Signal-to-noise ratio in MRI

T W REDPATH, ​PhD​, ​FInstP

Department of Bio-Medical Physics and Bio-Engineering, University of Aberdeen, Foresterhill,

Aberdeen AB25 2ZD, UK

energy is SNR is still required to avoid poor image

Clinical MRI quality depends crucially on the with not
quality. very small, the population di​ff​erence increases
by their e​ff​ect on coil Q factors, but by a
signal-to-noise ratio (SNR) available from the more in In this commentary, the physics underlying the
fundamental consideration of the electrical system
transducer, in this case the receiving coil, noise EMF direct proportion to B0​, as does the size of the
which they induce in the receiving coil. which yields a
small radiofrequency (RF) voltage The nuclear variation of SNR with static field strength will be
magnetic resonance (NMR) signal when nuclear
nuclear magnetic dipole moment. Secondly, the briefly
magnetic dipoles are excited to pre- EMF induced in
the receiving coil increases as the cess within it. reviewed, practical methods of optimizing Larmor
square of the Larmor frequency f​0​, in other words The
frequency at which the dipole moment SNR will be
latest MR scanners have considerably as the square of
outlined, including a discussion of the precesses is
B​0​, the static magnetic field higher SNR than the
prototype machines of the strength [1]. This increase ​ that the rate of ​principles of
proportional to B0 so

is due to two separate ​late 1970s and early 1980s, quadrature and phased-array RF change of flux linked

owing to the use of and independent factors. Firstly, to the RF coil also increases coils, and methods of

the separation of higher field strength magnets and to measuring SNR for accept- in proportion to f0​.
Therefore, by Faraday’s law ​ance testing and quality
improvements the energy levels of spin-up and

spin-down protons in RF receiving coil design. assurance will be of electromagnetic induction, the

However, modern is proportional to B0​, so that their signal produced discussed. by a dipole magnetic

population ​imaging techniques often demand very moment of constant magni- tude,
​ will increase in
proportion to its precession frequency. The
high speed di​ff​erence is larger at high values of B0​.
combination of these two factors ​Signal-to-noise
Since the ​and spatial resolution, so that the highest

possible ratio of the energy di​ff​erence to thermal ratio and static magnetic ​gives the f0​2
dependence. ​field ​ oise in the MR image
strength N approximately resistive. Since only the of the physics

is caused by the ther- The arguments presented in this underlying the relative sizes of signal narrow

section are mally driven Brownian motion of electrons bandwidth around the Larmor frequency and noise

within based on the electromotive forces (EMFs) EMFs. This avoids the need to consider is used for

induced the body’s conducting tissue, and within the signal observation, only noise voltages specifically the

receiv- in RF receiving coils. EMF is measured in reduction of coil Q when the patient in this narrow

volts. ing coil itself. This noise is sometimes called frequency band produce noise in is placed inside the

The voltage measured across the terminals of a coil. RF eddy currents within the image. As the

Johnson noise. At mid and high fields, the patient coil Larmor frequency increases, so the patient’s

tuned with a capacitor is approximately Q will be the conducting tissue are therefore dealt does the

dominant noise source unless the coil times any EMF frequency of the eddy currents which produce
​ the

image noise. Therefore, by Faraday’s ​Received 5

produced in it, owing to the is very small. The patient
January 1998 and in revised form 6 March 1998,
​ accepted
noise EMF is caused by voltage multiplication e​ff​ect 20 March 1998. ​law of electromagnetic induction, the
noise EMF
of coil tuning (Q is the random RF currents circulating

