Progress In Electromagnetics Research M, Vol.

7, 109–122, 2009

A SIMPLE NUMERICAL METHOD TO COMPUTE

THE SIGNAL-TO-NOISE RATIO OF A MAGNETIC
RESONANCE IMAGING SURFACE COIL

R. Rojas and A. O. Rodriguez

Centro Nacional de Investigacion en Instrumentacion e Imagenologia
Medica
Universidad Autonoma Metropolitana Iztapalapa
Ave. San Rafael Atlixco 186, Mexico, DF 09340, Mexico

Abstract—The quality of a magnetic resonance image can be reliably

measured by the signal-to-noise ratio. This widely accepted parameter
is a function of the magnetic field generated by the coil and the electric
field produced by the sample to be imaged. A simple numerical method
is proposed to calculate the coil signal-to-noise ratio of a circular-
shaped coil and a spherical phantom. The phantom is composed of two-
concentric sphere simulating a brain-skull model. The electromagnetic
fields produced were then numerically computed by solving Maxwell’s
equations with the finite element method implemented in a commercial
software tool. The electric and magnetic fields were used to numerically
determine the signal-to-noise ratio using the quasi-static approach.
The numerical results demonstrated that this simple method is able
to calculate the signal-to-noise ratio of surface coils with simple coil
geometries involving a simulated phantom.

1. INTRODUCTION

The Radio Frequency (RF) resonator coils are a fundamental device

of a Magnetic Resonance Imaging (MRI) system. The quality of the
MR images strongly depends on the signal-to-noise ratio (SNR) of
the acquired signals normally obtained with a receiver coil [1]. The
SNR is the widely accepted parameter for measuring coil performance.
This parameter involves both the MRI signal and the associated noise
generated by different sources [2].
Corresponding author: A. O. Rodriguez (arog@[Link]).
110 Rojas and Rodriguez

Analytical expressions can be derived from the classical

electromagnetic theory for the simplest cases of surface coils [3].
The classical electromagnetic theory poses mathematical challenges
usually extremely difficult to solve for more complex geometries. The
complexity of the problem increases when the coil is nearer the sample
and the electromagnetic properties of the environment modifies the
distribution of SNR. An analytical solution even for the unloaded coil
case is extremely difficult to derive. Alternatively, numerical solutions
have been proposed in the past with good results [4].
The simulation of the electromagnetic properties of MRI coils
represents a good alternative to both the derivation of analytical
expressions and, the trial-and-error scheme to study the coil
performance. A number of numerical approaches have been developed
to solve various electromagnetic problems and applications [5].
Numerical solutions to Maxwell’s equations can be classified into
two main approaches: 1) the integral equation model, and 2) the
differential equation model [6]. The are a number relevant numerical
methods: a) The Finite Integration Technique (FIT) requires the
discretisation of the integral forms of Maxwell’s equations to transform
them into a set of matrix equations. The current equations contain
topological information, and the constitutive equations are expressed
in terms of matrices that depend on dielectric and magnetic properties
of the media and, relate the voltages of the fluxes. b) In the
Finite Difference (FD), Maxwell’s equations are solved by dividing the
region of interest into uniformly-sectioned small rectangular grids and
then employing a set of linear equations together with an iterative
method. c) The Finite Difference Time Domain (FDTD) is an
extension of the method described in b) and is applied in the time
domain. This involves spatially sampling the electric and magnetic
field distributions over the volume of interest by enclosing it in a
rectangular box. The box is divided into many small rectangular cells
and the calculation is performed over a given time period. This method
is easy to implement but loses accuracy for arbitrary geometries.
d) The Method of Moments (MoM) is a frequency domain method
and in computational electromagnetics it has become synonymous
with the integral-differential operator equations. It operates in terms
of a volumetric equivalent current that accounts for the effect of
the permittivity and conductivity of an inhomogeneous body. This
method is popular for arbitrarily-shaped and inhomogeneous dielectric
bodies. e) The Finite Element Method (FEM) is a differential
equation approach in which the complete problem, source geometry,
excitation, scatterers and boundary constraints, is discretised in a
variable manner. FEM divides the region of interest into irregular
Progress In Electromagnetics Research M, Vol. 7, 2009 111

triangular cells. It does not loose accuracy for abritray geometries.

