonant ∂x
frequency of the species of interest, and fT M S is the reso- ∂Bz
nant frequency of a reference substance (TMS: tetramethyl- Gz = (42)
∂z
silane). The effect of chemical shift is observed in images
where more than one chemical is present. The value of δ is To generate a one dimensional image we simply acquire
very small, usually on the order of a few parts per million and the NMR signal in the presence of a spatially varying mag-
is dependent on the local chemical environmental in which netic field which is added to the uniform field [6-7]. If a
the nucleus is situated. Fat (CH2 ) is a well-known example linear gradient in the Z-direction is employed, the resulting
of a chemically-dependent component which is chemically magnetic field parallel to the uniform field is
shifted about a 3.35 ppm in Larmor frequency from water
∂Bz
(H2 0) protons. A large range of δ values exist for biological Bz = B0 + z (43)
objects giving rise to many resonant frequencies. It is valid ∂x
to assume that the resonant frequency range of a spin system Therefore, the variation of the resonance frequency with
can be expresses as position can be expressed
ωmax
|ω − ω0 | 6 (37) ω0 (z) = γBz = γ(B0 + Gz z) (44)
2

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

278 A.O. RODRÍGUEZ

From Eq. (44) we can appreciate that the gradient gives however, we are usually interested in exciting a thin slice
a linear variation of frequency with position. The mag- of the sample. This can be done by simultaneously apply-
netic field now has varying amplitude for spins along the ing a magnetic field gradient and a selective pulse [10]. A
X-direction. For a one dimensional object, the FID is the suitable shaped selective RF pulse (amplitude modulation in
sum of all the individual contributions from spins at differ- time) is required to excite only a limited section of the fre-
ent positions. The decomposition of the FID into a frequency quency spectrum. In the presence of a magnetic field gradient
spectrum is obtained using the Fourier transformation of the such a pulse will only excite a slice of the sample. The thick-
MR signal. The effect of a linear gradient applied across a ness of the slice is proportional to the spectrum width. This
two-dimensional sample is shown in Fig. 6. effect can be seen from the spatial variation of the transverse
magnetization after application of a rectangular pulse of du-
8.2. Selective excitation ration th in the presence of a Z-gradient, Gz . The magnitude
of the transverse magnetization is [10]
When applying common NMR techniques, the excitation [Mx2 + My2 ]1/2 = m sin(θ)[cos2 (θ)
spectra of the applied RF pulses (non-selective or hard pulse)
are many times broader than the spin absorption spectrum, so ×(1 − cos2 (γBef f th ))2 + sin(γBef f th )2 ]1/2 (45)
that the whole sample is excited by the RF pulse. In MRI,
and its phase is defined by
· ¸ · ¸
Mx cos(θ)(1− cos(γBef f th ))
φ= tan−1 = tan−1 (46)
My sin(γBef f th )
where Bef f is the effective field, and θ is the angle Bef f
makes with Z-axis which are given by
£ ¤1/2
Bef f = B12 + Gz (z − z0 )2 (47)
· ¸
B1
θ = tan−1 (48)
Gz (z − z0 )
For small values of B1 such that
B1
¿1 (49)
Gz (z − z)0
this reduces to
· ¸
2B1 M0 γGz (z − z0 )th
|M |= sin (50)
Gz (z − z0 ) 2
γGz (z − z0 )th
θ= (51)
2
Thus the shape of the slice (sinc) corresponds to the
Fourier transform of the applied pulse (top hat), and the width
of the selected slice is inversely proportional to Gz . The phase
variation with Z means that there will be little total transverse
magnetization at the end of the pulse application. The phase
variation can, however, be eliminated by applying a gradi-
ent, Gz of opposite polarity for a time, th /2. For general
pulse shapes the form of the excited slice follows the pulse
spectrum only for low angles, and more powerful techniques
must be used to design pulses to excite rectangular slices with
large flip angles. The relation between the slice select gradi-
ent used and the orientation of the slice plane produced is
summarized in Table III.

TABLE III. Image orientation and the gradient coils used.

Applied slice select gradient name slice plane orientation
Gx sagittal parallel to y-z plane
F IGURE 6. a) Application of a linear gradient field to a sample, b)
projection of the sample spin distribution, and c) selective excita- Gy coronal parallel to x-z plane
tion of a plane of spins in a cylindrical sample. Gz transverse parallel to x-y plane

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

 PRINCIPLES OF MAGNETIC RESONANCE IMAGING 279

Multi-dimensional RF pulses can also be applied in a

wide variety of MR applications in which transverse mag-
netization is to be excited, or to be refocused within a well-
defined spectral or/and spatial region.

