Hazards
In general, MRI examinations are considered relatively safe for the patient and less
harmful than, for example, X-rays. Hazards of routine MRI include heating, the attraction
of loose ferromagnetic objects, peripheral nerve stimulation, and claustrophobia [4 ]. Special care has to be taken with respect to implants. Decisions as to whether
or not a patient with an implant can undergo MRI have to be made on an individual
basis. There are three classifications for implants with respect to MRI: MRI-safe,
MRI-unsafe and MRI-conditional [5 ]. The information is usually provided in the implant user instructions (according
to the international electrotechnical commission IEC 60 601–2-33 2010) or third-party
services [6 ]
[7 ].
MR-safe implants are not considered to pose a significant risk for patients undergoing
MRI. Implants are categorized as MR-unsafe mostly because of the use of ferromagnetic
materials. Patients with MR-unsafe implants cannot undergo MRI. MRI-conditional implants
can be scanned under certain conditions (see instructions for use [8 ]). Still, specific handling is required, e. g. limits of the magnetic or radiofrequency
field or a certain specific absorption rate must not be exceeded [5 ].
Hazards of implants with respect to MRI include malfunction (e. g., pacemakers) or
harm by displacement, heating, or induced voltages. These effects are caused by the
strong, static magnetic field B
0 ≈ 1 T, weaker but rapidly changing magnetic fields (gradients) B
G ≈ 10–3 T at a frequency in the kHz range, and radiofrequency (RF) fields B
1 ≈ 10–6 T with a frequency in the 10–200 MHz range. In the following, we briefly describe
the hazards caused by the different fields with a focus on implants.
Static magnetic field
Most modern MRI systems feature a magnetic field
B
0 of 1.5 or 3 T. As per convention, we assume the main component of the static magnetic
field to be in the z-direction along the bore (Equation 1).
The static magnetic field can attract loose ferromagnetic objects. Therefore, care
has to be taken that no ferromagnetic material is brought near the MRI. In the past,
fatal accidents have occurred when large metallic objects were attracted by the magnetic
field and hit patients [4 ]
[9 ]. This includes items of daily use in the hospital such as oxygen cylinders ([Fig. 1 ]), patient beds, scissors, and cleaning utensils. Accidents have also been caused
on rare occasions, e. g., by respiratory equipment of firefighters.
Fig. 1 Photograph of an oxygen bottle in a 3 T whole-body MRI unit. A non-MR-safe oxygen
bottle was carried into the MRI room and attracted by the strong magnetic field causing
it to be catapulted into the MRI unit.Abb. 1 Foto einer Sauerstoffflasche in einem 3T-Ganzkörper-MRT. Eine nicht MR-sichere Sauerstoffflasche
wurde in den MRT-Raum getragen und durch das starke Magnetfeld so angezogen, dass
sie in die MRT-Röhre katapultiert wurde.
In general, materials are divided into three different classes depending on the magnetic
susceptibility (χ): diamagnetic (χ < 0), paramagnetic (χ > 0) and ferromagnetic (χ > > 1)
materials. The magnetic susceptibility is a property that indicates how strongly a
material will be magnetized by an external magnetic field [10 ]. Implants containing ferromagnetic material will exert a force, leading to displacement
of the implant and thus malfunctioning or tissue injury.
Gradient fields
When exposed to changing magnetic fields, a voltage is induced into a conductor. This
voltage results in a current if the resistance is finite. For MRI, changing magnetic
fluxes occur while approaching
B
0
(the magnet) and by rapidly switching, spatially varying magnetic field
B
G
(“gradients”). Gradients are applied in each Cartesian direction (Equation 2):
While entering the bore is relatively harmless, vertigo has been reported, especially
at ultra-high fields of 7 T [11 ]. Gradients may induce electrical stimulation of nerves [12 ] or even heating of tissue [13 ].