Then FEM can accurately model arbitrary geometries with small mesh
elements used to describe complex geometries and larger mesh elements
used in more uniform regions. The trial function is a combination of
a set of base functions defined over subdomains (the mesh elements)
that comprise the entire problem domain. Thus, the field equations
are determined in terms of polynomials with unknown coefficients
defined in the mesh nodes or along element edges. These unknown
coefficients are then determined solving a matrix equation system.
Since the shape of the triangular cells can be arbitrary in FEM, we
have used it to numerically simulate the electromagnetic behavior of
different geometries (polygonal shapes) of a single loop RF surface and
arrays coil designs [7, 8]. FEM is suitable to solve complex problems
frequently found in electromagnetic theory.
In this paper, we proposed a simple numerical method to compute
the SNR generated by a single loop coil interacting with a pixel-based
model of two concentric spheres emulating a brain-skull model. The
SNR is computed via the calculation of the electric and magnetic fields
of the coil-sphere system in a three-dimensional Cartesian space. We
used a commercial software tool based on the FEM to solve Maxwell’s
equations.

2. METHODS

2.1. The Mathematical Model

A mathematical model to numerically compute both the electric and
magnetic fields was derived using Maxwell’s equations. It is important
to highlight that the model presented here was then transformed
into descretised form to be solved with the FEM and it is solely
for the quasi-static approach. The crucial criterion for the quasi-
static approximation to be valid, is that the electromagnetic fields are
practically the same at every instant as if they had been generated
by stationary sources. This implies that the currents and charges
generating the electromagnetic fields vary slowly in time.
Under these assumptions, Maxwell’s equations can be simplified
and their numerical solution can be obtained from:
XN Z Z
Azi [5Ni · 5Nj + ωµ(ωε − jσ)Ni Nj ] dxdy
i=1 Ω
Z Z
=− µNi Jze dxdy (1)
Ω
112 Rojas and Rodriguez

where Ω is the integration region, J e is a current density generated

externally, σ is the medium conductivity (S/m), µ is the medium
permeability (H/m), ε is the medium permittivity (F/m), ω is the
Larmor frequency and, ρ is the electric charge density (C/m3 ). Azi is
the vector potential along the z -direction for the nodal number, i. Ni
and Nj are the expansion nonzero functions associated with the node,
i. A complete derviation of Equation (1) is in Appendix A. The term
on the right-hand side of Equation (1) can be defined as Gij and the
left-hand side as bi . Finally, Equation (1) becomes
N
X
Gij Azi = bi (2)
j=1

Equation (2) represents only the solution along the z -direction,

however similar expressions can be found for the x and y axes. A set
of linear equations can be formed for this direction and solved using
a matrix system. To numerically compute the magnetic and electric
fields, the vector potential Azi and Equations (A9) and (A10) together
with a software tool were used.

2.2. Electromagnetic Simulation

The graphic interface of the commercial software tool COMSOL
multiphysics [9] was used, to build a spherical phantom to mimic the
brain and skull and the circular coil. The three-dimensional structure
was developed using regular geometrical figures such cylinders, cubes,
etc. The skull and the brain were represented in the model of Fig. 1
by concentric spheres of 13 cm radius and 12 cm of radius, respectively.
The MRI surface coil had the following dimensions: thickness of 0.1 cm,
inner radius of 6.0 cm and external radius 9.0 cm. The origin of
co-ordinates system was located at the inner sphere centre. Fig. 1
shows the pixel brain-skull model and the surface coil model used
to numerically compute both the electric and magnetic fields. The
conductive spheres representing the head produce changes in the
impedance as a function of the sample electrical properties: σ and
ε. RF coils are normally made out of copper, so σ = 5.998 × 107 S/m
and εr = 1 were used. The three-dimensional model was surrounded
with an outer sphere of 20 cm of radius representing boundaries for the
simulation process, where the gap between the sphere arrangenment
and the boundary was assumed to be filled with air. Different electric
and magnetic properties for each element in the model were used as the
frequency was modified in the simulations. Electrical conductivity and
permittivity values for brain and skull were taken from [10]. Current
Progress In Electromagnetics Research M, Vol. 7, 2009 113

