NMR IN BIOMEDICINE

NMR Biomed. 2007; 20: 304–325

Published online in Wiley InterScience
(www.interscience.wiley.com) DOI:10.1002/nbm.1149

Review Article
Radio frequency coil technology for small-animal MRI

F. David Doty,* George Entzminger, Jatin Kulkarni, Kranti Pamarthy and John P. Staab
Doty Scientific Inc., Columbia, SC, USA

Received 14 June 2006; Revised 13 December 2006; Accepted 19 December 2006

ABSTRACT: A review of the theory, technology, and use of radio frequency (RF) coils for small-animal MRI is presented. It
includes a brief overview of MR signal-to-noise (S/N) analysis and discussions of the various coils commonly used in
small-animal MR: surface coils, linear volume coils, birdcages, and their derivatives. The scope is limited to mid-range coils,
i.e. coils where the product (fd) of the frequency f and the coil diameter d is in the range 2–30 MHz-m. Common applications
include mouse brain and body coils from 125 to 750 MHz, rat body coils up to 500 MHz, and small surface coils at all fields.
In this regime, all the sources of loss (coil, capacitor, sample, shield, and transmission lines) are important. All such losses
may be accurately captured in some modern full-wave 3D electromagnetics software, and new simulation results are
presented for a selection of surface coils using Microwave Studio 2006 by Computer Simulation Technology, showing the
dramatic importance of the ‘lift-off effect’. Standard linear circuit simulators have been shown to be useful in optimization of
complex coil tuning and matching circuits. There appears to be considerable potential for trading S/N for speed using phased
arrays, especially for a larger field of view. Circuit simulators are shown to be useful for optimal mismatching of
ultra-low-noise preamps based on the enhancement-mode pseudomorphic high-electron-mobility transistor for optimal
coil decoupling in phased arrays. Cryogenically cooled RF coils are shown to offer considerable opportunity for future gains
in S/N in smaller samples. Copyright # 2007 John Wiley & Sons, Ltd.
KEYWORDS: small-animal MRI; surface coils; circular polarization volume coils; phased arrays; low-noise preamplifiers;
decoupling

INTRODUCTION software tools have been available commercially from

several vendors for at least 5 years for applications where
The radio frequency (RF) coil is recognized as being of coil losses are not too important, whereas the available
critical importance for obtaining maximum signal- options are still not always fully satisfactory for
to-noise ratio (S/N) in MRI, so naturally many papers mid-range coils.
have appeared on this subject. However, the vast majority Another distinction between many clinical MR volume
of articles on MRI RF coils over the past two decades coils and those for small-animal research is that the
have focused on large coils, where sample losses former are often intended for a rather narrow range of
dominate, or on microcoils, where sample and capacitor loadings, whereas the latter are often desired to operate as
losses are negligible. Few have addressed the mid-range optimally as practical from the nearly unloaded condition
coils, seen in the majority of small-animal applications, to the case of being stuffed to the limit. This, along with
where all the sources of loss (coil, capacitor, sample, the higher operational frequencies, makes tunability more
shield, and transmission lines) are important. The market of an issue. Often the tuning/matching circuitry must be
for human MRI is more than an order of magnitude larger somewhat remote from the coil, and this makes
than that for small-animal (or pre-clinical) MRI, so it is optimization complex.
not surprising that advanced and generally sufficient The focus of this (somewhat pedagogical) technology
review is on mid-range coils, where the product (fd) of
*Correspondence to: F. D. Doty, Doty Scientific Inc., 700 Clemson the frequency f and the coil diameter d is generally in the
Road, Columbia, SC 29229, USA.
E-mail: [email protected]
range 2–30 MHz-m. This would include, for example,
Contract/grant sponsor: NIH; contract/grant number: R44 mouse brain and body coils from 125 to 750 MHz, rat
EB000445-03. body coils up to 500 MHz, and small surface coils at all
Abbreviations used: BHP, balanced high-pass; CF1T, center-fed fields. Coils in this range, except near the upper end, are
one-turn; CP, circular polarization; CST, Computer Simulation Tech-
nology; E-PHEMT, enhancement mode pseudomorphic high electron often perceived by research managers to be relatively
mobility transistor; MWS, Microwave Studio; NF, noise figure; pre- simple to optimize and build. Hence, many MR research
amp: preamplifier; RF, radio frequency; SENSE, sensitivity encoding;
S/N, signal-to-noise ratio; SQT (Doty), symmetric quarter turn; T/R, programs have electronics and machine shop facilities
transmit/receive. devoted to building such coils.
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 305

We begin by presenting a brief overview of MR S/N single resistor R0 of temperature Tn in series with a
analysis, followed by brief discussions of the various coils lossless sample coil and capacitor, then for given MR test
commonly used in small-animal MR (surface coils, linear conditions (sample, B0, T%2, TS, and method), the following
volume coils, birdcages). We include some new simulation equation can be derived from either eqn (1) or the
results on surface coil optimization and show the principle of reciprocity (2–6):
importance of the often overlooked ‘lift-off effect’.
B1 VS
Finally, we briefly discuss advanced pseudomorphic S=N / pffiffiffiffiffiffiffiffiffi ; (3)
high-electron-mobility transistor preamplifiers (preamps) i R0 Tn
and mismatching requirements for phased arrays from a where i is the sample coil current and B1 is the mean RF
numerical circuit simulation approach and potential future CP field generated within the sample by current i. As the
progress from the use of cryogenically cooled RF coils. Johnson noise voltage is proportional to (R0Tndf)1/2, the
signal voltage, from eqn (3), is proportional to B1VS/i.
This expression for signal voltage is valid irrespective of
THEORY the noise source temperature (5), but it is often not
particularly convenient. Equation (3) is easily cast into the
S/N in complex coil circuits following form, which is more useful in practical probe
design and evaluation, where power is dissipated in
The dependence of the S/N following a single 908 pulse numerous losses of uniform temperature, Tn, in a complex
when all the resistive losses (coil, capacitors, sample) are circuit:
at the same temperature, Tn, can be expressed in a number B1 VS
of ways. We have found the following often most useful in S=N / pffiffiffiffiffiffiffiffiffiffi ; (4)
coil design (1,2): P T Tn
" pffiffiffiffiffiffiffiffiffiffiffi#" pffiffiffiffiffi# where PT is the total transmitter power required to
h!2 2pm0 nS g Ix ðIx þ 1Þ T2% generate B1 (power applied at the disconnected preamp
S=N ¼ 3=2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi port, so that transmission line losses are properly
12 kB TS Tn þ TP
included). Both eqns (3) and (4), usually without the
& ðhE hf QL VS Þ1=2 v3=2 (1) Tn in the denominator, are commonly referred to as
statements of the principle of reciprocity. (Note that for
where ! h is Plank’s constant divided by 2p, m0 is the
linear polarization, B1 is half the peak RF field strength.
permeability of free space, kB is Boltzmann’s constant, ns
Also, the B1 field strength is often expressed as the
is the number of spins at resonance per unit volume, g is
reciprocal of the pw90, the time required to rotate the
the magnetogyric ratio, Ix is the spin quantum number, T %2
magnetization 908.) However, the above expressions fail
is the effective spin-spin relaxation time, TS is the sample
when the various losses are at significantly different
temperature, Tn is the probe noise temperature, TP is the
temperatures. Clearly, one cannot ignore the Tn factors in
effective preamp noise temperature, hE is the RF
cryo-probes, which are now in widespread use for liquids
efficiency (the fraction of power dissipated in the sample
NMR and appear to be coming soon for solids NMR and
coil), hF is the magnetic filling factor of the sample coil,
micro-imaging.
QL is the matched, loaded, circuit quality factor, VS is
Equation (4) is easily extended, in a more generalized
the sample volume, and v is the Larmor precession
statement of the principle of reciprocity, to handle
frequency, gB0.
complex circuits where various losses are at different
The primary problem with eqn (1) is that it has not been
temperatures as follows (7):
easy to calculate the magnetic filling factor with good
accuracy except for very simple cases until rather recently. B1 VS
S=N / pffiP
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; (5)
Filling factor is traditionally defined as the magnetic energy Pn Tn
in the transverse component of the magnetic field
throughout the sample divided by the total magnetic where Pn is the transmit power dissipated in the
energy, U, throughout all space (recall U ¼ I2L/2 for a nth resistance of temperature Tn when generating B1,
simple coil). However, for the concept to apply properly and the summation is over all resistances (sample, coils,
to circular polarization (CP), it is better to define hF in capacitors, shields, and transmission lines) in the circuit.
terms of the transverse rotating field component B1: One way to show this is to transform each loss into an
R 2 equivalent resistor, Rn, in series with the sample coil (7).
B dVS In large coils at high fields, the denominator summation is
hF ¼ S 1 : (2) dominated by the power dissipated in the sample, which
m0 U
is usually at 310 K for in vivo applications. As sample loss
If we assume that TP is negligible compared with Tn is proportional to the integral of sE2 over the full sample,
(a reasonable assumption with state-of-the-art tuned where s is the sample conductivity and E is the electric
preamps, except perhaps for phased arrays and cryop- field, optimization of coils for these cases boils down to
robes, as we will see below) and losses are confined to a minimization of the integral of E/B1.
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
306 F. D. DOTY ET AL.