8.3. The k-space concept

This is a very important concept in MRI since it allows us to

manipulate how the data are acquired, manipulated, and re-
constructed for viewing. The k-space can be simply defined
as the abstract platform onto which data are acquired, posi-
tioned, and then transformed into the desired image. There is
no such agent in any other imaging modality (x ray-imaging,
Positron Emission Tomography, Computerised Tomography,
Ultrasound). The choice of the letter k is based on the tradi-
tion among physicists and mathematicians to use that letter
to stand for spatial frequency in other similar equations, and
has no particular significance at all.
The MR signal as a function of time produced by the slice,
omitting relaxation effects [10], is given by
Z Zt
S(t) = ρ(r) exp(iγ r · G(t0 )dt0 )dr (52)
0

where ρ(r) is the spin density at position r. Mansfield and

Grannell [7] realized that there is a similarity between the
MRI signal and the Fraunhofer diffraction pattern of scatter-
ing plane waves. A diffraction pattern is generated and its
Fourier transform is the image of the object. In our situation, F IGURE 7. k-space trajectories corresponding to well-known
the plane wave is more of a theoretical concept rather than imaging schemes: a) Projection Reconstruction (PR), b) Two-
a real wave, but it can help in the study and design of pulse Dimensional Fourier Transform (2D-FT), c) Echo-Planar Imaging
sequences. Mansfield and Grannell introduced the concept of (EPI). d) The Modulus Blipped Echo-Planar Imaging (MBEST). e)
Spiral Echo-Planar Imaging (SEPI), e) Square Spiral Echo-Planar
the reciprocal space wave vector
Imaging (SSEPI).
Zt
k=γ r · G(t0 )dt0 (53) For a time independent gradient G, which is applied for a
0 time td , kmax can be written as:
where k is a vector in k-space. From Eq. (52) and Eq. (53)
the signal can be written as: kmax = γGtd (56)
Z
S(k) = ρ(r) exp(ik · r)dr (54) Hence, from Eq. (56) the image resolution can be increased
by augmenting the gradient amplitude or the time duration,
It is not very difficult to see that the parameter k describes to sample higher spatial frequencies.
a trajectory scanning through k-space as the time t of Eq. (54)
changes. k-space contains all the information required to
form an image and allows us to depict graphically most of
9. Magnetic resonance scanner
the MRI techniques. Some k-space trajectories of common
A number of processes must be completed to produce MR
imaging techniques are shown in Fig. 7. The spatial res-
images. These processes include nuclear alignment, RF ex-
olution achievable in MRI is established by the wavelength
citation, spatial encoding, and image formation. In simple
λ = 2π |k| − 1. This parameter indicates that spatial reso-
terms, an MRI system consists of five major components:
lution does not depend on the wavelength of the RF at the
a) a magnet, b) gradient systems c) an RF coil system, d) a
operating frequency. The resolution of an image in the X-
receiver, and e) a computer system. Fig. 8 shows a block
direction, can be measured as a function of the maximum
digram of the main components of MR scanner for clinical
spatial frequency kmax sampled in the experiment:
applications. All of which must be taken into account when
π designing a suitable site and all of which have their own spe-
∆x = (55)
kmax cial problems.

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

280 A.O. RODRÍGUEZ

former that surrounds the imaging subject. Three sep-

arate coils are needed in order to produce a linear vari-
ation of the Z-component of the magnetic field along
each of the three Cartesian directions (see Table III).
Many clinical MR systems are capable of producing
40 mT m−1 gradients to this end.
c) RF coil system. In MRI is necessesary to irradiate the
sample under test with an RF field (B1 ), in order to
flip the magnetization away from its equilibrium state
and generate a detectable NMR signal. This is usu-
ally done with an RF transmitter which is responsible
for pulse shape, duration, power, and timing (repeti-
tion rate). Since the imaging subject is excited with
an RF field, each spin produces a sinusoidal signal at
a frequency dependent on the local magnetic field. To
detect the signal coming from the spins is necessary a
device to couple the nuclei to some external circuitry.
These devices are called RF coils, or RF resonators,
or RF probes. RF coils can be divided in two main
groups: volume and surface coils. Volume coils are
typically cylindrical-shaped structures, and the most
efficient volume coil to the present time is the so called
bird-cage coil. Surface coils can be subdivided into
single-loop coils and array coils (phased- array coils
and array of independent coils for ultra-fast imaging
F IGURE 8. Diagram of whole-body magnetic resonance imager. schemes).