In most cases, the unintentional nerve excitation is an uncomfortable but transient
sensation for the patient (e. g., twitching or tingling). However, external devices,
such as electrodes, may heat up significantly and cause fourth-degree burns [14 ]. To avoid burns, skin-skin contact (e. g., touching hands, thighs) or skin-metal
contact has to be prevented [4 ]. For active implants, induced voltages could result in damaged circuits or malfunction
like unwanted impulses (e. g., pacemaker) [15 ]. To account for the effects of switching gradients, some MR-conditional implants
are specified to a maximal allowed slew rate. The slew rate of a gradient is a measure
for how fast a gradient is turned on (in units of Tesla per meter per second (T/m/s)).
It is calculated by the ratio of the maximum value of the gradient and the time necessary
to reach this gradient amplitude.
RF pulses
Radiofrequency (RF) pulses applied by every MRI are partially absorbed by the tissue
and converted to heat. The rate, how much energy is absorbed, is given by the specific
absorption rate (SAR) in units of W/kg and depends on the patient's size and body
weight. To avoid heating, the allowed SAR must not be exceeded.
Conductive materials, like some implants, may reflect and absorb the RF pulses and
thus increase heating [16 ]. Each implant is approved for a given magnetic field strength and SAR, e. g. 1.5 T
and SAR < 2.0 W/kg for cardiovascular devices [17 ] or 1.5 T and SAR < 0.1 W/kg for deep brain stimulations [16 ]. However, it was realized that SAR is not the optimum measure to access implant
safety. Instead, the root mean of the
B
1
field (B
1 +rms in units of mT) was suggested. B1 +rms does not depend on the patient and is used for accessing implant safety today
[18 ]
[19 ].
Imaging artifacts
Often, implants contain magnetic materials that distort the static magnetic field
and shield RF pulses. This distortion is caused by the magnetic susceptibility (χ)
that is much different compared to the surrounding tissue (see the previous section).
This distortion results in an altered resonance frequency of the spins in the affected
volume. These frequency changes may have several effects: a) the spins are not excited
and/or detected, b) a fast T2* decay and c) errors in the spatial localization. To
understand these effects, we need to consider the physical background of an MRI sequence.
Magnetization and resonance frequency
The nuclei (protons) of hydrogen atoms like in water (H2 O) possess a nuclear spin, which gives rise to a magnetic moment. The spin is a quantum
mechanical property, which we do not need to consider in much detail for the purposes
of this paper. It is essential to consider, however, that the spins have two stationary
states in a magnetic field, parallel (“up”) or antiparallel (“down”). One milligram
of water, 1 mm3 , contains approx. 1020 hydrogen atoms and thus spins and their magnetic moments.
However, most of these spins are MR-invisible and do not contribute to the MRI signal
because they are almost equally in the up and down state and effectively cancel each
other out (Boltzmann distribution of the Zeemann states). Only a small fraction of
all spins, called polarization P, contributes to the magnetization and thus signal
(Equation 3). The approximation only holds for room temperature
At room temperature, P and thus the magnitude of M increase approx. linearly with the strength of an outer magnetic field B0
. This is the reason why ever stronger magnets are being built.
Still, the polarization is very small – no more than about three parts per million
(ppm) effectively contribute the signal at 1 T (or ≈ 10 ppm at 3 T). Thus, 99.999
percent of all spins are invisible at current magnetic fields. This implies that 100 000-fold
enhancements are possible by increasing polarization – a fact that hyperpolarization
techniques exploit [20 ]. Still, the small fraction of spins contributing to the signal is sufficient for
MRI as we know it today.
Leaving the microscopic scale, let's consider that the spins, e. g., of 1 mm3 water, form a macroscopic magnetization vector M = [Mx My Mz ]. Some aspects of this macroscopic magnetization may be compared to the needle of
a hiking compass. One gram of water has a nuclear magnetization of 3.1 A/m in a magnetic
field of 1 T. With B
0 = [0, 0, Bz ], the magnetization becomes M = [0, 0, Mz ].
When an oscillating magnetic field
B
1 is applied at the right frequency ω
L
and perpendicular to
B
0 , magnetization M can be excited or flipped by an arbitrary degree α. The energy difference between
the up- and downstate is dependent on the Larmor frequency ω
L
:
Where γ is the magnetogyric ratio, a constant of nature proportional to the magnetic
moment of the element under consideration (γ(1 H) = 42 MHz/T 2π), and t
RF is the duration of the RF pulse. Note: ω
L
is directly proportional to B
0 .