In most mid-range coils, losses in the sample coil, Full-wave 3D electromagnetic software
capacitors (4), and transmission lines are also significant,
and the complete circuit must be optimized. Equation (5) It appears that most 3D software validations on MRI coils
can be cast into the following form, which is have not confirmed the accuracy of the software for
generally more easily related to the results from linear complex, mid-range coils where all types of losses are
circuit simulation software in the analysis of complex significant. In mid-2002, we carried out rather detailed
circuits: evaluations of three leading ‘full-wave 3D electromag-
netics’ (EM) packages and concluded that for most
vc VS mid-range NMR and MRI RF coil problems, Microwave
S=N / pffiffiffiffiffiffiffiffiffiffiffiffiffiP
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; (6)
LC VC Pn Tn Studio (MWS) 4.3 by Computer Simulation Technology
(CST) was better suited and more accurate than the other
where LC is the sample coil inductance, VC is the coil software we evaluated (HFSS by Ansoft, and XFDTD by
volume, and vc is the voltage generated across the REMCOM) (8). Others have more recently shown that
sample coil when pulse power PT is applied at the Ansoft HFSS also gives accurate results for some aspects
impedance-matched port. The LC in the denominator may of small-coil problems (9). The CST software is based on
be initially surprising. However, the derivation of the a discretized solution of the integral formulation of
above is straightforward for a specific sample coil type, Maxwell’s equations; hence, the method is referred to as
and it has been experimentally validated in numer- finite integration technique (FIT) (10). To solve these
ous experiments (7). Equation (6) shows that the relative equations, a calculation domain is defined enclosing the
S/N (for a given sample coil, T2%, etc.) in a complex application problem. An important part of obtaining
circuit [containing various coils, capacitors, and trans- accurate solutions with reasonable mesh sizes lies in the
mission lines between matching elements, sample coil, detailed handling of mesh elements containing several
transmit/receive (T/R) switch, and preamp] is indicated different materials, especially when one material is a
simply by the voltage induced at the sample coil by a lossy metal and another material is a dielectric. CST has
given power applied at the disconnected preamp input put considerable effort into optimizing the calculation of
port. the mean effective fields and losses within these
Thus, the RF ‘coil’ optimization problem consists of troublesome cells which generally cover most of the
two major parts: (1) minimization of the integral of E/B1 surfaces of conductors in complex structures or may even
for the sample coil (where the numerator is integrated be divided by several thin sheets of conductors. The
over the full sample, and the denominator is integrated discretized mesh equations can be solved either in the
only over the homogeneous field region); (2) maximizing time domain by a transient finite difference time domain
the efficiency of delivering RF power to the coil when (FDTD) approach or in the frequency domain using
viewed from the transmit perspective – even if the coil is second-order harmonic relations. Until quite recently, we
for receive only. The first task can only be fully addressed have usually used the transient method, primarily because
using full-wave software with effective algorithms for it provides a broad-spectrum solution but also because the
handling conductor surface losses, even though pertur- frequency domain solver was not well developed until late
bation methods permit accurate workbench measurement in 2006. We observe that the CST time-domain solver
of the magnetic filling factor (4) and QL may be easily usually gets the RF copper losses right within '10% (the
measured. The second task is best addressed using typical limit of our experimental accuracy) for thick
common linear circuit simulators, such as ARRL Radio conductors and wires, and also for foil conductors when
Designer, SPICE, Ansoft Designer, or GENESYS. the current densities are not too different on opposite sides
Although the simplified analytical approaches usually of the foil. However, where current densities are radically
presented in the professional literature are useful in different on opposite sides of foils, as often arises in
providing insights, we find the numerical tools to be birdcages for example, it has often underestimated copper
superior in practice – a point we will emphasize. losses significantly. Recent tests of the frequency-domain
It is necessary to keep in mind several assumptions in solver with tetrahedral meshing indicate it is both faster
the above analyses: (1) the preamp’s noise temperature is and more accurate for high-Q coil problems. The
low compared with that of the coil circuit; (2) T2% is not eigenmode solver is several orders of magnitude faster
adversely affected by coil magnetism – an issue that is than the other solvers, but it does not handle lumped
generally of no consequence in design of coils larger than elements, nor does it handle losses well.
80 mm but often becomes of critical importance for coils In all our simulations, the tuned coil is excited with a
smaller than 12 mm; (3) the frequency-domain filter broadband 50 V pulse source of Gaussian distribution,
bandwidth is equal to 1/(pT2%). Note that this noise 14.14 V peak, centered very near resonance, which
bandwidth may be established by exponential multipli- delivers 0.5 W to the coil when matched to 50 V. Also, the
cation of the signal in the time domain before the Fourier simulation space (which includes the coil, sample, and
transform or by signal processing (such as line broad- shield) has copper boundaries. The mode frequencies are
ening) after the Fourier transform. usually calculated within 2% of the experimental values
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 307

(even for double-resonance coils), and the calculated B1 ATC700B series were in the range 15–30 ppm, the
magnitude often agrees within 5% with the MR Cornell Dublier MCM series were in the 30–50 ppm
experiment. While the B1 magnitude can sometimes be range, and the ATC100B capacitors were in the range
off by up to 25%, the E/B1 integral is probably generally 5–10 ppm.
accurate within a few percent.

Detailed RF circuit modeling

RESULTS AND DISCUSSION
Figure 1 depicts a useful circuit model, with node
Surface coils reference numbers, for a commonly used T/R surface coil
with remote tune/match, and a photo of a typical coil is
The surface coil is widely used in MRI, as it is often a shown in Fig. 2. The sample coil LS is shown with a node
convenient and effective way of obtaining higher explicitly at its center so the voltage at this point is
localized S/N than can be obtained with volume coils, directly available to confirm that the coil is balanced. To
but it is important to appreciate that the advantages in
small-animal applications are often not as great as is seen
in human applications, where the volume coil is almost
always sample-noise dominated. Several excellent review
articles have appeared (11–13), along with many other
articles (14–22), several book chapters (23,24), and an
elementary book devoted to the subject of mid-sized MRI
coils (25). We will not attempt to repeat here much of
what is already quite accessible. Rather, our emphasis
will be on some implementation and optimization details
that are less well covered in the literature.
Surface coils become advantageous for MR when
sample losses are dominant and/or the region of interest is
very near the surface. A 12 mm surface coil, for example,
will be superior to a well-optimized (short) birdcage of
20 mm diameter for mouse brain 1H MRS at 7 T only for
features within 6 mm of the surface. At greater depths, the
small birdcage has higher S/N in addition to its strong
advantage in RF homogeneity. Here, coil (not sample) Figure 1. Circuit model for the balanced T/R surface coil
losses are dominant for both the surface coil and the with remote tune/match, as discussed in the text. Trans-
birdcage, but the surface coil achieves higher S/N over a mission lines are designated by TRLx, match capacitors
by CMx, balance capacitors by CBx, and tune capacitors
small region near the surface because of its higher filling by CTx.
factor.
One usually prefers to use the surface coil as
receive-only if possible, as more efficient pulse-sequence
options are available with the uniform excitation field of a
body transmit coil. However, there are times where there
is no space for a suitable body coil, or perhaps it is
preferable to restrict the transmit region to the receive
region for specific absorption rate (SAR) reasons, as in
double-resonance cross-polarization or high-power
1
H decoupling. For these cases, the T/R surface coil is
appropriate.
For surface coils smaller than '50 mm, it is important
that their capacitors have low magnetism. In most cases, it
is sufficient to simply specify that the standard nickel
barrier in the capacitor terminations be omitted, but with
very small coils it may be necessary to be even more
discriminating. We recently measured the mean bulk
susceptibility, x, of a number of ‘non-magnetic’ (nickel- Figure 2. A 20 mm balanced T/R surface coil with heavy,
magnetically compensated, parallel conductors. The thin
free) chip capacitors. The susceptibilities of the TEMAX acrylic coating over all the parts is not obvious, and the
series CHB capacitors were in the range (4 to 0 ppm (E-6 remote tune/match network, included in Fig. 1, is not shown
SI volumetric units), whereas susceptibilities of the here but is at the remote end of the feed cable.
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
308 F. D. DOTY ET AL.

reduce the clutter, stray capacitances are not shown coefficient in the simulation from the actual value (for
explicitly, but they are always included appropriately in example, 0.145 dB/m for Belden 1855 at 200 MHz) to
the model. The sample losses appear in the Q of LS. For a zero and noting the increase in the voltage across the
12 mm 300 MHz unsegmented coil, for example, a typical sample coil. (It is not straightforward to obtain the voltage
value (XL/RS) might be 240 unloaded and 100–180 data at nodes that are not ports in some simulators, such
loaded. (Remember that the matched QL of a circuit is as GENESYS, but there are usually work-arounds.
half the unmatched Q0 of a tank circuit, and the For example, very high impedance ports, which have
unmatched Q0 of a tank circuit would be half that of negligible load on the circuit, can be defined at internal
the coil and the capacitor separately if, for example, their circuit nodes where the voltage data are needed.)
Q values were equal.) As shown in eqn (6), the object of Likewise, the effects of the losses in all the capacitors
the circuit optimization is to achieve the highest possible and L1 are readily determined by changing their Q values
voltage across the sample coil (nodes 12 and 13) when from actual values to infinite and noting the change in the
excited with a given power at the 50-V port labeled RF. voltage across the sample coil. With a little experience,
All the elements are modeled with realistic Q values one can fairly quickly come up with appropriate values
and lead inductances. The matching and balancing is most for all the components that achieve the desired objectives:
conveniently achieved capacitively, so most, but not all, tuned to the desired frequency, nearly zero voltage at the
of the coil tuning capacitance, CC, is placed directly center of the sample coil, total S/N loss from all
across the sample coil with minimal lead length. A components other than the sample plus the coil itself
reasonable estimate of the quality factor QC of a low-loss less than 10% (half of which will probably be in CC and
ceramic capacitor of capacitance CP pF at f0 MHz is given perhaps a third in TRL1), no parasitic mode nearby, and
by the following: adequate tune and match adjustability with CT2
and CM2.
QC ) 1:5E07 CP(0:85 f0(1:35 ; (7) It should be emphasized that an incorrect choice of
(For example, a hi-Q 10 pF chip may have QC of matching capacitor CM1 could result in half the S/N
'450 at 500 MHz. The Q values of the zero- being lost in TRL1, even with the coil balanced and the
temperature-coefficient types, such as ATC700B, are circuit tuned and matched to 50 V by CM2. This problem
'20% lower.) The small tuning variable CT1 may need to is often exacerbated by the use of a rather light-weight RF
be nearly a centimeter away to prevent shimming difficul- line for improved flexibility (such as RG174, 0.32 dB/m at
ties, in which case the leads between it and CC may 200 MHz) for perhaps the first 15 cm connected to the
need to be accounted for. They are shown in Fig. 1 as coil. Note that lower losses for a given RF line diameter
transmission line TRL2, which, for example, may be a will usually be seen if TRL1 is a 75 V line, such as Belden
balanced 100 V line, 8 mm long. 1855, even though the RF preamp is usually optimized for
In the older circuit simulators, the input consisted of a a 50 V source.
‘node list’ – a list of the circuit elements with the node A common mode (with nearly zero efficiency) will
numbers to which they were connected, along with always be present, and often it will initially be not too
additional parameters giving relevant characteristics, far from the desired differential mode. Its location will
such as propagation velocity, length, Q, etc. In the more usually be fairly well predicted if the stray capacitances in
recent simulators, the input is simply the circuit the coil model (from nodes 11, 12, and 13 to ground) are
schematic, such as Fig. 1, with descriptions of each of correct (often about 2 pF at node 11 for a 2 cm coil). Its
its elements. position is dependent on just about everything except CC
Although most clinical coils are not user tunable, it is and CT1, so it can usually be moved well away from
generally useful to have some remote tune/match adjust- the efficient mode without too much difficulty. These
ment capability in small-animal applications, as the complications emphasize the need for the kind of detailed
frequency of the surface coil can be shifted significantly circuit analysis that is best handled with standard circuit
by changes in its location relative to the transmit coil, simulators rather than analytical closed-form solutions,
sample, or shield. Sufficient remote tunability may be which usually include unrealistic simplifications. How-
achieved without significant loss in S/N if the voltage ever, there are limitations in this model, primarily because
standing wave ratio (VSWR) on the cable to the surface there is no attempt to include electric couplings to the
coil (transmission line TRL1) between the surface coil body transmit coil. As a result, common mode problems
and the remote tune network is kept moderately low. A may still arise that may be able to be addressed with cable
coil L1 in parallel with remote tune variable CT2 tuning traps (18) – a subject we will return to below.
out about half of the maximum value of CT2 allows a plus We avoid using coupling loops (in place of capacitive
or minus reactance adjustment and often doubles the balancing) because they are much more difficult to model
tuning range for a given loss in efficiency. Keeping the accurately, are less convenient with active decoupling,
length of TRL1 roughly equal to nl/2 is also often and usually offer no advantages. We also do not find it
helpful, although not essential. The effect of the loss in beneficial to use a balanced pair of lines for TRL1, as has
TRL1 on S/N is readily seen by changing its attenuation been recommended in the literature.
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 309