a) Magnet. The magnet aligns the nuclei into low energy d) Receiver system. To convert the received RF signal
(parallel) and high energy (anti parallel) states. There- from the RF coil into form suitable for an analog-to-
fore, a strong magnet is necessary to generate a high digital converter (ADC) or digitizer, some receiver cir-
magnetic field (B0 ), which should be uniform over a cuitry is often employed. The signal is first amplified
volume of interest. A high-field magnet provides better with a low noise amplifier, then is transmitted to a re-
SNR resolution both in frequency and spatial domains. mote location to form an image via a computer pro-
However, the main requirement for the B0 field, is that cessing. The rest of the process involves signal de-
its field uniformity should be very good. A few parts modulation using a superheterodyne style circuit. This
per million over a spherical volume, 50 cm in diame- is normally done with respect to the same frequency as
ter, are required for a great variety of clinical applica- the emitted RF radiation.
tions. The optimal field strength and the type of magnet e) Computer system. This system represents the inter-
for imaging are dependent on the application so per- face through which the user initiates measuring sys-
manent, resistive, or superconducting magnets may be tem functions (system test, display images, measure
used. For most clinical MRI systems B◦ varies from functions) and usually retrieves images. In particular,
0.05 to 3.0 Tesla, and permanent and superconducting for the reconstruction process, the computing require-
magnets are mainly used. Superconducting magnets ments varies according to the imaging method used,
are made of a niobium-titanium alloy and are cooled but almost universally some form of Fourier transform
to temperatures below 12 K by immersion in liquid he- (FT) is required. The best algorithm for FT is the fast
lium, whose boiling point is 4.2 K. The main magnet Fourier transform (FFT), which can be used for two-
rarely produces a field of sufficient uniformity by itself, or three-dimensional images. The computer system
so to maintain the magnetic field homogeneity a shim should also be able to display images on a high-quality
system is necessary. monitor.
b) Gradient coils. All type of MRI modalities require
to deliberately alter the field uniformity by applying 10. Overview of MRI methods and techniques
a magnetic field gradient Gz (r), which varies lin-
early with position r to spatially encode the NMR sig- The imaging methods will be divided into four categories:
nal. Such gradients are generated by passing currents point, line, planar (two dimensions), and three dimensional
through specially arranged coils of wire, placed on a imaging, depending on the manner in which the image data