round a number Q factor of the RF coil). Thus, tuning T he British Journal of Radiology, July 1998 704
C
amplifies of eddy current loops, thus producing o
m
randomly both signal and noise EMFs. However, the
m
ratio varying magnetic fields which induce noise e
n
voltages of the signal EMF to the noise EMF remains t
a
the in the RF receiving coil. The current amplitude r
y
same whether the coil is tuned or not. It is therefore :
S
flowing in these loops is, to a first approximation,
i
g
simpler not to base SNR arguments on coil Q
n
reasonably independent of frequency, since factors, or a
l
on the size of observed voltages across Johnson noise -
t
is white, and the impedance of the the terminals of o
-
tuned coils, but on consideration loops is n
o winding to reduce its resistance spatial resolution can
i be retrieved by image smoothing. As the underlying
s noise level increases and hence the noise it generates,
e or by eliminating coil resistance altogether by using a
r
high tempera- as the square root of the acquisition
a
bandwidth per pixel Df, it is important to choose as
t
low a ture superconductor (HTS) to form the winding
i
[4]. Immersion in liquid nitrogen is the usual
o
valueasispractical.However,toolowabandwidth results
i
n
in chemical shift artefacts and geometric method of
M cooling. Its boiling point is 77 K, and it is much
R cheaper and easier to handle than distortion. Following
I the introduction of surface coils [7] liquid helium. The
rapid increase of signal with field strength which
optimize SNR over a small superficial volume of
increases in proportion to f​0​, since the flux linkage interest, further innovations in RF coil is the only
reduces the number of signal averages possible in a factor contributing to improved conven- tional image
given time, as field strength is increased) [5]. is quality at high field. All other factors design have
changing faster for the same eddy current ampli- tude improved sensitivity. The main ones to note are: the
[2]. This argument also shows that a well fitting RF introduction of circularly polarized work against it,
with two notable exceptions. MRI of brain function
receiving coil, closely coupled to the ​Optimizing (fMRI) is based on blood oxygen- (CP) (i.e.
quadrature) coils [8]; highly homo- geneous RF field
signal-to-noise ratio ​precessing magnetization CP ‘‘birdcage’’ designs [9]; and ation level dependent
(BOLD) contrast. This is
should be used to optim- ize
​ SNR as a large signal will muchstrongeratveryhighfieldstrengthsthanat most
be induced in the For a particular scanner, the user can recently, phased array RF coils [10]. Quadrature coils
optimize SNR by appropriate choice of sequence, can give a ​√​2 improvement in 1 or 1.5 tesla, so that
spatial coil; and the smallest possible tissue volume is very high field strength imaging is often used. In
‘‘seen’’ by the coil, thus reducing noise. Good coils spectroscopic imaging, SNR by acquiring the signal
resolution and receiving coil. Once the sequence has independently through two orthogonally orientated
been chosen it is essential to choose an appro- are also coils, or the absolute frequency separation of
designed to minimize the quasi- electrostatic
capacitive interaction with the patient, priate chemically shifted lines is proportional to B​0​, so that it
field-of-view, matrix size and slice thickness. If these is through two modes of the same coil. This is
parameters are chosen to make the voxel which allows achieved because the noise voltages in each coil often
additional noise from the random electronic motion to easier to resolve metabolites at higher field strengths.
degrade the image [3]. volume so small as to give an However, for conventional imaging, the are more or
inadequate SNR, then image smoothing may be unable less uncorrelated, while the signals can be added after
to correct At imaging frequencies below about 10 appropriate phase-shifting. penalties to be paid for
MHz or so, especially with small coils, the resistance
increased B​0 ​are: noise levels increasing at least
of the the error. Edelstein et al [6] argued that MRI is
unlike transmission and emission tomography in coil linearly with f​0 as
​ outlined Phased array coils take this
winding itself, rather than the patient, can be the
approach further, with multiple overlapping RF
dominant source of Johnson noise. In this case, this
receiving coils above; increased chemical shift
respect, as with those techniques, poor SNR resulting
artefacts; signifi- cantly higher RF power deposition
from image acquisition at too high a significant gains
and tissue arranged so that they operate independently,
in SNR can be achieved either by cooling the coil
by virtue of their having negligible mutual inductance.
from all coil elements have to be added together, to
heating, increasing as the square of f​0 for
​ a particu- lar comparisons between di​ff​erent scanners and and with
pulse sequence; reduced longitudinal relaxation it all of the noise, so that the array is manufacturers
Signals are collected and images computed separ- [14]. Previously, manufacturers being synthesized by
ately from each coil, before they are merged, using the image merging software had used a multiplicity of
time (T​1​) contrast as tissue T​1 values approach that of phantoms, pulse into a conventional CP coil. For
​
tissue at the sequences and voxel volumes to measure
pure water; RF magnetic field inhomogeneity a variety
SNR, surface of the patient, SNR could in theory be
of possible algorithms, into a composite image. This
with the technique sometimes biased to favour
approach gives similar SNR to that e​ff​ects due to eddy
improved virtually indefinitely if coil noise is their
current and wavelength e​ff​ects; and higher capital,
equipment. The projection protocol uses a ignored.
installation and running costs available from each coil
This could be achieved simply by making single-shot,
individually, but over a much bigger field-of-view. It
could be said that [5]. Thus, at best, SNR will increase spin echo sequence with non-selective ​the phased
linearly with magnetic field strength. In practice, to
maintain they give the SNR of a surface coil, but the array coil elements smaller and smaller, RF pulses to
coverage of a conventional coil. However, at the the
chemical shift e​ff​ect at a constant number of pixels, give a one-dimensional projection of so that each
the frequency encoding bandwidth used has centre of
coupled more strongly to the precess- the signal from
the body or head, it is well known that a surface coil
will not give good results, so that at to increase in a syringe or test tube of water onto ing superficial
proportion to B​0​, thus reducing the gain in SNR with
magnetization, with minimal noise its axis. In some
frequency to only a √​ ​f​0 depen-
​ depth, a phased array
scanners, this sequence is not being picked up from
coil will not be superior to a conventional quadrature
coil by virtue of its dence. Furthermore, once the tiny volume of tissue available, so that this method