It is important to note that the excitation pulse has a finite bandwidth bw
p – that is a frequency range, where the excitation pulse affects the magnetization.
A higher bandwidth corresponds to a larger frequency range and therefore excites spins
across a broader range of Larmor frequencies. The bandwidth depends on the duration
and shape of the pulse.
After the excitation pulse, the magnetization vectors M precess about the z -axis. Any component of the magnetization vector in the xy -plane, [Mx My ], induces an alternating magnetic flux and thus a current in a nearby conductor (e. g.,
a coil): this is the MR signal.
Similar to the excitation, the detection process has a bandwidth bw
d , too, which depends on the rate of the discrete sampling of the continuous MR signal
where td is the dwell time and depicts the interval between two digitized samples. Shorter
sampling intervals result in a higher bandwidth but reduce the signal-to-noise ratio
(SNR) because less time was used to collect the signal. Different vendors indicate
the bandwidth differently, either the total bandwidth in Hz or the bandwidth divided
by the number of steps for the frequency encoding in bandwidth in Hz/Px.
After the excitation, the longitudinal or z-component of the magnetization Mz returns to its equilibrium value, M0 , while Mxy decays to zero. Both effects can be described with exponential functions with the
constant T1 for longitudinal and T2, T2* and T2i for transverse relaxation (Equation 7). Once Mxy is zero, no MR signal is recorded (while Mz may not yet be recovered). There are two fundamental effects contributing to the
transverse relaxation: the effect of static B
0 inhomogeneities, described by an exponential decay with T2i , and other components, described by T2. While the T2 decay is irreversible, loss
by T2i decay can be recovered by refocusing the magnetization in spin echoes.
Spin echoes are formed by flipping the transverse magnetization around × or y. This
way, the effect of the inhomogeneities is reversed.
For spatial encoding, i. e., imaging, additional magnetic fields,
B
G , are applied whose strengths vary with the position in space:
Gx , Gy and Gz are the strength of the gradient fields in T/m and x, y and z are the spatial positions
in m.
B
G is applied at different points in time: during the pulse (“slice selection”), between
excitation and detection (“phase encoding”), or during readout (“readout encoding”).
It is essential to realize that the magnetic field's linear variation (induced by
B
G ) causes a spatial dependency of the Larmor frequency. As a result, the Larmor frequency
becomes a function of the position, and this relation of spatial location and resonance
frequency allows the reconstruction of images. Using this effect for imaging was awarded
the Nobel Prize in 2003.
It should be noted that when we talk about the strength of an MRI device, we refer
to the magnetic flux density, B , in T and not to the magnetic field strength, H , in A/m. The permeability, a proportionality factor, connects B and H.
Impact of implants
As mentioned above, implants are either paramagnetic, diamagnetic, or ferromagnetic,
a classification depending on their magnetic susceptibility χ. Like implant materials,
each tissue in the body also has a different magnetic susceptibility ([Table 1 ]).
Table 1
Typical magnetic susceptibilities (χ) for different materials and the corresponding
magnetic properties.Tab. 1 Typische magnetische Suszeptibilitäten (χ) für verschiedene Materialien und die entsprechenden
magnetischen Eigenschaften.
material
magnetic susceptibility (χ)
magnetic property
silver
–20 × 10–6
diamagnetic
water, soft tissue
–9.05 × 10–6
diamagnetic
bone
–8.86 × 10–6
diamagnetic
magnesium
11.7 × 10–6
paramagnetic
titanium
182 × 10–6
paramagnetic
air
0.36 × 10–6
paramagnetic
iron
~105
ferromagnetic
magnetic stainless steel
~103
ferromagnetic
Material brought into a magnetic field may enhance, weaken, or not affect the magnetic
field locally ([Fig. 2 ]).
Fig. 2 Schematic of the influence of diamagnetic or paramagnetic/ferromagnetic material
on the magnetic field. Diamagnetic materials weaken the magnetic field, while paramagnetic
and ferromagnetic materials enhance the magnetic field.Abb. 2 Schematische Darstellung des Einflusses von diamagnetischem oder paramagnetischem/ferromagnetischem
Material auf das Magnetfeld. Diamagnetische Materialien schwächen das Magnetfeld,
während paramagnetische und ferromagnetische Materialien das Magnetfeld verstärken.