Surface coil decoupling of the sample coil, and current in LS is minimized during
transmit, as desired to minimize its effect on the trans-
Although the use of separate transmit and receive mit field. One can sometimes dispense with L2 and L3
coils permits uniform excitation with better local S/N, and have adequate decoupling with the diodes placed
the interactions between the coils often leads to problems. directly across CC, but usually this approach is not as
If the receive coil is not ‘decoupled’ during the transmit satisfactory (15).
pulse, its resonance dominates and destroys the hom- Figure 3 further illustrates the need for good models
ogeneity of the transmit field (15). (Note the distinctly and accurate circuit simulations. The efficiency of this
different meaning here for ‘decoupled’ from its more circuit is seen to suffer badly if L2 and L3 are not small
common meaning in double-resonance NMR spec- compared with LS or if CCþCT1 is not large compared
troscopy.) With a linear transmit coil, the coils may with CB1 and CM1. Because of the rather large off-state
sometimes be adequately decoupled by orienting the capacitance in the diodes and other parasitics (again, not
surface coil so that its B1 is orthogonal to that of the shown), efficient conditions cannot be achieved above
transmit coil. Another approach that often works well '200 MHz with a surface coil larger than '20 mm
(when there is adequate space) is the self-shielded ('20 nH) unless it is capacitively segmented, i.e. a
surface coil, also called a gradiometer or quadrupolar capacitor (CS) must be inserted opposite CC to negate
coil (22), consisting of two loops, one above the other, about half of LS. For the T/R coil of Figs 1 and 2, one can
with oppositely directed currents. Such a coil also has a go to considerably higher inductance and frequency
reduced depth of field, which is often a disadvantage, but before CS is required. However, the segmenting capacitor
it can have higher S/N because of reduced far-field losses may be desired for reducing dielectric losses in the
when used on large samples (22). However, the most sample for fd greater than '5 MHz-m, as will be shown in
commonly used method of coil decoupling is detuning – the next section. When optimally implemented, this
either passively or actively. passive detuning circuit, shown also in Fig. 4, typically
Figure 3 illustrates an effective approach for passive degrades coil S/N by '5%.
detuning in small coils at high frequencies (15,20). The It was noted above that the switching diodes in Fig. 3
switching (PN) diodes D1 and D2 (we have recently used look like shorts during an intense transmit pulse.
Vishay type BAS16D, which have sufficiently low However, the transmit field may not have sufficient
magnetism) look like small capacitors ('1 pF each) intensity during its full duration for sufficient detuning
during receive and like shorts ('1 V) during an intense using the passive method of Fig. 3, so active detuning may
transmit pulse. The small coils (L2, L3) and very short be required. Active detuning is also useful in some B1
(4–10 mm) transmission line (TRL2) between the diodes field mapping techniques.
and CC are resonant with CC plus CT1 when the diodes An active detuning circuit is shown in Fig. 5 in which
are shorted. Thus, there is a high impedance at the the sample coil is segmented. A single PIN diode D1
ends of the surface coil LS (nodes 12, 13) in the absence (such as type MA4P505-1072, 500 V, 0.7 pF, 1 V at
10 mA, 1.5 W, 2 ms minority carrier lifetime) may be
used. Inductors LB1, LM1, and LM2 are used in
balancing and matching so that a ‘DC’ path is provided
for biasing the diode using a ‘DC’ detune current pulse
superimposed on the RF line. Other components, as in

Figure 3. Circuit model for the capacitively segmented, Figure 4. A 20 mm balanced surface coil with passive
passively detuned surface coil with remote tune/match. detuning. The diodes are hardly visible.
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
310 F. D. DOTY ET AL.

One method of making a cable trap has been coiling a

short portion of the RF cable (TRL1) into one or two
small turns (depending on the frequency), exposing the
shield at the ends of this coil, and placing a capacitor
across it that tunes it to the resonant frequency of the
surface coil. An approach that has several distinct
advantages is to place a coaxial resonator, tuned to the
frequency to be suppressed, around the cable (26). This is
essentially a compact, lumped-element version of the
(cumbersome) tri-axial ‘bazooka’ (27). An adjustable,
split version of the coaxial resonator has been patented by
IGC Medical Advances (28).

Surface coil optimization

Figure 5. Circuit model for the actively detuned segmented For coil fd below '4 MHz-m (e.g. a 14 mm coil at
surface coil with remote tune/match. 300 MHz), coil resistance losses are often dominant, so
minimization of these losses by the use of very heavy
Fig. 3, may again be useful in reducing losses in TRL1 or conductors is clearly beneficial. For example, two
moving a common mode. magnetically compensated (copper-clad aluminum)
As with the passively detuned circuit, CC must be large parallel wires (of 1.6 mm diameter), as shown in earlier
compared with the off-state capacitance of D1 plus figures, was found to give '30% higher S/N than a single
parasitics. Also, the magnitude of the reactance of CC 1 mm wire for a 20 mm 128 MHz coil. Here, the benefit
should be small compared with the magnitude of the was little more than half due to increase in QL. The
reactances of LB1, LM1, and LM2, all of which, of increase in magnetic filling factor hF from the reduction in
course, must be air core, as they must be non-magnetic the intense B1 near the conductor surface was nearly as
and be predictable in the magnetic field B0. Optimization significant. To understand how this can be consistent with
using a circuit simulator is still straightforward: simply eqn (3), as it is clear that the ratio B/i is not improved (but
set the losses in any component to zero and note the actually slightly degraded) by the addition of the second
increase in the voltage across the sample coil to see how parallel wire, one must recognize that the reduction in R is
much the S/N is being degraded by that component. With considerably greater than simply inverse with Q, as the
a little care, total S/N loss from all tuning components can inductance is also reduced by the addition of the parallel
be under 15% even for the small coil. wire, and R ¼ L/Q. [The point here is that eqns (1–6) all
While we have discussed ways to deal with the provide useful and valid insights.] Of course, for larger fd,
magnetic interactions of the transmit coil with the surface the reduction in coil resistance becomes less important,
coil, we have thus far looked at the surface coil circuit but the reduced inductance of the heavy coil continues to
models only in isolation from the transmit coil. Care is be beneficial, as it reduces near-field E/B1 within the
always taken to minimize and balance the electric fields sample and thus reduces sample losses.
from the transmit coil, but these electric fields and their A portion of the E field within the sample, the so-called
standing waves in the sample still often have significant ‘conservative E field’ (because it arises from the scalar
interactions with the surface coil and its cable. These potential, rather than from dB/dt), may be decreased by
interactions produce voltage and current signals on the reducing the coil’s inductance. Hence, multi-turn coils are
shield of the surface coil cable that would not be captured almost never optimal for fd above 2 MHz-m. Capacitive
in the above circuit models, which assume that the shield segmentation allows one to reduce the effective induc-
of this cable is everywhere at ground potential. The tance of a conductor element and thus sometimes reduce
common-mode shield current and voltage distributions the losses from the conservative E field. However, the
may be similar to what is seen on dipole antennas (18), series resistance of the capacitor is always greater than
characterized by the free-space wavelength. One that of a piece of copper of similar size, and the
approach to suppressing shield currents is to insert one concentrated E fields very near the capacitors may
or more cable traps (18), where the object is to insert a increase total sample losses. Accurate simulations or
higher impedance to common-mode cable signals with experiments are generally required to determine the
little effect on differential-mode signals. Cable traps may optimum.
also be a method of dealing with common-mode Table 1 presents some results from a number of
resonances that are captured in the circuit models simulations, using CST MWS 2006, of 16 mm coils at
presented above, although we find the circuit optimization different frequencies, with two different loadings, varying
methods described above more effective. segmentation, and several different air gaps between the
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 311

Table 1. Segmentation, spacing, and loading of a 16 mm surface coil

f (MHz) No. coil segments C (pF) QC Load Air gap (mm) QL Sample losses (%) H, A/m, @ 0.5 W
200 1 31 550 Light 0.6 148 22 69
200 2 62 300 Light 0.6 81 32 51
200 2 61 310 Mod. 1.6 93 20 60
300 1 13.5 640 Light 0.6 132 35 59
300 2 27 360 Light 0.6 78 48 42
300 2 28 350 Light 1.6 118 14 53
300 2 26.5 360 Mod. 1.6 105 31 50
500 1 4.5 820 Light 0.6 119 55 47
500 2 9.5 440 Light 0.6 105 39 42
500 3 14 310 Light 0.6 86 30 38
500 2 9.8 420 Light 1.6 131 26 46
500 1 4.4 820 Mod. 1.6 93 66 44
500 2 9.2 450 Mod. 1.6 76 54 40
750 2 4 530 Light 0.6 93 57 38
750 2 4.1 530 Light 1.6 119 43 42
750 2 4.2 530 Light 2.0 132 38 43
750 3 6.1 370 Light 1.6 100 34 38

coil and sample. In all cases, there were two parallel appropriate values are used for the capacitor resistances,
1.6 mm wires as shown in Fig. 6. The ‘light load’ is a (2) circuit losses external to the surface coil and its
sphere of 30 mM saline, 20 mm in diameter (4.2 g), with capacitors are ignored, (3) the coil and sample calculation
center '9 mm from the central plane of the two-wire space are enclosed in a large enough copper box to make
surface coil, as shown in Fig. 6. The ‘moderate load’ is a boundary effects negligible, and (4) the transient solver
sphere of 50 mM saline, 40 mm in diameter (33 g), with was used with no symmetry planes and adequate mesh
center '20 mm from the plane of the coil. The light load refinement in the vicinity of the coil. The H field
is somewhere between the loading of a mouse head and a magnitude (A/m, 2B1/m0) at 0.5 W excitation (tuned,
rat head, and the moderate load is midway between the matched, and balanced) is given on the z axis, 8 mm from
loading of a mouse and rat body. In all cases, (1) the center of the surface coil. As all the losses are at the
same temperature and all the coils are driven at 0.5 W,
the S/N for a given voxel size is simply proportional
to this calculated H. The calculated sample losses as a
percentage of total are also listed, which permits
estimation of the unloaded matched Q, QU. (Careful
NMR and MRI experiments on a number of cases similar
to those simulated here indicate the CST MWS 2006
software can generally be trusted to get the H field
magnitude right within '8% and the homogeneous mode
frequency within '2% for problems of this type.)
Note that performance is significantly degraded if the
space between the sample and the coil is suboptimal, i.e.
the ‘lift-off effect’ (22). In practice, coils in the 1–3 cm
size range are often mounted on a teflon substrate with
copper shielding patches between the capacitors from the
sample. The higher dielectric constant of teflon (2.1)
compared with air increases the spacing needed for
optimum performance, but the floating shield patches
may reduce the required space a little. These details were
omitted to speed up the simulations without significantly
altering the conclusions. Another difference between
the simulated cases and the common practice is the sol-
der joint connections, which may add about 15% to coil
losses. Also, the coil is usually given a thin acrylic
coating, which has negligible effect.
Figure 6. The surface coil and ‘light load’ simulation geo- The results show that there is often more to be gained
metry, as described in the text. The lumped element capaci- from increased spacing from the sample than from
tors are shown as small chamfered disks. increased capacitive segmentation. However, segmenta-
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
312 F. D. DOTY ET AL.