decreasing tissue con- ductivity and increasing T​1 are

​ has not found general seen by the coil element. In
taken into account, multicoil technology, but only by
virtue of its fitting the patient more closely, thus practice, cost and acceptance. The method has the
minimizing it can be argued that the real gain in SNR
advantage of engineering complexity limits this
with f​0 is even less (the first e​ff​ect causes the patient
approach. giving the SNR per unit volume of water,
to the volume of tissue from which eddy current noise
nor- When combining signals from two or more
is received. In e​ff​ect, to image at depth, the signals
generate more noise than expected, and the second malized to 1 Hz receiver bandwidth. As a result,

elements of a phased array RF coil, it is important

T he British Journal of Radiology, July 1998 705
comparisons are straightforward, andT are una​ff​ec- to
W
realize that the SNR performance of Rthe array ted by
e
slice profile and longitudinal relaxation
d time can be
p
degraded if noise from di​ff​erent elements
a is (T1​)
t
e​ff​ects. However, the result is still strongly ​correlated.
h
Consider a tissue region which is dependent on the Hz bandwidth. For argue that noise correlation is

amount of RF loading used approximately equidistant possible even if accurate results this method still

from two coil elements, when making the requires a knowl- mutual inductance is zero [10, 12].

measurement. Loading refers to so that it gives rise to Redpath [13] edge of slice profile. The strength of the

roughly the same signal size the use of saline, or some method is has derived a formula for the degree of

other electrically con- in both. Combining the signals correlation that the maximum possible SNR

from both elements ducting material, to stimulate the achievable for based on the measurement of voltage

e​ff​ects of eddy will roughly double the signal in this feed through the phantom can be calculated for

region, while currents in the patient. Loading, conventional from one element to another. In my

therefore, intro- the combined noise will increase by opinion, RF coils, providing the RF field can be

√​2 if it is duces eddy current noise which reduces the assumed Jesmanowicz and Hyde’s position could

meas- completely uncorrelated, so that SNR will be probably to be homogeneous over the whole phantom,

ured SNR. improved by about ​√2​ . If the noise is for be reconciled with that of other authors if, instead

completely Recently, McRobbie has proposed making comparison with the measured result. The calcu- of

SNR correlated no improvement in SNR will be seen, basing their arguments on zero mutual induct- lation

measurements using a standard phantom with as both of the maximum achievable SNR assumes ance, they

signal and noise will be doubled on com- known based it on zero voltage feed through that the RF coil

relaxation times, and a loading annulus of bining the itself is noiseless. from one element to the other, with

outputs from the two elements. known electrical patient or For transmit–receive coils, the maximum

conductivity and geometry so that There is some SNR loading phantom in position within the coil

controversy regarding theoretical RF loading e​ff​ects array. achievable from any patient or test object can be

are standardized [15]. aspects of noise correlation, This is because Jesmanowicz and Hyde appear to

with Jesmanowicz and Conventional selective estimated from the net forward power needed to be

two-dimensional spin echo Hyde convinced that noise under the misapprehension that zero mutual produce a

cannot be correlated if images are acquired with 90​° ​RF pulse of known shape [16]. The inductance

measurements nor- elements have no mutual between elements implies that the coils advantage of

inductance [11]. Others malized to unit volume and 1 this approach is that no assumptions cannot interact.
 :
However, as argued by Redpath need to be made S
i
about RF homogeneity, or about [13], this is not the g
n
case. Zero mutual inductance the size, shape or
a
l
composition of the patient or only implies that the
-
coils will not interact if the phantom being imaged, as t
o
detailed knowledge of patient is absent. In practice, -
n
however, provided RF loading is not needed. o
i
Furthermore, the SNR phased array coils are designed s
e
according to the estimated from the forward power
r
measurement principles outlined by Roemer et al [10], a
t
noise does include noise coming from the coil itself, as i
o
correlation does not appear to significantly degrade i
n
well as that from the patient or phantom. For phased M
R
array coil performance. quality assurance and
I