Diamagnetic material, such as biological tissue and calcium, will weaken the magnetic
field, while paramagnetic and ferromagnetic material, like gadolinium and iron, will
enhance the magnetic field locally. The susceptibilities of different tissues are
similar enough to have little impact on conventional MRI, although specialized sequences
generate contrast based on different susceptibilities (susceptibility-weighted imaging,
SWI) [21 ]. Problems occur in the presence of metal implants or at the interface between air
and tissue (e. g., nasal cavities, intestines).
For implants, due to the different magnetic susceptibility compared to surrounding
tissue, the magnetic field is distorted, causing changes in the Larmor frequency of
nearby spins. This may have one or all of the following effects:
Insufficient excitation because ω
L
is not within bw
p
Low signal detection because ω
L
is outside of bw
d
Low signal detection because the transverse magnetization decays very fast (large
T2', dephasing)
Misregistration of spins because ω
L
is different than expected ([Fig. 3 ]).
Failure of frequency selective pulses.
Fig. 3 Schematic view of signal pile-up induced by implants: The superposition of B0 (dashed line) and BG cause the field strength to vary linearly with the position (red line, top). When
a pulse is applied, all spins within the excitation bandwidth of the pulse (fp ± bwp /2) are excited. As a result, only spins within a slice are excited (shaded areas).
If the field is distorted, e. g., by an implant (red line, bottom), the variation
of the field strength is not linear due to Bimplant. Therefore, other spins outside
that slice may contribute to the signal, too, causing so-called pile-up artifacts.Abb. 3 Schematische Darstellung der durch Implantate induzierten Signalanhäufung: Die Überlagerung
von B0 (gestrichelte Linie) und BG bewirkt, dass die Feldstärke linear mit der Position variiert (rote Linie, oben).
Wird ein Puls angelegt, werden alle Spins innerhalb der Anregungsbandbreite des Pulses
(fp ± bwp /2) angeregt. Dadurch werden nur Spins innerhalb einer Schicht angeregt (schattierte
Bereiche). Wird das Feld z. B. durch ein Implantat verzerrt (rote Linie, unten), ist
die Variation der Feldstärke aufgrund des Implantats nicht linear. Daher können auch
andere Spins außerhalb dieser Schicht zum Signal beitragen, was zu sogenannten Pile-up-Artefakten
führt.
All of this will lead to signal loss or signal pile-up and cause spatial distortions
in the measured slice.
In addition to image distortion, failure of spectral fat suppression often occurs
in tissue near metal implants for the same reason (distortions of the magnetic field
and thus Larmor frequencies). Spectral fat suppression methods use the fact that the
Larmor frequencies (chemical shift) of water and fat are different by approximately
3.5 ppm (e. g., 220 Hz for 1.5 T). A narrow (low bw
p ) pulse is applied at the frequency of fat to saturate its magnetization to suppress
the fat signal. In the proximity of metal implants, the field and thus frequencies
are distorted (e. g., by 30–80 Hz). As a result, fat is not excited by the narrow
saturation pulses. Similar effects occur in other techniques, such as Dixon-type fat
water imaging [22 ] ([Fig. 4 ]).
Fig. 4 Reconstruction of a fat-suppressed, abdominal T1-weighted 3D gradient echo Dixon
MRI of a patient with an artificial hip joint (left). The implant caused signal loss
(*) as well as failure of fat-water separation (arrows). No such artifacts are apparent
on the other side.Abb. 4 Rekonstruktion eines fettunterdrückten, abdominalen, T1-gewichteten 3D-Gradientenecho-Dixon-MRTs
eines Patienten mit einem künstlichen Hüftgelenk (links). Das Implantat verursacht
einen Signalverlust (*) sowie ein Versagen der Fett-Wasser-Trennung (Pfeile). Auf
der anderen Seite sind keine derartigen Artefakte zu erkennen.