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

 PRINCIPLES OF MAGNETIC RESONANCE IMAGING 281

points are acquired. This is to say, it is usually by the sim- acquisition. This is done by using some form of selective ir-
ple expedient of replacing one stage of the imaging process radiation as described previously. Imaging sequences may be
by a selective excitation. It is clear from their ability to re- divided into two groups: a) frequency encoding of the spin
ceive the signal, simultaneously from the entire region of in- system, and b) phase encoding. We shall start with the sim-
terest, that planar and three dimensional techniques will be plest technique which uses frequency encoding.
more efficient than point and line methods, in terms of image
signal-to-noise ratio (SNR) per unit time, and will therefore 10.2.1. Backprojection reconstruction
be preferred in most situations.
This method is normally used in X-ray CT, and was intro-
duced into MRI by Lauterbur [6]. The signal obtained from
10.1. Point and line methods
a sample in the presence of an applied gradient corresponds
The point methods have the advantage of being very simple to a one-dimensional projection of the object onto the axis of
experimental methods. Usually they do not require a great the gradient. An image can be constructed by altering the di-
deal of hardware and the field homogeneity only has to be rection of this gradient to collect a number of projections of
good over a very small region. One example of these point the object, and then processing these profiles using a recon-
methods is the so called FONAR (Field Focused Nuclear struction algorithm. The total spin density contributing to a
Magnetic Resonance) method, introduced by Damadian [11]. frequency ∆ω = γGφr, to each ray (r,φ), is given by the line
It is well known, that the FID obtained from a sample lo- integral:
cated in an inhomogeneous magnetic field will decay rapidly. Z
This occurs because different regions of the sample are lo- P (r, φ) = ρ(x, y)ds (57)
cated in different magnetic fields. And, the signal therefore Γ,φ
has a wide range of frequency components which rapidly de- where s is the distance along the ray direction. Each value of
phase giving a shortened FID, and correspondingly a broad P(r,φ) is called a ray sum, and the set of values for a given
lineshape on Fourier transformation. A sample which is suf- gradient orientation φ as the projection at angle φ. Eq. (57)
ficiently large to extend beyond the homogeneous region of a represents a NMR spectrum, which is the Fourier transform
magnet will therefore give rise to a signal (to a first approxi- of the FID obtained from the gradient Gφ . A set of projec-
mation) with two components: one which is long-lived from tions P(r,φ) is acquired by rotating the gradient angle. These
the central homogeneous region, and one which is short-lived are then filtered, and back projected to form an image.
from the surrounding non-homogeneous region. A selective
pulse with an adequately narrow bandwidth can be used to 10.2.2. Fourier imaging
excite only those spins inside the resonance aperture (signal-
producing region). An image can then be produced point by The Fourier method was first developed by Kumar and et. al,
point, by moving the sample relative to the magnet or by re- in 1975 [13], and can be considered to be a typical two di-
locating the resonance aperture. mensional spectroscopy technique. The most popular class
Hinshaw [12] introduced another point method which of planar methods are currently the two dimensional Fourier
employs three sinusoidal oscillating orthogonal field gradi- techniques, for short 2D-FT. The basic idea is to apply a 90◦
ents, which define a small region at their intersection. Only RF pulse, and a slice selection gradient to generate an FID
the spins within this region are in a time-invariant magnetic which evolves first under the influence of a field gradient, Gy
field. This region called the sensitive point, generates a signal for a time ty , followed by a second gradient, Gx with dura-
which is acquired using a narrow band filter. This is a straight tion t, which is applied at a right angles to the first. The signal
forward technique to implement since there is no need for is phase encoded during the application of the first gradient,
complicated image reconstruction and little dependence on and sampled for times tx during the spatial encoding, read
gradient linearity. This method allows localised NMR mea- gradient, Gx . Excluding relaxation effects, the signal is:
surements to be made without the necessity of forming an im- ZZ
age, so it could be used to measure T 1 . But, the long imaging S(tx , ty )= ρ(x, y) exp [iγ(Gx xtx +Gy yty )] dxdy (58)
time is its major disadvantage. These methods are an exten-
The experiment is then repeated with an incrementing
sion of the point methods. The basic idea is to isolate a line
period of evolution under the gradient Gy . In these experi-
within a three dimensional object and subsequently to distin-
ments, k-space is filled with one line at a time. After the full
guish between signal emanating from different points along
data acquisition is done, a two dimensional Fourier transform
the selected line. These methods are insensitive to magnetic
(2D-FT) yields the spin density distribution. The sequence of
field inhomogeneities though to a lesser degree than the point
RF pulses and gradient pulses is shown in Fig. 9. Edelstein
techniques.
and et al. [14] modified the Fourier imaging method, to in-
troduce another method called spin-warp imaging, in which a
10.2. Two-dimensional techniques phase variation is generated by changing the gradient ampli-
Most MRI experiments are sequential plane techniques in the tude with the phase encoding carried out for a constant time
sense that a slice of magnetization is excited prior to data period. This has the advantage that effects remain constant

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

282 A.O. RODRÍGUEZ

F IGURE 10. A timing diagram for the 2D-FT Spin Warp scheme.
F IGURE 9. The sequence of RF pulses and gradient pulses for 2D-
Fourier Transform Imaging.
rents. These problems can be overcome by using another
throughout the experiments. A timing diagram for the 2D-FT type of partial sampling scheme, which makes use of slightly
spin-warp imaging sequence is shown in Fig. 10. more than half of the k-space data in one encoding direc-
tion [17-18]. This enables us to construct a coarse phase
10.3. Full and half Fourier imaging map from the symmetric data around the origin, which will
be employed afterwards to correct the phase errors, see
Whenever the four k-space quadrants are scanned, filling the Fig. 11(e). Some Half Fourier sampling strategies are illus-
whole of the space, this is designated as a Full Fourier tech- trated in Fig. 11.
nique. The Fourier transform generates information con-
taining real and imaginary parts with a well defined phase. 10.4. Three-dimensional techniques
Taking the modulus of the two parts, gives rise to modulus
images in which all phase effects are removed. However, It is possible to extend the two dimensional techniques to pro-
these methods can require long imaging times. To abbreviate duce three dimensional imaging. Three-dimensional data sets
the long imaging times of the Full Fourier methods, we can can be generated in many different ways from Fourier tech-
scan only half of the k-space [15-16], and then reconstruct niques, for example, by multislice imaging or by using ad-
the other half using the Hermitian symmetry property of the ditional phase encoding. The multislice technique can pro-
Fourier transform of the MR signal, S(-k) = S*(k) [15]. Then, duce images of different sections. The slice of the object
in some cases only half the number of the experiments are goes through a cycle during the interval TR, during this time
required. This method requires zero filling of the data before a different slice is selected and then imaged. Most of the
Fourier transformation. Filtering is needed to avoid ringing imaging time is spent in waiting for T 1 recovery of magneti-
at the edges produced by the truncating step function. Phase zation (TR). A 3D FT image can be acquired by repeating a
variation causes problems since the Hermitian symmetry is 2D-FT process a certain number of times without slice selec-
lost. Such phase variation results from field inhomogeneity, tion, where the repetition number depends on the number of
misalignment of the signal detection, and gradient eddy cur- slices thatarerequired.Theinformationinthe Z-axisisencoded