acceptance testing transmit– receive

​ coils, often the
2. Hoult DI, Lauterbur PC. The sensitivity of the
most frequently used coils ​Measuring
advantages, in that considerable prior knowledge

signal-to-noise ratio o​ n many systems, a zeugmatographic experiment involving human ​of the

combination of the standard ​phantom method of SNR expected result is available. ​samples. J Magn Reson
measurement and the The projection method of SNR
1979;34:425–33. 3. ​ Redpath TW, Hutchison JMS.
measurement was proposed with the aim of bringing
Estimating patient dielectric losses in NMR imagers. Magn
some order forward power method would appear to Reson Imag 1984;2:295–300. 4. Hall AS, Alford NM,
have
Button TW, Gilderdale DJ, ​Discussion ​Gehring KA,

T he British Journal of Radiology, July 1998 706

Young IR. Use of high temperature ​It is surprising that,
C
o given MRI has been in ​superconductor in a receiver coil
m
m for magnetic reson- ance
​ imaging. Magn Reson Med
e 1991;20:340–3. ​clinical use since the early 1980s, there
n
is not a ​5. Hoult DI, Chen C-N, Sank VJ. The field
t
a
dependence ​widely accepted method of performing
r
y SNR ​of NMR imaging II. Arguments concerning an opti-
measurements for acceptance testing and quality ​mal Froncisz W, Kneeland JB. ​field has particular

field strength. Magn Reson Med 1986;3:730–46. ​assurance. advantages. ​Noise correlation. Magn Reson Med

This problem needs to be addressed. As ​6. Edelstein 1991;20:36–47. 12.

​ Carlson JW. Power deposition and noise
correlation in NMR samples. Magn Reson Med 1989;
WA, Glover GH, Hardy CJ, Redington ​a first step, 10:399–403. 13. Redpath TW. Noise correlation in multicoil

manufacturers should include a non- ​RW. The intrinsic receiver ​

Acknowledgment ​systems. Magn Reson Med
signal-to-noise ratio in NMR imaging.
​ Magn Reson Med
1992;24:85–9. 14.
​ Edelstein WA, Bottomley PA, Pfeifer
1986;3:604–18. s​ elective one-dimensional projection
LM. A signal- ​The author is grateful to the Aberdeen
spin echo ​7. Ackerman JJH, Grove TH, Wong GG, Gadian Royal ​to-noise calibration procedure for NMR imaging

DG, ​sequence [14] in the standard imaging library, for

Hospitals NHS Trust for their continuing support
Radda GK. Mapping of metabolites in whole ani- ​SNR
systems. Med Phys 1984;11:180–5. ​of clinical MRI

testing. The standard phantom [15] and ​mals by 31P research at Aberdeen Royal ​15. McRobbie DW. The
NMR using surface coils. Nature ​power measurement
absolute signal-to-noise ratio ​Infirmary. ​in MRI acceptance

techniques [16] appear to ​1980;283:167–70. ​o​ff​er the testing. Br J Radiol 1996;

69:1045–8. 16. Redpath TW, Wiggins CJ. Achievable
basis of standard SNR measuring ​8. Hoult DI, Chen signal-to- noise ratio: a useful new index for acceptance
testing, quality control and routine imaging. In: Proceedings
C-N, Sank VJ. Quadrature detection in
​ the laboratory frame.

Magn Reson Med ​protocols. ​1984;1:339–53. ​Some MR

References ​of the International Society of Magnetic

applications, such as ultra fast imag- ​9. Hayes CE, Resonance in

​ Medicine, 5th Scientific Meeting; Vancouver,
1997. 1. Hoult DI, Richards RE. The signal-to-noise ratio of
Edelstein WA, Schenk JF, Mueller OM, ​ing, and some the nuclear magnetic resonance experiment. J Magn
Berkeley: International Society for Magnetic Resonance in
quantitative techniques, demand the ​Eash M. An Medicine, 1997:1486. Reson 1976;24:71–85.

e​ffi​cient, highly homogeneous radio- ​highest possible

SNR, and therefore are best done ​frequency coil for

whole-body NMR imaging at ​on high field systems. For

conventional imaging, ​1.5T. J Magn Reson

1985;63:622–8. 10.
​ Roemer PB, Edelstein WA, Hayes CE,
Souza SP, ​the variation of SNR with magnetic field

strength ​Mueller OM. The NMR phased array. Magn Reson

is less strong than is often supposed. However, for

Med 1990;16:192–225. ​spectroscopic and functional

brain imaging, high ​11. Jesmanowicz A, Hyde JS,

 T he British Journal of Radiology, July 1998 707

View publication stats View

publication stats