Reduction of imaging artifacts
Several approaches were suggested to reduce implant-related artifacts. To some extent,
standard sequences can be used with some parameter optimization. Additionally, special
MRI sequences were developed for artifact reduction. In general, artifacts are dependent
on the field strength of the MRI device, the used sequence, the chosen parameters,
and the properties of the implant itself, such as orientation and type of material.
General considerations
Lower magnetic field strengths exhibit less metal-induced artifacts than higher field
strengths. The field homogeneity (also without an implant) is higher in lower magnetic
fields. Thus, lower field strengths were recommended for imaging implants [23 ]
[24 ]
[25 ]
[26 ]
[27 ]
[28 ]
[29 ]
[30 ]
[31 ]
[32 ]
[33 ]
[34 ].
As expected, the material's susceptibility was found to affect the artifacts: larger
susceptibility differences caused larger distortions.
The orientation of the implant with respect to B
0 may play a role, too [35 ]. It appears beneficial to position the long implant axis parallel to B
0 , although it must not act as an antenna as heating may occur.
The phase encoding direction is less affected by metal artifacts. Thus, it may be
advantageous to switch the phase and frequency encoding direction to improve the visibility
of the nearby tissue. If phase oversampling is needed to avoid a folding artifact,
the measurement time increases [22 ]
[25 ].
Some sequence types are more robust with respect to distortions than others. Generally,
spin echo (SE) sequences are benign because the echo can be acquired fully, even if
T2* is short (compared to gradient echo sequences, where the signal may decay with
T2* during spatial encoding before the acquisition is started). As a result, some
of the inhomogeneities can be refocused with spin echo sequences. Although spin-echo-based
sequences have benefits, some sequences like phase contrast imaging or fast imaging
methods often use gradient echoes. Here, it is advantageous to reduce the time between
excitation and image acquisition, TE [28 ]
[29 ]
[32 ]
[36 ]
[37 ]. In the extreme, ultra-short or zero-TE sequences can reduce artifacts and even
image hard tissues [38 ]
[39 ].
In the presence of field inhomogeneities, the spins precess with a frequency offset
of Δf compared to the non-distorted case. The spatial distortion depends on the frequency
offset and the slice thickness and negatively on the excitation bandwidth (bwp ) [22 ]. Besides this, the spatial distortion also depends on the receiver bandwidth (bwd ). A higher receiver bandwidth reduces the spatial distortion ([Fig. 5 ]) as well as an increasing slice selection gradient, which results in a decreased
slice thickness. Additionally, increasing the matrix size or reducing the voxel size
is helpful for reducing metal artifacts. However, a drawback of those reduction techniques
is an increased scan time and reduced signal-to-noise ratio (SNR) [24 ]
[25 ]
[34 ]
[36 ]
[37 ]
[40 ]
[41 ]. A higher slice selection gradient ensures that each slice is encoded with a higher
frequency range. As a result, spins with higher or lower Larmor frequencies (e. g.,
caused by field inhomogeneities) are still excited in the correct slice and contribute
to the signal. This especially reduces signal loss. Again, higher bandwidth reduces
SNR [23 ]
[24 ]
[25 ]
[32 ]
[36 ]
[40 ]
[41 ]
[42 ]
[43 ]
[44 ]. For fat suppression near metal implants, spectral fat suppression sequences should
be avoided. Instead, T1-based suppression techniques, such as short tau inversion
recovery (STIR), are more robust. STIR exploits the fact that the T1-relaxation time
of fat is shorter than that of tissue and can be nulled with an inversion pulse. STIR
is beneficial compared to the spectral fat suppression near implants but results in
a lower SNR. Additionally, a second RF pulse is necessary, which increases the specific
SAR.
Fig. 5 Schematic of different receiver bandwidths (bwd ) on spatial distortions (Δx). In the case of a frequency offset due to field inhomogeneities,
a higher receiver bandwidth results in smaller spatial distortion than lower receiver
bandwidths.Abb. 5 Schematische Darstellung unterschiedlicher Empfangsbandbreiten (bwd ) auf die räumliche Verzerrung (Δx). Bei einem Frequenzunterschied aufgrund von Feldinhomogenitäten
führen höhere Empfängerbandbreiten zu geringeren räumlichen Verzerrungen im Vergleich
zu niedrigeren Empfängerbandbreiten.