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

 PRINCIPLES OF MAGNETIC RESONANCE IMAGING 283

tial distribution) of identical nuclei that different levels of

magnetic shielding due to their chemical environments as
shown in Eqs. (35) & (36). The possibility of observing body
chemistry at the same time as producing an image is a bonus
which, if applied clinically, could uniquely pinpoint regional
metabolic disorders in tissues and organs.Via chemical shift
imaging, the clinician and the researcher are able to glean
information of a truly chemical and biochemical nature in a
completely non-invasive manner.

11. Contrast and signal-to-noise ratio (SNR)

An important aim of imaging for diagnostic purposes is to be
able to distinguish between diseased and neighbouring nor-
mal tissue. MRI offers important advantages when compared
with other imaging modalities due to its excellent soft tis-
sue discrimination of the images. MRI has an abundance of
signal-manipulating mechanisms, as the signal is dependent
of a wide variety of tissue parameters. A contrast in the im-
age can then be created to meet certain demands. Image con-
trast is defined in terms of differences in image intensity as
follows:

|ImA − ImB |
CAB = (59)
Imref

where ImA and ImB are the image intensity of tissues A

F IGURE 11. k-space can be scanned in different ways: a) symmet-
and B, and Imref is a normalising value. Im is a function of
rically about the centre, b) alternate phase-encoding lines, c) non- spin density σ, T 1 , T 2 ,T ∗2 and diffusion coefficients D. If the
positive phase-encoding steps, d) in the readout direction, and e) data acquisition parameters are chosen such that the T 1 effect
partial scanning in the readout direction. The horizontal direction is dominant, then
is the readout time. The shaded areas correspond to the experimen-
tally acquired data and the other lines are zero. CAB ≈ f (T1 ) (60)

in the same way as it is encoded along the Y-axis by the use and the resulting images is said to carry T 1 contrast or a T 1
of an additional phase encoding gradient. The Fourier trans- weighting. Similarly for T 2 contrast and spin density con-
form is applied successively three times to reconstruct the 3D trast:
image.
CAB ≈ f (T2 ) (61)
10.5. Chemical shift imaging
and
MRI and MRS have been developed somewhat indepen-
dently, and the information provided by these two techniques CAB ≈ f (σ) (62)
is complimentary. MRI exclusively uses the proton (1 H) res-
onance, to yield spatially-resolved images with no discrimi- Spin density contrast is linearly proportional to the tis-
nation among signals arising from hydrogen nuclei existing sue spin density difference, whereas T 1 , T 2 contrasts have
in different chemical groups (no chemical shift resolution). an exponential dependence on the tissue T 1 , T 2 values. Nor-
MRS is able to yield NMR spectra in which chemical shift mal soft tissue usually have a small variation in spin density,
resolution is very important. Such spectra contains no spa- but have quite different T 1 values. Therefore, T 1 -weighted
tial information. Thus it has long been desired to obtain imaging is an effective method to obtain images of a good
both spectrally and spatially-resolved information in a sin- anatomical definition. Many disease states are characterized
gle study. These two sources of biological information can by a change of the tissue T 2 value, and T 2 -weighted imaging
be put together in the so called chemical shift imaging or is a sensitive method for a disease detection. Fig. 12 shows
spectroscopic imaging. This imaging technique is referred T 1 -weighted and T 2 -weighted images of brain obtained at
to the process of selectively imaging (or obtaining the spa- 1.5 Tesla.

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

284 A.O. RODRÍGUEZ

The size of the spatial features that can be distinguished

in a MR image can be defined as the resolution, and does not
depend on the wavelengths of the input RF field.