tion is often needed just to make a detuning circuit work losses are very dominant. However, when coil losses are
efficiently, as noted above. The simulations show that it is dominant, S/N from the combination would usually be
very important to have the coil spaced away from the lower, as the coil interactions decrease both the unloaded
sample; a teflon foam substrate of thickness 5–10% of the Q values and the filling factors of both coils.
coil’s diameter is normally optimum for the T/R surface The loop/D combination is also often used for double
coil, but a larger separation may be useful in further resonance (e.g. 1H/31P or 1H/19F). In many cases, it
minimizing electric field interactions that are more likely permits higher S/N at both frequencies than the
to be troublesome in receive-only applications. It has alternative of double-tuning a single loop, though many
recently been again suggested that coaxial loops offer methods to this latter approach have often been used (17).
some S/N advantage (29,30), but the arguments and data Coplanar, tightly coupled loops can also be well suited for
are not yet convincing. double resonance (32), although it can be difficult to
Perhaps the most surprising result is that the most obtain good efficiency on both coils (16). With either the
common coil-quality metric, QU/QL, is often misleading for multiply tuned single loop or the tightly coupled coplanar
surface coils (and, by analogy, for array coils). For example, loops, detailed circuit simulations (8) are generally
this ratio is approximately 2 for the second 300 MHz coil essential to achieve the desired balance of RF efficiencies
listed in Table 1, whereas it is under 1.2 for the third at the different frequencies.
300 MHz coil, both of which are with the same load. Yet the Catheter coils, intra-cavity coils, and surgically
second of these coils has 25% higher S/N, whereas implanted coils, like conventional surface coils, provide
estimation of the relative merit of these coils simply by their greatly enhanced local sensitivity compared with volume
QU/QL ratios would suggest the first has 30% higher S/N. coils (21,32,33). They have seen increasing use in human
This point can hardly be over-emphasized. The QU/QL ratio applications, including double-resonance prostate coils
is more often useful for homogeneous volume coils, but (32,34), and there are indications that they may begin to
even there it also is often misleading, as we note below when see use in small-animal applications.
comparing Litz coils with birdcages.
The RF magnetic field from the loop surface coil over
most of its effective region is normal to its plane. A pair of
back-to-back D coils (also called a butterfly coil) can be Linear-polarization volume coils
used to generate an RF magnetic field just below their
surface that is orthogonal to that of the loop surface coil Solenoids were the mainstay of NMR receiver coils in the
and, for proper orientation of the D coils, is also early days for a very simple reason: they permit the
orthogonal to B0(19,31). Hence, the combination of the highest S/N when sample losses are not dominant and B0
loop and pair of D coils, as shown in Fig. 7, can be used homogeneity is not critical (35). They also can achieve
for CP, which may improve S/N in cases where sample very high B1 homogeneity. With the advent of super-
conducting magnets, saddle coils supplanted solenoids
for most NMR spectroscopy for two compelling reasons:
they permit much higher spectral resolution, and they are
compatible with automatic sample exchange (35,36).
Perhaps the greatest advantage of the multi-turn saddle
coil (37) is that two of them can be oriented orthogonally
with excellent performance of each in double-resonance
experiments.
As magnets progressed to higher fields, it was seen that
the S/N disadvantage of the saddle coil compared with the
solenoid was not as great as was originally thought,
especially for large, single-tuned, 1H applications where
sample losses dominated. The Alderman–Grant coil (38),
as shown in Fig. 8, demonstrated that capacitive
segmentation was a very effective method of dramatically
reducing sample losses because the voltage builds up only
over half of the inductance of a similar one-turn saddle
coil before it is reversed by a segmenting capacitor. Also,
the quadrupolar symmetry of the conservative E field
reduces its average value throughout the sample.
Numerical optimization by Kost et al. (39) showed that
Figure 7. Quadrature surface coil from the combination of the optimum subtended angle of the window was about
a pair of D coils and a loop. The current direction is shown by 908 for best B1 homogeneity, although it still left much to
the arrows. be desired.
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 313

Figure 9. Primary foil pattern (side 1) for the basic CF1T Litz
coil. Crossovers on the back side of the laminate are partially
shown with solid lines, and tuning capacitors are placed
across the central gap.

the simplest, significant improvement on the Kost

optimization (39) of the slotted resonator (45), or
single-turn saddle coil (46), to which its inductance is
similar. There are two parallel paths on each side of this
coil with an insulated crossover (not fully shown) such
that each path, by symmetry, has the same inductance.
Hence, each path carries the same current, irrespective of
the azimuthal locations, axial locations, or frequency. The
azimuthal locations are chosen for a balanced optimiz-
ation of B1 homogeneity, hF, and Q. With typical foil
widths, the optimum mean azimuthal locations of the
inner and outer loops are 918 and 1568. Compared with
the Kost coil, the diameter of the homogeneous sample
volume (6% rms inhomogeneity) is increased from 65%
to 78% of the coil’s diameter, the Q value is increased by
'15%, and the coil has excellent transparency to
Figure 8. The Alderman–Grant saddle coil.
transverse RF flux. The filling factor is increased by
over 40%, primarily because of the large increase in
Both multi-turn saddle coils and solenoids continue to the homogeneous volume. As it is not capacitively
be the coil of choice for many applications, especially in segmented, its performance is usually suboptimal for
NMR spectroscopy (40,41). The micro-solenoid, which single-resonance, fixed-frequency applications above
has only very recently been well understood (8,40), is '5 MHz-m. However, it is often the best choice for fd
usually preferred for sample sizes below 3 mm; and the in the range 3–11 MHz-m when a wide tuning range is
multi-turn saddle coil is usually the volume coil of choice desired, such as 31P to 13C for mouse and rat applications
for fd in the range 2–6 MHz-m for double resonance, and at 4.7 to 14 T. (Recent simulations of more complex
sometimes as high as 10 MHz-m. The Alderman–Grant versions of this coil (43), with more parallel paths and
coil has often been used for the 5–15 MHz-m range, more crossovers, show that they offer little advantage.) Of
usually with a mean subtended angle of '1208 for course, there is significant capacitance between the two
improved orthogonal flux transparency when an orthog- parallel paths at the crossovers, and this introduces a
onal coil is needed for double resonance. parasitic high-frequency mode that can be near the
1
The primary limitations of the Alderman–Grant coil H frequency in double resonance. Also, as the coil has
are limited B1 homogeneity and poor transverse flux rather low inductance (for example, '26 nH for a short
transparency. These limitations may be solved by using 30 mm coil inside a large shield), lead losses can be fairly
parallel conductors with insulated crossovers in a way substantial if not properly addressed when multi-nuclear
that forces a more optimal surface current distribution tuning is desired. As with the surface coil, detailed RF
(42,43). At the same time, both QL and hF may be circuit modeling is useful. A balanced transmission line,
improved, and the capacitive segmentation may be such as a twisted pair, from the coil to the matching
extended for improved tuning with larger samples at network generally gives better results than other options.
higher frequencies. Coils with parallel conductors with The 24 mm linear coil shown in Fig. 9 achieves 20 ms
1
insulated crossovers have been denoted Litz coils. H pw90 for a hard 50 W pulse for mouse head imaging at
The primary surface pattern, as shown in Fig. 9 laid out 4.7 T, which is better than obtained with birdcages.
flat, is the basic version of what has been denoted the In-plane resolution, as shown in Fig. 10, of 175 mm is
center-fed one-turn (CF1T) Litz coil (42,43,44). This is readily obtained for a T2-weighted, spin-echo, multi-slice
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
314 F. D. DOTY ET AL.

600 MHz to 160 mm at 125 MHz. In some cases, they

have been tunable from 19F to 1H.
The primary disadvantage of the SQT coil is the extra
axial space required at the ends (needed to achieve the
reduced current concentrations there), as this prevents it
from being effective for head coils, whether for mouse or
man. Another disadvantage is that the segmenting chip
capacitors needed at the center of the coil may lead to
shimming artifacts for coils below 20 mm diameter. Yet
another drawback is that since much of the surface is
covered by the copper foil pattern, it is not as easy to see
the sample through the coil, a desirable feature in working
with small animals. This is not important in closely
shielded RF coil modules, as the shield naturally is
opaque, but the more convenient approach for small
animals in horizontal-bore magnets is a platform with
removable RF shield and full access to the coil and
animal. Hence, the birdcage, or a derivative as discussed
in the next section, is often selected even where the SQT
might provide higher S/N, such as for mouse liver up to
400 MHz and rat liver up to 200 MHz.
Figure 10. Live mouse brain at 4.7 T, linear Litz coil (Bio- The SQT coil has often been used for the 1H channel in
medical MR Laboratory, Washington University, St Louis, double-resonance for fd in the range 5–20 MHz-m with an
MO, USA). orthogonal CF1T Litz coil for the low-frequency channel.
Couplings between the orthogonal coils generally limit
experiment, no contrast agent, 128 & 128, 0.4 mm slice this approach to 25 MHz-m.
(12 nL voxel), TR ¼ 3 s, TE ¼ 20 ms, NEX ¼ 1.
It is possible to further improve B1 homogeneity and
extend the useful range by capacitive segmentation and CP volume coils
more parallel current paths in which the currents are
optimally controlled. Figure 11 illustrates the foil The birdcage is arguably the ideal volume coil for
patterns, laid out flat, for what has been designated the generating uniform CP, which generally allows a 40%
symmetric quarter turn (SQT) (Doty) Litz coil (the increase in S/N compared with linear polarization from
current paths execute a quarter turn around the B1 axis the same coil, as well as a factor of 2 reduction in SAR
before their inductance is capacitively negated). The (48–51). Because the birdcage and its common variants
homogeneous B1 region extends radially to 84% of the [low-pass, balanced low-pass (BLP), balanced high-pass
coil diameter and axially to the inside of the arcs on which (BHP), high-pass] have been discussed and analyzed in
the end-segmenting capacitors, CE, are mounted. (Note numerous papers over the past two decades (4,51–55), it
that the isolated closed loops in the ends of the central flux will not be reviewed here. Rather, we will mention only a
windows improve B1 homogeneity by reducing the peak few points that are of particular interest in small-animal
fields otherwise seen in these areas. Otherwise, the coil is applications and then look a little more closely at a recent
quite similar to one previously described in detail (42).) variant that is particularly useful for many small-animal
Both B1 homogeneity and S/N of this linear coil are often applications, the LitzcageTM.
better than that of the 16-rung CP birdcage for closely Much smaller capacitances are generally needed in
shielded cases with long samples (extending well beyond birdcages for small-animal applications than in common
the coil) with fd below 20 MHz-m (47). For example, a human applications. Consequently, the effects of stray
16 cm SQT coil at 3 T proved superior to CP birdcages capacitances are often greater in small-animal coils, and
and transverse electric magnetic (TEM) resonators of the problem is exacerbated by the fact that the Q values
similar dimensions (47). It is instructive to point out that are much higher. Accommodating wide ranges of loads
the unloaded to loaded Q ratio in the CP coils exceeded 3, presents challenges for small CP coils at high fields.
and still the linear coil achieved higher S/N. Another Tropp (53,55) has shown, both theoretically and
useful attribute of this coil is that it may be tuned over a experimentally, that symmetry in the birdcage is not
relatively wide range (compared with the birdcage) with critical for the case of the human head at 3 T and higher,
little degradation in B1 homogeneity or S/N, as it does not where the fd product is 30 MHz-m or greater, as in such
rely on capacitive phase shifts to achieve the desired cases dielectric resonance effects within the sample have
current distribution. This coil (or one very similar) has dominant effects on the field profiles. However, this is not
been used in numerous applications from 18 mm at the case for most small-animal applications, where fd is
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 315

Figure 11. Foil patterns, side-1, side-2, and superimposed, for the SQT Litz coil. End segmenting capacitors, CE,
and central segmenting capacitors CC1 and CC2, are placed as shown.