Special sequences with high robustness against spatial distortions
One problem regarding imaging near metal implants is the failure of slice selection
and readout encoding. For the acquisition of an MR image without slice selection and
readout encoding, single point imaging (SPI) can be used. In SPI sequences, only phase
encoding is applied, and only one point in k-space is acquired for each excitation.
This method makes it possible to minimize the time between excitation and acquisition
but leads to very long scan times of 40 minutes or longer which are rarely practical
for in vivo use [45 ].
To reduce in-plane distortions, view-angle tilting (VAT) was suggested in conjunction
with spin echo sequences. VAT exploits the fact that the distortion is dependent on
the gradient strength of the slice selection gradient. To compensate the off-resonance
effects of the slice selection gradient, an additional gradient is used during the
readout, equivalent to the one used for slice selection. The additional gradient results
in a tilting of the readout direction ([Fig. 6 ]). The readout direction’s tilt reduces the in-plane artifacts, but with a drawback
of blurred edges [40 ]
[42 ]
[46 ]
[47 ].
Fig. 6 Schematic of the principle of view-angle tilting (VAT). a non-distorted field, b distorted field without correction, and c distorted field with VAT. The distortion in b results in a signal void and signal pile-up in the measured slice due to the inhomogeneities.
The angle of the readout direction c reduces the distortion but results in blurring edges (blue arrows).Abb. 6 Schematische Darstellung des Prinzips des view-angle tilting (VAT). a keine Feldverzerrung, b Feldverzerrung ohne Korrektur und c Feldverzerrung mit VAT. Aufgrund der Inhomogenitäten führt die Verzerrung in b zu einer Signalauslöschung und Signalanhäufung in der gemessenen Schicht. Der Winkel
der Ausleserichtung c reduziert die Verzerrung, führt aber zu unscharfen Kanten (blaue Pfeile).
A specific optimization of parameters (high bandwidth, thin slices) was initially
introduced by Olsen et al. as a metal artifact reduction sequence (MARS) [48 ]. More recently, modified spin echo sequences were combined with a view-angle tilting
technique [49 ].
VAT is only able to reduce in-plane distortion. Slice Encoding for Metal Artifact
Correction (SEMAC) was suggested to reduce through-plane distortions, which can be
applied additionally to VAT. SEMAC is based on a conventional 2 D spin-echo sequence.
It utilizes an additional phase-encoding step in the slice-selection direction, which
reduces slice distortions. Knowledge of slice distortion can be used in the postprocessing
step, where each slice is coded and reconstructed individually. The number of slice
encoding steps is dependent on the extent of the geometric distortion. Like the other
methods mentioned, this sequence's drawback is the increased scan time of around 10
minutes [37 ]
[43 ]
[50 ]. By optimizing for the particular implant, the scan time can be reduced to around
5 minutes [51 ].
Another sequence suggested for reducing through-plane distortions was Multi-Acquisition
Variable Resonance Image Combination (MAVRIC), based on a 3D fast spin echo sequence.
Here, multiple frequency-selective excitations are used instead of a single excitation
pulse. All excited slabs are then analyzed in the postprocessing step to reduce artifacts.
Like SEMAC, MAVRIC increases the scan time to around 20–25 min [22 ]
[37 ]
[52 ]
[53 ].
A combination of SEMAC and MAVRIC was introduced by Koch et al. [53 ]. MAVRIC SL uses the selective excitation pulse like SEMAC, but the pulse profile
is the same as for MAVRIC [24 ].
It is a common phenomenon that manufacturers give different names to relatively similar
sequences. For example, optimized spin echo sequences with VAT are called O-MAR [54 ] (Philips) and WARP [55 ] (Siemens), or O-MAR XD [56 ] and Advanced WARP [57 ] with the addition of SEMAC. Readers are advised to consult available online databases
to identify the appropriate sequence for their system.
It should not go unmentioned that attempts were recently made to reduce artifacts
using artificial intelligence (AI) [58 ]
[59 ].