12. Ultra-fast imaging

The development of faster magnetic resonance imaging has
mainly been motivated by the long acquisition times com-
pared to physiological motion and patient tolerance. These
long times cause imaging artefacts, a limited number of appli-
F IGURE 12. A T 2 -wheighted image (left) and T 1 -weighted image
cations and increased costs. Respiration, pulsation peristalsis
(right) of the human brain. occur during the exam can result in substantial distortion of
the image. Artefacts from patient motion can drastically de-
grade the image quality, due to discomfort originated by ex-
In any physical measurement there is always present ams with long duration times. This exacerbated in patients
noise either random or systematic which diminish the image with pain and suffering from claustrophobia. Shorted scan-
quality. Random noise often arises in an imaging system be- ning times can reduce the cost of a conventional MRI exam.
cause of spontaneous fluctuations such as the thermal noise MRI offers alternatives to reduced-time acquisition
(Brownian motion) of free electrons inside real o equivalent schemes. In the most commonly used imaging sequence,
electrical components. This noise is sometimes called John- multislice two-dimensional Fourier transform (2D-FT), the
son noise. The degree to which noise affects a measurement scan time Tscan is given by the following simple relationship:
is generally characterized by the SNR. This is the ratio of the
amplitude of the signal received to the average amplitude of Tscan = NY T R N EX (66)
the noise. The signal is the voltage induced in the receiver
coil by the precession of the magnetization in the transverse where N is the number of phase-encoding samples, TR the
plane. The noise is generated by the presence of the patient pulse repetition time, and NEX the number of excitations
in the magnet, and the background electrical noise of the sys- (also termed number of acquisitions or number of averages)
tem (coil and electronics). It is also known that the variance used for signal averaging.
of the fluctuating noise voltage is
√ 12.1. Fast Fourier imaging
σthermal = 4κT R BW (63)
The spin-warp imaging technique owes its popularity in the
where R is the effective resistance of the coil loaded by the clinical environment to its robustness, however in its simplest
body, and BW is the bandwidth of the noise-voltage detecting form it requires long imaging times. In an effort to reduce
system. Both the signal and noise are detected by the receive scan time, Haase and co-workers [19] introduced an imag-
coil, and the bandwidth is determined by the cutoff frequency ing method called Fast Low Angle Shot: FLASH. This is
of the analog low pass filter. The SNR can be defined as, basically the gradient echo version of the 2D-FT technique
signal S(k) with a very short repetition time TR. Reduction of the echo
SN R = = (64) time necessitates the use of a low angle RF pulse for slice se-
resistance Ref f
R lection, typically around 15◦ , for maximum SNR. There are
ρ(r) exp(ik · r)dr other techniques which are based on the same principles, such
SN R = (65) as the Snapshot/Turbo FLASH and Gradient Recalled Acqui-
Rbody + Rcoil + Relectronics
sition in the Steady State (GRASS) techniques [20]. GRASS
From Eq. (65) can be appreciated that to increase the is a very similar technique to FLASH, which only differs in
SNR can be done by a) reducing the effective resistance: this that an additional phase encoding gradient is applied after the
can be achieved by reducing the coil and electronics resis- sampling period. This gradient has an opposite sign to the
tance, this demands better coil designs with low resistance first one, so it can rephase any transverse magnetization re-
values, b) increasing the MR signal with high-field MR scan- maining at the end of the experiment. Consequently, the SNR
ners. High field MRI promises to improve anatomic imag- is improved and image artefacts are removed.
ing quality by factors, and to bring metabolic and functional
imaging to the forefront of diagnostics modalities. There are 12.2. Echo-planar imaging
systems available with field strengths up to 22 Tesla, but for
clinical purposes the highest field is 3 Tesla. Although it is Echo Planar Imaging (EPI) was introduced by Mansfield
possible to purchase whole-body systems of 7 T or 8 T for in 1977 [21], and is a true snapshot technique which enables
MRI applications in humans. The SNR is also an important the formation of a complete image in 30-100 ms. The rapid
parameter to measure performance of a MR imager. nature of this technique is due to the fact that only one RF