Figure 1. Three-dimensional skull-brain model. The inner sphere

represents the brain cavity, the outer sphere represents the skull-head,
and the circular ring represents the copper coil. The external sphere
represents the system envelope (spherical bound).

densities at different radio frequencies were applied at one time, then

the software tool calculated the electric and magnetic fields of the
sphere-and-coil model. To numerically solve our problem, tetrahedrals
were used to form the mesh to fully cover the entire three-dimensional
model.

2.3. Mesh Construction

The FEM divides the three dimensional domains into a mesh of small
tetrahedric subregions or finite elements. Usually, four nodal points
are associated with each element at the corners of the tetrahedral. An
advantage of using tetrahedral elements is that they can be made to
fit any shape of domain boundary to a very good accuracy, which can
easily be increased y increasing the number of elements (Fig. 2(a)).
The element size and shape are defined by the geometry under study,
each element and its nodes are numbered with different sequences [11].
Fig. 2(b)) shows the final mesh configuration used in this work. The
solution of three-dimensional electromagnetic problems requires great
computing power. Extra care should be taken when selecting the type
of mesh. An arbitrary mesh is not recommended since for the coarsest
mesh the iterative process may not easily converge. A minimum
requirement is that there should be at least two mesh elements per
wavelength in the geometry according to the Nyquist criterion. The
highest frequency in this work was 100 MHz with a wavelength of 3 m.
The radiation-matter interaction depends on the material properties
and the wavelength, and the mesh has to resolve the wave across the
whole geometry.
114 Rojas and Rodriguez

(a) (b)

Figure 2. In FEM modeling the computational domain is discretised

into tetrahedral elements where each tetrahedral corresponds to an
element (a). The mesh construction with the tetrahedral meshing for
the three-dimensional pixel model is in (b).

2.4. Signal-to-noise Ratio Calculation

A set of nuclear spins precessing inside the sample, induce an EMF
at the coil proportional to the resonant frequency, ω that depends on
principal magnetic field B0 . The number of spins is proportional to the
sample volume V, and the coupling between nuclei and the coil depends
on the magnetic field of the RF, B1 . The noise is proportional to the
effective resistance Refec including the interaction with the body. Hoult
and Richards proposed [1]:
V ω 2 B1
SNR = (3)
Ref ec
The noise level is determined by dissipative power losses in
the system, such as conductor, radiation, and overall body losses.
The fluctuation dissipation theorem states that there exists a direct
relationship between electrical resistance and noise. The thermally
activated motions of the charge carries in dissipative media produce
random electric and magnetic fields which can be detected as noise.
Apresistance R produces an RMS noise voltage Vn given by
Vn = 4κT ∆f R, where κ is Boltzmann’s constant, T is absolute
temperature, ∆f is the receiver bandwidth, and R = RA + RB , where
RA and RB are the real part of the input impedance of the RF coil itself
and the sample respectively [7]. From Equation (3) and the definiton
of the RMS voltage, the SNR can be rewritten as:
V ω 2 B1
SNR = p (4)
4κT ∆f (PA + PB )
Progress In Electromagnetics Research M, Vol. 7, 2009 115

where the power losses are PA (coil) and PB (biological sample). The
energy dissipation per volume unit is approximately PA = I 2 RA and
PB = I 2 RB [4], then:
dP
= σE 2 (5)
dV
Then, a good aproximation to compute the SNR at a particular
cpoooinrdt can be obtained combining Equations (4) and (5):

B1 (magnetic field)
SN R ∝ (6)
E (electric field)

Equation (6) implies that it is necessary to compute the magnetic

field B1 (r) and the electric field E. Once both fields were computed as
described in Section 2.2, bi-dimensional mapping data were represented
by matrices: [B1 (i, j)] and [E(i, j))] where (i, j) corresponds to the
Cartesian coordinates of the magnetic and electric fields at a particular
position. Then, the SNR can be written
B1 (i, j)
SNR(i, j) ∝ (7)
E(i, j)

Equation (7) implies that the SNR is computed by diving every

matrix entry in an arithmetical fashion rather than using a matrix
operation. These matrices were finally used with specially written
programmes in MATLAB (V. 6.5, MathWorks, Inc) to determine the
SNR. All simulations were performed on a Windows-operated PC. An
alternating current of 1 A was applied to the circular coil for simplicity
at the frequency of 100 MHz.