usually less than 20 MHz-m and often as small as 12-section birdcage, partly because it is possible to attach
5 MHz-m (mouse at 200 MHz). For such cases, tuning two adjustment variables to nodes at 458 with respect to
symmetry is quite important. the feed planes, which simplifies the symmetrization
Even with perfect symmetry, at least 12 rungs are problem when tuning to different loads. Corrections in the
generally required for adequate B1 homogeneity in a 12-rung birdcage, on the other hand, tend to mix more
closely shielded small birdcage when a relatively large with all tune and match adjustments, which complicates
region of uniformity is needed. Such coils typically have the process. While the 458 nodes are available in the
an easy tuning range of less than 1% with good 16-section birdcage, it has twice as many distinct
homogeneity and channel separation, while sample capacitors and usually about half the tuning range.
tuning shifts can be as large as 8% for the small, heavily Crozier et al. (56) observed that capacitor losses are
loaded birdcage. However, the eight-section birdcage is usually rather significant in small birdcages, and for this
about twice as robust (tunable and correctable) as the reason the small eight-section coil (at 200–750 MHz)
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
316 F. D. DOTY ET AL.

usually has higher Q and S/N than the 12- or 16-section The TEM coil is a favorite with many researchers in
small coil, an observation confirmed by many others (57). high-field human applications (53,60), and the strip-line
They also showed that a significant increase in the usable variant has been used in some small-animal applications
fd limit and Q of the eight-section high-pass birdcage (61). A recent analysis using multi-conductor trans-
could be obtained by using two bands in parallel in each mission line theory achieved remarkable agreement
section rather than a single wide rung (56). With a single between experiments and theory in predicting unloaded
wide rung, most of the current flows near the edges, so mode frequencies for a number of linearly driven cases,
removing the copper from its center has little affect on including 7.5 cm at 200 MHz and 13.4 cm at 300 MHz
copper losses, while it reduces electric field couplings to (62). Apparently, it is quite difficult to achieve
the sample and thus improves the Q. Still its homogeneity satisfactory quadrature operation in such coils for
is that of the eight-rung coil, and it is degraded by the fact small-animal applications with slotted RF shields. This
that the current in a birdcage always crowds to the worst is at least partly because the inhomogeneous modes are
side of the rungs. For example, when the rotating B1 is closer than in the birdcage and its derivatives (53). Our
aligned with x, the current crowds to the edges of the experience suggests that this difficulty in achieving clean
rungs closer to the xz plane. When the phase is aligned quadrature tuning (at least for fd >12 MHz-m) also often
with y, the current crowds to the edges closer to the yz extends to the balanced low-pass birdcage, which is
plane. In an eight-section birdcage, the effect is quite topologically similar to the TEM and strip-line coils.
pronounced. While additional modes not present in the isolated BHP
Varian has demonstrated that small birdcages with even birdcage also appear when it is coupled and matched, the
more than 16 rungs can be produced by integrating the matching methods presented below are more robust and
capacitors into the double-clad low-loss laminate (58). predictable with this topology.
These MillipedeTM coils have demonstrated exceptional The homogeneity of the Crozier coil may be improved
B1 homogeneity in small coils with mineral-oil samples by inserting an insulated crossover at the center of each
(58). They have also demonstrated impressive images in a pair of rungs, as shown in Fig. 12, in what has been
multiple-mouse application with high dosage of contrast denoted the Litzcage (63,64). The central crossover (by
agent at 7 T (59), but details of the RF performance of the symmetry) forces the current to be equal in each of the
current products are not readily available. One advantage two parallel paths in that section, which gives a significant
of this approach is that it reduces the susceptibility improvement in B1 homogeneity over the alternative
artifacts from chip capacitors in very small coils. eight-section CP coils (either a single wide rung or the
Another recent avenue being pursued by Varian is Crozier version), in which the current concentrates on the
(effectively) ultra-thickening of the conductors in the side closer to the rotating B1 axis. With the crossovers
conventional birdcage by bending and extending the between the parallel paths, the currents are always equally
conductor foil edges in the radial direction (57). They divided between the two parallel paths, and B1
report an eight-rung, 550 MHz, 7 mm example in which homogeneity is improved for cases well below dielectric
the radial thickness of the conductors is 30% of the coil resonance conditions, i.e. for fd less than '20 MHz-m.
inner radius r1, and the shield radius r2 is 3r1. Here, Figure 13 compares the calculated B1 map for the Crozier
compared with a reference thin-foil 12-rung case with the coil and Litzcage for a 25 mm coil at 300 MHz.
same r1 and r2, a 50% improvement in Q with a 20% loss From an RF circuit perspective, the homogeneous
in filling factor (for a constant sample volume) was mode is almost indistinguishable from that of Crozier’s
obtained, giving a 15% gain in S/N, but with a substantial parallel-rung eight-section birdcage (56), which is of
loss in B1 homogeneity. Similar approaches have been course quite similar to the conventional birdcage.
discussed and evaluated by various researchers (43), but However, the 30% reduction in stray capacitance in the
usually there is not enough space available between the Litzcage (relative to the eight-rung birdcage) allows it
coil and an outer coil or the shield for such an approach to to tune '15% higher. The Litzcage has homogeneity and
be practical. S/N at least as good as that of the ideal 12-rung birdcage

Figure 12. An approximation of one surface of the high-pass Litzcage foil pattern.
Insulated crossovers are shown with solid lines.
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 317

Figure 13. B1 field in the central xy plane for the Crozier coil (left) compared with the Litzcage (right) for
low fd ( f ¼ 300 MHz, d ¼ 25 mm). The mean field strength is '10% higher for the Litzcage, and
inhomogeneities near the rungs are less. Contour increments are approximately 2.5% of the central value.
while retaining the tuning ease and robustness of the (the range that keeps the loaded peak-to-peak relative
eight-rung birdcage. The eight-section Litzcage has been rung current errors below 15%) is very tight for two-point
used at fd up to 41 MHz-m (20.5 cm, 200 MHz), but the quadrature drive in small, high-frequency birdcages. For
16-section BHP birdcage is generally a better choice an eight-rung BHP birdcage, mean capacitor value
beyond 25 MHz-m, especially in larger coils, where the accuracy must be within 1.5%. A short 18 mm coil of
extra modes caused by the crossovers can get in the way. this type (for mouse brain studies) at 750 MHz requires
Also, with the central crossovers, accurate 3D full-wave tuning capacitors of '3.9 pF – including stray, which
simulations are much more computationally intensive. varies from 0.2–0.5 pF, depending on the sample. Hence,
1=2
As discussed above, S/N is proportional to B1/Pi . the stray variability exceeds the required tolerance by
(This also is one of the best methods of evaluating the more than a factor of two, which makes this coil with
accuracy of MRI RF coil simulation software.) NMR two-point-drive problematic. Moreover, the maximum
measurements on a 21 mm diameter, 20 mm length, useful tuning range for a small eight-section BHP
750 MHz Litzcage yielded a 908 pulse length of 22 ms for birdcage with standard two-point-drive is '1.3%, well
a square 50 W pulse on a pure water sample in an 18 mm under typical sample-induced tuning shifts. Four-
diameter NMR tube with a QL of 100 (65), which was point-drive networks can achieve up to 5% tuning range
lower than expected for this coil. We suspect that higher with good symmetry and efficiency, but they have more
than expected RF eddy current losses in the external RF parasitic modes, so it is quite useful to have a good circuit
shield contributed to the discrepancy. The external, model to be better able to deal with such.
gradient-transparent, RF shielding is seldom recognized Figure 14 illustrates a simple circuit model that often
as a significant source of signal loss, but in fact that can be gives the accuracy needed for the small, eight-rung, BHP
the case for small-animal coils with closely spaced birdcage or Litzcage. To represent the nearest-rung
external shields. We have found that the standard method couplings (LC), each rung includes two ideal transfor-
(overlapping slotted shields on double-clad Duroid mers, one on either side of the central plane, e.g. rung
laminate) can add very high losses under some conditions. 2 includes two transformers, {2, 12, 3, 13} and {13, 23,
Lower shield losses can often be obtained using 14, 24}. A transmission line (TRL) at each end of each
single-layer gapped foil with discrete chip capacitors rung completes its self-inductance and furnishes most of
across the gaps in the regions where the azimuthal- the stray capacitance per rung, (e.g., {11, 12} and {14,
RF-current densities are high. For an excellent study on 15} in rung 2). All of the major parasitics are included.
shield slotting requirements, see Ref. (66). Most of the losses appear as corrected attenuation
coefficients in the TRLs representing the rungs. Appro-
priate values for the characteristics of the TRLs and the
RF circuit models for small, tunable, CP coils rung couplings LC can be determined adequately by
conventional methods.
Several factors conspire to make it difficult to achieve One channel of an effective quad-balance network is
high B1 homogeneity in small-animal coils. It is shown in Fig. 15. The two series quarter-lambda’s
sometimes difficult to achieve the necessary precision force the needed symmetry, greatly improving tuning
in the capacitor mounting, or in the uniformity of the range with good symmetry and making it easier to achieve
shield spacing around the coil. The capacitor accuracy adequate channel isolation. The unlabeled capacitors are
required to place the resonance within the tuning range simply eddy-current-blocking capacitors (RF shorts). Lm
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
318 F. D. DOTY ET AL.

Figure 14. A useful RF circuit model for the eight-section small-animal BHP birdcage. Elements between nodes 1 and
5 correspond to rung 1, between nodes 11 and 15, rung 2, etc. Nodes 1 and 5 on the right are connected to the same on
the left.

is used to move the common mode well away from the and physiological monitoring systems (such as those
differential mode. LT tunes out half of the sum of the made by SAII, Stony Brook, NY, USA; http://www.i4-
tuning variable CT and the mean match variable CM, sa.com/index.htm) may be set up, possibly along with
thereby doubling the useful tuning range. The half- surgical procedures (64), on the live animal. The
lambda feed line allows placement of the variable assembly slides inside an external RF shield.
capacitors well away from the coil for maximum A 25 mm diameter 300 MHz Litzcage similar to
openness and access around the coil. With low-loss that pictured in Fig. 16 permitted 50 mm isotropic
coaxial lines, the total signal loss added by the balancing resolution (0.125 nL voxel) on four fixed mouse embryos
network is typically a few percent. An example for the simultaneously, one of which is shown in Fig. 17(67). It is
horizontal bore is shown in Fig. 16. Complex life support important to note that only slightly lower performance

Figure 15. One channel of a four-point-drive network.

Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 319

speed and in improving S/N, especially in the periphery

(68–70). Speed-up factors of 2 to 6 are commonly
achieved, and higher speed-ups have been achieved. A
number of techniques have been developed to take
advantage of the improved localized S/N from an array of
surface coils, and perhaps the most successful has been
SENSE (sensitivity encoding) and related techniques (71).
When sample losses are dominant, the phased array may
provide a significant increase in S/N near the surface. Also,
the speed-up is achieved without severe demands on the
gradient hardware (as in echo planar imaging (EPI)). Ideally,
the S/Nsen of the SENSE technique relative to the
conventional S/Nfull using a volume coil is given by

S=N full
S=N sen ¼ pffiffiffi (8)
Figure 16. Doty Litzcage for rat head in horizontal bore.
g R
where g is a coil geometry factor that is always greater
than 1, and R is the speed-up factor. In general, there will
be systematic errors in the coil sensitivity calibration that
further degrade S/N, and performance will also be
degraded by couplings between the coils. Various
methods of both inductive and capacitive decoupling of
adjacent coils have been utilized, but the most effective
general method is preamp decoupling, in which ultra-
low-noise preamps with high reflection coefficients and
special mismatching networks either detune or severely
over-couple the coils (72).
Speed-up has been demonstrated in rat brain imaging at
11.7 T using a four-channel rat head array (73) with a
novel approach to adjacent coil decoupling. There have
also been other demonstrations of phased arrays in MR
microscopy (74) and small-animal imaging (75), and the
major vendors have been supplying four-channel and
eight-channel high-field scanners for small-animal
applications for several years. SENSE and GRAPPA
(generalize auto-calibrating partially parallel acqui-
sitions) both offer clear advantages compared with EPI
at very high fields. Still, the general applicability of
phased arrays in small-animal applications seems some-
Figure 17. Mouse embryo with 50 mm isotropic resolution what limited, as resolution is usually paramount and is
obtained with a 25 mm Litzcage at 300 MHz using contrast
agent GSA-Gd-DTPA. Modified from Wadghiri et al. (67) often limited by S/N. This is particularly true in MRS
with permission. (76,77).
When two coils tuned to the same frequency are
would have been achieved with the linear Litz coil in this coupled, the mutual inductance causes the resonance to
case. The voxel volume reduction of two orders of split into two modes on either side of the original
magnitude compared with Fig. 10 was primarily from (1) frequency, and sensitivity at the original frequency is
the use of a high level of contrast agent, (2) the increased quite low. Roemer et al. (69) showed that preamps with
image acquisition time (here, 14.6 h for the complete 3D low input impedance may be used to reduce the resonant
data set, rather than a few minutes), (3) the use of a 3D current and coupling in surface coils for a particular type
gradient echo sequence, and (4) the higher field. of matching network (similar to that shown for passive
detuning in Fig. 3) in which a low preamp impedance is
transformed to a high impedance in series with the actual
Phased arrays surface coil. The effect is essentially equivalent to placing
a preamp of low input impedance between nodes 2, 3 in
In human MRI applications, phased arrays have proven to Fig. 3, or a preamp of high input impedance between
be extremely advantageous in trading excess S/N for nodes 12, 13. They note that when a half-lambda line,
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
320 F. D. DOTY ET AL.

practical transformation methods, and standard tuned solutions are beginning to appear. Local high-gain
MOSFET (metal-oxide-semiconductor field-effect tran- preamps with optical links are being investigated (81).
sistor) preamps are used, the effect on efficiency at It currently appears that the NF of a direct modulation
64 MHz is to insert about 5 V in series with the surface optical link is likely to be '29 dB, so a local coil preamp
coil. The majority of the resistance comes from the circuit with '40 dB of gain is needed to prevent significant
elements other than the field-effect transistor and thus has degradation in NF (81). Achieving the needed dynamic
the same effect on S/N as putting a real 5 V resistor in range in the high-gain, low-noise preamp is a major
series with the coil. This may be acceptable on a large challenge. Local coil preamps with all the control, power,
coil, but it can be devastating in a small coil where the and signal on a single coaxial cable are currently more
series resistance is initially '0.5 V. Simulations indicate practical and promise a substantial benefit by allowing
that the loss in a lambda/4 cable (type 1855) between a more flexibility in dealing with cable modes without these
maximally decoupled small surface coil at 200–300 MHz measures degrading S/N (82).
(26 nH in series with 10 pF tuning, no matching) and a The need for special preamps integrated into the array
typical ‘zero-impedance’ ('2 V input impedance) pre- coil package near the animal adds a substantial amount to
amp results in '2.5 dB loss in S/N in just the cable for its cost. Although this cost is clearly justified in many
light-load applications. The impedance transformations human imaging applications, other options may often be
and noise figure (NF) in commercially available preamps superior to phased arrays in small-animal coils. These
can lose another 2 dB. options include conventional CP volume coils with
Others have noted the limitations of Roemer’s initial improved optimization for the region of uniformity, more
approach to preamp decoupling for small coils and have fully optimized surface coils, and cryogenic coils. But
shown that there are better and more flexible schemes that having said this, there are nonetheless many places
simultaneously achieve low resonant currents in the where phased arrays provide substantial advantages in
surface coils and improved noise match (72), but the small-animal studies, and it is important to appreciate
analytical approaches still often include significant their requirements.
assumptions: lossless matching components and infini-
tesimal transmission line lengths. With small coils at high
fields it is important to fully include all losses in the
various transformations (at the coil, at the preamp input, Ultra-low-noise mismatched preamps
and there between) in detailed numerical circuit
simulations. To achieve significant S/N gains for many interesting
It is possible that some published performance gains cases – for preamp-isolated phased arrays, or for cryo-
with phased arrays of small coils have been based on genically cooled coils – ultra-low-noise preamps are
comparisons with suboptimal conventional coils. Clearly, needed, and perhaps at unusual input impedance.
there are many ways to inadvertently obtain performance The tuned preamps normally used in 1H MRI typically
from a small reference surface coil or volume coil that is achieve NF of about 0.4 dB (not including input line
below optimum; and having a small coil sample-noise losses) when noise matched, though long RF lines can
dominated may be a sign of excessive conservative E-field easily add 1 dB to the NF even when VSWR is low. The
losses (8), as seen in Table 1. cable losses are greater with phased arrays when the
There is a need for improved methods of assessing the reflection coefficients at both ends of the line are high.
performance of phased arrays, as g factors and noise Some appreciation for the real effects of mismatching and
correlations tell one only how quickly the coil degrades low loss transformations may be gained from a reported
with increasing speed-up factors. The g factor gives no case at 0.5 T (21 MHz). Here, the effect of a typical low
insight as to how the starting point performance (R ¼ 1) loss 908 hybrid and minimal nl/2 cable lengths (probably
compares with a well-optimized homogeneous coil. '5 m) was to reduce S/N by 3.3 dB for a factor-of-4
Moreover, as it is usually impractical to operate a phased impedance mismatch to a preamp with 0.5 dB NF (when
array in transmit mode, phased arrays have seldom been optimally noise matched) (83). In this particular case, the
evaluated from the perspective of reciprocity. The mismatch was not deliberate for coil decoupling but
importance of being more quantitative about S/N rather was a result of an absence of suitable coil matching
performance of phased arrays has recently stimulated circuitry.
several groups to develop improved methods of calculat- For the case of coil decoupling, much greater
ing ultimate attainable S/N for large MR phased arrays mismatching is required. For coil coupling coefficients
(78–80), where losses in the coils, capacitors, transform- of 0.01 (a typical value for adjacent coils without
ation networks, and preamps are all more easily made extremely careful inductive or capacitive decoupling)
negligible. with individual loaded QL of 50, a minimum of a
The common-mode or ‘cable’ problems, briefly factor-of-10 over-coupling would be needed for low noise
discussed above under surface coils, become more correlation between the coupled coils. Extreme over-
challenging with phased arrays, but more effective coupling places severe requirements on the preamp’s NF,
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 321

whether the coil is matched by parallel or series or mixed point in the line and all losses between there and the coil
elements (84). for the mismatched condition. (A complication in circuit
The lowest NF is currently obtained with enhancement- simulation of phased array coils is determining the circuit
mode pseudomorphic high-electron-mobility transistors component losses accurately for the severely mismatched
(E-PHEMTs), a fairly recent gallium arsenide device. The case, as the power actually absorbed at the excitation port
ATF-58143 has NFmin below 0.13 dB (at 258C, zero in the simulation will be a very small fraction of the
magnetic field) at frequencies below 500 MHz when incident power and it will be strongly influenced by
matched at the optimum (frequency-dependent) input component losses.) That impedance can then be used as
reflection coefficient (85). It also has very low the input port impedance in the preamp simulation, from
intermodulation distortion and saturated output power which its NF may be accurately determined for that
of 12 dBm. Accurate simulations of complex circuits source impedance. Figure 19 shows the NF and gain of the
containing active devices are readily carried out with preamp of Fig. 18 as a function of input resistance as
standard circuit software if their S-parameters and noise calculated by the GENESYS ‘core’ linear circuit
parameters are known at the frequencies of interest. The simulator for the case when the preamp input matching
pulse-protected simplified preamp circuit shown in is optimized for minimum NF with a 20 V source. Input
Fig. 18 can achieve NF under 0.3 dB NF with over absolute S11 (reflection coefficient) at optimum noise
20 dB gain at 300 MHz when noise matched for line match at 300 MHz is '0.3 and input impedance is 18 V
impedances of 25–100 V, but stabilization is more at 288.
complicated than suggested by this simplified schematic. The S/N loss in the preamp-decoupled coil is the sum
Equation (4) is still valid for the unmatched or detuned of the losses (in dB) in the coil mismatching circuit (with
coil as long as all the losses between the preamp and transmission lines) and the NF of the mismatched preamp.
the coil are included in the coil circuit simulation and the With proper optimization, this sum may be below 1.5 dB
preamp noise temperature is small compared with the even at high fields with adequate coil decoupling; but with
dominant coil noise temperature. The coil/preamp system insufficient attention to the circuit details, it can easily
optimization may be addressed by picking a convenient exceed 6 dB for small coils. Because of their higher
point in the transmission line between the coil and preamp Q values, the demands in the impedance transformation
for splitting the coil circuit problem from the preamp network can be an order of magnitude more stringent for
problem. Simulating the coil circuit (including trans- small coils than for large coils.
mission lines, matching, etc.) from that point to the coil The effective, over-coupled Q may be seen by
allows one to determine the coil circuit impedance at that combining (connecting) the two separate simulations

Figure 18. An ATF-58143 E-PHEMT 300 MHz pulse-protected preamp can achieve total NF under 0.3 dB (excluding
input cable losses). The source inductor L6 (4 nH) and the output RC network improve stability. Two sets of crossed
diodes with inductors L3 (13 nH) and L4 (25 nH) may be needed for adequate input pulse protection. Noise matching
is largely determined by C8 (2.2 pF) and L1 (65 nH).
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
322 F. D. DOTY ET AL.

Note that the simple alternative of reducing the

effective QL of the surface coil by an order of magnitude
by inserting a suitable resistor in series with it would
reduce the S/N by '10 dB, and applying Roemer’s
original (2n-1)l/4 approach with a preamp of very low
input impedance to small surface coils could result in a
6 dB (or more) loss in S/N, even though tuned preamps of
2 V input impedance are now commercially available
with fairly low NF for frequencies up to 300 MHz (86).
Clearly, it is essential to carry out detailed circuit
simulations of the coil/cable/matching/preamp system for
small phased arrays.