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

 PRINCIPLES OF MAGNETIC RESONANCE IMAGING 285

excitation is needed per image. It thus allows the acquisi-

tion of a two-dimensional image from only one FID. Fol-
lowing the excitation of a slice, the signal is sampled under
the influence of two orthogonal gradients. A gradient Gx
(X-axis) is modulated rapidly to generate a series of gradi-
ent echoes. This forms the frequency encoding part of the
experiment. In conjunction with Gx , another gradient Gy
(Y-axis) is applied: this gradient is either blipped or continu-
ous and performs the phase encoding part of the experiment.
This combination of gradients serves to encode each point
in the image with a different degree of phase evolution be-
tween echoes and frequency during an echo. A train of phase
encoded echoes is thus acquired. After some data manipula-
tion required because of the effect of evolution of sequential
echoes under positive and negative read-out gradient lobes, it
is possible to apply a 2D-FT to create an image. Generation
of a 128 × 128 image requires the production of 128 gra-
dient echoes, with each echo being sampled using 128 data
points. EPI has undergone many modifications since its in-
troduction in 1976. The original forms of EPI were called
Double Echo Planar Imaging (DEPI) and Fast Low Angle
Excitation Echo Planar Technique (FLEET) [22]. DEPI is a
two shot experiment with Gy having positive and negative
starting phase in alternate shots. The data sets are corrected
F IGURE 13. A timing diagram for MBEST technique.
by splicing together all the echoes formed in totally positive
and totally negative gradients. Two images are produced,
which can be re-registered and added to yield an improve- Three dimensional EPI techniques can also be obtained as
ment in SNR. Flip angles of 45◦ and 90◦ were used to mini- an extension of EPI in two dimensions. In the Echo Volumar
mize the acquisition time. These techniques originally used Imaging (EVI) technique [28], the signal is sampled under
a constant phase encoding gradient. Ljunggren [23] first pro- the influence of three gradients: one of them has a constant
posed replacing the continuous phase encoding gradient by polarity and the other two are modulated. The trajectory of
a blipped gradient: Gy blips are applied in the spaces be- k-space is extended to three dimensions which are covered as
tween echoes. The resulting trajectory through k-space is a in two- dimensional EPI. A time of over 100 ms is needed to
square raster scan. Correction of the time evolution can be complete a scan and form a 3D EPI image. Other examples
achieved by simply reversing the data which makes up ev- of 3D EPI are: multislice EPI and Phase Encoded Volumar
ery alternate echo prior to FT. The scan still starts at the k- Imaging (PEVI) [29].
space origin, and zero filling is still required. This single shot
technique is called Blipped Echo Planar Imaging (BEPI) or 12.3. Parallel imaging: SENSE and SMASH techniques
Blipped Echo Planar Single pulse Technique (BEST) [24-25].
BEST is susceptible to the effect of magnetic field inhomo- Reduction of acquisition time without diminishing the SNR
geneities which cause phase errors. Therefore, it is important is a paramount challenge in MRI. To meet this challenge, two
to ensure that the transverse magnetization is completely in important schemes have been recently introduced: a) Sensi-
phase at the start of the experiment. This implies the use of a tivity encoding for fast MRI (SENSE) [30], and b) Simultane-
180◦ pulse which will be applied immediately following slice ous Acquisition of Spatial Harmonics (SMASH) [31]. These
selection. Data acquisition begins at the point of maximum imaging schemes are based on the difference in sensitivities
signal, when the magnetization is completely refocused. The between individual coil elements in a receive array to reduce
modulus version of this experiment, called MBEST, is found the number of gradient encoding steps required for imaging.
to be essentially identical to BE21ST, except for the addition Since they make use of the individual coil sensitivities, they
of a large negative Gy gradient pulse prior to data acquisition. are called parallel imaging techniques. Each coil in an ar-
k-space is fully scanned and the FT data produces a modulus ray receives a signal from a region of interest. By appropri-
image which is considerably more robust and less suscepti- ately combining the images from each coil, an overall image
ble to phase errors. The MBEST timing diagram is shown in of higher sensitivity and/or broader field of view can be ob-
Fig. 13. EPI is a compact technique which is easily adaptable tained than with a single coil element. The individual coil
as an imaging module applied after any suitable spin prepa- sensitivities are used to reconstruct a full image from a partial
ration experiment, such as Inversion Recovery [26] and Spin k-space acquisition. A g factor (noise amplification factor) is
Echo EPI [27]. introduced due to the image reconstruction process.

Rev. Mex. Fı́s. 50 (3) (2004) 272–286

286 A.O. RODRÍGUEZ

Parallel imaging is useful for any application where min- facts, and improve image quality. The g factor is a limitant
imum acquisition time is paramount. Applications include issue when using parallel imaging.
real-time imaging, first-pass bolus contrast imaging, cardiac
imaging, and EPI. These imaging sequences heavily depend
on the coil sensitivity, so the interest to develop new coils Acknowledgment
for parallel imaging has reborned again. These imaging tech-
niques have a very promising future since are able to generate I would like to thank Sir Peter Mansfield and Professor
an image in less than a few milliseconds, reduce motion arte- Richard Bowtell for illuminating conversations.