3. RESULTS AND DISCUSSION

It was possible to derive a discretised expression as a function of the

vector potential along the z -direction from classical electromagnetic
theory, which was reported by the commercial package COMSOL we
found no evidence of similar work having been published previously.
Discretised expressions for the x- and y-directions can also be derived.
Equations in each direction were numerically solved using COMSOL,
which employs the finite element method together with tetrahedrical
elements to simulate electromagnetic fields in a three-dimensional
space. The tetrahedral elements can take different sizes to fit a
particular arrangement to be studied with a certain degree of accuracy.
The accuracy depends on the number of tetrahedrals used in a
particular mesh.
116 Rojas and Rodriguez

The mesh in Fig. 2(b) was used to cover the head-coil setup
with a reasonable number of tetrahedral elements for the numerical
computation of both fields as summarized in Table 1. A single
tetrahedral elements had a volume of 2.5 cm3 . The size of the elements
is in good concordance with the Nyquist criterion as mentioned in
Section 2.3. Since only regular figures were used the dimensions of the
mesh elements across the setup did not vary drastically. Most mesh
generation schemes use a mesh refinement technique to represent fine
structures with much smaller elements. The mesh refinement allows
us to obtain a more accurate solution. However, this tends to increase
the number of nodes and elements, thus demanding overwhelming
computation capabilities. One regular refinement can be sufficient to
obtain a good approximation. All numerical simulations were done
using a coarser mesh with the same size as summarized in Table 1.

Table 1. Number of elements used for all simulations.

Structure No. mesh elements
Brain sphere 1240
Skin sphere 1217
Surface coil 3875
Boundary sphere 1244
Total number 23 267

As a demonstration of the presented methodology, the magnetic

and electric fields were numerically computed using the three-
dimensional model of Fig. 1 and mesh configuration of Fig. 2(b). Bi-
dimensional plots in the x-y plane of both fields generated by the RF
coil and the sample at 100 MHz are shown in Figs. 3(a) and (b). The
numerically-simulated fields in the coronal orientation show that the
field uniformity decreases at the centre of the model for both electric
and magnetic fields. The intensity of the electric and magnetic fields
is higher near the coil boundaries, and decreases towards the coil
centre. On the other hand, a good coil design should produce a very
low electric field and a high magnetic field to be able to generate a
good quality image. This is an expected pattern typically found in the
design of surface RF coils normally used for various MRI applications.
Additionally, our approach is able to show section through at almost
any location and in any arbitrary plane as shown in the top row of
Fig. 3(a).
With the field mapping data of Fig. 3 and Equation (7), SNR
maps were finally computed for a resonant frequency of 100 MHz. The
Progress In Electromagnetics Research M, Vol. 7, 2009 117

(a) (b)

Figure 3. The electric and magnetic fields simulated at 100 MHz and
8 cm away from the coil plane.

(a) (b)

Figure 4. Bi-dimensional mappings of the numerically-acquired SNR

are shown in axial (a) and coronal (b) orientations for the resonant
frequency of 100 MHz.