Cryogenically cooled RF coils

Bruker and Varian have been producing cryoprobes for

high-resolution NMR on liquid samples of 3–5 mm
Figure 19. NF and gain of the 300 MHz E-PHEMT preamp
of Fig. 18 as a function of source resistance when optimized
diameter for more than six years, and their factor-of-four
for 20 V. gain in S/N is revolutionizing the field of NMR (87).
Recently, similar, linear-polarization, 5 mm micro-
and applying a small ‘tickler’ signal (via a large resistor or imaging cryoprobes for use in vertical-bore 400 MHz
a small mutual coupling) to the sample coil. For adequate magnets have been introduced, and S/N gains up to a
decoupling, the coil/preamp circuit usually needs to show factor of four are seen with small, low-loss samples.
an effective Q at the output of the preamp (the S21 Bruker has also very recently announced preliminary
response, where S21 is the forward transmission results from a cryogenic RF platform for in vivo mouse
coefficient) below 5 – possibly below 2, depending on brain in horizontal-bore magnets that uses a cryogenically
coil coupling coefficients. (Note that the standard cooled high-temperature superconductor surface coil
‘impedance matched’ condition will often show an (88). They reported a factor of 2.4 gain in S/N compared
effective Q of nearly twice the coil’s QL because of the with a 20 mm quadrature room-temperature surface coil
high reflection coefficient of low-noise preamps.) at 9.4 T with active detuning. Other approaches for the use
There are a number of ways to over-couple the coil, but of high-temperature superconductor coils in MR micro-
some are better than others (72,84). For a small scopy have also been explored (9,89–91).
series-matched 300 MHz light-load 16 mm surface coil As noted above, the noise voltage is proportional to the
(with QL of 60), a factor-of-20 over-coupling (i.e. coil square root of the product of the resistance and
transformed to 2.5 V resistive at the preamp input port) temperature. Reducing the temperature of a copper coil
gave an effective QL of 8 at the output of a 50 V preamp and its RF shield from 300 K to 50 K reduces its RF
(similar to that of Fig. 18) with an NF of 2.8 dB at 2.5 V. resistance by typically a factor of 4, but the Q values of the
With a half-lambda low-loss cable (type 1855, 0.41 m) capacitors may not improve much. Hence, when sample
between the surface coil and the preamp, the losses in the losses are small, the total Q may improve by about a factor
cable are '1 dB. However, when the same coil is of 3. However, the noise temperature of the coil and
parallel-matched to 120 V resistive, a similar effective QL capacitor losses is also reduced by a factor of 6, so the
is achieved with the 20 V preamp of Fig. 18. In this total noise voltage from the coil, capacitors, and shield
case, the preamp NF (at 125 V) is '1.25 dB and cable may be reduced by more than a factor of 4. To take full
losses are only 0.15 dB. One can also use a series- advantage of the reduced noise temperature of the coil and
parallel-match, similar to that shown in Fig. 3, with capacitors, the preamp (with its input transmission line)
impedance 50 V at 858 phase angle, to get similar must have a noise temperature that is much less than 50 K
effective QL, but both cable losses and NF are (0.7 dB). Cooling the preamp to reduce its NF below
significantly increased compared with the parallel- 0.2 dB is generally justified; but it is just as important to
matched (high-resistance) case of a preamp of moderately position it very close to the coil, as 1 m of low-loss
low impedance. The use of l/8 rather than l/2 cables for cable at 300 MHz adds a comparable amount to the
the above, with proper adjustments in the tuning, reduces effective NF (7).
cable losses by about a factor of 3. However, the NF of Both experiments and simulations show that sample
gallium arsenide field-effect transistor and PHEMT losses are nearly comparable to coil and capacitor losses
preamps increases when placed in fields above '7 T, for the case of rat head 1H imaging at 400 MHz with a
so it may be preferable to use a full lambda cable length so short, 38 mm Litzcage or Crozier birdcage, so cryogenic
the preamp can be positioned in a region of lower field. cooling of the coil should increase its S/N by 40%. For
Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 323

mouse brain at 300 MHz, sample losses in a small REFERENCES

birdcage may be only 10% of coil and capacitor losses, so
a factor of 3 gain in S/N may be possible. Sample losses 1. Abragam A. Principles of Nuclear Magnetism. International
increase somewhat more rapidly than quadratically with Series of Monographs on Physics 32. Oxford University Press:
London, 1961.
frequency, so for 13C MR on rat brains at 9.4 T, there 2. Doty FD. Probe design and construction. In Encyclopaedia of NMR
appears to be the potential for a factor of 4 increase in S/ Vol. 6. Wiley: Chichester, 1996; 3753–3762.
N. Relative coil and capacitor losses in small double- 3. Hoult DI, Richards RE. The Signal-to-Noise Ratio of the NMR
Experiment. J. Magn. Reson. 1976; 24: 71–85.
resonance CP coils are typically 2–3 times those in 4. Doty FD, Entzminger G, Hauck C, Staab JP. Practical aspects of
single-resonance coils. Hence, there is even more birdcage coils. J. Magn. Reson. 1999; 138: 144–154.
potential for gain in S/N on 13C in double-resonance 5. Hoult DI. Sensitivity of the NMR Experiment. In Encyclopaedia of
1 13 NMR Vol. 7. Wiley: Chichester, 1996; 4256–4266.
H/ C applications using cryogenically cooled coils. 6. Tropp JS. Radiation damping and reciprocity in magnetic reson-
Undoubtedly, technical progress can be expected in this ance reception. Presented at 12th ISMRM. Poster 1646. Kyoto,
area in the future. 2004.
7. Doty FD, Shevgoor S. Beating reciprocity S/N expectations in
triple-resonance narrow-bore MAS by cryogenic cooling of critical
circuit components in the OptiMASTM probe. Poster presented at
CONCLUSION Rocky Mountain Conference, Denver, 2005. http://www.dotynmr.-
com/PDF/OptiMAS_ENC05.pdf
8. Doty FD, Kulkarni J, Turner C, Entzminger G, Bielecki A. Using a
The technological advances in mid-range MR RF coils of cross-coil to reduce RF heating by an order of magnitude in
the past two decades have helped fuel the rapid growth in triple-resonance multinuclear MAS at high fields. J. Magn. Reson.
the field of pre-clinical or small-animal MRI. In this 2006; 182: 239–253.
9. Brey WW, Edison AS, Nast RE, Rocca JR, Saha S, Withers RS.
regime, all the sources of loss (coil, capacitor, sample, Design construction and validation of a 1-mm triple-resonance
shield, and transmission lines) are important. All such high-temperature-superconducting probe for NMR. J. Magn.
losses may be accurately captured in some modern Reson. 2006; 179: 290–293. http://www.bme.ufl.edu/documents/
edison_research_article2_research.pdf
full-wave 3D electromagnetics software, and these tools 10. Weiland T. Time domain electromagnetic field computation with
permit improved coil optimizations. Standard linear finite difference methods. Int. J. Numer Model. 1996; 9: 295–319.
circuit simulators, such as SPICE and GENESYS, have Also see: www.cst.de
11. Ackerman JJH. Surface (local) coils as NMR receivers. Concepts
been shown to be useful in optimization of complex coil in Magnetic Resonance 1990; 2: 23–42.
tuning and matching circuits. Such simulators are also 12. Hyde J. Surface and other local coils for in vivo studies. In
useful for optimal mismatching of ultra-low-noise Encyclopaedia of NMR Vol. 7. Wiley: Chichester, 1996;
4656–4664.
preamps based on the E-PHEMT, as needed for coil 13. Haase A, Odoj F, Von Kienlin M, Warnking J, Fidler F, Weisser A,
decoupling in phased arrays. Nittka M, Rommel E, Lanz T, Kalusche B, Griswold M. NMR
Many coil applications that were considered quite probeheads for in vivo applications. Concepts in Magnetic Reson-
ance 2000; 12: 361–388.
challenging just a few years ago are now considered 14. Ackerman JJH, Grove TH, Wong GC, Gadian DG, Radda GK.
routine, but the relentless quest for increased speed and S/ Mapping of metabolites in whole animals by 31P NMR using
N continues to bring new challenges and new opportu- surface coils. Nature. 1980; 283: 167–170.
15. Edelstein WA, Hardy CJ, Mueller OM. Electronic decoupling of
nities. Both full-wave field simulations and linear-circuit surface-coil receivers for NMR imaging and spectroscopy. J Magn
simulations are likely to play an increasingly important Reson 1986; 67: 156–161.
role in future advances in MR RF coil technology, 16. Mispelter J, Tiffon B, Quiniou E, Lhoste JM. Optimization
of 13C-1H double coplanar surface-coil design for the WALTZ-16
especially in phased arrays and in cryogenically cooled decoupling sequence. J. Magn. Reson. 1989; 82: 622–628.
RF coils, both of which currently seem to hold more 17. Fitzsimmons JR, Brooker HR, Beck B. A Comparison of
promise than advances in magnet technology. While there double-tuned surface coils. Magn. Reson. Med. 1989; 10: 302–
309.
appears to be limited potential for technological advances 18. Peterson DM, Beck BL, Duensing GR, Fitzsimmons JR. Common
in room-temperature mid-range surface and volume coils, mode signal rejection methods for MRI: reduction of cable shield
there is considerable potential for trading S/N for speed currents for high static magnetic field systems. Concepts in
Magnetic Resonance 2003; Part B, 19B(1): 1–8.
using phased arrays, especially with larger samples. 19. Misic GJ, Hurst GC, Holland GN, Patrick JL, Orlando PT. Surface
Finally, cryogenically cooled RF coils have revolutio- coil system for magnetic resonance imaging. 324/318. U.S. Pat.
nized the field of NMR spectroscopy of liquid samples 4,920,318, 1990.
20. Hyde J, Rilling RJ, Jesmanowicz J. Passive decoupling of surface
over the past five years, and they have been shown to offer coils by pole insertion. J. Magn. Reson. 1990; 89: 485–495.
considerable opportunity for future gains in S/N in 21. Zhou X, Maronpot RR, Cofer GP, Hedlund LW, Johnson GA.
1 Studies on bromobenzene-induced hepatoxicity using in vivo MR
H MRI in small-animal brain studies and in low-gamma
microscopy with surgically implanted RF coils. Magn. Reson.
applications on larger regions. Med. 1994; 31: 619–627.
22. Suits BH, Garroway AN, Miller JB. Surface and gradiometer coils
near a conducting body: the lift-off effect. J. Magn. Reson. 1998;
Acknowledgements 135: 373–379.
23. Chen C-N, Hoult DI. In Biomedical Magnetic Resonance Tech-
nology. Adam Hilgar: New York, 1989; chapters 4 and 5.
This work was partially supported by NIH R44 24. deGraaf RA. In In Vivo NMR Spectroscopy: Principles and Tech-
EB000445-03 and by Doty Scientific, Inc. niques. Wiley: Chichester, 1999; chapter 8.

Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
324 F. D. DOTY ET AL.