1. F. Bloch, W.W. Hansen, and M.E. Packard, Phys. Rev. 69 21. P. Mansfield, J. Phys. C: Solid State Phys. 10 (1977) L55.
(1946) 127.
22. R. Ordidge, Ph D Thesis, University of Nottingham, 1981.
2. F. Bloch, Phys. Rev. 70 (1946) 460.
23. S. Ljunggren, J. Magn. Reson. 54 (1983) 338.
3. F. Bloch, W.W. Hansen, and M.E. Packard, Phys. Rev. 70
24. A.M. Howseman et al., Brit. J. Rad. 61 (1988) 822.
(1946) 474.
4. E.M. Purcell, H.C. Torrey, and R.V. Pound, Phys. Rev. 69 25. B. Chapman et al., Magn. Reson. Med. 5 (1987) 246.
(1946) 37. 26. M.K. Stehling, R.J. Ordidge, R. Coxon, P. Mansfield, Magn.
5. N. Bloembergen, E.M. Purcell, and R.V. Pound, Phys. Rev. 73 Reson. Med. 13 (1990) 514.
(1948) 679. 27. I.L. Pykett and R.R. Rzedizan, Magn. Reson. Med. 5 (1987)
6. P.C. Lauterbur, Nature 242 (1973) 190. 563.
7. P. Mansfield, P.K. Grannell, J. Phys. C 6 (1973) L422. 28. P. Mansfield, R.J. Ordidge, R. Coxon, J. Phys. E: Sci. Instrum.
21 (1988) 275.
8. E.L. Hahn, Phys. Rev. 80 (1950) 580.
9. J.B. Weaver, Y. Xu, D.M. Healy, and J.R. Driscoll, Magen. Re- 29. A.M. Blamire, Ph D Thesis, University of Nottingham, 1990.
son. Med. 24 (1992) 275. 30. K.P. Pruessman, M. Weiger, M.B. Scheidegger, and P. Boesiger,
10. P. Mansfield, PG. Morris, NMR Imaging in Biomedicine, Sup- Magn. Reson. Med. 42 (1999) 952.
plement 2 in Advances in Magnetic Resonance (Waugh, J.S., 31. D.K. Sodickson and W.J. Manning, Magn. Reson. Med. 38
Editor), Academic Press, New York, 1982. (1997) 591.
11. R. Damadian, M. Goldsmith, L. Minkoff, Physiol. Chem. Phys. 32. A. Abragam, Principles of Nuclear Magnetism (Clarendon
10 (1978) 285. Press, Oxford, 1989).
12. W.S. Hinshaw, Phys. Lett. A 48 (1974) 87. 33. C.P. Slichter, Principles of Magnetic Resonance, 3rd Edition,
13. A. Kumar, D. Welti, R.R. Ernst, J. Magn. Reson. 18 (1975) 69. (Springer-Verlag, Berlin, 1992).
14. W.A. Edelstein, J.M.S. Hutchinson, G. Johnson, and T. Red- 34. D.G. Gadian, Nuclear Magnetic Resonance and Its Applica-
path, Phys. Med. Biol. 25 (1980) 751. tions to Living Systems. 2nd Ed., (Oxford, University Press,
15. N.S. Cohen and R.M. Weisskoff, Magn. Reson. Imaging 9 1995).
(1991) 1. 35. P.G. Morris, Nuclear Magnetic Resonance Imaging in Medicine
16. H. Fischer, F. Schmitt, H. Barfuss, H. Bruder, 7th Ann. Meet. and Biology (Clarendon Press, Oxford, 1986).
Soc. Mag. Res., San Francisco, 1988. 36. C.N. Chen, D.I. Hoult, Biomedical Magnetic Resonance Tech-
17. C.H. Oh, S.K. Hilal, J.B. Ra, Z.H. Cho, Books of Abstracts, 6th nology (Adam Hilger, IOP Publishing, Britain, 1989).
Ann. Meet. Soc. Mag. Res., New York, 1987. 37. M.A. Foster, J.M.S. Hutchinson, (Editors), Practical NMR
18. L.E. Crooks, M. Arakawa, J.D. Hale, J.C. Hoenninger, J.C. Imaging, (IRL Press, Oxford, 1987).
Watts, L. Kaufman, D.A. Feinberg, Books of Abstratcs, 5th 38. E.M. Haacke, R.W. Brown, M.R. Thompson, R. Venkatsen,
Ann. Meet. Soc. Mag. Res., Quebec, 1986. Magnetic Resonance Imaging, Physical Principles and Se-
19. Haase, A., Frahm, J., Matthaei, D., Hanicke, W., Merboldt, K. quence Design, (Wiley-Liss, New York, 1999).
D., J. Magn. Reson. 67 (1986) 258. 39. Z.P. Liang, P.C. Lauterbur, Principles of Magnetic Resonance
20. P.V.D. Meulen, J.P. Groen, and A. M. C. Tinus Brutink, Magn. Imaging, A signal processing perspective, (IEEE Press, New
Reson. Imag. 6 (1988) 335. York, 2000).

Rev. Mex. Fı́s. 50 (3) (2004) 272–286