numerically-computed SNR is shown in the form of bi-dimensional

maps in Figs. 4(a) and (b) for the 100 MHz case. These numerical
results compare very well with those reported in the literature using
other numerical approaches [4]. A major drawback is the size of
the mesh, and necessarily the number of elements implied in the
simulation. The selection of the right mesh size may determine
118 Rojas and Rodriguez

the feasibility of this method, since the finer the mesh the more
computational power required. The FEM becomes even more
demanding as the resonant frequency is increased limiting the study of
surface coil performance at magnetic fields lower than 100 MHz with a
standard PC.
The spherical brain-skin model is one of the most simple models.
However, this procedure can be applied to more realistic human-head
models. It still remains to be investigated if by refining the mesh a
significant accuracy can be obtained with our method. Consequently,
the use of a computer cluster is mandatory as has been shown by other
research groups.
Despite all those technical limitations imposed by the FEM and,
the limited computational capacity it was possible to numerically
compute the SNR of a setup involving the biological sample. This
is an important advantage since the actual biological sample can be
included, offering more realistic results to study the coil performance
without a great deal of computer resources. This approach offers a
simple and easy-to-follow methodology to calculate the SNR of an MRI
surface coil. A good alternative to investigate coil performance should
include pixel-based models of organs and tissues for more realistic
situations and to be able to be compared with experimental results.

4. CONCLUSIONS

The numerical method presented in this work offers a simple way

to compute the SNR of common geometry surface coils for MRI
involving a simulated phantom too. The results show the feasibility
of using a commercial software package to numerically simulate the
electromagnetic properties of surface coils for MRI. This method can
extended to other organs and tissues together with more complex coil
configurations.

ACKNOWLEDGMENT

R. R. gratefully acknowledges the National Council of Science and

Technology of Mexico (CONACyT) for a PhD scholarship and grant
numbers: 1-35119, 1-35106-21. We would also like to thank Dr
Benjamin Wilton for careful reading of the manuscript.
Progress In Electromagnetics Research M, Vol. 7, 2009 119

APPENDIX A. CALCULATION OF TIME-HARMONIC,

QUASI-STATIC MAGNETIC AND ELECTRIC FIELDS

From Maxwell-Ampere’s law combined with the Lorentz force equation

applied to current density, J, for a quasi-static system, an equation
relating the magnetic and elecrtic fields can expressed as follows:
∂D
5×H − = σ(E + ν × B) + J e (A1)
∂t
where H is the magnetic field intensity (A/m), D is the electric
displacement (C/m2 ), J is the current density (A/m2 ), σ is the medium
conductivity (S/m), E is the electric field intensity (V/m), ν is the
relative velocity of the reference system (m/s), B is the magnetic field
density (Wb/m2 ) and, J e is a current density generated externally
(A/m2 ).
Expressing the magnetic and electric fields in terms of the
magnetic vector potential (A) and the electric scalar potential (V )
B = 5×A (A2)
∂A
E = −5V − (A3)
∂t
The complete constitutive equations for the magnetic and electric
fields are:
B = µ(H + M ) (A4)
D = εE + P (A5)
where, M is the magnetization vector (A/m), P is the electric
vector polarization (C/m2 ). But remembering that the magnetization
and polarization are properties of the medium, we assumed for
simplicity that the human brain produces a negligible effect. Using
the constitutive Equations (A4) and (A5) in (A1) we obtain:
∂(εE + P )
5 × (µ−1 B − M ) − = σ(E + ν × B) + J e (A6)
∂t
Using Equations (A2) and (A3) in Equation (A6), it then transformed
into:
¡ ¡ ¢ ¢
¡ −1 ¢ ∂ ε − 5 V − ∂A ∂t + P
5 × µ 5 ×A − M −
µ ¶ ∂t
∂A
=σ −5V − + ν × 5 × A + Je (A7)
∂t
120 Rojas and Rodriguez

Considering a non-magnetic (M = 0) and non-polarized (P = 0)