25. Mispelter J. NMR Probeheads for Biophysical and Biomedical whole body magnetic resonance. In Encyclopaedia of NMR Vol. 2.
Experiments: Theoretical Principles and Practical Guidelines. Wiley: Chichester, 1996; 968–974.
Imperial College Press: London, 2006. 51. Tropp J. The theory of the bird-cage resonator. J. Magn. Reson.
26. Seeber DA, Jevtic J, Menon A. Floating shield current suppression 1989; 82: 51–62.
trap. Concepts in Magnetic Resonance 2004; Part B, 21B(1): 52. Tropp J. Mutual inductance in the birdcage resonator. J Magn
26–31. Reson 1997; 126: 9–17.
27. Matson GB, Vermathen P, Hill TC. A practical double-tuned 53. Tropp J. Dissipation, resistance, and rational impedance matching
1
H/31P quadrature birdcage headcoil optimized for 31P for TEM and birdcage resonators. Concepts in Magnetic Reson-
operation. Magn Reson Med 1999; 42: 173–182. ance 2002; 15(2): 177–188.
28. Seeber DA. Tuning system for floating radio frequency trap. U.S. 54. Leifer MC. Resonant modes of the birdcage coil. J. Magn. Reson.
Pat. 6,664,465, 2003. 1997; 124: 51–60.
29. Wong WH, Leung J, Funk A, Mehr K. Coils for high frequency 55. Tropp J. Image brightening in samples of high dielectric constant.
MRI. U.S. Pat. 6,980,000, 2005. J.Magn. Reson. 2004; 167: 12–24.
30. Eydelman G, Damadian RV, Giambalvo AJ. MRI antenna. U.S. 56. Crozier S, Luescher K, Forbes LK, Doddrell DM. Optimized
Pat. 6,847,210, 2005. small-bore, high-pass resonator designs. J. Magn. Reson. Ser. B.
31. Mehdizadeh M, Molyneaux DA, Holland GN. Quadrature 1995; 109: 1–11.
surface coils for magnetic resonance imaging. U.S. Pat. 57. deSwiet T, Wong WH, Romo M, Finnigan J, Burns S. NMR
4,918,388, 1990. resonators optimized for high Q factor. U.S. Pat. 6,667,674, 2003.
32. Karpodinis KG, Carvajal L, Chen AP, Jimenez B, Bruce N, Tropp 58. Wong WH, Sukumar S. Millipede imaging coil design for high
J, Kurhanewicz J, Vigneron DB. 3T prostate coils for 1H and field micro imaging applications. Presented at 8th ISMRM. Poster
31
P MR spectroscopic imaging. Presented at 14th ISMRM. Poster 1399. Denver, 2000.
2593. Seattle, 2006. 59. Dazai J, Bock NA, Nieman BJ, Davidson LM, Henkelman RM,
33. Farrar CT, Wedeen VJ, Ackerman JL. Cylindrical meanderline RF Chen XJ. Multiple mouse biological loading and monitoring
coil for intravascular MRI of atherosclerotic plaque. Magn Reson system for MRI. Magn. Reson. in Med. 2004; 52: 709–715.
Med 2005; 53: 226–230. 60. Collins CM, Liu W, Wang J, Gruetter R, Vaughan JT, Ugurbil K,
34. Tropp J, Calderon P, Carvajal L, Karpodinis K, Chen A, Vigneron Smith MB. Temperature and SAR calculations for a human head
D, Hurd R, Ardenkjaer-Larsen J-H. An endorectal dual frequency within volume and surface coils at 64 and 300 MHz. J. Magn.
13
C-1H receive only probe for operation at 3.0 tesla. Presented at Reson. Imaging. 2004; 19: 650–656.
14th ISMRM. Poster 2594. Seattle, 2006. 61. Bogdanov G, Kueppers G, King JA, Ferris CF. Actively tuned dual
35. Hoult DI. The NMR receiver: a description and analysis of design. RF resonator system for functional MRI of small animals. Pre-
Progress in NMR Spectroscopy 1978; 12: 41–77. sented at 9th ISMRM. Poster 1093. Glasgow, 2001.
36. Ginsberg DM, Melchner MJ. Optimum geometry of saddle shaped 62. Bogdanov G, Ludwig R. Coupled microstrip line transverse
coils for generating a uniform magnetic field. Rev. Sci. Instrum. electromagnetic resonator model for high-field magnetic reson-
1970; 41: 122–123. ance imaging. Magn. Reson. Med. 2002; 47: 579–503.
37. Zens AP. NMR probe coil system. U.S. Pat. 4,398,149, 1983. 63. Doty FD, Entzminger G, Rafique Z, Holte L, Welsh T. The
38. Alderman DW, Grant DM. An efficient decoupler coil design Litzcage: a high-field circular-polarization RF coil with improved
which reduces heating in conductive samples in superconducting tunability, B1 homogeneity, and S/N. Poster presented at the ENC.
spectrometers. J. Magn. Reson. 1979; 36: 447–451. Asilomar, 2002; http://www.dotynmr.com/PDF/RFLC_.pdf
39. Kost GJ, Anderson SE, Matson GB, Conboy CB. 64. Doty FD, Laws N, Holte L, Staab JP, Zempel J, Garbow JR. The
A cylindrical-window NMR probe with extended tuning range 4-point-drive Litzcage: a semi-open MRI RF CP coil with a
for studies of the developing heart. J. Magn. Reson. 1989; 82: wide tuning range. Poster presented at 11th ISMRM. Toronto,
238–252. 2003.
40. Li Y, Webb AG, Saha S, Brey WW, Zachariah C, Edison AS. 65. Doty FD, Entzminger G, Staab JP, Gravel JD, Plant HD.
Comparison of the performance of round and rectangular wire Optimization of homogeneity and S/N in a 750 MHz CP micro-
in small solenoids for high-field NMR. Magnetic Resonance imaging coil. Poster presented at the 44th ENC. Savannah,
in Chemistry 2006; 44: 255–262. http://www.bme.ufl.edu/ April 2003.
documents/edison_research_article1_research.pdf 66. Alecci M, Jazzard P. Characterization and reduction of gradient-
41. Fuks LF, Huang FSC, Carter CM, Edelstein WA, Roemer PB. induced eddy currents in the RF shield of a TEM resonator. Magn.
Susceptibility, lineshape, and shimming in high-resolution. Reson. Med. 2002; 48: 404–407.
J. Magn. Reson. 1992; 100: 229–242. 67. Wadghiri YZ, Schneider AE, Gray EN, Aristizabel O, Berrios C,
42. Doty FD, Entzminger G Jr, Hauck CD. Error-tolerant RF Litz coils Turnbull DM, Gutstein DE. Contrast enhanced MRI of right
for NMR/MRI. J. Magn. Reson. 1999; 140: 17–31. ventricular abnormalities in Cx43 mutant mouse embryos. NMR
43. Doty FD. Low inductance transverse Litz foil coils. U.S. Pat. Biomed. 2007; 20: this issue. DOI: 10.1002/nbm1113.
6,060,882, 2000. 68. Hayes CE, Roemer PB. Noise correlations in data simultaneously
44. Doty FD, Entzminger G Jr. Center-fed paralleled saddle coils for acquired from multiple surface-coil arrays. Magn. Reson. Med.
multinuclear double-resonance NMR or MRI. U.S. Pat. 6,175,237, 1990; 16: 181–191.
2001. 69. Roemer PB, Edelstein WA, Hayes CE, Souza SP, Mueller OM.
45. Schneider HJ, Dullenkopf P. Slotted tube resonator: a new NMR The NMR phased array. Magn. Reson. Med. 1990; 16: 192–225.
probe head at high observing frequencies. Rev. Sci. Instrum. 1977; 70. Hardy CJ, Bottomley PA, Rohling KW, Roemer PB. An NMR
48: 68–73. phased array for human cardiac 31P spectroscopy. Magn. Reson.
46. Doty FD. Parallel single turn resonator for NMR. U.S. Pat Med. 1992; 28: 54–64.
4,641,098, 1987. 71. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE:
47. Doty FD, Bass W, Yang Q, Entzminger G, Wang JH, Doty GN, sensitivity encoding for fast MRI. Magn Reson Med 1999; 42:
Holte LL, Smith MB. The SQT2 Litz coil: an exceptionally robust 952–962.
knee coil for 3T. Presented at 10th ISMRM. Honolulu, 2002. http:// 72. Reykowski A, Wright SM, Porter JR. Design of matching networks
www.dotynmr.com/PDF/KNE_SML.pdf for low noise pre-amplifiers. Magn. Reson. Med. 1995; 33:
48. Hayes CE, Edelstein WA, Schenck JF, Mueller OM, Eash M. 848–852.
An efficient highly homogeneous radiofrequency coil for whole- 73. Lanz T, Dodd S, Steinberger T, Silva A, Koretsky A. A 4-channel
body NMR imaging at 1.5 T. J. Magn. Reson. 1985; 63: 622– rat head array for 11.7T. Presented at ESMRMB. Oral 81. Basle,
628. 2005.
49. Edelstein W. RF systems and coils for MRI & MRS. In Encyclo- 74. Zhang X, Webb A. Design of a capacitively decoupled transmit/
paedia of NMR Vol. 6. Wiley: Chichester, 1996; 3950–3954. receive NMR phased array for high field microscopy at 14.1 T.
50. Hayes CE. Birdcage and other high homogeneity RF coils for J Magn Reson. 2004; 170: 149–155.

Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm
 RF COIL TECHNOLOGY FOR SMALL-ANIMAL MRI 325

75. Sutton BP, Ciobanu L, Zhang X, Webb A. Parallel imaging for 83. Sorgenfrei BL, Edelstein WA. Optimizing MRI SNR for quad-
NMR microscopy at 14.1 T. Magn Reson Med 2005; 54: 9–13. rature unmatched RF coils: two preamplifiers are better than one.
http://mrel.beckman.uiuc.edu/'lciobanu/pics/p9.pdf Magn. Reson. Med. 1996; 36: 104–110.
76. Dedeoglu A, Choi JK, Cormier K, Kowall NW, Jenkins BG. MRS 84. Miller JB, Suits BH, Garroway AN, Hepp MA. Interplay among
analysis of alzheimer’s disease mouse brain that express mutant recovery time, signal, and noise. Concepts in Magnetic Resonance
human APP shows altered neurochemical profile. Brain Res. 2004; 2000; 12: 125–136.
1012: 60–65. 85. Avago Technologies. ATF-58143 LN enhancement mode pseudo-
77. Wadghiri YZ, Sigurdsson EM, Sadowski M, Elliott JI, Li Y, morphic GaAs HEMT. 2004. http://www.avagotech.com/pc/down-
Scholtzova H, Tang CY, Aguinaldo G, Pappolla M, Duff K, loadDocument.do?id¼4251
Wisniewski T, Turnbull DH. Detection of alzheimer’s amyloid 86. MR Product Information. Microwave Technology, Fremont, CA,
in transgenic mice using MR microimaging. Magn Reson Med 2006. www.mwtinc.com
2003; 50: 293–302. 87. Kovacs H, Moskau D, Spraul M. Cryogenically cooled probes: a
78. Ohliger MA, Grant AK, Sodickson DK. Ultimate intrinsic signal- leap in NMR technology. Magnetic Resonance Spectroscopy 2005;
to-noise ratio for parallel MRI: electromagnetic field consider- 46: 131–255.
ations. Magn. Reson. Med. 2003; 50: 1018–1030. 88. Haueisen R, Marek D, Sacher M, Kong F, Ugurbil K,
79. Wiesinger F, Boesiger P, Pruessmann KP. Electrodynamics and Junge S. Cryogenic probe setup for routine MRI on
ultimate SNR in parallel MR imaging. Magn. Reson. Med. 2004; small animals at 9.4 T. Presentation # 80, at ESMRMB. Basel,
52: 376–399. Switzerland; September 2005.
80. Lattanzi R, Grant AK, Ohliger MA, Sodickson DK. Measuring 89. Hurlston SE, Brey WW, Suddarth SA, Johnson GA. A
practical coil array performance with respect to ultimate intrinsic High- temperature superconducting Helmholtz probe for
SNR: a tool for array design and assessment. Presentation # 424, at microscopy at 9.4 T. Magn. Reson. Med. 1999; 41: 1032–
the 14th ISMRM. Seattle, 2006. 1038.
81. Yuan J, Wei J, Qu P, Shen GX. A direct modulated optical link 90. Nouls J, Izenson M, Greeley H, Bagley M, Rozzi J, Johnson GA. A
for MR coil array interconnect. Presented at the 14th ISMRM. high-temperature superconducting volume coil for MR micro-
Poster 2617. Seattle, 2006. scopy at 9.4 T. Presented at 14th ISMRM. Poster #221. Seattle,
82. Oppelt R, Vester M. A low input impedance MRI preamplifier 2006.
using a purely capacitive feedback network. Presented at the 14th 91. Darrasse L, Ginefri J-C. Perspectives with Cryogenic RF probes in
ISMRM. Poster 2026. Seattle, 2006. biomedical MRI. Biochimie. 2003; 85: 915–937.

Copyright # 2007 John Wiley & Sons, Ltd. NMR Biomed. 2007; 20: 304–325
DOI: 10.1002/nbm