medium, a non-moving geometry (ν = 0) and no external electric
gradient potential (V = 0, this is the ground boundary condition
used for symmetry boundaries) and reordering terms of the equation,
Equation (A7) becomes:
∂A ∂2A ¡ ¢
σ + ε 2 5 × µ−1 5 ×A = J e (A8)
∂t ∂t
A time-harmonic function using the Euler’s formulation can be written
in the form: A(r, t) = Re [A(r) exp j(ωt + φ)], where φ is the
phase angle, and A(r) exp jφ is the phasor, using this function in
Equation (A15) we obtain:
¡ ¢
ω(jσ − ωε)A + 5 × µ−1 5 ×A = J e (A9)
However, A is the magnetic potential so 5·A = 0. Now, Equation (A9)
can be rewritten along the z -direction
52 Az + ωµ(jσ − ωε)Az = µJze (A10)
The FEM does not actually solve Equation (A10). The numerical
solution corresponds to the values of known quantities at the the
nodes or edges of the discretised domain. The solution is finally
obtained by solving a set of linear equations using scalar functions (e.g.,
Equation (A10)) for each direction in the Cartesian plane and in their
corresponding integro-differential formulation. These equations are
solved either by minimisation or using the weighted residual method
(WRM) that requires of functions W (x, y) denominated weighted
functions. If we applied a weighted function to Equation (A10) and
using surface integral, it becomes
Z Z Z Z
£ 2 ¤
W (x, y) 5 Az −ωµ(ωε−jσ)Az dxdy = W (x, y)µJze dxdy
Ω Ω
(A11)
Using the vector identity 5 · (W 5 Az ) = W 52 Az + 5W · 5Az
and Gauss’ theorem can help us to simplify equation (A.11) which can
be rewritten as follows
Z Z
[5W · 5Az + W ωµ(ωε − jσ)Az ] dxdy
Ω
I Z Z
∂Az
= W dl − W µJze dxdy (A12)
C ∂z Ω

FEM assumes that the electric field vanishes at the boundary of

the region of interest, so the solution only considers the interior of the
Progress In Electromagnetics Research M, Vol. 7, 2009 121

region of interest. To find the numerical solution of Equation (A12),

for the electric field, a Azi value is assigned for each internal node,
where i denotes the nodal number (see Fig. 2). Additionally, it is
assumed that the field in each element is a linear extrapolation of the
three elements compromising the node. From these assumptions, the
potential Az can be written in a discrete formula
N
X
Az (x, y) = Ni (x, y)Azi (A13)
i=1

where N is the total number of internal nodes and Ni (x, y) is the

expansion nonzero function associated with node, i. Equation (1) can
be derived using Equations (A12) and (A13).

REFERENCES

1. Hoult, D. I. and R. E. Richards, “The signal-to-noise ratio of

the nuclear magnetic resonance experiment,” J. Magn. Reson.,
Vol. 24, 71–85, 1976.
2. Hoult, D. I., “Sensitivity and power deposition in a high-field
imaging experiment,” J. Magn. Reson. Imaging, Vol. 12, 46–67,
2000.
3. Ocegueda, K. and A. O. Rodriguez, “A simple method to calculate
the signal-to-noise ratio of a circular shaped coil for MRI,” Con.
Magn. Reson., Part A, Vol. 28A, 422–429, 2006.
4. Jin, J., Electromagnetic Analysis and Design, CRC Press, Boca
Raton, 1985.
5. Li, B. K., F. Liu, E. Weber, and S. Crozier, “Hybrid numerical
techniques for the modelling of radiofrequency coils in MRI,”
NMR Biom., in press, 2009.
6. Hand, J. W., “Modelling the interaction of electromagnetic
fields (10 MHz–10 GHz) with the human body: Methods and
applications,” Phys. Med. Biol., Vol. 53, 243–286, 2008.
7. Rojas, R. and A. O. Rodriguez, “Finite-element electromagnetic
simulation of a volume resonator coil for MR neuroimaging,” 27th
IEEE EMBS Conf., 1659–1662, 2005.
8. Rojas, R. and A. O. Rodriguez, “Numerical study of the optimal
geometry of MRI surface coils,” 29th IEEE EMBS Conf., 3890–
3893, 2007.
9. FEMLAB Reference Manual, 2nd Edition, COMSOL Multi-
physics, Burlington, MA, USA, 2005.
122 Rojas and Rodriguez

10. Gabriel, S., R. W. Lau, and C. Gabriel, “The dielectric properties

of biological tissues: III. Parametric models for the dielectric
spectrum of tissues,” Phys. Med. Biol., Vol. 41, 2271–93, 1996.
11. Fenner, R. T., Finite Element Methods for Engineers, 38, The
MacMillan Press LTD, London, 1975.